(Kyo Kageura) The Dynamics of Terminology A Desc PDF

THE DYNAMICS OF TERMINOLOGY
TERMINOLOGY AND LEXICOGRAPHY

RESEARCH AND PRACTICE
Terminology and Lexicography Research and Practice aims to provide

in-depth studies and background information pertaining to Lexicography and
Terminology. General works will include philosophical, historical, theoretical,
computational and cognitive approaches. Other works will focus on structures for
purpose- and domain-specific compilation (LSP), dictionary design, and training.
The series will include monographs, state-of-the-art volumes and course books in
the English language.
Series Editors
Marie-Claude L’Homme
Ulrich Heid
Consulting Editor
Juan C. Sager
Volume 5
Kyo Kageura
The Dynamics of Terminology.

A descriptive theory of term formation and terminological growth.
THE DYNAMICS OF
TERMINOLOGY
A DESCRIPTIVE THEORY OF TERM
FORMATION AND TERMINOLOGICAL GROWTH
KYO KAGEURA
National Institute of Informatics, Tokyo
JOHN BENJAMINS PUBLISHING COMPANY

AMSTERDAM/PHILADELPHIA
TM The paper used in this publication meets the minimum requirements of Ameri-
8
can National Standard for Information Sciences — Permanence of Paper for

Printed Library Materials, ANSI Z39.48-1984.
Library of Congress Cataloging-in-Publication Data

Kageura, Kyo, 1964-
The Dynamics of Terminology: A descriptive theory of term formation and terminological
growth. / Kyo Kageura.
p. cm. -- (Terminology and lexicography research and practice, ISSN 1388-8455; v. 5)
Includes bibliographical references and index.
1. Terms and phrases. I. Title. II. Series.
P305 .K34 2000

418--dc21 2002027966
ISBN 90 272 2328 9 (Eur.) / 1 58811 314 0 (US) (Hb; alk. paper)
© 2002 – John Benjamins B.V.
No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any
other means, without written permission from the publisher.
John Benjamins Publishing Co. • P.O.Box 36224 • 1020 ME Amsterdam • The Netherlands
John Benjamins North America • P.O.Box 27519 • Philadelphia, PA 19118 • USA
Contents
Acknowledgements vii
Introduction 1
PART I: Theoretical Background 7

Chapter 1. Terminology: Basic Observations 9
Chapter 2. The Theoretical Framework for the
Study of the Dynamics of Terminology 25
PART II: Conceptual Patterns of Term Formation 43

Chapter 3. Conceptual Patterns of Term Formation:
The Basic Descriptive Framework 45
Chapter 4. Conceptual Categories for the Description of
Formation Patterns of Documentation Terms 61
Chapter 5. Intra-Term Relations and
Conceptual Specification Patterns 91
Chapter 6. Conceptual Patterns of the
Formation of Documentation Terms 115
PART III: Quantitative Patterns of Terminological Growth 163

Chapter 7. Quantitative Analysis of the Dynamics
of Terminology: A Basic Framework 165
Chapter 8. Growth Patterns of Morphemes in
the Terminology of Documentation 183
vi
Chapter 9. Quantitative Dynamics in Term Formation 201
PART IV: Conclusions 247

Chapter 10. Towards Modelling Term Formation
and Terminological Growth 249
Appendices 273
Appendix A. List of Conceptual Categories 275
Appendix B. Lists of Intra-Term Relations and
Appendix C. List of Terms by Conceptual Categories 281
Appendix D. List of Morphemes by Conceptual Categories 295
Bibliography 303
Index 315
Acknowledgements
I started the research described in this book around 1990, when I began
working on conceptual descriptions of term formation patterns with Pro-
fessor J. C. Sager at the Centre for Computational Linguistics, UMIST.
Although the direction of my research has changed gradually and gained
in more mathematical technicality since then, all the theoretical aspects of
this book have their roots, either explicitly or implicitly, in my intensive
and intriguing discussions with Professor Sager. Just how indebted I am to
him for his constant encouragement and understanding, I find it difficult to
express.
From around 1995, as my theoretical standpoint on the study of termi-
nology took shape, I felt the necessity to augment my conceptual approach
with a quantitative approach on the descriptive front. I had the luxury of
thinking intensively about this in 1996, when I stayed with the Natural Lan-
guage Processing group at the Department of Computer Science, the Uni-
versity of Sheffield. I would like to thank Professor Yorick Wilks and all
the staff there.
One of the most important triggers for the development of an interpre-
tative framework for the quantitative part of my work was the International
Quantitative Linguistics Conference held in Helsinki in 1997, where I had
the opportunity to learn LNRE models from Dr. R. Harald Baayen of the
Max Planck Institute of Psycholinguistics, Nijmegen, the Netherlands. This
helped me to finalise the work described in Part III of this book. For that, I
owe him a great deal.
I knew the work of the "French connection" of computational termi-
nology well from their publications, but after meeting them face-to-face
in 1998 benefited greatly from the stimulation their intellectual input gave
to my work. Among them, I especially wish to thank Professor Chris-
tian Jacquemin of CNRS-LIMSI and Professor Béatrice Daille of IRIN, the
viii
University of Nantes. My research cooperation and many discussions with

them clarified what was lacking in my research agenda, and thus supported
this work indirectly.
The concrete descriptions of the formation and growth of the terminol-
ogy of Japanese documentation in this book were made possible by Profes-
sor Yasuhiro Ohta of Bunkyo University, Mr. Hajime Sasaki and Professor
Shunsaku Tamura of Keio University, who have made pioneering efforts
in listing the Japanese terminology of documentation. I greatly appreciate
their work.
Many people generously gave of their time to listen to my ideas and
discuss some of the topics contained in this book. Among them, I would
particularly like to express my gratitude to my colleagues Dr. Atsuhiro
Takasu, Dr. Akiko N. Aizawa, Professor Teruo Koyama, Dr. Masaharu
Yoshioka and Dr. Koichi Takeuchi at the National Institute of Informatics,
Japan, where I currently work, my former students Keita Tsuji and Fuyuki
Yoshikane, Dr. Satoshi Sekine of New York University, Dr. Sandra-Yamilet
Santana, Professor Hiroshi Nakagawa of the University of Tokyo, and Dr.
Toru Hisamitsu of Hitachi. My thanks also go to Professor Masao Naga-
sawa of Aichi Syukutoku University and Professor Eisuke Naito of Toyo
University, who introduced me to the interesting research area of informa-
tion studies, which led me to terminology.
In the process of writing this book, Professor Sager, Dr. Marie-Claude
L'Homme of the University of Montreal, and Professor Béatrice Daille read
the manuscript in detail and offered many relevant comments and criti-
cisms, not all of which I could accommodate in the present work. Ms.
Irma Nunez and Ms. Stephanie Coop read and proofed the draft carefully.
Finally, I would like to thank the many other people too numerous to
mention who have also played a role in the research reported in this book.
31 May 2002
Tokyo, Japan
Introduction
In recent times there has been a growing interest in the study of technical
terms (henceforth simply "terms" for succinctness), as can be witnessed by
the publication of textbooks (Cabré 1993; Felber 1984; Picht & Draskau
1985; Sager 1990), of collections of papers (Rey 1995; Sonneveld & Loen-
ing 1993) and of a journal (Terminology), as well as regular conferences on
terminology such as Terminology and Knowledge Engineering and Com-
puterm. Despite this fact, the study of terminology, i.e. the theoretical and
applied study of terms as coherent systems of lexical items endowed with
a singular creative dynamism, is as yet neither clearly defined nor is there
general agreement about its scope.
A related problem is the fact that, while work concerning what is tra
ditionally known as the "theory" or "principles" of terminology is pursued
simultaneously with, but independently of, terminology-related NLP appli
cations, little effort is being devoted to theories underlying the descriptive
analysis of terms. Besides, most of what currently passes for a theoretical
foundation of terminology amounts to little more than a simplified, a priori
theory of conceptual structures supported by largely prescriptive principles
of what "should be" rather than what is the actual usage of terms.
This situation seems to reflect basic characteristics of terms, i.e. terms
manifest themselves as concrete linguistic objects within a specialised dis
course and their number is constantly growing. The fact that terms are first
and foremost concrete linguistic objects makes it difficult to define the the
ory of terms at a proper level of abstraction. Many so-called theories about
terms are really only theories of something — for instance, of concepts —
that can be used to describe terms. In addition, many studies treat only
a very limited number of terms, mostly for exemplification. The fact that
terminology (and the number of terms) is constantly growing, on the other
2 Dynamics of Terminology
hand, fosters application-oriented studies of the computational treatment of

terms, but without satisfactory theoretical and/or descriptive foundations.
In short, theories of terms — as opposed to theories of something for
describing terms — are missing in the academic study of terminology. On
the side of applications or descriptions, there is a lack of solid descriptive
studies, as opposed to the application-motivated processing of terms. Given
this situation, research in the field of terminology needs to be broadened to
include concrete descriptive analyses of terminology based on an explicitly
stated theoretical position; only through the accumulation of this type of
study does it become possible to consolidate the theory of terminology.
This book is an attempt in this direction, focusing on the phenomenon of
term formation and terminological growth.
The organisation of the book
The book is divided into four main parts. Part I (Chapters 1 and 2) is de
voted to clarifying the author's view of the object of the study as well as
defining his theoretical standpoint. On the basis of close observation of
terms and examinations of the existing theoretical studies of terms, it will
be argued that a theory of terms or terminology should deal with the ter
minology of a domain in its totality, because it is only with respect to indi
vidual domains that the very concept of "term" is consolidated. It will also
be argued that a theoretical study of a terminology should be accompanied
by the descriptive study of a terminology, for proper descriptive studies are
theories of terms. The concept of the dynamics of terminological growth
is then introduced, and the overall framework is illustrated by means of the
description of conceptual patterns of term formation, complemented by the
analysis of quantitative regularities of terminological growth.
Parts II and III are devoted to the detailed development of the
theoretico-descriptive framework for the study of the dynamics of termi
nology and the required concrete description of the actual manifestation of
this dynamics. Throughout these two parts, the Japanese terminology of
documentation (introduced below) is used in the analyses.
As a first step towards the characterisation of the dynamics of terminol
ogy, Part II (Chapters 3-6) is devoted to the description of the conceptual
patterns which determine the formation of terms within the chosen domain.
The basic aspects to be observed at the conceptual level are the relationships
Introduction 3
between terms and their constituent elements, the relationships among the
constituent elements, as well as the type of conceptual combinations used
in the construction of the terminology. In Chapter 3, the basic descrip
tive framework, as well as the elements necessary for the description of the
conceptual patterns underlying term formation, are discussed. Chapter 4
is devoted to a presentation of the conceptual categories discovered in the
analyses of terms. Chapter 5, then, examines the conceptual relationships
between constituent elements of terms. Chapter 6 describes the character
istics of term formation patterns in the field of documentation, which are
based on the concrete descriptive devices introduced in Chapters 4 and 5.
The description of conceptual patterns of term formation, to be exam
ined in more detail, has a logical limitation. If, by describing conceptual
patterns, we try to give necessary and sufficient conditions for the formation
of terminology, we end up listing all the combinations of linguistic items in
the terminological data used in the study. This, however, obscures the ob
servation of the dynamics and would reduce the study to a natural history
of existing terms. The description of conceptual patterns, therefore, must
necessarily remain somewhat general, at the level where the broad regulari
ties of term formation patterns in a domain can most properly be described.
This immediately leads to the loss of the fine granularity of the description.
Part III (Chapters 7-9) explores the quantitative analysis of the patterns
of terminological growth, which compensates for the limitation of the de
scription of conceptual patterns of term formation and thus completes the
description of the dynamics of terminology. There is a statistical method to
observe the growth patterns of lexical items in their entirety within a certain
category or set of lexical items. Applying this method, it becomes possible
to give a detailed description of the dynamics of the potential growth pat
terns of the terminology of a domain. In Chapter 7, the statistical method
is presented together with factors that should be taken into account in the
application of the method. Chapter 8 is, then, devoted to describing the
growth patterns of constituent elements within the terminology. Chapter 9
details the growth patterns of terms for each subset of terms, the formation
patterns of which were observed in Chapter 6.
Part IV (Chapter 10) concludes the study. By examining the theoretical
standpoint introduced in Part I with respect to the findings of the concrete
description carried out in Parts II and III, it evaluates what was achieved
and what should be explored further.
The data
The present study takes the position that a theoretical work on terminology
should, as a logical requirement, be accompanied by a concrete description
of the terminology of a domain. As such, Parts II and III, together with
theoretical and methodological discussions, present the results of analyses
of terminology. For the concrete analyses, Japanese terminological data in
the field of documentation are used, a field of which the author has an in-
depth knowledge. The data are taken from Wersig & Neveling (1984), the
Japanese version of Wersig & Neveling (1976), a small but representative
terminological glossary in the field of documentation1. This is a multilin
gual glossary with the indication of related terms.
A few normalisations were applied to the Japanese entries of Wersig
& Neveling (1984), such as the normalisation of orthographic and minor
notational variants. Also, a single non-noun simple entry, (automatic),
was omitted. As a result, 1,228 terms were obtained.
There are no mechanically applicable criteria for delimiting constituent
elements or morphemes in Japanese, as Japanese lacks boundaries between
linguistic units such as spaces or hyphens. So the morphemes of terms
were identified manually, based on the criterion originally introduced by
Nomura & Ishii (1988), which reflects an average Japanese speaker's intu
ition about morphemes and was successfully applied to large-scale analysis
of Japanese terms. The method is briefly described as follows:
1. A minimal element is defined as the minimal linguistic element which
bears a meaning in current Japanese.
2. According to the origin of linguistic elements, i.e. wago (original
1
The idea of collecting terms from articles and texts was examined but discarded because
the available texts depended too much on the circumstantial tendencies and the data was
sparse from the point of view of collecting a representative set (sample) of terms. Among
reference sources such as glossaries, Wersig & Neveling (1984) was a semi-optimal choice
from the point of view of representativeness and balance of terms, partly because no better
alternatives existed. There were three other glossaries when the study was started: Mon-
busyou (1958) was too old and too prescriptive; Young (1988), being a translation of Young
(1983), was too biased to the U.S. library services and some Japanese terms were artificially
coined; and JIS Series (1989) was incomplete and, again, too prescriptive. A new glossary,
JSLIS (1997), which is better in its coverage and timeliness, has appeared since, but by
the time of its publication, the author's analysis based on Wersig & Neveling (1984) had
already been completed.
Introduction 5
Table 1. Number of terms by number of morphemes.

Length of terms Number of terms in the data (%)
1 morpheme 246 (20.0%)
2 morphemes 621 (50.6%)
3 morphemes 283 (23.0%)
4 morphemes 61 (5.0%)
Table 2. Quantities of morphemes.

Running morphemes Different morphemes Average use
2668 830 3.21
Japanese), kango (originated in Chinese) and gairaigo (originated in

[mainly] Western languages), the morphemes are defined as follows:
a. Wago/gairaigo: A minimal element constitutes a morpheme. But if
a minimal wago element is combined with a wago affix, this unit is
treated as a morpheme. A small number of Roman alphabet acronyms
are treated as gairaigo. The number of wago morphemes is 38 by
type and 83 by token. An example of wago morphemes is (hand).
The number of gairaigo morphemes is 199 by type and 602 by token.
An example is (computer).
b. Kango: A first-order combination of two minimal elements in the
terms constitutes a morpheme. However, a minimal kango element
attached to a morpheme (in an affix-like manner) is also treated as
an independent morpheme. Arabic numbers are treated as kango, as
their readings are mostly the same as kango numbers. The number of
kango morphemes in the data is 589 by type and 1979 by token. An
example is (book).
 Kango and wago: A first-order combination of a minimal kango and
a minimal wago Chinese character element is treated as a single mor
pheme. This is a modification from the original rule by Nomura &
Ishii (1988). The number of mixed morphemes is 4 both by type and
by token. An example is (staff).
Japanese "morphemes" as defined above are closer in status to words that
make up English compounds in that they mostly have independent mean-
ings (Kageura 1994). Table 1 shows a breakdown of the terms by number

of morphemes. Table 2 shows basic quantitative information about mor
phemes in the data. The full lists of terms and morphemes are given in
Appendices  and D, respectively.
Part I
Theoretical Background
Chapter 1
Terminology: Basic Observations
In this chapter, the basic status and nature of terminology within language is
clarified. Then, the traditional approach to the study of terminology is sum
marised and critically examined, and some recent developments in the study
of terms are briefly introduced. This chapter is intended to give readers the
basic background against which the theoretical position of the present study
is outlined. This will be elaborated in Chapter 2.
1.1 Basic observations
1.1.1 Terms and related notions
Any discussion about the basic status, nature and function of terms within
language must start with a provisional definition of "term" and immediately
related concepts. According to Bessé, Nkwenti-Azeh & Sager (1997) — a
compact and convenient glossary of the expressions of this field — "term"
and "terminology" are defined as follows:
term : A lexical unit consisting of one or more than one word which rep
resents a concept inside a domain.
terminology : The vocabulary of a subject field.
Two sets of expressions are important in these definitions, i.e. "lexical
unit" and "vocabulary" on the one hand and "concept inside a domain"
on the other. "Lexical unit" and "vocabulary" are conventional linguistic
notions defined adequately in many dictionaries of linguistics. They need
not concern us further at the moment.
It is unfortunate that, in the definition of "term", the authors use "word"

instead of "lexical item" since the whole purpose of a glossary of termi
nology is to distinguish terms from words. Similarly, the authors could
have avoided using the ambivalent "vocabulary" by simply stating "the set
of terms of a subject field". From a different point of view, however, the
above definition of "term" reflects the essential ambivalence in the relation
between "term" and "word". In one sense, "word" is used as equivalent
to "lexical item", in which case terms are a subset of words. In another
sense, "word" and "term" are used in a mutually exclusive sense (Miyajima
1981)1. The point will be elaborated shortly. At the moment, it suffices
to state that "term" refers to an individual item and "terminology" refers
to the collective object, in accordance with the distinction between lexical
unit and vocabulary.
The definitions of "concept" and some related concepts are given as:
concept : An abstract unit which consists of the characteristics of a number
of concrete or abstract objects which are selected according to specific
scientific or conventional criteria appropriate for a domain.
conceptual structure : A representation of the structure of the concepts
which belong to a subject field or domain.
characteristic : The semantic element which together with others consti
tutes the intension of a concept.
Lastly, "domain" and "subject field" are defined as follows:
domain : The subject field, area of knowledge, discipline, production pro
cess, or method in which a concept is being used.
subject field : An area of knowledge which is established for the purpose
of grouping into conventional categories the concepts considered as be
longing together.
Other glossaries of terms exist such as ISO 704 (1987), and most text
books give basic definitions of most of the terms listed above (Cabré 1993;
Felber 1984; Picht & Draskau 1985; Sager 1990). In fact, the definition
and characterisation of "term" in itself has been one of the main topics
among terminologists (e.g. Akhmanova 1974; Desmet & Boutayeb 1994;
Miyajima 1981; Shelov 1982; and a special issue of the journal Terminol-
1
In the following, "word" will be used in either of these two senses. Which of the two
is intended should be clear from the context.
Basic Observations 11
ogy [vol. 5, no. 1, 2000]). Any definition is controversial and provisional

outside a specific context, and the validity of any definition of term should
be supported by an explicitly declared theoretical standpoint and the con
crete description of the phenomena based upon it. The above definitions
are sufficient for the immediate purpose of opening the discussion.
1.1.2 Terms, words and parole
By defining a "terminology" as the set of terms of a subject field, terms

have been placed firmly in the realm of parole, the realisation of language,
as opposed to langue, the system of language (Kageura 1995; see also Tem
merman 1997). When Sager (1998) says "as linguistic signs, terms are a
functional class of lexical items", he makes the same distinction since for
him "functional" clearly refers to communicative rather than syntactic func
tion.
This attribution provides us with the first clear distinction between
terms and words, for a word can be recognised and defined at various other
linguistic levels, including the phonological, orthographical, grammatical
and semantic levels, many of which belong to the language system (Bauer
1983; Bloomfield [1933] 1984; Crystal 1992; Geeraerts 1994; Hockett
1958; Lyons 1968; Mugdan 1994; Nishio 1988; Sapir 1921; Tanaka 1988).
Some of the definitions of "word" are as follows:
1. semantic definition, e.g. "one of the smallest, completely satisfying bits
of isolated 'meaning' " (Sapir 1921: 34);
2. orthographic definition, e.g. "a unit which, in print, is bounded by spaces
on both sides" (Bauer 1983: 7);
3. phonological definition, e.g. "[a unit] bounded by successive points at
which pausing is possible" (Hockett 1958: 166);
4. grammatical (functional) definition, e.g. "a minimum free form"
(Bloomfield [1933] 1984: 178).
It is, of course, technically difficult to give necessary and sufficient
conditions for the specification of actually existing items called "words" by
means of a characterisation at any one of these levels alone. In the case
of terms, however, it is logically impossible to give necessary and suffi
cient conditions to specify terms at anything other than the level of parole.
This is clearly shown by the fact that there is no way of modifying the
above definitions of "word" to suit the definitions of "term"; we can speak

of, for instance, the syntagmatic patterns of terms, on condition that terms
are recognised at the level of language facts, but we cannot talk about the
"term" taking an independent position at the syntactic level.
The contrast between "word" and "term" can be illustrated by another
simple thought experiment using the distinction of language competence
and performance (Chomsky 1964). In the case of words or lexical units in
general, we can, for instance, speak of word formation at the level of lan
guage competence, discarding real-world factors and focusing on the possi
ble forms of theoretically infinite complex words, analogous to Chomsky's
syntactic theory (e.g. Kageyama 1982). It would be nonsense to speak of
the formation of terms in an analogous way, because, by their very essence,
terms have a concrete social existence as a functional class of lexical items
which manifests itself in the actual communication activity or performance
in a domain.
From a very different point of view, Sager, Dungworth & McDonald
(1980: 231) point out the essential difference of the theoretical perspec
tives in which "word" and "term" are recognised, by means of the different
importance given to the phenomenon of "lexicalisation" in the studies of
"word" and of "term".
The question of lexicalisation, so very elusive in general lan
guage, and therefore often put aside in linguistic analysis, is
fundamental [my emphasis] to the description of special lan
guages.
In other words, when terminology becomes the focus of research, the ques
tion of lexicalisation, or more specifically, terminologisation, should be
considered essential (even though the concept of terminologisation is not
the main target of the research); by contrast, when the attention of research
is centred on the word, the question of lexicalisation need not be essential.
Figure 1.1 illustrates the position of "term" within the distinction of
parole and langue. It is this position of "term" that is important in consoli
dating the concepts "term" and "terminology".
The position of "term" in Figure 1.1 immediately shows that the factual
identity of "term" and "word" at the level of the language system, i.e. terms
manifesting themselves as lexical units2, does not guarantee that studying
2
This restriction may be considered too narrow as it excludes extra-linguistic signs from
Figure 1.1. The position of "term" in language.
terms in the same manner as studying words is theoretically valid (Kageura

1995). While the study of words can extend from the investigation of word
forms at the level of langue, or language system, through the semantic
study of word formation patterns to the concrete study of word use or word
change within a language community, etc., not all the corresponding inves
tigations of terms constitute by themselves the proper study of terminology.
A naive study of linguistic aspects of terms on the basis of their factual iden
tity with, or similarity to, words would not be a proper study of terms but
rather a study of lexical items which happens to be based on terminological
data. For instance, a description of the formal structure of terms may well
be indistinguishable from a description of the formal structure of complex
general nouns. In general, the position of "term" in Figure 1.1 indicates
that, if we pursue the description of terms in parallel and by analogy to
what Lyons (1981) calls "theoretical synchronic microlinguistics", the very
essence or raison d'être of terms can never be addressed.
Some additional considerations which anchor the description to actual
sets of terms are therefore necessary for the study of linguistic aspects of
the class of terms. But it is essential for maintaining the granularity necessary for a scientific
study.
terms to be the study of terms. For instance, observing the ratio of various
forms of terms in a representative corpus of terminology, though simple,
would be useful to attribute the observation of forms of terms to the study of
terminology (cf. Pugh 1984). Any theoretical work on terminology should
be at least conscious of the fact that the level at which the category "term"
and "terminology" is consolidated is different from the levels at which the
category "word" is recognised, even though as empirical objects terms are,
like words, manifested as lexical items. We will come back to this point
later in 1.4.
1.1.3 Characteristics of terms
As the concept "term" belongs to the level of parole, the basic characteris
tics of terms should be and can only be observed at the level of parole. So
let us examine what seem to be the basic characteristics of terms in actual
usage.
"As linguistic signs, terms are a functional class of lexical units" (Sager
1998), and the basic function of terms is to express more sharply delineated
meanings identified as necessary within a particular domain by the com
plexity and number of concepts that have to be clearly distinguished. From
the angle of specialised discourse, we can state that some meanings of lex
ical units are consolidated by clarification and narrower determination in
order to satisfy the degree of specification required by the domain in which
they are used. Roughly speaking, it is in this way that lexical units become
the 'terms' of the domain.
Thus the division between general words and terms as empirical ob
jects is not rigid. As Sager (1998) states, "it can happen that non-specialists
consider a word to be a term which is, however, only a general word for
the specialist; equally, it can happen that specialists use terms which their
non-specialist audience take to be words in the general language". What is
more, "the possibility of many lexical units to function both as words and
as terms may even be a question of individual choice and interpretation of
the speaker and listener". Individual terms constantly interact and inter
sect with general words because they share the same linguistic forms (see
also Budin & Oeser 1995). So the particular range of terms representing a
domain is fluid.
On the other hand, however, in the modern world where science and
technology become more and more specialised, we tend to regard terms and
terminologies as having a clear and independent status within our language
experience. This tendency is accelerated because scientific and technical
discourse is spreading more and more into our daily communication due
to the arrival of the "information age". That terminology has a clearly ob
servable existence reflects an aspect of the truth, because the terminology
of a domain is a representation of the systematic part of the knowledge
of the domain, as can most typically be observed, for instance, in the ter
minology of mathematics. The terminology of a domain, therefore, taken
independently, can be regarded as having its own structure, representing the
"concepts" of the domain3.
The fact that terms are located within the tension between the need
for efficient communication and the requirement of representing the con
cepts of a domain makes terminology somewhat unique as a linguistic phe
nomenon. To the extent that the functional requirement of terminology is to
gain the precision necessary for expressing restricted meaning, terminology
tends towards stronger systematisation of its internal structure, not only in
its function of creating stable relations between lexical item and meaning,
but also in linguistic form. This makes terminology approach the nature of
the lexicon of artificial languages, which are characterised by rigidity and
systematicity of reference and designation (Hoffmann 1979; Sager 1994).
At the same time, to the extent that terminology shares its linguistic form
with the general vocabulary, it tends towards using the full flexibility of
natural language, not only in its lexical-formal dynamics but also in its ca-
pacity of establishing dynamic relations between lexical items and mean-
ing. This dynamic force, inherited from natural language, is strengthened
by intersecting with general-language words in real discourse. Figure 1.2
illustrates this specific characteristic of terminology within the dimension
of natural/artificial language.
3
As such, from the point of view of linguistics, "concepts" are clearer and more re
stricted meanings specific to a domain, though this alternative to a definition includes an
element of circularity.
Figure 1.2. The position of terminology in natural and artificial language.
1.2 The traditional theory of terminology
Now that we have seen the basic position and characteristics of terminol
ogy, it is convenient to examine the existing perception of what constitutes
the theory of terminology. We start the discussion with the examination
of the "Vienna" school of terminology (Felber 1984). It is undeniable that
this school, originally based on the work of Wüster (Wüster 1959/60), con
tributed to opening the research field of terminology. The "Vienna" school
has also strongly asserted claims for the independence of terminology as
a separate discipline, with its own theory and methods. As a result, some
textbooks of terminology devote significant space to the explanation of the
"Vienna" school or Wüster's ideas (Picht & Draskau 1985; Cabré 1993). In
addition, many of the current developments started from a critical exami
nation of the propositions made by the "Vienna" school.
1.2.1 Basic claims of the traditional theory

According to Felber (1984), there are three characteristics specific to the
theory of terminology (Felber 1984: 98):
1. "Any terminology work starts with concepts. It aims at the strict delimi
tation of concepts. The sphere of concepts is independent of the sphere
of terms."
2. "Only the terms of concepts, i.e. the terminologies, are of relevance to
the terminologist, not the rules of inflections and the syntax."
3. "The terminological view of language is a synchronic one, i.e. for ter
minology the present meanings of terms are important. For terminology
the system of concepts is what matters in language."
In short, the traditional theory of terminology addresses the relation
between concepts and terms, starting from concepts and focusing on the
present state of the conceptual structure and its representation. In this
framework, it is "concept" that takes a crucial role. "Concept" — "the
cornerstone of the GTT (general theory of terminology) and the starting
point of any terminology work" (Felber 1984: 102) — is defined as an
"element of thinking", which "consists of an aggregate of characteristics",
which "themselves are concepts" (Felber 1984: 103).
In support of these claims concerning the theoretical study of terms,
Felber (1984: 98-99) points out three peculiarities of the nature of termi
nology:
1. "Terminologies are deliberate creations. In common language the stan
dard is the usage of language... In terminology the free play of language
would lead to a chaos."
2. "The standardisation of single terms requires unified translinguistic
guidelines."
3. "Preference of the written form to the phonic form."
1.2.2 Restrictions of the traditional theory

While appreciating the historical importance of the "Vienna" school in con
solidating both scientific and practical concerns with the phenomena of
terms, the "traditional theory" is not devoid of crucial limitations. We can
immediately see the self-restrictiveness of the traditional theory of termi
nology both in the theoretical framework and in the peculiarities of the
perception of the nature of terminology4.
4
See also Temmerman (2000) for a critical examination of what we call here the tradi
tional theory of terminology.
First of all, the precedence of concepts over terms is dubious. If "the

sphere of concepts is independent of the sphere of terms", then anything
can be put into the sphere of concepts and there would be no guarantee that
the concepts thus understood would have any meaningful relation to the
sphere of terms in the first place.
In addition, despite the fact that utmost importance is given to "con
cept" as "the cornerstone of the GTT (general theory of terminology) and
the starting point of any terminology work", the nature of "concept" as
sumed in the traditional theory is not very rich. "Concept" is defined as an
"element of thinking" which "consists of an aggregate of characteristics",
which "themselves are concepts". This characterisation of concept is essen
tially the same as a widely-known descriptive mechanism of meaning, i.e.
that meaning consists of semantic features (Lyons 1977), and falls far short
the current theories of concepts and their formalisations in linguistics (e.g.
Pustejovsky 1995; Jackendoff 1983; 1987; 1990; Wierzbicka 1996) and in
cognitive science (e.g. Keil 1979; Rosch 1978). A recently published col
lection of major papers on concepts, though oriented to cognitive science,
covers a wide variety of views on concepts, among which the view of con
cepts as bundles of features is only a very basic one (Margolis & Laurence
1999). Even if we accept that concepts represented by terms have rigid con
notation, regarding a concept as a mere aggregate of characteristics does not
seem to have much to commend itself.
Secondly, the claim that "only the terms of concepts are of relevance"
to terminology, while the rules of inflections and the syntax are not, cannot
be accepted without reservation. It is true that these rules do not constitute
the consolidating factor of the concepts "term" and "terminology", as the
rales of inflections and syntax belong to the theory of the language sys
tem. However, the actual distribution of the morphological or syntactic
rules in terminology belongs to the sphere of parole and thus may reveal
terminology-specific characteristics. More importantly, as will be argued,
"concept" cannot in itself be the essential consolidating factor of "term" ei
ther. In view of this, the morphological and syntactic study of terms cannot
be categorically excluded from the theoretical study of terms.
Thirdly, the terminological view of language need not necessarily be a
synchronic one. It is easy to imagine a diachronic study of terms, and there
has been no argument which would persuade us that this would not be a
theoretical study of terminology; it simply depends on what sort of theory
one aims at. The fact that terms are observed in parole is a clear indication
that they can be studied diachronically.
The three peculiarities of the perception of the nature of terminology
also reveal the excessive restriction of the scope of terminology in the tra
ditional theory.
Among them, the second point, i.e. standardisation, is of little rele
vance to the theoretical study of terminology. The standardisation of terms
is by its very nature prescriptive and cannot be part of what we currently un
derstand by "theory". The process of standardising terms can theoretically
be studied as a(n external) factor to terminological phenomena, probably as
a kind of terminological socio-politics, but there is no room in any theory
of terms to incorporate this sort of study.
The claim that "terminologies are deliberate creations", on the other
hand, reflects an element of truth. As we discussed in 1.1.3, the unique
position of terminology is characterised by the combination of two contra
dictory factors, i.e. the quest for systematicity and flexibility. If, regarding
terminology as being close to artificial nomenclatures, the systematic as
pect of terms is dealt with in the study, it is useful to adopt the operational
characterisation that "terminologies are deliberate creations". Combining
the claim of systematicity in term creation with standardisation as can be
observed in Felber (1984: 98) is, however, not only irrelevant to the scien
tific study of terms but also harmful to the practical aim of standardisation.
This is because useful standardisation should be based on the observation
of the objects to be standardised. Linking the claim of the systematicity of
terminology to the quest for standardisation may well result in an estrange
ment of the standard from the reality. The validity of the assumption of
deliberateness must instead be examined with reference to the actual de
scriptive results of a scientific study.
A preference for the written form also seems to reflect a part of the
nature of terms. In comparison to words, the written forms of technical
terms are relatively more important than the phonic form. But this does
not automatically mean that consideration of the phonic form should not
be included in the theoretical framework. In practice it may be the spoken
language which encourages the creation of terminological variants. Al
though the present study is based on the written form, some interesting and
important aspects of terminological phenomena may be explained by their
phonetic characteristics.
In short, various aspects excluded from the traditional theory of termi

nology may be relevant at least to a certain aspect of terminological phe
nomena, and no in-depth argument has been given to validate the exclusion
of these aspects. As such, they may constitute a study of terms and might
well contribute to our knowledge of terms and terminology.
1.3 Recent trends in terminology
A growing number of researchers have recognised the limitations of the

traditional theory of terminology. For instance, Sager (1990) regards three
dimensions, i.e. cognitive, linguistic and communicative, as relevant to ter
minology. More recently, we have witnessed the growth of studies that
deal with the aspects of terms so far ignored in the traditional theory. This
development has partly been promoted by the requirement of automatic
processing of terms in texts.
For instance, Daille, et. al. (1996), Jacquemin (2001), Tartier (2001)
and Yoshikane, et. al. (1999) observe the morphological and/or syntactic
variation of terms in relation to the automatic processing of terms and their
variants. Tsuji (1998) shows that the actual survival of rival synonymous
terms is correlated to the length as counted by phonemes. Studies such
as Temmerman (1995; 2000), Pearson (1998) and Meyer & Mackintosh
(2000) analyse terms in use, thus explicitly or implicitly introducing the
diachronic perspective along which terms evolve. Seaborg (1994) reports
the history of the terminology of transuranium elements.
Within the so-called "Vienna" school itself, important new develop
ments are observed. Budin & Oeser (1995), for instance, observe "the
dynamic interaction between ordinary language and scientific terminolo
gies", thus effectively contradicting the strong proposition concerning the
nature and status of "concept" in the traditional theory. Budin (1996) talks
of the historical evolution of terminology, thus introducing the diachronic
viewpoint into the study of terms.
Along with these current developments of terminology, the nature of
concept defined in the traditional theory has been also critically examined.
Cabré (1995), Zawada & Swanepoel (1994) and Temmerman (1998/99;
2000) have argued that some terminological phenomena can be better de
scribed by using more flexible and powerful structures of concepts such as
prototype theory. They emphasise the flexible relationship between con

cepts and terms as well as the difficulty of defining the borderline of a con
cept. As such, they tend towards the natural language aspect of terminology
in Figure 1.2.
With this development, there are at present different perceptions of
the nature of concepts represented by terms. On the one hand, there are
researchers who, emphasising the artificial nomenclatural aspect of termi
nology, regard terms as systematic and deliberate creations reflecting the
systematic nature of concepts. This position is represented by the 'Vienna'
school. On the other hand, there are researchers who, emphasising the nat
ural language aspect of terminology, apply a more flexible framework of
concepts to the description of terminological phenomena. Although some
times presented as something exclusive to, or conflicting with, each other,
they are complementary, occupying opposite sides of terminology along the
horizontal axis of natural/artificial distinction indicated in Figure 1.2. As
far as the nature of "concept" is concerned, both the traditional view and
the recent view seem to be indispensable for capturing the full complexity
of terminological phenomena.
Recently, research in terminology has extended its scope, covering
from phonetic through formal to conceptual aspects. Diachronic as well
as synchronic studies have appeared, and a more advanced and flexible the
ory of concepts is applied to terminology. In addition to the individual
studies, new research paradigms of terminology are also searched for (e.g.
Cabré 2000; Pearson 1998; Temmerman 2000). The limitations of the tra
ditional theory of terminology seem to be overcome thanks to these recent
developments.
1.4 A missing piece for the theory of terminology
Despite the current development of research in terms and terminology, a

crucial piece, i.e. the position of the description of terms within the theory
of terminology, seems to be missing. In order to clarify this point, it is con
venient to examine the theoretical position of "concept" in the traditional
theory of terminology.
First of all, in the traditional theory of terminology, there is not a sin
gle explanation of the formal relationship between "concept" and "terms"
which makes it essentially different from the relationship between mean

ing and words in general linguistic semantics. The privileged relationship
between concept and terms is simply declared without argument, together
with the precedence of concepts over terms. As we argued, however, if a
system of concepts relevant to terms of a domain is to be constructed, the
existence of terms should be presupposed; a system of concepts constructed
without reference to terms cannot correspond to the structure of terminol
ogy. This also holds for the relationship between meaning and words. Thus
the relationship between concept and terms is formally equivalent to the
relationship between meaning and words.
In addition, the descriptive form of individual concepts in the tradi
tional theory is the same as a simple descriptive form of meanings, i.e. a
bundle of features or characteristics. These facts are not surprising; they
are consistent with the claim in the traditional theory of terminology that
the concept is the meaning of the term (Felber 1984: 103).
Another problem is related to the granularity of the description of con
cepts in the studies of terminology. Standard textbooks of terminology,
such as Felber (1984), Picht & Draskau (1985) or Cabré (1993), present
only partial structures of concepts, with small numbers of examples. With
only a few exceptions such as Pugh (1984) or Sager & Kageura (1995),
little work has been done on the concrete description of domain-dependent
conceptual structures which can be used to characterise specific termino
logical phenomena. More importantly, in what little concrete description of
terminological phenomena by means of concepts there is, we see that the
granularity of the descriptions of concepts does not exceed that of the de
scriptions of meanings in general linguistic semantics (Lyons 1977; Ogino
1987) or that of the descriptions of the more general "concepts" in natural
language processing and artificial intelligence (Carlson & Nirenburg 1991;
EDR 1989; Guda & Lenat 1990; Schank 1976; Sowa 1984; NTT 1997)
and/or in cognitive science (Keil 1979; Rosch 1978), all of which were
developed as general conceptual or semantic structures.
Concerning the position of "concept" in the traditional theory, we are
therefore facing the following situation:
1. The relation between concept and terms in the traditional theory of ter
minology is formally equivalent to the relation between meaning and
words.
2. The descriptive structure of concepts in the traditional theory of termi

nology, i.e. a bundle of characteristics, is the same as a simple descriptive
structure of meanings, i.e. the bundle of semantic features.
3. The granularity of conceptual systems described so far in the studies of
terminology does not exceed the granularity of semantic or conceptual
systems established in non-terminology-related studies.
Figure 1.3 illustrates this situation, i.e. the isomorphism between se
mantic studies of words and conceptual studies of terms. Given this,
we cannot distinguish semantically-oriented descriptions of words from
conceptually-oriented descriptions of domain-dependent terms from the re
sultant conceptual or semantic descriptions (Kageura 1995). If the concep
tual description of terminology constitutes the theory of terminology, then
this cannot be distinguished from the theory of linguistic semantics. What
researchers are doing within the traditional theory of terminology is thus no
different from what researchers are doing with lexical semantics in general
linguistics. The only difference resides in the fact that the former deals with
terms rather than general words5 ; the theory of terminology is supported
by the mere fact that it deals with terms.
A question naturally arises here about the status of the description of
terminological phenomena: Would the description essentially and only be
of terms of interest, or would it be of words in general that happened to be
obtained from the observation of terms? It is no more possible to distin
guish the study of terms from the study of some kind of lexical items on the
basis of terminological data. Note that we are seeing here the recurrence of
what was briefly mentioned at the end of 1.1.2, this time in relation to the
conceptual study of terms.
Unfortunately, this argument applies not only to "concept" in the
traditional theory of terminology but also to the recent studies (Budin
1996; Budin & Oeser 1995; Cabré 1995; Temmerman 2000; Zawada &
Swanepoel 1994). This is because they are concerned with the nature
of concepts for the description of terminology, borrowing the descriptive
framework of terms from studies of concepts which do not have anything
5
It was on the basis of this recognition that various studies about terms, such as the rules
of inflection or syntax, were defended in 1.2.2. Once it becomes clear that the relationship
between concept and terms in traditional theory does not have any privileged status, we lose
the argument for excluding various other aspects as a study about terms.
Figure 1.3. The conceptual study of terms and the semantic study of words.
to do with terminology in any inherent sense6; they are not concerned with
the position of concepts within the theory of terminology.
We are here facing an essential problem concerning the theory of ter
minology, as opposed to the theory of something that can be used for the
description of terms or terminology. The use of "concept" in the descrip
tion of terms, though essential, does not automatically make a study about
terms a theory of terms or terminology. The simple incorporation of more
advanced theories of concepts does not help either. The problem resides
in the theoretical framework of terminology in which "concept" is located,
rather than which theory of "concept" is used in the description of terms.
This recognition will be the starting point of the next chapter, where the the
oretical and methodological framework of the present study is elaborated.
6
Interestingly, the title of Temmerman's article (Temmerman 1998/99) is "Why tradi
tional terminology theory impedes a realistic description of categories and terms ...", where
the theory, which in fact is of concepts (which Temmerman compares with a theory of
understanding), is useful for the description of terminological phenomena. It is implied
here that terminological theory is the theory of something which is used for the description
of terminology. Incidentally, Bessé, Nkwenti-Azeh & Sager (1997) explicitly introduce
domain-specificity in their definition of concept (see 1.1.1) to avoid the problem examined
here at the level of definitions, but they do not say anything about how domain-dependent
concepts can be differentiated from ordinary concepts.
Chapter 2
The Theoretical Framework for the

Study of the Dynamics of Terminology
If we search for a theory of terms or terminology, as distinct from mere de

scriptions of individual terms using other theories such as theories of con
cepts, then it is necessary to lend more prominence to the so-far neglected
aspect of the position of "theory" in the study of terminology. This is all the
more important as methodological directions often depend crucially on the
theoretical standpoint. This chapter therefore starts by examining the con
ditions for the theory of terminology. Explanations of the basic hypothesis
for this study and of the methodological framework adopted in the pursuit
of this study will follow.
2.1 Conditions for the theory of terminology
Let us seriously consider the seemingly tautological claim that terminol

ogy (the study of terms) must be of terms and not just about terms (Sager
1998/99), not only as a requirement concerning the empirical range of the
phenomena to be studied, but also as a requirement for a theory of terminol
ogy itself. Then, not only should a theoretical foundation for a discipline
of terminology be based on the study of terms but also any resultant claim
derived from such a "theoretical foundation" should be of terms in some
meaningful way. In other words, the mere fact that the "theory" is derived
from some observations of terms as empirical data does not guarantee that
the theory is of terms. What is at stake here is the position of the theory
and the resultant descriptions of terms vis-à-vis the category of "term" or
"terminology" (in its nominalistic sense). We therefore start the discussion

in this section by clarifying the nature of the category "terminology" as dis
tinct from terms as empirical objects and then go on to examine the status
of a theory.
2.1.1 Terminology as an aspectual category
In 1.1.2, we saw that terms are functional variants of words. Thus "for
mally terms are indistinguishable from words" (Sager 1998/99). This of
fers a good, though implicit, starting point for understanding the status of
the "theoretical foundation" of terminology. That terms are functional vari
ants of words reflects the essential nature of terms, not only of terms as
empirical objects but also of "term" as a category, as we postulated from
the viewpoint of the epistemological conditions, on the basis of which a
discussion about terms becomes possible in the first place.
The essential point is that terms (and terminology), being a functional
variant of words at the level of parole, constitute an aspectual category of
the frame category of "lexical unit" or word"1. Thus "formally terms are
indistinguishable from words" simply because the very category of term is
not consolidated from the formal point of view. We may even go so far as
to claim that formally terms are words. This partly explains why the word
"word" was used in the definition of "term" by Bessé, Nkwenti-Azeh &
Sager (1997) (see 1.1.1).
Two problems immediately arise from the fact that terms constitute an
aspectual category. Firstly, the fact that a study is based on lexical items
that fall into the category of "terms" does not mean that any claims drawn
from the data are of terms and terms only. Secondly, some characterisations
of terms, even if they are true concerning terms, may not be meaningful.
For instance, assume that we discovered empirically, on the basis of the
observation of terms, that the part-of-speech categories for words can be
used to describe the formal patterns of terms. This observation, though
true, applies not only to terms but also to words in general. What is more,
it adds nothing new to our current state of knowledge, because this can be
deduced on the basis of the position of the category "term" with respect to
1
The situation is simplified slightly and a small group of terms, including the Latin
names of biology or geology and chemical and pharmaceutical names, is ignored.
Theoretical Framework 27
the category "word". When we try to characterise a phenomenon which is

aspectually consolidated, there is always the danger of falling into this trap.
This urges us to be conscious of the status or position of the claims ob
tained from the study; some claims concern the aspectual category because
that claim is made in such a way that it is valid for the aspectual category
either factually or de jure, while others are meaningful only within a par
ticular frame of reference.
Going back to the statement that "a theoretical foundation for a dis
cipline of terminology [should] be based on the study of terms", we can
now add another condition: Descriptions and claims obtained from a study
of terminology must be relevant to terms in a meaningful way. It must
be relevant to terms de jure (i.e. vis-à-vis the very concept of "term" and
"terminology") or it must make sense within a given context of discussion
and state of knowledge of both terms (the aspectual category) and words
(the frame category). The status of the theory is thus crucially related to the
range of generalisations that can be made from descriptions or observations
of the study.
2.1.2 The range of generalisations
Studies of aspectual categories should have a scope of generalisation ap

propriately matching meaningful discussions within the given frame of ref
erence and context. Three different possibilities can be recognised con
cerning the scope of generalisations based on the claims or descriptions
obtained from the study:
1. No generalisation is possible, i.e. the resultant description is only valid
for the particular data or examples that are used in the study.
2. The generalisation is applicable and valid only for the range of the as
pectual category, i.e. terms and terminology in the present case.
3. The generalisation is applicable to the frame category, thus collaps
ing the border between the aspectual category and the other categories
within the frame category.
To what extent a concrete observation can be generalised depends on
many factors, most importantly, the status of the data, the granularity of the
description and the overall setting of the study.
We should, for instance, recognise the difference between samples and

examples with respect to the status of the data (Hayashi 1993; Rea & Parker
1997). In quantitative analysis, this is understood as a matter of course,
but it is important for qualitative empirical research as well; from a proper
sample one can generally derive stronger conclusions than from arbitrary
examples. A good sample is necessary to draw a generalised conclusion
from weak assumptions. If, on the other hand, some concrete or detailed
assumptions concerning the target of the study exist, then logical discus
sion on the basis of a few examples might well be useful in examining and
honing these assumptions. Thus the granularity of the description may also
affect the status of the theoretical claim.
The meaningfulness of the study is also affected by its context. If, for
instance, the purpose of a study is to falsify the claim that terms do not have
ambiguity, then a single example is sufficient to invalidate the claim, while,
if the aim is to illustrate the general characteristics of terms, then arbitrary
examples, however many are provided, may not be sufficient.
All in all, such factors as the position of the theory and the status of
the data are at least as important for the theoretical study of terms and/or
terminology as such "theoretical" questions as "What is a concept?"2.
What is required, then, if one wants to describe, just as in many other
research fields, characteristics which are commonly recognised among
terms and preferably only among terms? An obvious answer is that the
study must be carried out in such a way that the claim is generalisable only
with respect to terms. Here, however, a difficult problem arises. Unless the
characteristic to be observed is known to be relevant only to terminology3,
generalised statements derived from terminological data tend to be appli
cable to words as well, because many viewpoints relevant to terms are also
relevant to words.
Here we are facing a dilemma. If the study is to be the study of terms
and terminology both logically and factually, it should give up any kind of
generalisations, effectively reducing the study of terms to terminography.
2
Note that the theory of physics never asks what the physical objects are outside the
theoretical description of the physical phenomena themselves, i.e. how they are. In termi
nology, there has been too much talk about "What is terminology?" or "What is a concept?"
unaccompanied by actual studies of terms.
3
In the traditional theory of terminology, "concept" is regarded as having this privileged
characteristic, which was falsified in 1.4.
On the other hand, if the study aims at certain kinds of generalisation, then
it tends to overgeneralise and the results of the study would no longer be
attributable exclusively to terms or terminology. What was discussed in 1.4
about "concept" was in fact an example of this general problem.
One way to deal with this difficulty is to introduce a comparative view
point and discuss and establish the borderline between terms and non-terms
or words, e.g. contrast the difference or non-difference between terms and
words from some common points of view. For instance, Miyajima (1981)
introduces a comparative point of view to the study of terms with general
words. What makes Temmerman (2000) interesting as a study of terminol
ogy is that the concrete descriptions in the study are related to a particular
aspect of the borderline of terms and non-terms.
Still, even in studies concerned with the borderline between terms and
words, the problem of the range of possible generalisations remains. Unless
one knows in advance the range of terminology to which a discussion on the
basis of exemplar terms is relevant, one has to continue analysing concrete
examples ad infinitum.
Would, then, the only task of the "theoretical" study of terms and ter
minology be to keep incorporating, or developing, various viewpoints in
accordance with the development of related fields for describing termino
logical data? And would such a "theoretical" study continue to analyse
individual terms without knowing the range of possible generalisations?
If one is not satisfied with such a prospect, then it is necessary to ex
amine what needs to be taken into account in order for a study based on
terms to become a theoretical study of terms. This should be addressed first
and foremost with respect to the logical status of the theory vis-à-vis ter
minology. To do this, it is necessary to go back to the fundamentals which
determine the status of the theory and to ask "What is terminology?" but
this time from a different perspective. This is our next task.
2.1.3 "What is a term/terminology ?" — revisited

In order to examine the logical status of the theory of terminology, it is
necessary to differentiate the concepts of terminology and term from term
and terminology as empirical objects. In this section, the former shall be
referred to in italics. This convention will also be applied to other related
terms such as vocabulary. This differentiation is essential for the following
discussion, because for a study about terms to be a study of terms de jure,

it should be linked up with the very concept of term or terminology. Let us
elaborate this point.
When we read articles or papers of a particular domain, we can recog
nise some lexical items in texts as technical terms. But how does this hap
pen? Or, rather, under what conditions does it become possible to talk about
individual terms at all? Upon closer inspection of the logical conditions that
make it possible to talk about terms, we find out that the very fact that we
can talk of terms in texts presupposes the existence of something that can
be generalised under the concept terminology, i.e. the concept of the vocab-
ulary of a domain (see 1.1). This is a nominalistic conclusion derived from
a simple thought experiment.
The same conclusion can also be drawn from the fact that the basic
definiens in the definition of term given in 1.1 (i.e. term is a lexical unit
... which represents a concept inside a domain) is in fact "a concept inside
a domain", for the concept concept cannot logically precede the concept
term as the essential determining factor of term (see also the discussion in
1.2.2). Thus, terminology should logically precede term and terms, because
an individual term is not a better linguistic representation of a domain, com
pared to the totality of relevant linguistic phenomena (e.g. a set of all terms
or terminology).
We can thus state: As an empirical object, terminology manifests it
self first and foremost as a set of individual terms at the level of parole,
while the empirical manifestation of terms is always and already preceded
and supported epistemologically by the consolidation of the concept term,
which is in turn preceded by the concept terminology. In any discussion of
terms, terminology is the first player (though invisible).
So what is terminology? An immediate answer is that it is a logical
construct required for the observation of terminology as an empirical object
— a circular argument. Going one step back, the concept terminology can
be equated with the concept the vocabulary exclusive to a domain — a
definition. The same type of questions should be asked about the concepts
vocabulaiy and domain, which forces us to go into an infinite regression if
this line of examination is pursued further.
We can nevertheless stop here because the determining concepts of ter-
minology, i.e. vocabulary and domain, are both external to terminology it
self; their definitions do not presuppose the existence of terminology. From
the study of terminology, therefore, it can be assumed that vocabulary and

domain can be externally identified in advance. As long as the empiri
cal manifestation of the concepts vocabulary and domain can be identified,
therefore, the empirical manifestation of terminology can also be identified,
without going into circularity. In fact, this is what is commonly done when
terminological dictionaries are edited, without bothering too much about
the question "What is a term/terminology?".
The main points so far are:
1. Terminology precedes term as well as terms as empirical objects, logi
cally and de jure, and terms as empirical objects in turn precede termi
nology as an empirical object.
2. Terminology is externally supported by the concepts vocabulary and do-
main.
3. From the point of view of the study of terms/terminology, vocabulary
and domain as empirical phenomena can be externally identified and
their empirical characteristics can be assumed to be known.
2.1.4 Theories of terminology
The epistemological precedence of the concept terminology over term, indi

vidual terms as empirical objects and terminology as a set of terms implies
that it is essential to treat the empirical phenomena corresponding to the
concept terminology in the theoretical study of terms/terminology. This
constitutes a minimum requirement for a study about terms to be a theory
of terms de jure and is independent of choices of the "theoretical" view
points or methodological devices in the study. In addition, because terms
as empirical objects are concrete items consolidated at the level of parole,
studies of terminologies are subject to the following conditions:
1. The theoretical study of terminology is, first and foremost, bound to
individual domains.
2. Within individual domains, the study of terminologies should, as its
main target, treat the terminology in its totality instead of individual
terms as an arbitrary set of examples.
If one wants to undertake a theoretical study of terminology, therefore,
one has to start from a set of terms which is representative of the terminol-
ogy of a domain4. If one wants to carry out a theoretical study of terms from
the point of view of concepts, the emphasis should be put on the concepts
of the target domain, rather than concepts in general.
Two types of theoretical studies can be distinguished that satisfy the
condition for a theory of terminology:
1. Studies which are concerned with characteristics of individual terms.
In these studies, ideally speaking, the descriptions should be based on
some characteristics which are common to and only to the terms of the
domain. Or alternatively, a comparative point of view should be intro
duced to show that the characterisations of terms are differentiated from
those of lexical items in general.
2. Studies which are concerned with characteristics of the terminology of
a domain as a whole. Though these studies can constitute theories of
terminology de jure, how meaningful they can be depends greatly on the
granularity of the descriptions.
The present study takes the second approach, i.e. to characterise the termi
nology of a domain.
2.2 The dynamics of terminology: target and method
The objective of the present work is the description of the dynamics of ter
minology or, more precisely, the dynamics of the terminology of a domain
in its totality, not of individual terms. Below, the basic perception of the
two key concepts, i.e. "terminology" and "dynamics", are clarified first.
The methodological framework will then be sketched.
2.2.1 The dynamic system of terminology

In Chapter 1, we observed the tension of the two factors in terminology,
i.e. the inclination towards rigidity and the need for flexibility. Roughly
speaking, the former shows itself in the form of autonomous and system
atic terminology, which gains its own status as a linguistic phenomenon of
4
It may be argued that this has been the target of the mainstream study of terminology,
which in reality, however, is not the case. Most studies of terminology have only treated
individual terms or a small number of exemplar terms, and most textbooks of terminology
fail to bring into the foreground the study of terminology itself.
Figure 2.1. The difference between terms, terminology and discourse.
the domain independent of the individual situation of discourse. The lat

ter, the need for flexibility, meanwhile, can be more clearly observed in the
use of terms in the actual performance of specialised discourse5. Corre
spondingly, from the point of view of individual terms, the restrictive and
regulative aspects are more clearly observed with reference to the system of
terminology, while the flexible aspects are more clearly visible with refer
ence to their actual use in the discourse. Figure 2.1 illustrates the sphere of
the system of terminology and the sphere of discourse which interact at the
level of individual terms or, rather, whose interaction produces the actual
manifestations of individual terms.
Two approaches to the study of terminology can therefore be recog
nised. One is the study of terminology in relation to the use of terms in
discourse; the other is the study of the system of terminology on its own.
The main concern of the present study is the latter, i.e. the system of termi
nology on its own.
5
This does not mean that the regularities of terms can only be recognised in the former
aspect and the uses of terms in actual discourse do not have any rigidity or systematicity.
Compared to general discourse, specialised discourse is, in general, more restrictive. Per
haps even the scope and available images for metaphor may be more restricted in special
languages than in general languages. So the distinction here is theoretical rather than real.
Accepting the distinction between the spheres of discourse and of the

system of terminology, and regarding the system of terminology as the tar
get of the study, the basic hypothesis of the study of the dynamics of termi
nology can be formulated as follows:
Hypothesis: Some systemic/systematic factors in the existing
terminology of a domain determine the formation of new terms
and the growth of terminology.
Note that the concept of dynamism here is essentially synchronic, as we
assume the systemic/systematic factors in the existing terminology without
referring to a time scale. The dynamics of terminology addressed in the
present study is, thus, the dynamic potentiality (of creating new terms) ob-
served in the synchronic slice of the internal structure of terminology6. The
basic question to be answered in the course of the study can be postulated
as follows:
Is there any systematicity, or motivated patterns, observed in
the construction of the totality of existing terms of the target
domain which can be regarded as reflecting the creative dy
namics of the terminology?
A few qualifications of the concept of dynamics may be useful. Firstly,
we are concerned with term formation and terminological growth and ex
clude from the study phenomena, such as meaning shifts or the metaphoric
use of terms, which are better studied with respect to discourse. Secondly,
because terminology belongs to parole, the creative dynamics is concerned
with, so to speak, the realistic possibility of the existence of terms given the
present state of the terminology of the domain, rather than general restric
tions of the possible form or characteristics of terms.
One final note: The clear boundary of a domain may be questioned.
In the present study, the existence of operationally clear boundaries for the
chosen domain is taken for granted, as this is a useful and necessary pre
supposition, at least at the current stage, where a concrete descriptive study
of terminology is needed.
6
The concepts "synchrony" and "diachrony" in language, introduced by Saussure
(1916), have long been a topic of debate (Culler 1976; Ducrot & Todorov 1979: 137-
144). Admitting that they are methodological concepts rather than concepts concerning the
factual state of languages, we here assume that when a single state of language is implicitly
or explicitly assumed — in other words, when the different language states along the axis
of time are not explicitly introduced — the study is synchronically oriented.
2.2.2 Methodological framework

2.2.2.1 Basic assumptions
A few simplifying assumptions should be clarified before introducing a
concrete methodological framework. First, as in the traditional theory of
terminology, we assume the existence of regularity at the level of 'concept'
and its correspondence with linguistic representation patterns:
Assumption: The dynamics of the system of terminology can
be observed in the formation patterns of terms as manifested
in their linguistic construction at a given time, the regularity of
which can in turn be attributed to some extent to the system-
aticity of the dynamics of the concepts of the domain.
This represents an affirmative rephrasing of the following proposition
formulated by Rey (1995: 48):
To the extent that a terminological system, even if it matches a
coherent conceptual system, is incapable of reflecting its inter
nal relationships, terminology is autonomous with respect to
epistemology. To the extent that a terminological system, even
if it is formed from unmotivated and common language words,
denotes a conceptual system and exists only for denoting it,
terminology is autonomous with respect to linguistics.
In relation to this claim and as a natural result of the emphasis placed on
the system of terminology, it is important here to stress that terminological
regularities should be analysed not only syntagmatically but also paradig-
matically, or, in other words, not only by their regularity of descriptive
patterns but by their classificatory patterns.
A further assumption is made concerning the concept "domain" and
the uniformity of terminology.
Assumption: The domain we start and end with remains
within the range of what Kuhn called normal science (Kuhn
1962). Correspondingly, it is assumed that the terminological
data available at the starting point of the study is basically uni
form and there is no difference in status among terms.
This assumption may look rather too simplistic, because domains, like
information science, consist of many subdomains, and even a small domain
such as artificial intelligence may well consist of a mixture of different

viewpoints and may have a history with many theoretical leaps. However,
it can also be said that a terminology of a domain — even if it consists of
heterogeneous subdomains and takes shape as a result of a non-monotonic
dynamism of the domain — comes to obtain a fair degree of regularity as a
mass in itself.
2.2.2.2 Framework for describing the dynamics of terminology

In order to exploit the dynamic potentiality of term formation and termi
nological growth in a theoretically proper manner, two complementary ap
proaches are adopted, i.e. conceptual analysis and quantitative modelling.
2.2.2.2.1 Conceptual regularities
Under the basic hypothesis of the study and the above assumptions, the
conceptual patterns of the formation of terms observed among the exist
ing simple and complex terms can be regarded as reflecting the basic reg
ularity of the creative dynamism attributed to the system of terminology.
So the first task of characterising the dynamics of terminology is to reveal
and describe the conceptual regularities in existing terms. The basic points
to be observed at the conceptual level are the relationships between terms
and their constituent elements as well as the relationships among the con
stituents and the type of conceptual combinations used in the construction
of terms.
Recall that it was argued in 2.1 that the overall terminology of a domain
should be the basic object to which a theory of terminology is attributed.
Thus, the conceptual descriptions should deal with the overall conceptual
patterns of terminology construction, with corresponding linguistic forms
as manifested by the terms of a domain; the descriptions together would
constitute a theory of conceptual patterns of term formation for the target
domain. Further details of the descriptive framework will be discussed in
Chapter 3.
Other than the practical limitations imposed by the simplifying as
sumptions adopted above, the description of the conceptual patterns of term
formation is logically limited to a relatively broad characterisation. As dis
cussed in 2.1.2, if, by describing conceptual patterns, we try to give neces
sary and sufficient conditions for the formation of terminology, we would
end up listing the combinations of linguistic items in a terminological sam-
ple used in the study; in the process of description, we would obscure the
observation of dynamics.
So the description of conceptual patterns must necessarily remain
somewhat general, at a level where the general regularities of term forma
tion patterns in a given domain can most properly be described. This, how
ever, immediately means that we lose the fine granularity of the description.
As a result, the range of the possible terms expected within the conceptual
patterns of term formation would necessarily remain rather broad. This is
not desirable — even if the form of conceptual description is so defined
that the resultant characterisations can be said to be of terms de jure — for
a proper theory should aim at gaining due granularity.
The natural question that arises at this point is: Beyond the level of
the granularity logically and theoretically required in the conceptual ap
proach, is there any way of describing more fine-grained regularities of the
dynamics of terminology? The answer is yes, though the type of regularity
that can be grasped becomes inevitably different from that observed in the
conceptual approach.
2.2.2.2.2 Quantitative patterns
In order to analyse the fine-grained regularities of the dynamics of terminol
ogy beyond the level that can be addressed by the description of conceptual
patterns, we will explore the quantitative patterns of the occurrence of lex
ical or conceptual items as a mass, within the conceptual patterns.
The current development of quantitative linguistics has shown that,
given a certain amount of data, it is theoretically possible to describe the
growth patterns of lexical items in their entirety within a certain category
(Good 1953; Good & Toulmin 1956; Chitashvili & Baayen 1993; Kageura
1998b), including the potential lexical items. Applying this framework,
and regarding individual lexical items as representing more specific con
cepts within the general conceptual categories, it is possible to give a more
detailed description of the dynamics of the potential directions of growth
of the terminology of a domain.
The quantitative analysis must be based on the basic features or view
points that reflect the conceptual and linguistic categorisations relevant to
the phenomena of term formation of the domain, for it also needs to be
linked with the totality of the terminology of the domain. In the present
study, priority will be given to the conceptual categorisation. In that sense,
we should keep in mind the following claim that "the quantitative and quali
tative sides of human speech are correlated and interdependent" (Piotrowski
1968; cited in Tuldava 1995: 133). This is not only relevant for the obser
vation of language phenomena but also for the methodological restrictions
of the quantitative analysis.
2.2.2.2.3 A representative sample
Analyses cannot be carried out properly without a representative sample of
the terminology of the target domain. As described in the Introduction, rep
resentative terminological data of documentation will be used in the present
study. So the present work would be a theory of term formation and termi-
nological growth in the field of documentation.
2.2.2.3 Limitations
The current approach has two major limitations. The first is of the kind
discussed in the previous chapter with respect to the traditional approach.
The internal characterisation of terminology — even if it properly describes
the essential nature of the terminology of the domain — does not in itself
provide sufficient conditions to distinguish the terminology of the target
domain from the general vocabulary or the terminologies of some other
domains. This is an essential limitation which is caused by the facts that
terms are a functional variety of lexical items recognised at the level of
parole only and that they constantly interact with general words. What
is necessary in order to differentiate the terminology of one domain from
others, then, is to describe the terminologies of a number of domains and
then compare them. There is no alternative to this. What is important at
the first stage is to carry out the descriptive study of the terminology of a
domain in such a way that the resultant description can at least be said to
be attributed to the terminology of the domain de jure. At a later stage, it
may become necessary to compare the formation patterns of terminologies
of different domains or of general words, but this can only be done after
the formation patterns of the terminology of individual domains have been
clarified.
The second limitation is directly related to the choice of theoretical
position and methodology. We only focus on the systematicity of terminol
ogy, and thus various aspects relevant to the observation of the totality
of dynamic terminological phenomena are deliberately excluded. We will

return to this point shortly.
2.3 The position of the present study
2.3.1 Structural approach

The methodological framework defined above puts the present work in the
tradition of the structural approach in two senses. As the word "structural
ism" is used in many ways in various fields, it is necessary to clarify what
is meant by structural approach.
Firstly, this term is used to refer to the methodological aspect of the
study. So, instead of subscribing to the philosophical implication of the
term structuralism in the social sciences and humanities, the term struc
tural approach here should be understood in analogy with the mathematical
sense of the word (Toyama 1996; Piaget 1969), even though a rigid mathe
matical formalisation of the description of the dynamics of terminology is
not intended7.
The methodology adopted in the present study is structural in this sense
because the regularities of conceptual patterns defined in the domain of
concepts are mapped to the regularities of the linguistic representations as
terms, and vice versa, as in the algebraic structure. Though quantitative
modelling is not commonly regarded as belonging to the structural ap
proach (Deuleuze 1973), it still has an affinity to the structural approach
as commonly understood in that it reveals the synchronic structure of the
target.
Secondly, the approach to concepts in the present study can be said to
be "structural" in the sense of structural semantics (Greimas 1966; Lyons
1977; 1995), i.e. the semantics which uses componential analysis or the
concept of semantic fields to reveal and describe the semantic relationships
between linguistic (mainly lexical) items 8.
7
At least in the masterpieces of structuralism, the term "structure" has, in its formal
aspect, a clear mathematical implication (Lévi-Strauss 1958; Jakobson 1980).
8
Temmerman (2000) criticises the structural approach in this sense, in connection with
the prescriptive nature of the traditional theory of terminology. However, it should be em
phasised that the structural approach has not been applied to the descriptive study of ter
minology in its proper sense. In contrast with Temmerman's work, which is explicitly
Studying conceptual patterns is close to conventional structural seman

tics, but the level at which "structure" is addressed is different in two re
spects. Firstly, structural semantics is inclined to study the semantic fea
tures of individual items and individual semantic fields in which a set of
items is treated as a whole. The present study analyses the totality of the
conceptual structure represented by the terminology of a domain. Secondly,
the theoretical aspects addressed by structural semantics are sometimes said
to belong to the language system, while the current study, treating termi
nology which is clearly a parole-based concept, does not belong to what
is recognised as the language system. In fact, the structural approach does
not limit the target of the study to the phenomena of phonetics, morphology,
syntax, etc., nor does it limit the theoretical range to the language system, as
is commonly understood from the word "structural" in the sense of Amer
ican structural linguistics (Bloomfield [1933] 1984). We may recall here
that Foucault (1968), who clearly contrasted his analysis of enoncé with
structural linguistics in the narrow sense (or theoretical micro-linguistics,
in Lyons' sense), was widely held to be a structuralist, against his will but
not without reason.
2.3.2 The concepts of dynamics and the present study

The concept of dynamics addressed in the present study is very limited,
i.e. the dynamic potentiality observed in the formation patterns of terms
within the system of terminology at a synchronic slice. Let us clarify the
position of the present study within the wider potential research areas of the
dynamics of terminology. This can be clarified by the cross-classification
of the distinctions between synchronic and diachronic on the one hand and
between the system of terminology and individual terms in the discourse
on the other9.
Because the concept "terminology" is first and foremost consolidated
at the level of parole, it is easier to postulate the diachronic studies of the
dynamics of terminology. Firstly, we can recognise the study of the di
achronic evolution of terminology as a set (cf. Budin 1996). This type of
non-structural, the present study is explicitly and deliberately structural.

9
In most textbooks of terminology, there are descriptions of the patterns of formation or
forms of terms (Cabré 1993: 188-191; Felber 1984: 171f; Picht & Draskau 1985: 106-112;
Sager 1990: 61-87). However, they do not address the point examined here.
study would deal with the story of the historical origin of the terminology
of a domain and the actual historical evolution of the terminology to the
present state. It might also talk about its possible evolution for the future.
This type of study contrasts with the present study in that the former ad
dresses the diachronic concept of dynamics.
We can also trace the historical profile of individual terms (cf. Sakakura
1966). In this case, the concept of dynamics would be more strongly related
to the actual historical discourse in which a term has been used. When in
history a particular term is created, in what discourse and how the concept
which the term represents has changed, etc., would be some typical ques
tions to be answered in this type of research.
The study of the synchronic dynamics of individual terms in discourse
seems a slightly twisted notion, because the introduction of discourse tends
to imply the introduction of the real-world time scale. We can conceive,
however, the study of a common metaphoric transformation or a meaning
shift observed in individual terms in the present state of discourse as an
example of the study of the synchronic dynamics of terms in discourse, on
condition that these dynamics are attributed to the historically abstracted
structure of discourse.
Figure 2.2 illustrates the distinction between the structural study of the
dynamic potentiality of terminology, the synchronic study of the dynamics
of individual terms, the historical study of the evolution of individual terms
and the historical study of the evolution of terminology. In the reality of ter
minology, these aspects interact with one another. For instance, the actual
use of individual terms in discourse may change the structural dynamics of
terminology, while the structural dynamics in its turn restricts the direction
of the creation of new terms and the use of existing terms.
So the overall dynamics of terminology would only be theorised as the
interaction of these different perspectives of dynamics. Within the over
all configuration of the study of the dynamics and growth of terminology,
the present study, i.e. the study of the structural dynamics of terminology,
constitutes a small but essential part10.
10
Related discussion on the relation between synchronic and diachronic aspects in the
structural approach can be found in Vachek (1983) and Jakobson (1980).
42
Dynamics of Terminology
Broken line: potentiality; Solid line: actual manifestation or realisation

Circle: terminology; Square: individual term; Arrow: dynamic change
Figure 2.2 Studies of the synchronic and historical dynamics of terminology and terms
Part II
Conceptual Patterns of Term

Formation
Chapter 3
Conceptual Patterns of Term Formation:

The Basic Descriptive Framework
In Chapter 2, it was argued that the terminology of a domain should be the

basic unit of study, if a study is to be, at least by right, a study of terminol
ogy. It was also argued that term formation patterns should be interpreted
as realistic possibilities of existence rather than theoretical potentialities of
possible forms. In this chapter, the consequences of this with respect to the
descriptive framework of conceptual patterns will be presented. Firstly, a
basic assumption is introduced and then the descriptive forms of term for
mation patterns are examined. The introduction of the basic elements or
devices used for the description of the conceptual patterns of term forma
tion will follow.
3.1 Basic assumptions
A few basic assumptions should be made for analyses of term formation

patterns. Firstly, it is assumed that the general linguistic rules of word for
mation (Bauer 1983; Ishii 1987a) equally apply to term formation. In this
study, this aspect is not addressed, as, in 2.2.2.1, we have the assumption
that the main feature determining the systematicity of terminological dy
namics is the system of concepts of a domain.
The next and most important assumption here, though in fact already
introduced briefly in 2.2.2.2, is concerned with the patterns of linguistic
construction and the underlying linguistic operations.
Assumption: The regularities of terminology construction are

reflected in the relationships between terms and their con
stituent elements.
Although simple and complex terms are historically created not only
by compounding but also by abbreviation, metaphorical transformation,
etc., these methods of individual term formation will not be taken into
consideration. Instead, it is assumed that they are created according to
the simple building blocks of compounding on the basis of the constituent
elements1, while the creation of individual terms is carried out within, and
controlled by, the overall structure of terminology. This is an obvious sim
plification, but we can nevertheless adopt this assumption on the basis of the
rationale that, whatever profile individual terms have, the collective dynam
ics of terminology functions in such a way that the resultant terminologi
cal structures have the overall systematicity as manifested in the building
blocks of their constituent lexical items.
3.2 Word formation and term formation
In 1.4, it was argued that in many cases conceptual studies of terms cannot
be distinguished from semantic studies of words. To avoid this problem and
to make a study's descriptions be of terms, a proper descriptive framework
for term formation is required.
3.2.1 Term formation as word formation

Under the "building block" assumption, typical studies of semantic pat
terns of word formation by compounding take the following form (Down
ing 1977; Ishii 1986; Nomura 1975; Saiga 1957; Warren 1978; Yumoto
1977; 1979):
1
This does not mean that the concept represented by a complex term is completely and
transparently determined by the building blocks of the concepts represented by its con
stituent elements. This strong version of the "building block" metaphor in compounding is
criticised by Langacker (1987). For a summary discussion related to this point, see Ungerer
& Schmid (1996). Note incidentally that the very form of the description of conceptual
patterns of term formation in the present study partly reflects Langacker's "scaffolding"
metaphor of term formation.
Conceptual Patterns of Term Formation 47
1. They introduce semantic characteristics or categories for words and/or

constituent elements of words.
2. They introduce classes of intra-word relations, i.e. relations between
constituent elements of complex words.
3. They describe the restrictions or tendencies of the combinability of con
stituent elements in terms of semantic categories and classes of intra-
word relations. In some cases, they also describe patterns that derive
semantic categories of words from the categories of their constituent el
ements and intra-word relations.
In many cases, semantic descriptions of complex words explain the restric
tion of compounding patterns.
Replacing the word "semantic" with "conceptual", and "word" with
"term", we obtain the first version of the description of conceptual patterns
of term formation (Ishii 1986; Sager, Dungworth & McDonald 1980). Such
descriptions:
1. present conceptual characteristics or categories for terms as well as con
stituent elements of terms.
2. present classes of intra-term relations which can be used to describe the
relationships that constituent elements have within complex terms.
3. present the restrictions or tendencies of the combinability of constituent
elements in terms of conceptual categories and classes of intra-term re
lations; in some cases also describing patterns that derive conceptual
categories of terms from the categories of their constituent elements and
intra-term relations.
As this framework is formally equivalent to that of the description of se
mantic patterns of compounding in general words, it is crucial that concepts
should be exclusively linked with terms. Unfortunately, as was already ar
gued in Chapter 1, the quest for privileged connections between concepts
and terms results in mere tautology (concepts are something which are rep
resented by terms and terms are something that represent concepts). Some
thing is missing from this descriptive framework of conceptual patterns of
term formation, i.e. there is no guarantee that the overall description of term
formation patterns can be interpreted as the description of term formation
in its own right, as distinct from the description of formation patterns of
some complex lexical elements that happen to be based on the analysis of
technical terms as data.
Take the worst exemplar case within this framework, in which only the
conceptual restrictions of formation patterns are described. The resultant
description will then consist of a list of statements such as: Constituent
elements that represent such and such a type of concept may be combined
as a nucleus with elements representing such and such a type of concept
as a determinant. Even if the description is obtained through the analysis
of a set of terms of a domain, it cannot be claimed that such a study is
exclusively about the formation patterns of terms of that domain and not
about general word formation that happens to be based on terms.
If some clearly domain-dependent phenomena are observed in the de
scription, it can be seen that the description at least reflects the nature of
term formation as distinct from word formation in general. Even then, the
overall form of description does not guarantee that the resultant description
is of terms.
3.2.2 Towards describing term formation patterns

A better descriptive framework of term formation includes an explicit intro
duction of the overall quantitative tendencies of formation patterns, on the
basis of a representative sample or corpus of terminology (e.g. Pugh 1984;
Kageura 1993). Here, the inclusion of the overall tendencies in the descrip
tion is a theoretical necessity for the description of term formation per se.
In this case, the quantitative tendencies within a domain logically reflect
the nature of term formation patterns, to the extent that the terminological
data reflects the terminology of the domain. Note that by introducing quan
titative aspects, the description of term formation can now be interpreted
within the totality of terminology to which individual formation patterns
belong. This represents a shift of the focus of term formation from individ
ual terms to the terminology as a whole, in accordance with the discussion
in 2.1.
However, the status of this type of study crucially depends on the types
of conceptual categories and intra-term relations introduced in the study.
If they are introduced on the fly, then the basic categories of observation
with respect to which quantitative tendencies of concepts and intra-term
relations are interpreted are not guaranteed to be essentially relevant to the
terminology of a domain.
Individual conceptual categories by themselves are not guaranteed to

be linked to a domain. Since a domain has a structure, a better represen
tation of a domain within the conceptual sphere is a conceptual system in
its totality, and not individual conceptual features or categories or a set of
disconnected conceptual categories. In accordance with the shift from in
dividual terms to terminology, individual concepts should be considered
as part of the conceptual system in its totality. Note that making individ
ual conceptual categories finer and more strongly domain-dependent does
not solve the logical problem concerning the theoretical status of the study,
though it might well help improve the granularity of the description.
Unlike disconnected individual conceptual categories, an overall con
ceptual system defined as a reflection of one domain cannot logically be
used in another domain, except when the domain is embedded in other
domains. So, introducing the description of an overall conceptual system
constitutes a minimum condition for a study of terminology that satisfies
the requirements examined in the previous chapter. Taking this into ac
count, we can adopt a modified three-step approach to the description of
conceptual patterns of term formation as follows:
1. Introduce a conceptual system and conceptual categories or characteris
tics for terms as well as constituent elements of terms.
2. Introduce classes of intra-term relations (and redefine them as concep-
tual specification patterns, as we will define shortly) that reflect the over
all tendencies of term formation in this particular terminology.
3. Describe the preferential combination patterns of combinations of con
cepts by means of conceptual categories, intra-term relationships and
conceptual specification patterns within an overall conceptual system
that reflects the overall structure of the terminology of the domain.
If the conceptual categories are introduced in such a way that all the cate
gories are properly related and located within the overall conceptual struc
ture assumed to represent the domain, the description would satisfy the
minimum requirements for achieving the status of a theory proper to termi
nology.
The introduction of a conceptual system has another important impli
cation. In the description of semantic patterns of word formation and anal
ogous descriptions of term formation patterns, the phenomenon of word
formation is viewed mainly from a syntagmatic point of view. However,
Figure 3.1. Description of conceptual patterns of term formation.
according to our claim that term formation should be described within the
overall terminological structure, the classificatory aspect of term formation
is essential. The conceptual system — as opposed to fragmentary con
ceptual elements — in the description of term formation patterns formally
allows us to take into account the classificatory aspects of the dynamics of
term formation.
The difference between this descriptive framework and the semantic
description of word formation can be illustrated as in Figure 3.1. Note that
this descriptive framework still presents only the minimum necessary con-
ditions for the study of term formation with respect to the theoretical status
of a descriptive framework. It does does not guarantee a sufficient charac
terisation of terminological phenomena, as distinct from general phenom
ena of word formation, at the level of actual description. This is one of
the reasons why the description of conceptual patterns of term formation
should be complemented by a quantitative description of terminological
growth, which sheds more detailed light on different aspects of the dynam
ics of term formation.
3.3 Elements for the description of term formation
3.3.1 Linguistic elements

At the beginning of this chapter, it was assumed that the basic general lin
guistic rules of word formation also form the background of term forma
tion. This is a mere truism, because terms are a type of word distinguished
by their function. As such, at this stage it becomes necessary to introduce
linguistic notions for the description of term formation patterns.
The first element is the distinction between head and modifier in com
plex words which in their narrow sense refer to the syntactic or formal level.
They will be used interchangeably with nucleus and determinant, which are
used in Sager (1990).
For a complex term consisting of only two constituent elements, it is
not difficult to determine the nucleus and determinant, though there may
be some practical ambiguities. In the case of Japanese, the element on the
right is the nucleus, except for coordinate constructions. For instance, in
(information retrieval), (retrieval) is the nucleus and
(information) is the determinant. Figure 3.2 illustrates this.
For a complex term consisting of three or more elements, the funda
mental problem of determining the "basic-level" nucleus arises (which is
different from the problem of structural ambiguities) now that classifica-
tory factors of term formation are taken into account. Take, for instance,
(automatic data processing), which consists of three el
ements, i.e. (automatic), (data) and (processing). The
dependency structure of this term is (automatic) (data)
(processing) With respect to the determinant (automatic),
(data processing) is the nucleus, while within (data
Determinant and Nucleus of Complex Terms with Two Constituent Elements
Determinant and Nucleus of Complex Terms with Three Constituent Elements
Figure 3.2. Determinant and nucleus of complex terms.
processing), (processing) is the nucleus and (data) is the deter

minant. This is illustrated in Figure 3.2.
From the classificatory point of view, considering the overall dy
namism of the terminological space, it is possible and even preferable to
determine the "basic" nucleus, i.e. the nucleus which belongs to the more
active and essential level of the concept system2. For instance, one might
argue that (data processing) should be the basic conceptual nu
cleus vis-à-vis the concept system in the field of documentation, as
(processing) alone is much too general.
Though the problem of the basic nucleus sheds light on a theoreti
cally important aspect of describing term formation, the concept of basic
nucleus will not be used in the present study. There are a several reasons
for this. At this stage of the research, it is preferable to first observe the
complete building block patterns based on the constituent elements. Intro
ducing the concept of basic nucleus is not a prerequisite for the description
of conceptual patterns of term formation but can be superimposed upon the
description on the basis of constituent elements at a later stage. In addition,
2
"Basic" here should be understood as something analogous to the basic-level of cate
gorisation, which has aroused much interest in cognitive science. See Murphy & Lassaline
(1997) for a brief explanation of the basic-level in cognitive science.
there is a procedural ambiguity in determining the basic nucleus, as can be

understood from the debate over the concept of "basic-level" in concep
tual categorisation in cognitive science. Introducing the concept of basic
nucleus properly, with respect to the classificatory characteristics of term
formation within the concept system, is in itself an independent research
matter which will be put aside for the time being.
To compensate for this, however, it is essential to take into account
the level of compounding within the formal dependency structure. The dis
tinction between nucleus and determinant cannot be carried out adequately
without referring to the dependency structure of complex terms. For in
stance, a term "  ", which has three constituent elements, with the
structure <A <B C > > has a different determinant-nucleus pattern from
a term with the structure < < A B> C>. In order to account for these
differences, the dependency structure will be first decomposed into binary
determinant-nucleus combination, and then the formation pattern of terms
will be reconstructed. The actual method of description will be explained
in 3.3.2.2.
A distinction between the constituent elements that can be terms and
others will also be made, though this distinction may not be an important
factor in observing term formation patterns. Correspondingly, the distinc
tion between intra-term relations that can lead to the formation of a term
and those that cannot will be introduced as well.
Lastly, syntactic classes or part-of-speech categories will not be used in
the analysis. This is because basic syntactic classes used in the description
of sentence structures are not very useful in the analysis of complex terms3.
Also, since the basic research aim here is to describe conceptual patterns
of term formation, the use of syntactic classes does not fit into the overall
research aim, even though in some other languages the careful distinction
of the characteristics of syntactic classes from conceptual categories may
be more difficult.
3
According to traditional syntactic classes, most constituent elements are simply cat
egorised as nouns in Japanese. With a few exceptions, such as Uchiyama, et. al. (2001),
studies that try to overcome this problem apply semantic analysis. In other languages where
many different syntactic classes can take part in complex word formation, one may have ad
ditional technical problems of assigning conceptual categories to linguistic items due to the
complexity of syntactic classes. Theoretically, however, the discussion with respect to the
status of concepts should hold because it is related more to the position of the theory than
to the empirical nature of language.
All in all, the formal and linguistic elements to be used in the descrip
tion of term formation patterns are as follows:
1. Distinction between determinant and nucleus.
2. Representation of dependency structure of complex terms. This includes
the distinction between intra-term relations (or conceptual specification
patterns) that form a term and those that do not, the latter of which will
be called local or secondary combinations.
3. Distinction between constituent elements that can be terms and others.
These are subsidiary to the conceptual elements used in the description.
3.3.2 Conceptual elements

In 3.2.2, we examined the basic framework for describing conceptual pat
terns of term formation. There are two basic descriptive elements, i.e. the
conceptual system and categories or characteristics that can be attributed
to terms and constituent elements; and intra-term relations that describe the
conceptual patterns of combinations between constituent elements. The ba
sic position of the present study with respect to these two elements will be
elaborated here.
3.3.2.1 Conceptual system and categories

In the present work, a concept is first and foremost regarded as a descriptive
and explanatory device for capturing characteristics of term formation pat
terns. This, however, does not mean that addressing conceptual patterns of
term formation can be carried out freely, ignoring existing concrete work
on concepts. Existing studies of concepts will be briefly outlined here in
order to explain the standpoint of the conceptual system and the categories
used in the present study. Specific descriptions of the concept system and
concept relations will be discussed in Chapters 4 and 5.
3.3.2.1.1 Position and definition of concept
The word "concept" is frequently used in day-to-day conversation. As Jack-
endoff (1990: 7) pointed out, there is a fundamental tension in the use of
the word in ordinary conversation:
On the one hand, it is something out there in the world: "the

Newtonian concept of mass" is something that is spoken of as
though it exists independently of who actually knows or grasps

it. ... On the other hand, a concept is spoken of as an entity
within one's head, a private entity, a product of the imagina
tion that can be conveyed to others only by means of language,
gesture, drawing, or some other imperfect means of communi
cation.
The academic study of concepts is generally considered to have origi

nated with Aristotle, and various aspects of concepts are still actively stud
ied in philosophy, psychology, linguistics, artificial intelligence and library
science, either directly or indirectly. While discussion about the nature
of concepts continues, most concrete studies concerning concepts seem to
adopt the standpoint that concepts are mental objects. For instance, all
the psychological studies take this position by definition (cf. Margolis &
Laurence 1999; Lamberts & Shanks 1997), and the major standardised def
inition in the field of documentation (ISO 5127-1 1983; Young 1983) and
terminology (Felber 1984; ISO 704 1987; Sager 1990) reflects it as well.
Researchers in linguistics and artificial intelligence also tend to take this
position (Jackendoff 1983; 1987; 1990; Schank 1976; Sowa 1984).
Following are typical definitions reflecting this mentalistic position:
Concepts or conceptual categories ... are mental representa

tions of objects, entities or events, stored in memory (Roth &
Frisby 1986: 19)
Concepts are mental constructions serving to classify individ
ual objects of the inner or outer world by way of more or less
arbitrary abstraction (ISO 704 1987: 1)
[Concepts are] constructs of human cognition processes which
assist in the classification of objects by way of systematic or
arbitrary abstraction (Sager 1990: 22).
If terminology is a social construct, the explanatory device reflecting

terminology should also be something social, rather than purely mental.
Referring to the Aristotelian view of concept is thus justified. On the other
hand, even though terminology is a social construct by definition, the cre
ation and formation of individual terms are the result of human cognitive
processes. Thus, concepts used as descriptive and explanatory devices in
this study should reflect human cognitive activity as limited by the target
domain. Referring to cognitive or psychological studies of concepts is also
justified on this basis.
What is adopted here as a descriptive device is, be it a mental or social
construct, something that "exists", in order that the concept should have a
minimal explanatory substance. Leaving the problem of what "concept"
is to the safe hands of philosophers and psychologists while keeping the
problem of what "concept as used in the study of terminology" is to the
present study, concrete studies concerning concepts will be freely referred
to in the discussion of concept structure and type.
It is relevant here to mention the distinction between the term "con
ceptual" and "semantic". In many cases, these two terms are not distin
guished; thus Collins & Quillian's (1969) "semantic" memory is frequently
discussed in the study of concepts (e.g. Roth & Frisby 1986: 39-43). On
the other hand, Jackendoff (1983: 95), for instance, describes "semantic"
as a term which is used to refer to linguistic properties which explain such
phenomena as synonymy, anomaly, etc. of sentences and "conceptual" as
a term which refers to the level at which linguistic and non-linguistic in
formation are mutually compatible (although he himself does not rigidly
distinguish these two terms in his own writing). Trying to distinguish these
two is neither useful nor necessary from the perspective of the present study,
as should be clear from the discussions in Chapter 1. In this book, the
term "conceptual" is preferred, as the term "semantic" gives the impression
that anything referred to by this term is closely related to or dependent on
language structure as perceived from a "theoretical synchronic microlin-
guistic" point of view (Lyons 1981) and excludes other aspects related to
"concept".
3.3.2.1.2 Concept systems and concept categories

The organisation of concepts is a focus of much research, especially in
cognitive psychology, which studies the mental representation of concepts.
Smith & Medin (1981) as well as Laurence & Margolis (1999) review influ
ential ideas of concept representation, mainly based on psychological stud
ies but not excluding relevant studies in philosophy, psycholinguistics, lin
guistics and artificial intelligence. Ungerer & Schmid (1996) offers a more
linguistically-oriented discussion of the cognitive organisation of concepts.
According to the classical view of concept organisation, the representa

tion of a concept is a summary description of an entire category of objects,
entities, processes, etc. with features (singly necessary and jointly suffi
cient) delimiting the range of the concept. This view, or the revised ver
sion of this (i.e. the neoclassical view, which in general discards sufficient
conditions and introduces a richer structure), is expressed or assumed by
many researchers in linguistics, such as Lyons (1968), Katz (1972), Miller
& Johnson-Laird (1976) and Jackendoff (1983).
Difficulties facing the classical view — e.g. the existence of unclear
cases, the difficulty of identifying necessary and sufficient features of a
concept, the existence of such psychological phenomena as typicality ef
fects, etc. — lead to alternative views about concept organisation, such as
the probabilistic view or the exemplar view4.
Different views of the organisation or representation of concepts are
discussed in relation to various cognitive phenomena, such as the categori
sation of objects, but none seem to give a complete explanation of the vari
ous phenomena. For instance, despite various criticisms, Keil (1979) argues
that certain ontological concepts fit the classical view. Ungerer & Schmid
(1996) mentions that the organisation of some scientific concepts fits the
classical view.
Another important aspect of concepts is the relationship between con
cepts and the structure of concepts, examined by philosophers such as
Frege, Carnap and Quine and researchers in artificial intelligence and psy
chology such as Masterman, Collins & Quillian, Woods (Way 1991: 174-
200). By far the most important and the most commonly perceived structure
is the hierarchy, similar to the structure of biological taxonomy or library
classification systems, constructed according to the generality of concepts,
i.e. generic relationships (Collins & Quillian 1969; Miller & Johnson-Laird
1976; Rosch, et. al. 1976; Keil 1979; Miller 1998; Murphy & Lassaline
1997). The exact nature of the hierarchical organisation, however, is still a
point of debate (Murphy & Lassaline 1997).
In the present study, the following basic assumption is adopted: A
concept system is organised as a hierarchy of concept categories which can
be clearly defined. This is a most natural assumption given the "building
block" assumption coupled with the distinction of the nucleus and the de-
4
Here the discussion follows Smith & Medin (1981). Laurence & Margolis (1999) gives
a slightly different classification and terminology.
terminant. It is also assumed that individual concepts represented by terms

and the constituent elements of terms can be categorised without ambiguity
within the concept system. This standpoint is a natural consequence of the
assumption of the present study, as discussed in the previous chapters, and
close to the nature of the concept in the traditional theory of terminology.
Note again that what matters here is the nature, structure and position of
concepts as used for the description of term formation patterns and not the
theory of concepts itself.
3.3.2.2 Intra-term relations and conceptual specification patterns

Besides generic (and to some extent part-whole) relations (e.g. Cruse
1986), there are a few common types and characteristics of concept rela
tions. Other relations among concepts are recognised and listed by var
ious researchers (e.g. Lyons 1977; Sowa 1984; Cruse 1986; Mel'cuk &
Zholkovsky 1988; Sager 1990). Each researcher recognises different sets
of relations, depending on the background and purpose of the research. As
a result, there is no widely agreed structure of concepts other than hierarchy
based on generic (and partitive) relations.
In the present context, we are only concerned with conceptual relations
as far as they are represented by intra-term relations. The status of intra-
term relations is different from those of conceptual relations in general or
relational concepts. Before establishing the relevant conceptual relations
from the point of view of describing conceptual patterns of term formation,
it is necessary to examine the proper form of introducing intra-term rela
tions. So only the form of describing intra-term conceptual relations will
be addressed here, leaving the examination of actual types of intra-term
conceptual relations to Chapter 5.
Examining the conceptual relations introduced in the existing studies
of semantic or conceptual patterns of word or term formation reveals that
they are defined as binary relations between a determinant and a nucleus
(Downing 1977; Yumoto 1979; Ishii 1986; Pugh 1984; Saiga 1957; Sager,
Dungworth & McDonald 1980). In fact, most of them only deal with com
plex words or terms with only two constituent elements.
As for intra-term relations, such studies as Saiga (1957), Downing
(1977), Yumoto (1979) and Pugh (1984), in principle, define the relations
as the status or role of the determinant with respect to the nucleus. On the
other hand, Sager, Dungworth & McDonald (1980) sometimes shifts this
Figure 3.3. Term formation within the overall terminological/conceptual system.
viewpoint and describes the relations as the role of the nucleus with re
spect to the determinant, if only for the purpose of description. Ishii (1986)
takes a verbal morpheme-centred approach, i.e. the relations are seen as
the position or role of the non-verbal morphemes with respect to the verbal
morphemes5.
The intra-term relations are defined in this study as the status or role of
the determinant with respect to the nucleus. This is a natural consequence
of our standpoint that we observe term formation not just from its syntag-
matic combination patterns but from the point of view of the overall system
of terminology formation. To capture the patterns of term formation within
the overall classificatory system of terminology formation, it is a basic tenet
to see term formation as the specification of concepts within a conceptual
class, as represented by the nucleus, by means of modifications represented
by the determinants. Figure 3.3 illustrates this.
5
This is possible because Ishii (1986) dealt only with complex terms that have verbal
morphemes as a constituent element. Note that this verb-centric view is common in auto
matic processing of complex words (e.g. Finin 1980; Fujita 1984; McDonald 1982).
The next important theoretical consideration is whether binary concep

tual relations are sufficient or not. As we have already clarified that we ac
count for the dependency structure of complex terms with more than three
constituent elements, formal relations between binary relations within com
plex terms are taken into account methodologically. However, it might be
the case that a certain type of binary relations strongly require further speci
fication by means of a particular type of binary relations for the construction
of terms with more than three elements (as represented by a dashed arrow
in Figure 3.3). In order to make presentation of such phenomena possible,
it is useful to extend the binary intra-term relations to possible n(> 2)-
ary conceptual specification patterns (Kageura 1997a), as described by the
stable and strongly motivated combinations of binary intra-term relations.
This is a theoretical extension, and most conceptual specification patterns
might in fact be identical to intra-term relations.
Chapter 4
Conceptual Categories for the Description of

Formation Patterns of Documentation Terms
As a first step towards the description of the conceptual patterns of the

formation of documentation terms, in this chapter we establish conceptual
categories and a conceptual system consisting of these categories, starting
from the examination of existing categories and examining the terminol
ogy of documentation described in the Introduction. The conceptual cate
gories, thus established, are assigned to individual linguistic items, i.e. to
constituent elements of terms as well as to terms. In this and the following
two chapters, the description is specific to the terminology of documenta
tion, though general aspects are also discussed as far as possible. Targeting
the terminology of a specific domain is a necessary requirement for the
theoretical study of terminology, as clarified in 2.1.4.
4.1 Conceptual categories in existing studies
There are many studies in various fields concerned with conceptual cat
egories. One of the earliest efforts to establish a set of conceptual cate
gories was made by Aristotle, whose categories include Substance, Quan
tity, Quality, Relation, Time, Position, State, Activity and Passivity (Aristo
tle 1963). Since the time of Aristotle, every science has every so often felt
the need to identify the conceptual categories with which it operates, and
philosophers of science, notably Hempel (1952) and Achinstein (1968),
have tried to define the concepts with which natural science operates.
Conceptual structures, however simplified, also underlie documenta
tion thesauri and classification systems built by library and information sci-
entists (e.g. Aitchson & Gilchrist 1997; Ranganathan 1962; Chan 1994).
Ranganathan's famous PMEST categories in his Colon Classification sys
tem — i.e. Personality, Material, Energy, Space and Time — can be re
garded as the broadest conceptual categories. Datta & Farradane (1978)
propose a simple scheme for general classification consisting of three basic
types, i.e. Entities, Activities and Abstracts, with a fourth type, Proper
ties. Dahlberg (1978), emphasising the verbally observable units of knowl
edge, introduces the following four categories: Entities (principles, imma
terial objects, material objects), Properties (quantities, qualities, relations),
Activities (operations, states, processes) and Dimensions (time, positions,
space).
In linguistics, the conceptual approach has reappeared with the recog
nition of the limitation of formal approaches. For instance, Jackendoff
(1983; 1987; 1990) — who in his conceptual semantics assumes the ex
istence of a unified abstract level (conceptual level) of linguistic meaning,
perception, etc. — introduces the following categories (which he calls onto-
logical category features): Object, Place, Path, Action, Event, Sound, Man
ner, Amount, Number, Property, Smell and Time (Jackendoff 1987: 148-
152). Some linguistic thesauri also introduce conceptual categories. Koku-
ritsu Kokugo Kenkyusho (1961), which is a linguistic thesaurus based on
semantic principles, first divides the categories according to syntactic word
class and then introduces the conceptual categories: Abstract Relations,
Agents of Human Activity, Human Activity - Mind and Action, Products
and Tools, Natural Objects and Natural Phenomena.
Terminology is one of the sub-fields of linguistics which emphasises
the importance of concepts and conceptual structures. Sager (1990), in his
introductory book to terminology, recognises four basic categories, i.e. En
tities, Activities, Qualities and Relations. Pugh (1984) subdivides the entity
concepts into Material, Abstract, and Neutral. In addition, she introduces
two entity categories, i.e. Representational and Software, which are depen
dent on the field of information processing that she investigated. Sager &
Kageura (1995) further elaborate terminological concept systems along the
same line.
In the field of artificial intelligence, computational linguistics and cog
nitive science, many conceptual categorisation schemes are established un
der various names. Most of them are intended for specific systems, but
some are intended for general use. Initial efforts in artificial intelligence
Conceptual Categories for Documentation Terms 63
were geared toward establishing universal concept primitives, e.g. Master-

man (1962) and Schank (1976). While many studies are only concerned
with practical applications, some claim theoretical validity. Some provide
extremely large structure (e.g. Cyc, as described in Lenat, et. al. [1990] and
Lenat [1995]), while others prefer concise structuring of the basic concep
tual system.
Sowa (1984) is an effort to establish a theory of concepts applicable
to artificial intelligence. The main concern of Sowa (1984) is with the the
ory of the formal structure of concepts and its representation, but he also
gives conceptual categories in the appendix. The top-level categories are:
Attribute (quality of an entity), Entity (physical objects as well as abstrac
tions), Event (acts by agents as well as happenings like explosions), Infor
mation (anything that can be communicated), Measure (no definition given)
and State (with duration, as opposed to events, which are in flux).
Ishikawa, Sakamoto & Sato (1986) describe a scheme of semantic cat
egories used in a large-scale machine translation system. The categories
are: Nations and Organisations, Animate Objects, Inanimate Objects, Intel
lectual Objects, Parts, Attributes, Phenomena, Sense and Feeling, Actions,
Measurements, Spaces and Topography and Time. Though they paid at
tention to the nature of concepts in the process of establishing the scheme,
the resultant set seems to be of a highly applied nature, as exemplified by
the inclusion of the categories "Parts" and "Nations and Organisations" in
parallel with the other categories.
EDR (1989) represents another attempt to build a large-scale concep
tual dictionary. The broader levels of the conceptual categories are Entity
Concepts — which are subdivided into Concrete Objects (living things, part
of living things, artifacts, organisations), Location, Abstract Objects (intel
lectual products, other abstract objects) — and Event Concepts — which
are divided into Phenomena, Transference, Action and Activity, Change,
Character and Disposition, and Others.
Carlson & Nirenburg (1991) try to establish an ontological model, with
application to computational linguistics in mind. Their categorisation sys
tem at the broadest level consists of: Free Standing Entities and Properties.
The former are divided into Objects (entities existing primarily in space,
e.g. Physical-Object, Mental-Object, Social-Object) and Events (entities
existing primarily in time, e.g. Change-Location, Apply-Force, Mental-
Events). Properties are divided into Attributes and Relations.
WordNet is a recent — and one of the most successful — attempts

at organising lexical items by conceptual structure (Fellbaum 1998a). In
the current context, what is called "unique beginners" in noun elements in
WordNet can be regarded as top-level conceptual categories. Miller (1998)
describes 11 unique beginners, i.e. Entity, Abstraction, Psychological Fea
ture, Natural Phenomena, Activity, Event, Group, Location, Possession,
Shape and State.
It is noted that some categorisation systems are very similar to each
other. For instance, Sager (1990) and the second level of Carlson & Niren-
burg (1991) are almost the same. Also, at a relatively generic level, there
are a few categories shared by many or most of the studies cited above.
Some of the most widely used categories are Entity, Event, Attribute and
Relation, though with some differences in name and scope. These broad
conceptual categories match our own naive intuition about conceptual cate
gorisations very well, though minor variations can exist. On the other hand,
it is also noted that, even among the relatively general studies of concep
tual categories examined above, there are many variations in the levels of
specificity and emphasised categories.
The broad conceptual categories that are widely recognised across
studies constitute a useful starting point for any new trial of conceptual
categorisation. However, the fact that there are diversities in levels and in
types of narrower categories indicates the necessity of adjusting the levels
of specificity of categorisation according to the nature of the data and the
aims of the study. Bearing these in mind, let us turn now to establishing
the conceptual categories for the description of conceptual patterns of term
formation in the field of documentation.
4.2 Conceptual system for documentation terms
As mentioned in the previous chapter, this study adopts a pragmatic view

of the nature and structure of concepts, i.e. it analyses concepts in such a
manner and degree that the formation of terms in the field of documentation
can be explained. This, however, does not mean that ad hoc conceptual cat
egories can be freely introduced under the guise of subject-specificity. The
guiding principle for this effort is as summarised clearly by Rosch (1978:
29), "Maximum information with least cognitive effort is achieved if cat
egories map the perceived world structure as closely as possible". In the
current study, what is aimed at is to obtain categories that map the concep
tual regularity of formation patterns of documentation terms as closely as
possible. Below, we first introduce broader concept categories in a rather
top-down manner, referring to the schemes briefly reviewed in the previous
section, and then establish finer categories through examining documenta
tion terms and their constituent elements.
4.2.1 Broad conceptual categories

In order to establish the broad conceptual categories to which others may be
associated or subordinated, we will first briefly describe a simplified cogni
tive process, which justifies a set of conceptual categories. In the absence of
any other reliable mode of introduction, it is felt that this provides for some
intuitive control over what might result in yet another abstract scheme for
ordering knowledge.
In the building of a discipline, cognition is concerned with ordering the
items of our perception and experience, which implies relating and compar
ing these items. As a first step of cognition, moving from perception and
observation to creating knowledge, we identify similarities among objects
of our experience and group similar objects into categories. This process
of categorisation consists of grouping items of experience on the basis of
some perceived similarities into Entity concepts and then naming the con
cepts. With this objective, we look for identical or similar features and
exclude features which we consider irrelevant to the subject field, which
may be different from our general attributions. The next step, which now
properly belongs to establishing the field of the subject, is to group con
cepts into larger classes by some criteria relevant to the subject, e.g. form,
function, origin.
In the process of consolidating the items of our experience into enti
ties and classes of entities, we recognise features of items or categories,
Datta & Farradane's (1978) "precepts", which are then abstracted into con
cepts of the category Qualities. Qualities are thus, in the first instance, seen
as dependent concepts, i.e. as constituent elements of other, independent
concepts only. At the next level of abstraction they can be considered as
categories of their own, e.g. "red", "blue" and "colour". Some quality con
cepts tend to be quantified for the sake of a tighter conceptual delimitation,
e.g. "length" or "heat" can be associated with fixed extensions and mea-
surements. Thus, at this level, quantity is not considered to be different

from quality, but constitutes a subclass of quality.
By the process of identifying the constituents of items, we build up
a static system of concepts, the structure of which is established by a new
class of concepts, i.e. Relations. The same relations which hold together the
concepts of general knowledge, e.g. "inclusion", "difference", "similarity",
are used in the special field of documentation, but with different frequencies
and scope; there are also subject-specific ones, such as "synonymy" and
"disjunction". The closer-knit structure of special fields is characterised
by a narrower definition of certain relations, e.g. "accuracy", which can
represent a quantified relation.
Beside building up a static field of entities with their properties and re
lations, we also conceptualise and categorise the Activities of a special sub
ject field in order to permit the creation of a dynamic knowledge structure.
Activity concepts themselves have properties and form particular relations
with each other and with other concepts.
This very simplified model of concept formation has yielded four broad
categories common to many concept schemes and roughly equivalent to
Sager's (1990) four categories and Carlson & Nirenburg's (1991) second-
level categories. Referring to the concepts represented by documentation
terms and their constituent elements, we can define and subdivide these
four categories as follows:
Entity: The class of concept obtained from the abstraction of items of our
direct experience and reflection and perceived as having separate exis
tence in time and space. This category is conventionally divided into
material and nonmaterial, as observed in some of the schemes in 4.1.
In the field of documentation, we additionally identify many concepts
by their symbolic representational function, which places them at the
intersection between material and nonmaterial objects. Among nonma
terial entities, we identify a set of concepts which belongs to the meta
language level of conceptualisation1. They do not represent entities but
assist in the classification of entities. Their importance in describing
term formation was first recognised by Pugh (1984), who called them
"neutral entities". We extend this category and redefine it as Classifica-
1
This can be regarded as a mapping of linguistic function to the conceptual system, i.e.
quasi-concept.
Conceptual Categoriesfor Documentation Terms 67
tory entities. In the field of documentation, this leaves abstract entities,

which themselves form a subclass. Thus, we have four subcategories:
Material Entity: A type of entity concept characterised by derivation
from physical objects, e.g. "human", "machine".
Representational Entity: A type of entity concept characterised by
derivation from physical objects with a symbolic function relevant
to the field of documentation, e.g. "symbol", "copy", "character".
Abstract Entity: A type of entity concept not directly derived from
physical objects nor directly observable physical phenomena, e.g.
"information", "method".
Classificatory Entity: A type of entity concept which serves to order
and classify other concepts according to complexity, function, etc.
and is devoid of any physical association, e.g. "element", "unit".
Activity: The class of concept obtained from abstraction of separately
identifiable processes, operations or events carried out by or with en
tities, e.g. "publishing", "access".
Quality: The class of concept derived (i) from an analysis of components
and features of other concepts and (ii) by contrasting concepts and iden
tifying their distinctive characteristics. Quality concepts are at a basic
level considered constituent characteristics of concepts and, at a second
level of abstraction, as independent concepts, e.g. "white", "whiteness",
"colour".
Relation: The class of concept obtained from abstraction of physical, tem
poral or other ontological relationships among objects and from logical
relationships among entities, properties and activities, e.g. "difference",
"relation".
For convenience, we shall henceforth use the following abbreviations
for these categories: ME for Material Entities, RE for Representational
Entities, AE for Abstract Entities, CE for Classificatory Entities, AC for
Activities, QL for Qualities and RE for Relations.
4.2.2 Sub categorisation of conceptual categories

Because the conceptual categories identified so far in this chapter are rela
tively broad and yield groups with considerable internal diversity, even in
the relatively small field of documentation, it was felt useful to attempt
a further subdivision of conceptual categories in order to capture more

closely the nature of the basic concepts that binds the formation patterns
of terms of the field. Mainly, it was considered essential to develop the
typology of concepts to the extent that the study of the formation of com
plex terms could be carried out on the basis of the relevant coherent sets of
concepts.
The structures arrived at and presented here are therefore experimen
tal in two senses. Firstly, regarding their overall classificatory significance,
they are expected to confirm a need for a division into general and subject-
specific conceptual categories, in continuation of this type of division al
ready established at the higher level. Secondly, regarding their contribution
to the overall objective of this study we expect to reassess these subcate
gories in the light of their usefulness for the elucidation of patterns of term
formation.
The method used for creating subcategories was pragmatic. Groups
of single morphemes which represent similar concepts were assembled and
contrasted with other groups in order to define the extension of each class.
At the same time, the intensional coherence of the conceptual nature of
these groupings was analysed in order to confirm the inner consistency of
each group. So the subcategories are inherently restricted by the actual ter
minological data described in the Introduction. This reflects our standpoint
that we are concerned with the study of terminology and not the study of
concepts through terminological data2.
After a preliminary conceptual scheme was established, based on the
examination of the morphemes, it was validated against sample complex
terms extracted from the data. Regular adjustments, modifications of the
labels for each group and redefinitions of the classes led to the current set,
described below3.
2
At the same time, however, note also that we are trying to keep our criteria of concept
subcategorisation as general as possible. This is to avoid falling into the trap of tautol
ogy, i.e. establishing convenient categories purely to demonstrate the formation of terms in
the corpus to be consistent and coherent, which would exclude the chance of grasping the
dynamics of term formation.
3
In the process, in fact, it was decided that some ambiguities were collapsed. For in
stance, the place/organisation ambiguity of such morphemes as (library) was
recognised as irrelevant as it does not affect the description of regularities of term formation
patterns. For a related discussion, see 4.3.1.
The subcategories of each conceptual category are best described sep

arately. Before this is done, however, a few general observations can be
made which affect most, if not all, concept groupings arrived at.
(a) Separately-defined concepts and dependent concepts
A distinction can be made between morphemes representing concepts sepa
rately defined in the corpus and morphemes representing concepts which in
our corpus only appear as elements of complex terms. In general, concepts
independently defined in the corpus constitute the core set in the field; their
status is roughly divided into: (i) basic objects in documentation activity
and study, (ii) concepts which are essential in the treatment of basic objects
and (iii) central concepts of neighbouring fields relevant to documentation.
While we accept that this distinction is not fully satisfactory as it includes
an element of circularity if used in explaining the term formation patterns,
the conceptual distinction reflected in this phenomenon — which cannot be
straightforwardly fit into the ordinary classification of conceptual system
— in itself is still important. We are interested here in the consequences
of this distinction, rather than the discussion of individual cases of inclu
sion or exclusion from the corpus. Note that the result of this distinction is
considered to reflect the subject dependency (see also [e] below).
(b) Generality and speciality of concepts
A second important feature of subcategorisation is the degree of general
ity and specificity. This phenomenon makes grouping difficult and also
raises the question of the borders of the special subject field. For instance,
some concepts are highly specific in relation to a distinctive characteristic,
such as (volume: a material entity) and (summary: a representa
tional entity), while the general nature of the concept (index) makes
it difficult to fit it into a single conceptual position in relation to the mate
rial/representational distinction of the scheme4. At a different level, gen
eral quality or relation concepts can be regarded as abstract entities, e.g.
(relation), (form). While fully acknowledging the potential dif
ficulty of categorisation in some parts, the subcategorisation was carried
out so as to maximise the overall consistency of the attribution of concepts
represented by morphemes.
4
Though this might be better dealt with as a problem of polysemy, in the end,
(index) was classified under representational entity.
(c) Simple and complex concepts

A third general characteristic of almost all subcategories is a division into
simple and complex concepts. In the case of entities, these manifest them
selves sometimes as types of entities and their constituent parts and, in the
case of qualities and relations, as classes of properties or relations as well as
the properties and relations themselves. The criterion of division between
these two types of concept is provisionally considered to be subject-specific
and is based on what is considered to be a minimal functional or operational
unit in the field of documentation.
(d) Basic and functional or relational viewpoints of concepts
Some concepts, mainly entities, are differentiated from others by their inter
nal nature, while others are differentiated by their relation to other concepts
or functions. For instance, the distinction of the category 'part' from in
dependent concepts is based not only on the complex/simple point of view
but also on the basic/relational point of view. Some qualities are themselves
the manifestation of the relational or functional aspects of the concepts to
which they were originally attributed, while some are manifestations of
their internal nature. In this sense, relation concepts are considered to be
the direct manifestation of the very relations themselves, without being in
clined to any of the concepts among which the relations are established.
(e) General/subject-specific conceptual structure
In subdividing concepts into subcategories, the temptation had to be re
sisted to create what might be conceived of as a general knowledge struc
ture in the form of a thesaurus. The corpus presents a highly specific set of
concepts in one subject field. It is therefore to be expected that the types of
concepts identified here will reflect the nature of this subject field and that
certain areas of concepts will be heavily underrepresented if not altogether
absent from these subcategories. Any scheme must, however, acknowledge
this fact and leave room for other categories to be added to the knowledge
structure thus created.
These criteria are operative in the following subdivision of concept
categories, though we will not systematically refer to them. These criteria
may cause conflict in the categorisation of concepts, because different cri
teria may lead to different categorisations of concepts. The preferences of
these criteria were decided at each level of categorisation vis-à-vis the data
and the general rationale of the conceptual categories.
4.2.2.1 Subcategories of material entities (ME)

The subdivision of material entities appears at first sight to be fairly con
ventional; there are two large groups of animate and inanimate entities. An
imate entities are provisionally divided into a category of individual types
such as (author) and a category of groups and institutions such as
(centre). This corresponds to the distinction between simple and
complex items.
Inanimate concepts show the particular orientation of the subject field.
A large class of types of information carrier is first identified, thus made
divisible into documents and non-documents. Documents are further divis
ible into subtypes and constituent parts5.
The subtypes of documents can be defined by the viewpoints by which
concepts are identified and are not necessarily mutually exclusive at the
level of referents. Most document concepts are created from the mixed
points of view. Document concepts identified from their mode of publica
tion and formal structure, e.g. (journal), are consolidated as one group.
The other concepts which identify some aspects of content but in most cases
also specify other characteristics such as function or mode of publication,
e.g. (dictionary), are categorised into another group. The generic-
specific relation can be recognised between these two groups, but we treat
them as belonging to the same level, as we are concerned with the cate
gories for term formation and not a rigid hierarchical structure. Yet another
group identifies the physical form of documents, such as (book-form).
Then there are the groups of entities one would expect in general,
namely machines and implements such as (computer) and
materials such as (colour item). The former has similarity with ani
mate entities, as the concepts of this group can be active. Another coherent
group of places/locations, e.g. (region), was singled out.
5
Subcategories made by part-whole relation are different in nature from the subcate
gories established by generic-specific relation. We, nevertheless, collapse this distinction
in the same spirit as the incorporation of some partitive relations (such as disciplines and
their subdivisions) under the name of "hierarchy" in documentation thesauri (Aitchson &
Gilchrist 1997). This decision can be justified as we are not concerned with such conceptual
operations as inheritance of conceptual characteristics.
Table 4.1. Subcategories of material entity concepts.

Tag Explanation Example
ME1 Animate
ME11 — People
ME111 —— Types (author)
ME112 —— Parts (hand)
12 — Organisations (centre)
2 Inanimate
21 — Information Carriers
211 — Documents
2111 —— Types
21111 (by mode of publication/formal structure) (book)
21112 (by content/mode of publication) (dictionary)
21113 (by physical form) (book-form)
2112 ——— Parts (cover)
212 ——Non-Document Information Carriers (tape)
22 — Machines and Implements (computer)
23 — Materials (silver halite)
24 — Places/Locations (city)
25 — Parts of Inanimate Entities
251 ——Parts in General (pocket)
252 ——Parts as Places/Locations (edge)
Finally, parts of material entities were recognised at various levels,

which are presented as subcategories of relevant categories, as in docu
ment concepts. Parts of inanimate entities (which in the current context are
limited and constitute a coherent group both for morphemes and for terms)
were further divided into places/locations and parts in general.
These observations have led to the subcategorisations as illustrated in
Table 4.1. For convenience in regards to further reference to these cate
gories, we have added an identification tag to each. Examples are given for
the bottommost categories.
4.2.2.2 Subcategories of representational entities (RE)

We start the subcategorisation based on general/subject-specific distinction
and identify four categories, i.e. (i) broad concepts which are neutral to
specific subject field, e.g. (code); (ii) concepts which come from a
neighbouring field of linguistics that shares many representational concepts
with documentation, e.g. (word), (syllable); (iii) documentation
Table 4.2. Subcategories of representational entity concepts.

Tag Explanation Example
RE1 Broad Representational Entities
RE11 — General
RE12 — Specific
RE121 —— (by function and relation)
RE122 ——(by representational structure and type)
RE2 Linguistic Entities
RE21 — Set Concepts
RE22 — Element Concepts
RE3 Documentation Entities
RE31 — Types
RE311 —— Complex Units
RE312 —— Simple Units
RE3121 —— (by function and relation)
RE3122 —— (by structure)
RE32 — Parts of Documents
RE4 Software Entities
RE41 — Programme Units
RE42 — Programme Parts
RE43 — Software Elements
entities such as (entry); and (iv) software entities such as

(programme) which belong to data processing but appear in documentation.
The category of broad representational entities basically consists of
simple units. The dividing criteria (b) and (d) are convenient for establish
ing subcategories of broad entities. The category of linguistic entities can
be subdivided into concepts of sets and concepts of basic units, reflecting
the dividing criterion (c). In documentation entities, the two dividing cri
teria (b) and (c) are in evidence. As it seemed appropriate for the subject
field of documentation, it was decided to use the unit-part division first —
thus establishing the major division of independent concepts and their con
stituent parts — and then subcategorise the first group according to items'
relative complexity or simplicity as perceived from the point of view of the
subject field. Simple units thus established are further divided "by function
and relation" and "by structure", according to the criterion (d). Lastly, soft
ware entities are conveniently divided into three subgroups, i.e. programme
units, programme parts and software elements. Table 4.2 shows the sub-
categorisation scheme of representational entities for the terminology of
documentation.
Table 4.3. Subcategories of abstract entity concepts.
4.2.2.3 Subcategories of abstract entities (AE)

While the first definition of abstract entities was based on negative criteria,
i.e. anything which does not have material or representational attributes, a
closer analysis can lead to a positive definition, i.e. information in its broad
sense, as similarly defined in Sowa (1984). Three major subcategories were
identified here, namely concepts dealing with elements of information, e.g.
(sense); concepts dealing with systems of knowledge, e.g. (the
ory); and, lastly, concepts dealing with our ordering of and attitudes to
information, e.g. (prescription). These three are consolidated by a
mixture of the criteria (c) and (d). In some of these categories, the generic-
specific distinction can be observed. The subcategorisation scheme of AE
is shown in Table 4.3.
4.2.2.4 Subcategories of classificatory entities (CE)

The definition of classificatory entities given above can alternatively be ex
pressed: entities which are neutral with respect to the specific classes of
material, representational and abstract entity, and which are specified by
role, formal complexity, etc. In short, the determining characteristic of a
Conceptual Categoríes for Documentation Terms 75
Table 4.4. Subcategories of classificatory entity concepts.
classificatory entity cuts across the determining characteristics of the other

three entity categories. A typical example is (element). There are two
viewpoints by which classificatory entities are identified, i.e. formal com
plexity or structure and relation or role, in correspondence with the criteria
(c) and (d) above, respectively. Some concepts represented by morphemes
and many represented by terms are identified both by formal complexity
and relation.
Concepts specified only by formal complexity can be further divided
into two, i.e. single/simple items and complex items. The latter is then di
vided into complex items seen as one and complex items seen as composed.
Another specifically important distinction to CE can be made between
the concepts which can be independent in the field of documentation and
those which cannot. This distinction is the application of criterion (a)
above. Although we admit that introducing formal criteria in conceptual
categorisation is not totally satisfactory, it is considered to be conceptually
relevant here, as otherwise subcategories of CE cannot incorporate the dis
tinction of concepts important in the subject field in single morpheme form
and those which are not. Table 4.4 lists subcategories of CE.
4.2.2.5 Subcategories of qualities (QL) and relations (RL)
The two categories of qualities and relations share a number of features

revealed more clearly at the level of subcategorisation:
1. Both operate at two levels of abstraction. At the first level, we identify
the values of qualities and relations. For instance, (blue) and
(indirect) show specific values of quality and relation, respectively. At
the second, higher level of abstraction we encounter the abstracted labels
of the values and groups of values. (condition) and (relation)
can be seen as examples of such concept classes. The labels can be
divided into documentation-related and general.
2. Linguistically, both categories contain a number of dependent mor
phemes, the vast majority of which are general concepts, such as
(non-), which from the orthodox linguistic point of view can better be
defined from the point of view of linguistic functions.
3. Both categories contain a large number of concepts consisting of mor
phemes not separately defined in our corpus. This is an indication
that, singly, they cannot be distinguished from general concepts and can
therefore acquire special reference only through association in complex
terms.
4. There is a further group of simple morphemes among the labels and val
ues which share their form with general concepts but are given a subject-
specific definition in our corpus. For instance, (performance),
(constraint), etc., are redefined with the special reference they have in
the field of systems analysis.
Note that some of these features, e.g. undefined concepts and redefined con
cepts, are not concerned with what is commonly regarded as the inherent
part of the domain-oriented conceptual structure, but rather the reflection
of terminological structure.
4.2.2.5.1 Qualities (QL)
In the presentation of labels of types of qualities, it is considered convenient
to follow the order of grouping indicated above. The values cover a wide
range which can be subdivided into general categories such as "spatial",
"temporal", etc., and subject-field specific ones, such as "representational
mode", which are relevant to documentation. Table 4.5 shows the subcate
gorisation scheme of quality concepts.
Table 4.5. Subcategories of quality concepts.
4.2.2.5.2 Relations (RL)

This group of concepts has similar values as that of quality concepts, since
relation can be defined simply as a particular way of viewing a quality, i.e.
not from the angle of its intrinsic characterisation of a concept but rather
from the outside in connection with other concepts. This viewpoint, then,
adds a specific set of concepts which identify modes of relations, compar
isons between concepts and the relative status of concepts to each other.
Some dependent morphemes are polysemous and are therefore listed as a
single group at the beginning of the list of values of relations. Their specific
meaning only becomes clear in combination with the other concept in com
plex terms, i.e. they bear syntagmatic function corresponding to functional
elements, which we have conveniently classified here. Table 4.6 shows the
subcategories of RL.
Table 4.6. Subcategories of relation concepts.
4.2.2.6 Subcategories of activities (AC)

The conceptual nature of activity is complex, as activity inherently involves
various players, e.g. performers and objects, and has a temporal and aspec
tual dimension. There are many precedents of conceptual or semantic cat
egorisations of activities for various purposes, e.g. Fellbaum (1998b), Fujii
(1976), Jackendoff (1983), Kindaichi (1976), Okuda (1983), Ogino (1987)
and EDR (1989), to mention a few.
Three factors for the categorisation of activities were regarded as rel
evant here: (i) The types of phenomena introduced or represented by the
activity, e.g. production, change of state, etc., (ii) the conceptual fields
in which the activity concepts are situated, e.g. general and specific, or
documentation-related activity, computer manipulation, etc., and (iii) the
temporal nature of the activity, e.g. instant, durable, etc6.
In our corpus, a small number of concepts, like (action), have
a broad extension and cannot be classified properly from the above view
points. Most of them have little distinctive value in the field of documen
tation, while other concepts indicate various degrees of specific activities.
These broad activity concepts are indicators of activity or units of activity.
6
Here we resist the temptation to describe individual lexical elements representing ac
tivity concepts in the form of describing verbs in, for instance, lexical conceptual structure
or the combination of case frame and its semantic constraints, which are mainly used for de
scribing constraints on general combin ability of verbal elements (which in general represent
activity concepts) and other elements (cf. Fabre 1996; Takeuchi, et. al. 2001).
Table 4.7. Subcategories of activity concepts.
Other concepts are first categorised according to viewpoint (i), espe

cially with respect to the main concepts to which the activity concepts are
applied (linguistically, they roughly correspond to the grammatical objects
of transitive verbs and subjects of intransitive verbs). The subcategories ac
cording to this viewpoint are: (a) action, i.e. activities after the end of which
the original state of the concepts to which they are applied is restored, or
activities in which changes of state or position are irrelevant; (b) transfer

ence, i.e. activities which transfer the concepts to which they are applied;
(c) production, i.e. activities which leave some products, including such re
production activities, as (copy); and (d) state change, i.e. activities
which change the state of the concepts after application.
Then, each group (except action) is subcategorised further based on
the fields of the activities, which are, in general, observed in the conceptual
types of objects to which activities apply, e.g. representational entities, etc.
Lastly, each group is categorised according to the temporal nature of the
activities, i.e. instant, static durative and dynamic durative.
Table 4.7 lists the subcategories of activity concepts. The list of con
ceptual categories and subcategories, presented from Tables 4.1 to 4.7, is
also given in Appendix A.
4.3 Allocation of morphemes and terms to the conceptual system
The morphemes and terms in the data of documentation terms described

in the Introduction were allocated to the conceptual system. Note the dual
role of the conceptual system. On the one hand, by allocating terms to
the conceptual system, we are defining the conceptual fields within each
of which term formation patterns are observed; on the other hand, by al
locating morphemes to the conceptual categories, we are preparing for the
description of conceptually-motivated combinations in term formation. In
any case, what is aimed at by allocating morphemes and terms to the con
ceptual system is not to detail descriptions of concepts that morphemes or
terms represent, but to give them a classificatory framework within which
regularities of term formation patterns can be described.
We discuss here some important aspects of allocation and the general
qualitative and quantitative tendencies of the morphemes and terms with
respect to conceptual categories. The actual data is presented in Chapter 6
and in Appendices  (list of terms) and D (list of morphemes).
4.3.1 Allocation of morphemes

The morphemes were allocated independently to conceptual categories,
without referring in detail to their overall use in the corpus, because it is
procedurally necessary to first obtain morpheme-concept pairs in such a
Table 4.8. Morphemes with multiple allocations.
Table 4.9. Conceptually-indeterminate morphemes.
way that the allocation is not totally dependent on their specific use in and
only in the data. Most morphemes clearly fall under one of these cate
gories, but some ambiguities and doubtful cases were recognised7. These
are as follows:
1. For some morphemes, the distinction between material and representa
tional documentation entities was difficult; e.g. (index) and
(bibliography) can be material or representational depending on use. In
stead of assigning two categories, a single category has been allocated as
far as possible, in order not to lose the overall picture of formation pat
terns by being too fine-grained for individual cases. The validity of the
7
The ambiguities and doubtful cases discussed here are only with respect to the concep
tual system established here, i.e. only when they are expected to affect the overall descrip
tion of conceptual regularities of term formation patterns.
allocation must be evaluated later from the point of view of providing a

satisfactory explanation of the data in our corpus.
2. Though we tried to avoid multiple-allocations, multiple categories are
allocated to some morphemes. The 14 morphemes to which more than
two conceptual categories are assigned and which are actually used as
such in the corpus were listed in Table 4.8. Many morphemes take ac
tivity and entity concepts. As a result, the number of unique morpheme-
concept pairs became 845 (at the level of surface string, the number was
830, as shown in Table 2 on page 5). We will henceforth call these 845
morpheme-concept pairs simply morphemes.
3. Some morphemes were declared as conceptually indeterminate or void,
which is similar to Miyajima's (1973) concept of the "meaningless mor
pheme". This applies to those dependent gairaigo morphemes which
only occur in the determinant position in some specific terms and func
tion only as a discrimination symbol. They are listed in Table 4.9. Note
that they originally have their own meanings.
4.3.1.1 Tendencies of broad categories

We can now observe the basic quantitative nature of the terms and mor
phemes of the corpus with respect to the conceptual system. Table 4.10
shows the quantitative tendencies of broad categories and indicates the
number of morphemes by type and token8 (each with the ratio against the
overall number of types and tokens), average frequency, standard deviation
of frequency (Sdev), as well as the number of morphemes independently
used as terms and their ratio against the total number of morphemes of the
category.
Some general characteristics can be observed. Firstly, categories of
entity concepts have a higher average of use than other categories. This
means that either they include core morphemes, i.e. morphemes central to
the field of documentation, or morphemes generally used in constructing
complex terms irrespective of the domain. AE and CE are notable in that
they are small in number but frequent in use. The higher values for the
standard deviation of AE and CE indicate that they have greater internal
diversities. This is in a way understandable as they include strongly subject-
The tenus "type" and "token" here are used in the technical sense of quantitative lin
guistics (Herdan 1960).
Table 4.10. Quantitative data of morphemes for broad categories.

Category Type (%) Token (%) Average Sdev Term
ME 143 (16.92) 487 (18.25) 3.41 6.06 58 (40.56)
RE 117(13.85) 456 (17.09) 3.90 5.94 78 (66.67)
AE 72(8.521) 331 (12.41) 4.60 13.17 27 (37.50)
CE 35 (4.142) 158 (5.922) 4.51 9.61 16(45.71)
QL 146 (17.28) 314(11.77) 2.15 2.39 11 (7.534)
RL 132(15.62) 322 (12.07) 2.44 3.89 14(10.61)
AC 190 (22.49) 589 (22.08) 3.10 4.85 43 (22.63)
dependent subcategories as well as categories that seem to accommodate

versatile general morphemes (see Tables 4.3 and 4.4).
The high type ratio and low occurrence of QL and RL seem to reflect
their nature in term formation, i.e. they are commonly used for differentiat
ing concepts, but the range of application of individual morphemes is rather
limited.
4.3.1.2 Tendencies of subcategories

Table 4.11 shows the tendencies of subcategories of entity concepts. We
observe the tendencies at the second level of subcategorisation, because we
can obtain a good balance between the quantities of morphemes and the
qualitative characteristics of categories, including subject-dependent cate
gories (other than CE).
A few interesting points are observed. Firstly and expectedly, cate
gories related to documentation and information (i.e. ME21, RE31 and,
to some extent, AE1) constitute the major subcategories in terms of type
and token number of morphemes. Their average use is reasonably high,
but there are more "general" subcategories in which the average use of a
morpheme is even higher, i.e. RE11, AE11 and AE22. Also, categories re
lated to documentation and information include a high percentage of term-
constituting morphemes, but a few more general categories again have even
higher percentages of term-constituting morphemes.
The following informal arguments may explain these observations.
Terminology construction makes use of a large store of morphemes for core
subject-oriented concepts. At the same time, it uses some general or non-
core subject-related concepts, including morphemes from neighbouring
Table 4.11. Quantitative data of morphemes for entities.

Category Type (ratio) Token (ratio) Average Sdev Term
ME11 14(1.657) 54 (2.024) 3.86 6.72 3(21.43)
ME12 12(1.420) 53(1.987) 4.42 8.68 1 (8.333)
ME21 80 (9.467) 284 (10.64) 3.55 6.26 43(53.75)
ME22 13 (1.538) 60 (2.249) 4.62 6.79 7(53.85)
ME23 6(0.710) 11 (0.412) 1.83 1.60 0(0)
ME24 6(0.710) 9 (0.337) 1.5 0.84 1 (16.67)
ME25 12(1.420) 16 (0.600) 1.33 0.49 3(25)
RE11 4 (0.473) 51 (1.912) 12.75 7.72 4(100)
RE12 21 (2.485) 47(1.762) 2.24 2.12 11 (52.38)
RE21 4 (0.473) 18 (0.675) 4.5 4.51 4 (100)
RE22 16(1.893) 64 (2.399) 4 6.63 11 (68.75)
RE31 46 (5.444) 223 (8.358) 4.85 7.45 25 (54.35)
RE32 13 (1.538) 24 (0.900) 1.85 1.72 11 (84.62)
RE41 4 (0.473) 12 (0.450) 3 4 4 (100)
RE42 5 (0.592) 7 (0.262) 1.4 0.55 5 (100)
RE43 4 (0.473) 10 (0.375) 2.5 1.73 3(75)
AE11 9(1.065) 78 (2.924) 8.67 15.08 8(88.89)
AE12 6(0.710) 38(1.424) 6.33 12.58 3(50)
AE13 8 (0.947) 11 (0.412) 1.38 0.52 7 (87.5)
AE21 19 (2.249) 56 (2.099) 2.95 3.37 5 (26.32)
AE22 9(1.065) 112(4.198) 12.44 32.49 1(11.11)
AE31 10(1.183) 17 (0.637) 1.7 1.06 0(0)
AE32 6(0.710) 12 (0.445) 2 1.10 3(50)
AE33 5 (0.592) 7 (0.262) 1.4 0.55 0(0)
CE11 4 (0.473) 13 (0.487) 3.25 1.71 1(25)
CE12 20 (2.367) 129 (4.835) 6.45 12.45 10 (50)
CE21 5 (0.592) 8 (0.300) 1.6 0.89 5 (100)
CE22 6(0.710) 8 (0.300) 1.33 0.52 0(0)
disciplines. They do not constitute a large source for the terminology but
tend to be used repeatedly.
Table 4.12 shows the quantities of morphemes of QL and RL at the first
level of subcategorisation. The similarity of these two categories is obvious.
In both cases, labels of qualities and relations are small in number while
frequent in their use, i.e. in a way, they are closer to entity concepts (see
Table 4.10), while QL2 and RL2 accommodate large numbers of morpheme
types with much lower levels of use.
Table 4.12. Quantitative data of morphemes for qualities and relations.

QL1 36 (4.260) 100 (3.748) 2.78 3.23 11 (30.56)
QL2 110(13.02) 214(8.021) 1.95 2.02 0(0)
RL1 30 (3.550) 102 (3.823) 3.4 6.45 8 (26.67)
RL2 102 (12.07) 220 (8.246) 2.16 2.71 6(5.88)
Table 4.13. Quantitative data of morphemes for activities.

AC1 14(1.657) 31 (1.162) 2.21 2.42 0(0)
AC211 11 (1.302) 33 (1.237) 3 2.19 2(18.18)
AC212 10(1.183) 34(1.274) 3.4 4.95 2(20)
AC213 44 (5.207) 189 (7.084) 4.30 5.41 10 (22.73)
AC221 7 (0.828) 7 (0.262) 1 0 0(0)
AC222 11 (1.302) 33(1.237) 3 4.69 2(18.18)
AC223 7 (0.828) 21 (0.787) 3 2.89 4(57.14)
AC231 7 (0.828) 22 (0.825) 3.14 1.86 1 (14.29)
AC232 15(1.775) 51 (1.912) 3.4 6.05 5 (33.33)
AC233 14(1.657) 42(1.574) 3 2.69 4 (28.57)
AC241 12(1.420) 17 (0.637) 1.42 0.51 1 (8.33)
AC242 38 (4.497) 109 (4.085) 2.87 6.98 12(31.58)
Table 4.13 shows the quantities of morphemes of activity concepts.

Some common tendencies can be observed among subcategories related
to documentation concepts (AC222, AC223, AC232, AC233 and AC242),
with respect to their average use and the ratio of term-forming morphemes.
4.3.1.3 Summary of morpheme allocations

Looking across the broader categories, AC is the biggest category in terms
of the numbers of both different and running morphemes. QL and RL,
which accommodate good percentages of different morphemes, have a low
share in running morphemes, while AE and CE, which only accommo
date relatively small numbers of different morphemes, have high average
occurrences. RE, ME and AC also accommodate many morphemes rep
resenting central or important concepts in documentation, but the average
occurrences are not so high as those of AE or CE because the range of their
conceptual coverage is also wide.
If we look at the subcategories with a high average of use, we can

recognise the following three types:
a) subcategories which themselves represent the central or important con
cepts in documentation, e.g. ME21111, ME21112, REU, RE311,
AE11.
b) subcategories which accommodate morphemes with linguistically-
flexible applicability in term formation, e.g. QL11, RL21, AC2311.
c) subcategories which do not themselves represent central or important
concepts in documentation but accommodate important or central items
in documentation, e.g. AC213, AC2422.
In cases b) and c), the high average frequency is mainly due to a small
number of specific morphemes, such as (nature) (QL11) or (-ize)
(AC213). Although high occurrence does not mean a qualitatively specific
behaviour in term formation, in cases b) and c) — where the high average
occurrences are due not to the nature of the conceptual categories but to the
nature of some morphemes in these categories — it may be misleading to
draw conclusions only from the generalisations of term formation based on
the conceptual subcategories established here. Other criteria may have to
be considered.
4.3.2 Allocation of terms

The next step of analysis consisted of the allocation of conceptual cate
gories to the terms. Unlike the allocation to morphemes, no substantial
problems were encountered. The conceptual categories of individual terms
will be presented in Chapter 6.
Table 4.14 lists the number of terms belonging to each broad concep
tual category. ME and RE each constitute more than a fourth of the total
number of terms. These are followed by AE and AC, each of which con
stitute slightly more than 15 percent. The percentages for RL, QL and CE
are considerably lower. The low ratio of these three categories is under
standable: (i) CE is unstable, as CE concepts cut across ontologically more
stable classes of ME, RE and AE and (ii) QL and RL are recognised as
independent only at the second stage of abstraction.
Comparing these figures with the tendencies of morphemes observed
above, a few points can be confirmed. Firstly, QL and RL, which accommo
date relatively high ratios of different as well as running morphemes, only
Table 4.14. Quantitative data of terms for broad categories.

Category ME RE AE CE QL RL AC
Number of Terms 325 327 208 50 57 75 186
(Percentage) (26.47) (26.63) (16.94) (4.07) (4.64) (6.11) (15.15)
constitute a small proportion of terms. This confirms the role of QL and

RL in term formation, i.e. they are mainly used for differentiating related
or similar concepts and do not constitute independent concepts in them
selves. AC, although its share of terms is relatively high, also has a similar
tendency if we compare the ratio of AC morphemes to AC terms. ME,
RE and AE, on the other hand, show the reverse tendency, i.e. they have
higher ratios in terms than in morphemes. This indicates the centrality of
these concepts in documentation, as compared with QL, RL or AC. The
proportion of CE morphemes and terms is very small, with little variation.
These broad categories, though their overall tendencies could be recog
nised, allow considerable internal diversity; thus, observations at the sub-
categorisation level will also be useful. Table 4.15 presents the number
of terms at the subcategorisation level. Because many subcategories only
accommodate a small number of terms at the finest level of categorisation,
the middle-level categories were chosen, based on the size of the categories,
internal uniformity and subject specificity.
The four largest groups are ME211, AE2, RE311 and AC21. ME211
and RE311 are material and representational types of documents, respec
tively; they are subject-specific and constitute the core of the field of doc
umentation. AE2 contains subject headings required for classification, and
AC21 covers the main operations of documentation.
From a different point of view, RL2 and QL2 are notable as they ac
commodate only a few (RL2) or no terms (QL2: not listed in Table 4.18),
while RL1 (RL12 and RL13) and QL1 (QL12 and QL13) accommodate a
fair number of terms. This further specifies the above observation about
QL and RL. While there are a fair number of labels of qualities or relations
which constitute independent concepts (and thus terms) in documentation,
values of qualities or relations are used only as parts of complex terms and
rarely constitute independent concepts by themselves.
An observation of conceptual categories according to the number of
constituent elements of terms reveals additional characteristics of term for-
Table 4.15. Quantitative data of terms for subcategories.

Category Number (%) Category Number (%)
ME11 26 (2.12) AE1 60 (4.89)
ME12 38 (3.09) AE2 134 (10.91)
ME211 139 (11.32) AE3 14 (1.14)
ME212 54 (4.40) CE1 42 (3.42)
ME22 59 (4.80) CE2 8 (0.65)
ME23 0 (0.00) QL12 36 (2.93)
ME24 4 (0.33) QL13 21 (1.71)
ME25 5 (0.41) RL12 20 (1.63)
RE1 51 (4.15) RL13 29 (2.36)
RE2 45 (3.66) RL2 26 (2.12)
RE311 113 (9.20) AC1 20 (1.63)
RE312 60 (4.89) AC21 97 (7.90)
RE32 18 (1.47) AC22 22 (1.79)
RE4 40 (3.26) AC23 21 (1.71)
AC24 26 (2.12)
Table 4.16. Number of terms by length for each conceptual category.

Category Number of morphemes per term Total
One Two Three Four+
ME 58 (17.85%) 154(47.38%) 93 (28.62%) 20(6.15%) 325 (100%)
RE 78(23.85%) 184(56.27%) 55 (16.82%) 10(3.06%) 327 (100%)
AE 27 (12.98%) 74 (35.58%) 83 (39.90%) 24(11.54%) 208 (100%)
CE 16 (32.00%) 28 (56.00%) 5 (10.00%) 1 (2.00%) 50(100%)
QL 11 (19.30%) 32(56.14%) 6(10.53%) 8 (14.04%) 57 (100%)
RL 14(18.67%) 51 (68.00%) 9(12.00%) 1 (1.33%) 75 (100%)
AC 42 (22.58%) 98 (52.69%) 32 (17.20%) 14 (7.53%) 186 (100%)
mation vis-à-vis conceptual categories. Table 4.16 shows the number and
ratio of simple terms, terms with two morphemes, three morphemes and
four or more morphemes for each broad conceptual category. Some cate
gories, i.e. abstract entity (AE) and, to a lesser extent, material entity (ME),
tend to accommodate longer terms, while some, such as classificatory entity
(CE) or relation (RL), accommodate a higher ratio of shorter terms.
4.3.3 Allocation of embedded combinations (complex morphemes)

The allocation of conceptual categories cannot be complete without allo
cating categories to embedded or secondary combinations which appear in
complex terms with more than three constituent elements. Formally, there
are 458 embedded combinations in the corpus, i.e. 283 in three-morpheme
terms, in four-morpheme terms, in five-
morpheme terms, and in six-morpheme terms.
As the present study concerns the term formation patterns and not the
combination patterns of items, embedded combinations are, in principle,
regarded as single units, just like simple morphemes. However, not all
the embedded combinations can be regarded as such. Firstly, as will be
shown in Chapter 5, there are non-conceptual combinations, which will be
ignored. Secondly, some embedded combinations constitute a part of term-
forming conceptual specification patterns (see 3.3.2.2 and the next chapter).
Excluding these, there are in total 281 tokens of embedded combinations
that should be treated as complex morphemes. They comprise 201 types.
Table 4.17 lists the number of these embedded combinations by concep
tual category. The categories with an asterisk show that the combinations
consist of units which represent different subcategories of equal status. The
table also shows the number of types that occur as terms. In the descriptions
of term formation patterns in Chapter 6, these embedded combinations are
regarded the same as simple morphemes.
Table 4.17. Quantitative data of embedded combinations (complex morphemes).

Category Type (ratio) Token (ratio) Term
Material Entity 30 (14.93) 81 (28.83) 15 (50.00)
ME1* 1 (0.50) 1 (0.36) 0(0)
ME11 3 (1.49) 14 (4.98) 2 (66.67)
ME12 6 (2.99) 33(11.74) 4 (66.67)
ME21 15 (7.46) 24 (8.54) 7 (46.67)
ME22 4(1.99) 8 (2.85) 2 (50.00)
ME25 1 (0.50) 1 (0.36) 1 (100)
Representational Entity 32(15.92) 39(13.88) 17(53.13)
RE12 7 (3.48) 7 (2.49) 1 (14.29)
RE22 5 (2.49) 6(2.14) 1 (20.00)
RE31 18 (8.96) 24 (8.54) 14 (77.78)
RE32 1 (0.50) 1 (0.36) 0(0)
RE41 1 (0.50) 1 (0.36) 1 (100)
Abstract Entity 21 (10.45) 24 (8.54) 10 (47.62)
AE11 6 (2.99) 6(2.14) 2(33.33)
AE12 3 (1.49) 4(1.42) 1 (33.33)
AE21 7 (3.48) 7 (2.49) 4(57.14)
AE22 4(1.99) 6(2.14) 3 (75.00)
AE33 1 (0.50) 1 (0.36) 0(0)
Classificatory Entity 5 (2.49) 7 (2.49) 2 (40.00)
CE12 4(1.99) 6(2.14) 2 (50.00)
CE21 1 (0.50) 1 (0.36) 0(0)
Quality 36(17.91) 40 (14.23) 1 (2.78)
QL1 8 (3.98) 8 (2.85) 1 (12.50)
QL2 28 (13.93) 32(11.39) 0(0)
Relation 26 (12.94) 31 (11.03) 4(15.38)
RL1 6 (2.99) 6(2.14) 3 (50.00)
RL2 20 (9.95) 25 (8.90) 1 (5.00)
Activity 51 (25.37) 59(21.00) 5 (9.80)
AC1 2 (0.10) 3(1.07) 0(0)
AC2* 2 (0.10) 3(1.07) 0(0)
AC21 26 (12.94) 30(10.68) 3(11.54)
AC22 6 (2.99) 6(2.14) 1 (16.67)
AC23 6 (2.99) 6 (2.14) 0(0)
AC24 9 (4.48) 11 (3.91) 1(11.11)
Chapter 5
Intra-Term Relations and

Conceptual Specification Patterns
As the second and last step of preparation for the observation of the concep
tual patterns of term formation, this chapter introduces intra-term relations
and conceptual specification patterns. As in the previous chapter, some
related studies are briefly reviewed first, and then the binary intra-term re
lations and conceptual specification patterns arrived at for the corpus of
documentation terms are explained.
5.1 Intra-term (compound) relations in existing studies
There have been many studies of Japanese complex nouns. While many
analyse and describe complex words at the grammatical level (e.g. Nomura
1973; Saito 1981; Nagashima 1980; Okutsu 1975; Kageyama 1982; 1989;
Tamamura 1985), only semantically- or conceptually-oriented analyses and
descriptions are reviewed. Note that the studies reviewed are not limited to
the study of terms.
Saiga (1957), in his examination of the grammatico-semantic patterns
of complex words based on data collected from Japanese magazines, intro
duced the following relations:
- Coordinate relation, in which two constituent elements are coordinated
with the same status. This is further divided into synonymous coordina
tion and antonymous coordination.
- Subject-predicate relations, in which the first constituent functions as the
subject of the second element.
- Complementary relation, in which the first element functions as a com

plement of the second element.
- Modification relation, in which the first element modifies the meaning of
the second element. This relation is further divided into belonging, dif
ferentiation, motion or action, attribute, material and degree, according
to the types of modification.
- Auxiliary relation, where the second element changes the formal or syn
tactic nature of the first element.
Yumoto (1977, 1979), being critical of the grammatical approach, pro
poses a more semantically-oriented analysis and introduces semantic rela
tions based on noun-noun wago (original Japanese) complex words. He
first classifies complex words into three groups: (i) characterisation rela
tion, where the first element indicates the characteristic of the object re
ferred to by the second element; (ii) object-condition relation, where the
second element indicates the condition or the situation of being a particular
condition and the first element refers to the object which is in that condi
tion; and (iii) extraction of part/aspect relation, where the second element
indicates a part or an aspect of an object which is referred to by the first
element. He further divides the first relation, which is predominant, into
the following types:
- Characterisation by attachments: The first element indicates the type of
attachment of the second element.
- Characterisation by shape or colour: The first element indicates the
shape or colour of the object referred to by the second element.
- Characterisation by material: The first element indicates the material of
the object referred to by the second element.
- Characterisation by property or function: The first element indicates a
property or function of the object referred to by the second element.
- Characterisation by situation: This group is further divided into five
groups: (a) content/use relation, in which the second element indicates
a container or a building and the first element indicates the content or
use; (b) use relation, in which the second element indicates an artefact
and the first element indicates an object which has a direct relation to
the use of the second element; (c) positional relation, in which the first
element indicates the place of an object referred to by the second ele
ment; (d) temporal relation, in which the first element indicates the time
when an object indicated by the second element exists or occurs; and

(e) causal relation, in which the second element indicates a product or a
phenomenon and the first element indicates its cause.
- Characterisation by possession: The first element indicates an object
which possesses an object referred to by the second element.
Ishii (1986) analyses the role of constituent elements in technical terms
taken from 23 fields (Fujiwara & Fujiwara 1987) which include a wago
transitive verbal morpheme. He defines the following 14 roles of mor
phemes with respect to transitive verbal morphemes: object which is made
as a result of the action; action which becomes an object of the action; ob
ject other than the preceding two; instrument; agent; goal or destination;
source or starting point; place or point where an object passes; location;
result; manner; direction; time; and purpose. Though he treats technical
terms, he is more concerned with general intra-term relations than domain-
specific ones.
Much research has also been undertaken on English complex words
(some claim the results relevant to other languages). Jespersen (1942),
Hatcher (1960), Marchand (1969), Adams (1973), Downing (1977) and
Warren (1978) are devoted to the description and observation of compound
ing patterns and formation rules. Lees (1960) and Levi (1978) try to formu
late the rules of compounding within the framework of generative grammar.
Sager, Dungworth & McDonald (1980) and Pugh (1984) treat compound
ing in special languages. Rhyne (1976), Finin (1980), McDonald (1982)
and Leonard (1984) take a computational approach, trying to produce or
interpret compounds by computer. Here we briefly summarise the semantic
or conceptual studies.
Downing (1977) introduces a basic set of relations between conceptual
elements and the basic categories of head nouns. She suggests the follow
ing as an inventory of the most common relations between constituent el
ements: whole-part, half-half, part-whole, composition, comparison, time,
place, source, product, user, purpose, occupation.
Sager, Dungworth & McDonald (1980) gives a classification of re
lations between constituent elements of noun compounds of special lan
guages. The authors first group compounds into three, according to the
nucleus, i.e. those which designate (i) objects, (ii) properties, and (iii) pro
cesses and operations.
For the compounds which designate objects, the following nine rela
tions are defined: (a) comparison, i.e. the determinant compares the nucleus
to another object; (b) material, i.e. the determinant specifies the material of
which the nucleus is made; (c) inherent property, i.e. the determinant ex
presses an inherent property of the new concept which is not inherent in
the nucleus; (d) use, i.e. the determinant specifies the use to which the nu
cleus is regularly put; (e) origin/product, i.e. the determinant expresses the
product which is regularly associated with the nucleus, and the determinant
indicates the origin of the nucleus by specifying the primary product; (f)
instrument, i.e. the nucleus expresses the instrument which operates on the
determinant; (g) mode of operation, i.e. the determinant specifies the mode
of operation of the nucleus; (h) whole/part, i.e. the determinant is the whole
of which the nucleus is a part; and (i) identity, i.e. the determinant asserts
identity with the nucleus.
They also describe briefly the compounds designating properties and
those designating processes or operations. Compounds designating proper
ties are formed by the determinant specifying the concept which the prop
erty term is related, and compounds designating processes or operations are
usually created by the determinant specifying the subject, the object or the
instrument of the corresponding action.
Pugh (1984), analysing formation patterns of terms of information pro
cessing, introduced 18 intra-term relations:
- Destination: The new concept is a class of (the concept represented by)
the nucleus which is used for or intended for (the concept represented
by) the determinant1. This relation can be classified into direct and indi
rect function. An example of direct destination is "checking program",
and of indirect destination is "computer instruction".
- Means/Mode of Operation: The new concept is a class of the nucleus
which functions or occurs by means of or in the mode of the determinant.
This also can be divided into direct, e.g. "dichotomising search", and
indirect, e.g. "asynchronous computer".
- Affected Object: The new concept is a class of the nucleus which acts
on the determinant. This can also be divided into direct, e.g. "data pro
cessing", and indirect, e.g. "data processor", relations.
- Partitive: The new concept is a part represented by the nucleus of a
1
Henceforth the phrase "the concept represented by" is omitted for succinctness.
Conceptual Specifìcation Patterns 95
whole represented by the determinant, e.g. "flowchart symbol".

- Atomic Composition: The new concept is a whole represented by the
nucleus which is entirely composed or made up of the determinant, e.g.
"character set".
- Partial Composition: The new concept is a whole represented by the
nucleus which is partly made of the determinant, e.g. "block diagram".
- Origin: The new concept is a class of the nucleus which has as its source
or cause the determinant, e.g. "rounding error".
- Functional Role: The new concept is a class of the nucleus which is used
as or has the role of the determinant, e.g. "master file".
- Representation: The new concept is a class of the nucleus which repre
sents the determinant, e.g. "data flowchart". This relation is said to be
specific to information processing.
- Product: The new concept is a class of the nucleus which produces the
determinant, e.g. "system generation".
- Eponyms: The new concept is a class of the nucleus which is called after
the concept denoted by the determinant, e.g. "Fibonacci search".
- Integration: The new concept is the sum of the nucleus and the determi
nant, e.g. "adder-extractor".
- Conjunction: The new concept is a subset of the nucleus which coincides
with a subset of the determinant, e.g. "city-state".
- Local Situation: The new concept is a class of the nucleus which is
situated at/in the determinant, but is not perceived as a part of it. This can
be divided into physical location, such as "zone punch", and situational
quality, such as "peripheral transfer".
- Temporal Situation: The new concept is a class of the nucleus which oc
curs at the time represented by the determinant, e.g. "post-processing".
- Similarity: The new concept is a class of the nucleus which has the
physical form or appearance of the determinant, e.g. "graphic character".
- Nature: The new concept is a class of the nucleus which has a name
worthy quality or state which is the determinant, e.g. "flexible disk",
"mass storage".
- Related Concept: The new concept is a class of the nucleus which is
related to or pertains to the determinant, e.g. "block length".
The last category is a dustbin category, covering the relations that do not
fall under other categories.
All the above studies define intra-term relations as binary. Among

them, Saiga (1957), Yumoto (1977, 1979), Downing (1977) and Pugh
(1984) define the relation as the role or position of the determinant vis-
à-vis the nucleus, while Sager, Dungworth & McDonald (1980) sometimes
shifts the viewpoint. Ishii (1986) takes a verbal-element-centric viewpoint.
As discussed in 3.2.2.2, the present study will describe the intra-term rela
tion as the role of determinants with respect to the nucleus.
Although it was mentioned in 3.3.2.2 that there is not much agreement
concerning the types of conceptual relations in general, we can observe a
fair number of common intra-word relations, though the strict ranges of the
relations may be different. Table 5.1 summarises the common relations ob
served among Saiga (1957), Yumoto (1977, 1979), Downing (1977), Sager,
Dungworth & McDonald (1980) and Pugh (1984), with Pugh's relations
used as the basis2.
Yumoto does not have the relations equivalent to affected object and
product because he limits his observation to pure noun-noun compounds,
excluding elements of verbal origin. Sager, Dungworth & McDonald
(1980), which does not explicitly introduce the relations equivalent to lo
cal and temporal situation in the special language framework, neverthe
less suggests the existence of temporal and local situation as "patterns well
established in general language". Taking these into account, it is only
Pugh's "means/mode of operation", "representation", "eponyms" and "re
lated concept" which are not shared by at least two others. Among these,
"means/mode of operation" is shared by many other verbal-morpheme-
oriented studies such as Tamamura (1985), Kageyama (1989) and Ishii
(1986), and "related concept" is exceptional as it was defined in Pugh
(1984) as a dustbin category. This leave only two specific to Pugh's scheme.
Representation seems to be a subject-specific relation, while eponyms seem
to be more common among technical terms than among general words.
In summary, the following can be pointed out: (i) there are many rela
tions which are recognised in many, if not all, studies reviewed here; they
are considered to be central to complex words irrespective of language or
domain, (ii) some relations may be subject-specific, (iii) the definition of
boundaries of similar relations may differ according to the study.
2
Ishii (1986) is excluded here as his verb-centric viewpoints are limited in range and are
very different from the other five.
Table 5.1. Correspondences of intra-term relations.

Pugh Downing Sager, et. al. Saiga Yumoto
a. integration/ half-half identity coordinative juxtaposition
conjunction
b. destination purpose (user, use (motion or content/use
occupation) action) use
c. functional — — — (property or
role function)
d. affected — object/ complementary —
object instrument
e. product product origin/ complementary —
product
f. partitive whole-part whole/part belonging possession
g· atomic/partial composition material material material

composition
h. means/mode — mode of — —
of operation operation
i. similarity comparison comparison — shape/colour,
object-
condition
j. nature — inherent attribute attachment,

property or
property function
k. representation — — — —
1. local situation place — — positional
relation
m. temporal time — — temporal
situation relation
n. origin source origin/ — causal
product relation
0. eponyms — — — —
P· related — — (auxiliary) —
concept
5.2 Intra-term relations for documentation terms
5.2.1 Types of intra-term relations

The intra-term relations for documentation terms were established by the
following procedure: (i) Based on the observation of the relations above
(mainly based on Pugh) and the samples from the corpus, a working set
of intra-term relations was established; (ii) they were applied to a large
number of samples, then revised and reapplied, until the set became stable.
The definition of each relation is introduced below, together with examples
and the clarification of the range. In the definitions, we use "N" for the
nucleus and "D" for the determinant, for succinctness.
Firstly and before presenting the intra-term conceptual relations, it is
necessary to differentiate a linguistic relation, which does not in itself be
long to any conceptual relation but is used to connect concept-representing
items explicitly. This is the same as Saiga's (1957) auxiliary relation and
can be exemplified by the relation between (sciences) and (of) in
(science of sciences)3. For convenience, we define this relation
as connection introduction:
Connection Introduction [CIN] — The combination in which the second
morpheme is attached to the first morpheme, adding a formal character
istic to the first element and thus establishing a connection between the
unit of these morphemes and the succeeding morpheme.
This relation should properly be dealt with at the grammatical level, which
is not addressed in this book. Henceforth it is excluded from the major
observation4.
3
From this section onwards, the discussion is related to the construction of Japanese
terms taken from the data described in the Introduction. Accordingly, primary examples are
taken from the Japanese data. When the construction of a corresponding English term does
not reflect the construction of the original Japanese term, we give morpheme-to-morpheme
translation in English, followed by proper equivalent terms; otherwise, English equivalent
terms are given.
4
In the previous chapter, (of) was classified as representing a relation concept as we
did not want to make any exceptional cases in the analysis. Given the fact that the relation
expressed by this morpheme is qualitatively different from other relation concepts, these
morphemes should be differentiated from the other relational morphemes in the observation
of the conceptual patterns of term formation. This will be done simply by excluding the
relation CIN from the major observation.
Other than the formal relation, the following 23 intra-term relations

were established:
Juxtaposition [JXT] — The concept of the complex concept (henceforth
X) consists of two constituent morphemes having equal status. Example:
(reader printer).
Though some researchers apply subdivisions, it is not subdivided in our
scheme. This is because it was not perceived useful to logging the general
ity of term formation patterns.
Role [ROL] — X is a class of the concept represented by the nucleus
(henceforth NUC) which functions as or has the role of the concept
represented by the determinant (henceforth DET). Example:
(sample copy; advance copy).
Function [FUN] — X is a class of the concept represented by NUC whose
function or activity is the concept represented by DET. Example:
(editing person; editor).
Use [USE] — X is a class of the concept represented by NUC the use/the
aim or the purpose of which is represented by DET Example:
(expression criterion; notational base).
Destination [DES] — X is a class of the concept represented by NUC
which is intended for the concept represented by DET. Example:
(computer program).
These four constitute a broad group of functional relations.
Affected Object [AFO] — X is a class of the concept represented by NUC
which acts on or treats the concept represented by DET. Example:
(file maintenance).
Product [PRO] — X is a class of the concept represented by NUC whose
product is the concept represented by DET. Example: (index
making; indexing).
As in Pugh's (1984) affected object and product, these relations can be
divided into direct and indirect relations, but this distinction is not adopted
here. These two types can also be regarded as part of a broad group of
"functional relations".
The following three relations are broadly categorised as "partitive" re
lations. "Information Content and Representation" is rather specific to the
field of documentation; it was established to account for the compositional

relation with respect to meaning or information level.
Partitive [PAR] — X is a class of concept represented by NUC which
constitutes, is the part of or belongs to the whole represented by DET.
Example: (vocabulary primitive).
Constituents [CON] — X is a class of the concept represented by NUC
which is constituted of or whose constituent element is the concept rep
resented by DET. Example: (card catalogue).
Information Content and Representation [ICR] — X is a class of the
concept represented by NUC which represents or includes content or
information specified by the concept represented by DET. Example:
(term list; glossary).
Then the following two relations, i.e. means and manner, were recog
nised. They are similar in that both are used for indicating the internal
characteristics of activities. These are also close to function (FUN), af
fected object (AFO) and production (PRO), as all of these are related to the
complementary elements of activity concepts.
Means [MEA] — X is a class of the concept represented by NUC whose
means of operation is represented by DET. Example: (mag
netic tape).
Manner/Mode [MAN] — X is a class of the concept represented by NUC
whose manner or mode of operation is represented by DET. Example:
(binary search).
The following four relations express internal or non-environmental at
tributes. The relation 'Attributed Concept" is different from other relations
in that the nucleus, although representing the conceptual category to which
the whole term belongs, is nevertheless considered to extract a specific as
pect of concepts represented by the determinants. As such, the centrality
of the concepts represented by the nucleus and the determinant is reversed
in this relation. The relation "Nature" is established to accommodate gen
eral internal characteristics which cannot be clearly defined as other, more
specific relations. Thus it has a dustbin character, as in Pugh's "Nature".
Formal Attributes [FOA] — X is a class of the concept represented by
NUC whose formal or configurational nature (form, structure, organisa-
tion, shape, size, colour, etc.) is indicated by the concept represented by

DET. Example: (block diagram).
Quantity [QUA] — X is a class of the concept represented by NUC whose
quantity is indicated by the concept represented by DET. Example:
(two language) in (two language dictionary; bilingual
dictionary).
Attributed Concept [] — X is a class of the concept represented by
NUC which indicates the condition, aspect of measurement, nature, rel
ative position, etc., whose specific attribution or substance is indicated
by the concept represented by DET. Example: (noise ratio).
Nature [NAT] — X is a class of the concept represented by NUC which
has a name-worthy quality or state which is the concept represented by
DET. Example: (natural language).
The above four, and to some extent MEA and MAN, together can be re
garded as a broad group of relations that represent "internal or intrinsic
attribute relations", though this grouping is not very cohesive.
The following six relations were recognised as necessary to accommo
date the broad group of "external, environmental or relational relations":
Location [LOC] — X is a class of the concept represented by NUC which
is situated at/in the concept represented by DET. Example: (edge
hole; notch).
Time [TIM] — X is a class of the concept represented by NUC which
occurs at the time represented by DET. Example: (pre֊
coordination).
Origin [ORI] — X is a class of the concept represented by NUC whose
origin, source or cause is specified by the concept represented by DET.
Example: (derived word).
Scope [SCO] — X is a class of the concept represented by NUC whose
scope of application, use or coverage is indicated by the concept repre
sented by DET. Example: (universal bibliography).
Status [STA] — X is a class of the concept represented by NUC whose
status is indicated by the concept represented by DET. Example:
(basic entry).
Subject Specification [SSP] — X is a class of the concept represented by
NUC, whose subject, performer or agent is indicated by the concept
102 Dynamics of Termin ology
represented by DET. Example: (use=person demand; user

needs).
Lastly, the following relation which does not belong to any of these
was recognised:
Differentiation Tag [DTA] — X is a class of the concept represented by
NUC which is distinguished from other concepts with the same nuclei
by the existence of DET. Example: (Venn diagram).
This relation covers Pugh's eponyms. The positive meaning was not recog
nised in this sort of relation for two reasons. Firstly, because it is convenient
to accommodate meaningless morphemes. Secondly, the motivation of us
ing eponyms seems to be external to the conceptual system represented by
terminological structure.
The full list of intra-term relations, with abbreviations, is given in Ap
pendix B.l for reference. The relations are listed in the alphabetical order
of their abbreviations.
5.2.2 Some characteristics of intra-term relations

The 23 intra-term relations were assigned to the relations between con
stituent elements in complex terms. In the case of terms with three or more
morphemes, the structures of the terms were recognised and the relation
was assigned to the combinations at each level. Unlike the case with con
ceptual categories, multiple assignment was not allowed. Assignment was
relatively straightforward, and unique assignment did not cause problems.
The nature of intra-term relations is examined briefly here on the basis of
actual assignment, in order to identify the supporting knowledge needed for
establishing conceptual specification patterns based on intra-term relations.
5.2.2.1 Qualitative examinations

5.2.2.1.1 Uniformity of relations
Among the 23 relations, USE (use), AFO (affected object), PRO (product),
MEA (means), MAN (manner), ICR (information content and represen
tation), ORI (origin), FOA (formal attributes) and NAT (nature) were ob
served to have some internal divisions, while the others are fairly uniform.
The type of diversity in USE, AFO, PRO, MEA, MAN, ICR and ORI,
which are all related to case-like roles centred on activity-type concepts,
is related to the direct and indirect relations explained in Pugh (1984). In
all these cases, they can be clarified from the point of view of the roles
of the nuclei. For instance, both (reference book) and
(call number) are recognised to have intra-term relations of USE, though
the role of the nucleus vis-à-vis the determinant in the former case is that
of direct object, while the role of the nuclei in the latter is that of tools or
instruments. Other relations have the same sort of direct and indirect varia
tions. Collapsing the distinction between the direct and indirect relation is
justified because the intra-term relations are defined in terms of the role of
the determinant vis-à-vis the nucleus, which in turn is anchored to the over
all conceptual system; the term formation patterns are observed within the
conceptual system to which the different categories of nuclei are anchored.
The internal diversities of FOA and NAT come from the generic nature
assigned to them. FOA can, for instance, be further divided into size, shape,
composition, etc., but we did not adopt it here for the practical reason that
it was not necessary to go down to this level for clarifying the characteristic
patterns of term formation. NAT can include internal diversities because of
its dustbin nature.
5.2.2.1.2 Affnity of relations
Though at the level of definition the ranges of the relations are reasonably
clear, in the process of assignment it was recognised that some relations are
rather close to each other. The three pairs, i.e. FUN and USE, PRO and
PAR, and PAR and LOC, deserve some comments.
Firstly, the following examples show the affinity and the difference be
tween FUN and USE:
FUN : (retrieval system)
USE : (copying technique; reprography)
In the second case, the nucleus (technique) is used for (copying).
On the other hand, (retrieval system) can be interpreted in
two ways: (a) a system which retrieves something and (b) a system which
is used by somebody for retrieving something. In the first interpretation, the
nucleus is the central focus from where the role of the determinant is inter
preted, while in the second, a third party is implied. Because the naming
process in principle starts from the core concept (nucleus), the first inter-
pretation is theoretically better than the second one, a criterion which was
consistently maintained in the assignment.
The following example shows the affinity of PRO and PAR:
PAR : (entry word)
Taken independently, the determinant, (entry) is a product made up
of the nucleus (word), thus it is a product. On the other hand, the prod
uct and its constituent elements naturally make up a part/whole relation.
In such cases, the relation PAR is preferred because the dynamic aspect
implied by PRO is weak.
In a few terms in the corpus, PAR and LOC became potential can
didates, as in (library person; librarian). The relation PAR was
preferred in this and similar terms, as in the field of documentation "library"
is understood to be an organisation rather than a simple place.
5.2.2.2 Quantitative tendencies

As there are 621 two-morpheme, 283 three-morpheme, 61 four-morpheme,
15 five-morpheme and 2 six-morpheme terms in the corpus, there are 1,440
combinations at all the levels. Some types of combinations occur more than
once, thus different and running combinations can be distinguished. Only
the running occurrences are treated here because the different combina
tions are not very meaningful with respect to the intra-term relations. In the
present study, our interest is in the formation of terms; thus we made a dis
tinction between 982 term-forming combinations or top-level combinations
(TFC) and 458 embedded combinations (EC). The quantitative tendencies
are given in Table 5.2. Some relations such as USE and FUN are predom
inant in term-forming combinations and minor in embedded combinations,
while relations such as MAN or MEA show reverse tendencies.
Some of the differences of occurrence between term-forming com
binations and embedded combinations reflect the fact that some relations
are syntagmatically related. Table 5.3 lists the frequent patterns of combi
nations of intra-term relations in terms with three and four morphemes.
The structures "(MAN)USE", "(AFO)FUN" and "(MEA)USE" are no
table. They are in accordance with and responsible for the above observa
tions, i.e. USE and FUN are inclined to term-forming combinations, while
AFO, MEA and MAN are prevalent in embedded combinations.
Table 5.2. Quantitative nature of intra-term relations.

Relation Number of morphemes per term Total
Two Three Four+
TFC TFC EC ALL TFC EC ALL TFC EC ALL
CIN 0 0 17 17 0 32 32 0 49 49
INT 1 0 0 0 0 0 0 1 0 1
SUM 0 0 3 3 0 3 3 0 6 6
SEM 0 0 4 4 0 0 0 0 4 4
ROL 7 1 0 1 0 3 3 8 3 11
FUN 48 36 12 48 6 8 14 90 20 110
USE 37 56 4 60 17 4 21 110 8 118
DES 15 3 3 6 5 2 7 23 5 28
AFO 73 26 57 83 3 26 29 102 83 185
PRO 15 19 6 25 4 4 8 38 10 48
PAR 9 8 1 9 4 1 5 21 2 23
CON 25 8 6 14 1 2 3 34 8 42
ICR 51 14 12 26 4 2 6 69 14 83
MEA 16 4 29 33 4 12 16 24 41 65
MAN 58 17 45 62 6 24 30 81 69 150
FOA 63 18 18 36 1 7 8 82 25 107
QUA 0 0 3 3 0 0 0 0 3 3
 51 14 20 34 12 25 37 77 45 122
NAT 39 27 12 39 4 4 8 70 16 86
LOC 5 6 2 8 0 1 1 11 3 14
TIM 4 0 5 5 0 1 1 4 6 10
ORI 9 9 1 10 1 5 6 19 6 25
SCO 41 10 7 17 1 6 7 52 13 65
STA 47 4 13 17 2 2 4 53 15 68
SSP 0 3 1 4 3 1 4 6 2 8
DTA 7 0 2 2 0 0 0 7 2 9
TOTAL 621 283 283 566 78 175 253 982 458 1440
Table 5.3. Frequent combinations of intra-term relations.

Three-Morpheme Terms Four-Morpheme Terms
Pattern Frequency Pattern Frequency
(A-MAN-B)-USE-C 25 ((A-ACO-B)-MAN-C)-USE-D 6
(A-AFO-B)-FUN-C 20 ((A-FOA-B)-MAN-C)-USE-D 4
(A-MEA-B)-USE-C 18 (A-CIN-B)-MAN-(C-AFO-D) 4
(A-MAN-B)-PRO-C 9 ((A-FUN-B)-CIN-C)-ACO-D 3
(A-CIN-B)-NAT-C 7 (A-SCO-(B-AFO-Q)-PAR-D 3
(A-AFO-B)-AFO-C 7
5.3 Conceptual specification patterns
5.3.1 Types of conceptual specification patterns
The intra-term relations established above for the terminology of documen

tation were reorganised and reinterpreted as conceptual specification pat
terns. As explained in 3.3.2.2, the conceptual specification pattern is a the
oretical device that can accommodate η-ary combinations of morphemes in
order to describe possible formation patterns that consist of more than two
conceptual elements (thus morphemes).
A few procedural problems were recognised in the establishment of
conceptual specification patterns. Firstly, there is no structural or for
mal feature in the linguistic representations by means of which the em
bedded combinations can be distinguished as either parts of conceptual
specification patterns or as the combination of different patterns. Sec
ondly, conceptual specification patterns may have different linguistic re
alisations, so a qualitative rationale for establishing specification patterns
is required. Thus, though the quantitative tendencies of the co-occurrences
of intra-term relations give background support for the recognition of the
conceptual specification patterns, the analysis was basically carried out by
means of qualitative interpretation of the intra-term relations and their co
occurrences. As a result, the following conceptual specification patterns
were established, many of which consist of a single intra-term relation with
one-to-one correspondence. The conceptual specification patterns are cate
gorised under six broad groups.
1) Specification from the equality viewpoint
This has one-to-one correspondence to JXT (juxtaposition). We refer to
this specification pattern as EJXT, adding "E" for equality.
2) Specification from functional or allied viewpoints
The following specification patterns are recognised and can be broadly cat
egorised as functional. "F", for the functional viewpoint, is used in abbre
viations for the specification patterns of this group.
Specification of role: This corresponds to ROL (role). We refer to this
specification pattern as "FROL".
Specification of complementary elements of functions: The intra-term
relations AFO (affected object), PRO (product), MEA (means) and
MAN (manner) in the terms that represent activity concepts were to
gether recognised as constituting a single specification pattern, in which
a function concept is determined by such complementary elements as
object, product, means and manner. Other than the intuition that they
constitute a natural conceptual link of functions - objects/products -
means/manner, this is supported by the observation that (i) the com
bination of MEA or MAN always precedes AFO and PRO when they
co-occur in a complex term; (ii) they co-occur frequently; and (iii) even
when MEA or MAN is used without AFO or PRO in a term, the ex
istence of the objects or products are implied. For instance, we regard
(data processing: AFO), (index making: PRO),
(questionnaire survey: MEA), (multi
ple access: MAN) and (automatic data processing:
[MAN]AFO) as belonging to this same specification pattern. This rela
tion will be referred to as FCOM.
Specification of functional link: The intra-term relation FUN (function),
combined with the specification FCOM constitutes an integrated con
ceptual specification pattern, establishing a link of agents - functions -
objects/products - means/manner. This pattern is called here functional
link (FFUN). As the difference between FCOM and FFUN is the role of
the nucleus, FCOM can be regarded as the sub-pattern of FFUN, if we
follow the same argument put forward in the integration of direct and
indirect relations. These two were nevertheless distinguished, mainly
for convenience. Some examples of this pattern are:
(output unit: FUN), (information system: AFO),
(analogue computer: MAN), (in
formation transmission system: [AFO]FUN), (magnetic
storage device: [MEA]FUN) and (electric
data processing system: [MEA[AFO]]FUN). Many of the frequent com
binations of intra-term relations in Table 5.3 fall under this specification
pattern.
Specification of use: Another specification pattern identified as a func
tional or related type is the specification of use (FUSE), which is based
on USE. In accordance with the distinction of direct and indirect USE
relation, FUSE can also be subdivided into two: (a) the terms whose
nuclei have the role of the objects of the determinants; and (b) the terms
whose nuclei have the role of the tools or instruments of the determi-
nants. Though these two are clearly different from the viewpoint of
the role of the nuclei with respect to the determinants, the basic aspect
of the nuclei specified by these two are the same, i.e. both specify the
use of the core concepts represented by the nuclei. On the other hand,
the actual specifications differ between these two variations. When
the nuclei represent the tools or instruments of the determinants, the
intra-term relation USE can be extended with the specification pattern
FCOM, i.e. tools/instruments/methods - activities - objects/products -
means/manner, while such a link is not established when the nuclei have
the role of the objects. Some examples are: (deposit
collection: USE), (classifying method: USE) and 8
(octave expression method: [MAN]USE).
Specification of destination: The intra-term relation DES was recognised
as constituting a specification pattern by itself.
Among these functional specifications, FUSE and FDES are consid
ered to represent passive functional relations with respect to the status of
the nucleus.
3) Specification from part/whole viewpoints
The following three, each having one-to-one correspondence to an intra-
term relation, were recognised. These patterns include the prefix "P" for
part.
Specification of whole or affiliations: This specification pattern, referred
to as PPAR, corresponds to PAR (partitive).
Specification of constituent elements: This corresponds to CON (con
stituents). This pattern is called PCON.
Specification of information content and representation: This pattern
(PICR) corresponds to ICR (information content & representation).
4) Specification from the viewpoint of internal attributes
The following specification patterns were recognised. The patterns are re
ferred to by adding "I", standing for internal attributes, at the head of the
corresponding intra-term relations.
Specification of formal attributes: The intra-term relation FOA (formal
attributes) constitutes this pattern (IFOA).
Figure 5.1. Conceptual link in the specification of origin.
Specification of static manner: So far, the intra-term relations MAN

(manner) and MEA (means) were used for representing dynamic man
ner or modes. It was recognised that MAN in some terms indicates
static manner and constitutes a subgroup clearly distinguished from
other terms with MAN. To accommodate that, the specification of static
manner (IMAN) was defined. An example is (equivalence
relation).
Specification of nature: The pattern INAT corresponds to NAT (nature).
Specification of attributed concepts: The pattern IACO corresponds to
 (attributed concept).
Specification of quantity: The pattern IQUA corresponds to QUA (quan
tity). In the present corpus, there are only embedded specifications.
5) Specification from relational/external viewpoints
The following six specification patterns were defined under this broad
group. The specification patterns of this group are referred to by four letter
abbreviations beginning with "R", for relational viewpoint.
Specification of status: The intra-term relation STA (status) constitutes
the specification pattern RSTA.
Specification of location: The intra-term relation LOC (location) consti
tutes the pattern RLOC.
Specification of time: The intra-term relation TIM (time) constitutes the
pattern RTIM.
Specification of origin: The intra-term relation ORI (origin), which has
some internal variations, constitutes a specification pattern (henceforth
RORI), sometimes combined by the relation MAN (manner). The link
of intra-term relations in the specification of origin is illustrated in Fig
ure 5.1. Three different sub-patterns are recognised in this specification
pattern: (a) specification of original activities and manner of activities,
Table 5.4. Relations between intra-term relations and conceptual specification patterns.
Specification Intra-term relation Specification Intra-term relation
EJXT JXT INAT NAT
FROL ROL IACO 
FCOM AFO/PRO, MEA/MAN IQUA QUA
FFUN FUN, AFO/PRO, MEA/MAN RSTA STA
FUSE USE, AFO/PRO, MEA/MAN RLOC LOC
FDES DES RTIM TIM
PPAR PAR RORI ORI, MAN
PCON CON RSCO SCO
PICR ICR RSUB SUB
IFOA FOA ODTA DTA
IMAN MAN
(b) specification of source and (c) specification of producer. Examples

are: (a) (publication item: ORI), (serial publication
thing: [MAN]ORI), (b) (microfilm print:
[FOA]ORI), and (c) (author's abstract: ORI). Unlike varia
tions of FFUN, for instance, the above three cannot be manifested in a
single term. As such, their difference seems larger. They were never
theless treated under a single pattern, RORI, because (i) the conceptual
elements concerned with the variations of RORI are all linked by activity
concepts, and (ii) terms belonging to (b) and (c) are so few that it is not
practical to treat them as independent patterns at the same level as FROL
(specification of role), FFUN (specification of functional link), etc.
Specification of scope: The intra-term relation SCO (scope) constitutes a
specification pattern, RSCO, by itself.
Specification of subject: The intra-term relation SSP (subject specifica
tion) constitutes the conceptual specification pattern RSUB.
6) Other
Specification of differentiation tag: Lastly, the specification pattern cor
responding to DTA (differentiation tag) was recognised. This will be
referred to as ODTA.
The full list of conceptual specification patterns with abbreviations is
given in Appendix B.2 for reference. The patterns are listed in the alpha
betical order of their abbreviations.
Table 5.5. Occurrences of conceptual specification patterns in the corpus.

Specification Primary Secondary
pattern specifications specifications Total
EJXT 1 (0.10) 10(3.56) 11 (0.87)
F group 452 (46.03) 110(39.15) 562 (44.50)
FROL 8(0.81) 3(1.07) 11 (0.87)
FCOM 112(11.41) 33 (11.74) 145(11.48)
FFUN 147 (14.97) 55 (19.57) 202 (15.99)
FUSE 162 (16.50) 14(4.98) 176 (13.94)
FDES 23 ( 2.34) 5(1.78) 28 ( 2.22)
 group 126 (12.83) 24 ( 8.54) 150(11.88)
PPAR 21 (2.14) 2(0.71) 23(1.82)
PCON 34 ( 3.46) 8(2.85) 42(3.33)
PICR 71 (7.23) 14(4.98) 85 ( 6.73)
I group 245 (24.95) 90 (32.03) 335 (26.52)
IFOA 83 ( 8.45) 25 ( 8.90) 108 ( 8.55)
IMAN 18(1.83) 1 (0.36) 19(1.50)
INAT 67 ( 6.82) 16(5.69) 83 ( 6.57)
IACO 77 ( 7.84) 45(16.01) 122(9.66)
IQUA 0( ) 3(1.07) 3 ( 0.24)
R group 151 (15.38) 45(16.01) 196 (15.52)
RSTA 53 ( 5.40) 15(5.34) 68(5.38)
RLOC 11 (1.12) 3(1.07) 14(1.11)
RTIM 4(0.41) 6(2.14) 10(0.79)
RORI 25 ( 2.55) 6(2.14) 31 (2.45)
RSCO 52(5.30) 13(4.63) 65(5.15)
RSUB 6(0.61) 2(0.71) 8 ( 0.63)
ODTA 7(0.71) 2(0.71) 9(0.71)
TOTAL 982 (100.00) 281 (100.00) 1,263 (100.00)
5.3.2 Some characteristics of conceptual specification patterns

Table 5.4 shows the correspondences of the 21 conceptual specification pat
terns defined above with the intra-term relations. Some intra-term relations
such as MAN (manner), AFO (affected object), PRO (product) or MEA
(means) are used for constructing different specification patterns.
Actual occurrences of conceptual specification patterns can be divided
into two types, i.e. those used for forming terms (primary conceptual spec
ifications) and those used for forming embedded combinations (secondary
conceptual specifications). The number of occurrences of primary specifi
cations is 982, the same as the number of complex terms. Among them,
Table 5.6. Conceptual specification patterns and conceptual categories.

Conceptual Category of Terms
Spec. Pat. ME RE AE CE QL RL AC Total
EJXT 1 0 0 0 0 0 0 1
F group 126 67 138 4 2 3 112 452
FROL 2 4 0 1 0 1 0 8
FCOM 0 0 0 0 0 0 112 112
FFUN 94 35 16 2 0 0 0 147
FUSE 26 15 116 1 2 2 0 162
FDES 4 13 6 0 0 0 0 23
 group 40 71 8 5 0 0 2 126
PPAR 9 10 2 0 0 0 0 21
PCON 3 22 4 5 0 0 0 34
PICR 28 39 2 0 0 0 2 71
I group 39 67 16 15 38 53 17 245
IFOA 28 45 1 9 0 0 0 83
IMAN 0 0 0 0 0 18 0 18
INAT 11 22 13 5 0 12 4 67
IACO 0 0 2 1 38 23 13 77
IQUA 0 0 0 0 0 0 0 0
R group 58 43 17 9 6 5 13 151
RSTA 9 22 8 8 4 0 2 53
RLOC 5 3 0 0 0 1 2 11
RTIM 2 0 0 0 0 2 0 4
RORI 14 8 1 0 0 2 0 25
RSCO 28 10 8 1 2 0 3 52
RSUB 0 0 0 0 0 0 6 6
ODTA 3 1 2 1 0 0 0 7
Simple terms 58 78 27 16 11 14 42 246
TOTAL 325 327 208 50 57 75 186 1228
117 occurrences of conceptual specification patterns consist of the com

binations of two intra-term relations, and three occurrences consist of the
combinations of three intra-term relations. Among the embedded or sec
ondary combinations, 49 are combined by the intra-term relation CIN (con
nection introduction). Excluding them, the number of occurrences of sec
ondary conceptual specifications is 281. We already assigned conceptual
categories to these 281 embedded combinations.
Table 5.5 lists the number of occurrences of the conceptual specifi
cation patterns in the term-forming combinations and secondary combina-
tions. Although some of the specifications are occurrences of the same

combinations, only the token occurrences are observed.
Lastly, Table 5.6 shows the number of terms that take a given specifica
tion pattern as primary specification and represent a given broad conceptual
category. A strong correlation between some conceptual specification pat
terns and broad conceptual categories can be observed, e.g. between RE
(representational entity) and PCON (specification of constituent elements);
between AE (abstract entity) and FUSE (specification of use); between QL
(quality) and IACO (specification of attributed concepts), etc. The full in
vestigation of the relations between conceptual specification patterns and
conceptual categories is the main aim of Part II. It will thus be carried out
in the next chapter, where conceptually-motivated term formation patterns
are explored for each conceptual category to which terms belong.
Chapter 6
Conceptual Patterns of the

Formation of Documentation Terms
Now that all the necessary information has been gathered, it is possible
to describe the formation patterns of terms in the field of documentation.
In this process, the appropriateness of the descriptive devices, i.e. concep
tual categories and specification patterns, will also be evaluated. In the
following, the major tendencies of term formation in important conceptual
categories are observed first. This part introduces term formation patterns
specific to documentation terms with the full presentation of the data. The
description comprises the first major part of the pursuit of the dynamics of
the terminology of documentation. Then, the methodological aspects are
examined, both with respect to the description of documentation terms and
modelling term formation in general.
6.1 Term formation patterns in the field of documentation
The characteristic patterns of term formation are summarised here accord

ing to the conceptual categories to which terms belong. The major points
of observation are as follows:
1. The general tendency of term formation patterns in the major conceptual
categories is indicated by such aspects as the relative ratio of simple to
complex terms and the existence or absence of conceptual specification
patterns. The distinction between central or dominant and peripheral
patterns is introduced.
2. The individual characteristics of the major term formation patterns are

observed from the point of view of the interactions between conceptual
categories and conceptual specification patterns.
Given the nature of the categories and the volume of data, it is not
always helpful to describe the patterns at the bottommost level of concep
tual categories. Neighbouring or related categories are examined together
wherever relevant, and some of the smaller categories are simply excluded
from consideration. Note also that only the top-level or term-forming com
binations will be observed, as we are interested first and foremost in term
formation patterns and not in the relative combinability of constituent ele
ments in complex terms. The types of embedded or secondary specification
patterns will be referred to individually whenever useful.
In this section, a few simplified expressions and conventions are intro
duced, as follows:
1. Regarding concept classes, we shall speak of, for instance, "ME terms"
to refer to ME (material entity) concepts realised as terms.
2. Regarding conceptual specification patterns, we shall speak of, for in
stance, "FFUN terms" to refer to the terms formed by the specification
pattern FFUN (specification of functional link).
3. To refer to nuclei/determinants which are themselves combinations of
morphemes, we shall use "complex nuclei/determinants".
4. We shall refer to constituent elements of the top-level term construction,
i.e. both morphemes and embedded combinations, as "items"; this is a
convenient and useful way for describing term formation patterns.
5. For each conceptual category observed, we shall present a table con
sisting of the term-forming conceptual specification patterns, intra-term
relations that constitute specification patterns, the conceptual categories
of constituent items, and complex terms, together with frequencies in
round brackets. In tables, the nuclei that are themselves terms are in
dicated with an "*". The complex nuclei are regarded as terms when
the complex nuclei are themselves terms or when the embedded nuclei
are terms. Simple terms are omitted from the tables. For full lists, see
Appendix C.
6. Embedded combinations in the terms in the tables as well as in the
text shall be indicated with an "=". In the tables, only morpheme-
to-morpheme English equivalents shall be listed, to show the flavour
of Japanese constructions and also to save space. Appendix  lists

Japanese and corresponding English terms. Embedded combinations
shall be indicated with an "=" in English as well. In text, we also add
equivalent English terms (if they differ from morpheme-to-morpheme
translation).
7. To facilitate the mapping between the main text and the tables, we
shall consistently use the abbreviations of conceptual categories, speci
fication patterns and intra-term relations introduced in Chapters 4 and
5. In the text, we shall add descriptions of conceptual categories,
specification patterns and intra-term relations whenever is felt useful
for facilitating understanding. These will be provided in forms such
as "ME22 (machines and implements)", "machines and implements
(ME22)", "FFUN (specification of functional link)", "specification of
functional link (FFUN)", "DTA (differentiation tag)", "differentiation
tag (DTA)", etc. We shall not stick to a uniform style regarding which
element should be bracketed.
6.1.1 Material entity (ME) terms

Animate - people - types (ME112), materials (ME23) and parts as
places/locations (ME252) are not represented by independent terms. In
addition, documents - physical form (ME21113), documents - parts
(ME2112), places/locations (ME24) and parts in general (ME251) are omit
ted as they are represented only by small numbers of terms and no gener
alised patterns can be extracted.
6.1.1.1 Animate -people - types (ME111)

There are 26 terms (3 simple, 19 two-item and 4 three-item terms) in this
category. Table 6.1 shows the conceptual patterns of complex terms in
this category. Formation patterns for this category are very regular, i.e.
terms are formed solely by applying the specification of function (FFUN)
or specification of whole (PPAR) to the nucleus representing ME111.
In FFUN, either functions (FUN), objects or products (AFO or PRO),
or both (AFO or PRO and FUN) are specified. The concepts of determi
nants are mostly subject-specific, such as document production (AC233)
for FFUN with FUN only, while the determinants that specify function
in FFUN with AFO or PRO plus FUN tend to belong to general activity
Table 6.1. Formation patterns of ME111 complex terms.
concepts. The determinants of AFO and PRO relations are mostly subject-
specific, such as AE212 (specific subject field) or RE311 (complex docu
ment entities). As for PPAR, all the determinants represent organisations.
In fact, all of them represent subject-specific organisations, i.e.
(book=house; library) and its variations.
In short, the documentation terms representing people are formed by
specifying their function — either directly by subject-specific activities or
indirectly by subject-specific objects or products (this also applies to 3 sim
ple terms, all of which incorporate functions, e.g. [programmer])
— or by specifying the organisations (more specifically libraries) to which
the people belong.
6.1.1.2 Animate - organisations (ME12)

There are 38 terms (1 simple, 35 two-item and 2 three-item terms) in this
category. Table 6.2 shows the conceptual patterns of complex terms in this
category. All the nuclei belong to ME12, and the dominant specification
pattern is FFUN (specification of functional link).
Among FFUN terms, two different tendencies are observed: (i) gen
eral organisation concepts specified by subject-specific activity or entity
concepts, such as — (information centre); and (ii) libraries spec
ified by general or specific concepts, e.g. (mobile book=house;
mobile library). On the other hand, all but two terms constructed by other
specification patterns take (book=house; library) as a nucleus.
We can clearly observe two different levels at which term formation
is active: (i) when forming subject-specific organisations by taking general
organisation concepts and specifying their function; and (ii) when forming
various subcategories of libraries by specifying scope, the whole or origin.
Figure 6.1 illustrates this. Taking into account the fact that the concept
"library" takes a notable role also in the formation of ME111 terms, this
category may better be treated as an independent subcategory of the or
ganisation concepts in the field of documentation so as to more explicitly
reflect the conceptual regularities of term formation.
Figure 6.1. Hierarchy offormation of ME12 terms in documentation.
6.1.1.3 Inanimate - information carriers - types of documents (ME2111)

As mentioned, ME21113 (physical form of documents: 5 terms) and
ME2112 (parts of documents: 3 terms) are not observed here. There are
27 terms in ME21111 and 104 terms in ME21112.
6.1.1.3.1 ME21111: Documents by mode of publication/formal aspect
Among the 27 ME21111 terms, 9 are simple terms, 15 are two-item terms
and 3 are three-item terms. The fact that the ratio of simple terms is rel
atively high may indicate that ME21111 is central to the field, and thus
relatively many concepts are represented in short linguistic forms. Table
6.3 shows the formation patterns of the 18 complex terms.
Specifications of status (RSTA: 5) and of origin (RORI: 5) are the two
major specification patterns. Unlike ME111 and ME12, here the categories
of the nuclei vary, i.e. they are made up of both ME21 (documents: 14), of
which most are ME21111, and CE1212 (non-independent complex classi-
ficatory entity: 4).
Both RSTA and RORI terms have relatively stable formation patterns.
RSTA terms consist of ME21111 nuclei specified by RL27 (relative status)
determinants, as in - (primary document). RORI terms and those
combining the CE1212 nucleus with publication activities as determinants
are closely correlated; general objects become documents by the specifi
cation of publication and are further specified by the mode of publication.
These two types of conceptual specification are considered to belong to

different levels. Other specification patterns are concerned with the formal
aspects or external situations and take ME21111 nuclei.
6.1.1.3.2 Documents by content/mode of publication (ME21112)
There are 21 simple, 76 two-item and 7 three-item terms in this category.
Table 6.4 shows the patterns of ME21112 terms. PICR (information con
tent and representation) and RSCO (scope) are the dominant conceptual
specification patterns, followed by IFOA (formal attributes), FFUN (func
tional link), INAT (nature) and FUSE (use). The nuclei are ME21111 (25)
and ME21112(58).
Upon closer inspection, PICR can be divided into two groups, one
starting from ME21111 nuclei and the other from ME21112 nuclei. In the
former case, ME21111 is shifted to ME21112 by the specification of con
tent. The ME21111 nuclei are limited to a small number of linguistic items,
e.g. (list) and (book); they are specifically reserved for the construc
tion of ME21112 concepts by the specification of content. For these terms,
determinants are limited mostly to documentation-related entity concepts.
On the other hand, in the terms with ME21112 nuclei, which already imply
specific contents, the determinants are used to specify the subtypes of con
tents. Thus, the different categories of nuclei reflect the different levels of
conceptual specification of PICR terms.
All the nuclei of RSCO terms are ME21112. Only a small number of
items, i.e. (bibliography), (thesaurus) and (catalogue),
Table 6.4. Formation patterns ofME21112 complex tenns.
become the nuclei, with the common characteristic being that both content
and structure are clearly specified. This pattern takes a wide range of de
terminants. The formation of ME21112 terms by RSCO is active only for
specific central categories of documentation entities.
We can also observe interesting characteristics in IFOA and FFUN
terms. The nuclei of IFOA terms are limited to a subset of ME21112, m the
same manner as RSCO terms. This and the fact that IFOA are rarely used
in ME21111 terms indicate that formal attributes of ME2111 concepts oc
cur only after the content is specified. The FFUN terms follow a relatively
Table 6.4. Formation patterns of ME21112 complex terms (contd.).
fixed pattern, with ME21111 serving as core concepts represented by nu

clei. More specifically, the nuclei mostly take the form of (book) and the
determinant of the FUN relation is (report).
6.1.1.3.3 ME21111 andME21112
As the distinction between ME21111 (documents specified by mode of
publication/formal structure) and ME21112 (documents specified by con
tent/mode of publication) is based on the aspectual rather than the substan
tive viewpoint, they basically belong to the same category of "documents"
as material entity. The term formation patterns of ME21111 and ME21112
are linked through the ME21111 nuclei of ME21112 terms; the steps of
term formation patterns can be summarised as in Figure 6.2.
Figure 6.2. Major formation patterns of ME21111 and ME21112 terms.
6.1.1.4 Inanimate - information carriers - non-documents (ME212)

There are 54 terms (8 simple, 45 two-item and 1 three-item) in this category.
Table 6.5 lists the patterns of 46 complex terms. The dominant conceptual
specification patterns are IFOA (formal attributes: 13) and FUSE (use: 11),
and the nuclei are all ME212 (non-document information carriers). In con
trast with ME21112 terms, this category is characterised by the absence of
PICR (information content and representation).
The IFOA terms contain two types of determinants: (i) those specify
ing size or colour by QL (quality); and (ii) those specifying notable parts by
ME251 (parts). The FUSE pattern is characterised by the dominance of the
relation MEA (means), which implies that the basic functions of the con
cepts represented by the nuclei are already determined and do not require
explicit specification.
6.1.1.5 Inanimate - machines and implements (ME22)

There are 59 (7 simple, 40 two-item, 11 three-item and 1 four-item) terms
in this category. Table 6.6 lists the patterns of 52 complex terms. The dom
inant specification pattern is FFUN (functional link), used in 41 of the 52
complex terms. This shows that the basic level of concepts to be named is
that of machines or implements with specific functions. The simple terms,
e.g. computer), also support this observation. The categories

of the nuclei are ME22 (36) and classificatory entities (CE: 16).
FFUN terms can be divided into a few groups, each with its own ten
dencies. Firstly, FFUN with FUN (function) only are characterised by the
nucleus (device), a function-neutral machine. The AC (activity) de
terminants specify their function. Of the 4 FFUN terms with AFO (af
fected object) only, 3 are characterised by the nucleus (system)
with AE11 (basic information entity) determinants. The actual function of
"system" seems to be deliberately unspecified, so as to allow maximum
variety with respect to its activities. The FFUN terms with MEA (means)
or MAN (manner) only are characterised by nuclei which represent a ma

chine with a specific function, such as (computer) or
(camera), with QL (quality) determinants. Terms with both AFO and FUN
relations are characterised by CE (classificatory entity) nuclei, with AFO
determinants being document- or information-related entities.
Formation Patterns o f Documentation Terms 127
Though they are all formed by FFUN, the differences in their actual
representation, including that of linguistic items, reflect different levels of
term formation; terms with FUN or AFO are considered to be one step
broader than those with MEA or MAN.
6.1.2 Representational entity (RE) terms

All the bottommost categories other than RE3122 (simple documentation
entities with specific structure) accommodate independent terms. Below,
some related categories will be grouped for closer observation of term for
mation patterns.
6.1.2.1 Broad representational entities (RE1)

RE12 (broad representational entity with some aspects specified) is concep
tually subordinate to RE11 (general broad representational category), and
RE121 and RE122 are differentiated only by aspects of specification. So
terms of these categories are observed together. There are 10 (4 simple and
6 two-item) terms in RE11; 22 (6 simple, 14 two-item and 2 three-item)
terms in RE121 (broad representational entities with function or relation
being specified); and 19 (5 simple and 14 two-item) terms in RE122 (broad
representational entities with structure and type being specified). Table 6.7
lists the formation patterns of RE1 terms.
A few conceptual specification patterns are dominant, i.e. FFUN (func
tional link) for RE121 terms and PCON (constituent elements) and IFOA
(formal attributes) for RE122 terms. This is in accordance with the very
definition of these categories. The categories of the nuclei vary, i.e. RE11
and CE1222 (non-independent complex classificatory entities; more specif
ically, [set]) for RE11 terms, RE11, RE121 and CE for RE121 terms;
and RE11, CE1222 ( [set]), etc. for RE122 terms. Terms with a
CE1222 nucleus become RE11 by the specification of PCON with RE11
determinants, and become RE122 with RE122 determinants. This reflects
the nature of CE1222 as nucleus, i.e. instead of bearing an independent con
ceptual status, the concept of this category adds the formal characteristics,
the conceptual substance of which is specified by the determinants.
Figure 6.3 illustrates the relationships among the three categories of
RE1. Note also that the relative diversity of specification patterns in RE1
Table 6.7. Fonnation patterns of RE1 complex terms.
categories may show that this category accommodates concepts peripheral

to documentation.
6.1.2.2 Linguistic entities (RE2)

This category was divided into set concepts (RE21) and element concepts
(RE22), with the assumption that their formation patterns have different
characteristics. There are only 5 (4 simple and 1 two-item) terms in RE21.
So this category is not examined. On the other hand, 40 (11 simple and 29
two-item) terms belong to RE22. Table 6.8 shows the formation patterns
of RE22 terms. The specification patterns of RE22 terms vary; the con
siderable diversity of conceptual specification patterns without dominant
patterns may be in accordance with the non-central status of this category
Figure 6.3. Hierarchy of RE1 concepts as seen from the viewpoint of term formation.
Table 6.8. Formation patterns of RE22 complex terms.
in the field of documentation. All but 3 nuclei belong to RE22. The patterns
of 2 terms with CE (classificatory entity) nuclei are similar to those of RE1
(broad representational entity) terms with CE nuclei.
6.1.2.3 Documentation entities (RE3)

6.1.2.3.1 Documentation entities - complex units (RE311)
This is yet another central conceptual category in the field of documen
tation. It includes 113 (17 simple, 95 two-item and 1 three-item) terms.
Table 6.9 lists the formation patterns of this category. Two specification
patterns, i.e. information content and representation (PICR: 25) and formal
attributes (IFOA: 25), are dominant. The nuclei consist of RE311 (com
plex documentation entity: 88), RE122 (broad representational entity: 4),
CE121 (complex classificatory entity seen as one: 2) and CE122 (complex
classificatory entity seen as composed: 2).
The individual linguistic elements used as nuclei in the two major
specification patterns are mutually exclusive, though all but one belong
to RE311. The nuclei of PICR terms are limited to the morphemes that
strongly imply information content, e.g. (entry), (description),
etc., while the nuclei of IFOA terms have the characteristics that their spe
cific information content is either unimportant, e.g. (chart), or their
function supersedes or covers the content, e.g. (index), (abstract).
Other patterns are difficult to generalise. It is noteworthy that the spec
ifications of functional aspects (FFUN, FUSE, FDES and FROL) together
account for the formation of 15 terms.
6.1.2.3.2 Documentation entities - simple units (RE312)
RE3121 (simple documentation entities specified by function or relation)
accommodates 60 (8 simple, 49 two-item and 3 three-item) terms, while
RE3122 (simple documentation entities specified by structure) does not ac
commodate independent terms. Table 6.10 lists the formation patterns of
52 RE3121 complex terms.
The central specification pattern is FFUN (functional link: 13), fol
lowed by IFOA (formal attributes: 8) and PICR (information content and
representation: 8). The categories of the nuclei are RE3121 (simple doc
umentation unit: 30), RE11 (general broad representational entity: 10),
RE22 (linguistic entity: 5), RE121 (broad representational entity with func
tion or relation specified: 5) and CE11 (single/simple classificatory en
tity: 2). Broad representational entities (RE1) and linguistic entities (RE2)
are frequently used as the core concepts represented by the nuclei. Non-
RE3121 concepts are only used in the specification of functional aspects
(FROL, FFUN, FUSE and FDES) or in the specification of the whole or af-

Table 6.9. Formation patterns of RE311 complex terms (contd.).
filiations (PPAR). Thus, one of the main patterns of forming RE3121 con
cepts is to indicate functional aspects (or, in a few cases, "whole"), on the
basis of broad representational or linguistic entities. This contrasts with the
fact that PICR, IFOA, RSCO (scope) and RSTA (status) are only applied
to RE3121 nuclei. This and the fact that no independent terms represent
RE3122 indicates that simple units of documentation entities (RE312) are
first identified from a functional point of view. Thus, RE3121 terms are
formed: (i) at the level of broad representational or linguistic entities to
RE3121 by specifying functional aspects; and (ii) at the level of general
RE3121 to specific RE3121 by various specifications.
6.1.2.3.3 Documentation entities - parts (RE32)
There are 18 terms (11 simple and 7 two-item) in this category. Table 6.11
shows the patterns of complex terms. Among the complex terms, 3 take
PPAR (whole or affiliations), which specifies the whole of the part repre
sented by the nuclei. The low ratio of complex terms shows that RE32
terms are not very productive.
6.1.2.4 Software entities (RE4)

There are 21 (4 simple, 13 two-item and 4 three-item) terms in RE41 (pro
gramme units), 8 terms (5 simple and 3 two-item) in RE42 (programme
parts) and 11 terms (3 simple and 8 two-item) in RE43 (software ele
ments). Table 6.12 shows formation patterns of RE41, RE42 and RE43
complex terms. By far the dominant specification pattern in RE41 terms
is FFUN (functional link: 12). The nuclei are either CE1221 (independent
compound classificatory entities) or RE41 (programme units). RE42 is too
small for any meaningful generalisation. The specification patterns of the
8 RE43 complex terms are FUSE (use), FDES (destination), IFOA (for
mal attributes) and INAT (nature). Compared to RE41 terms, RE43 terms
lack the FFUN relation; the former are defined as having a positive role or
function by themselves, while the latter have only a passive role.
6.1.2.5 Summary of representational entities

There are some category shifts within representational entity terms, i.e. the
nuclei of RE11 (general broad representational entity), RE121 (broad rep
resentational entity with function or relation specified) and RE22 (element
concept of linguistic entity) terms are used for forming other RE terms,
especially RE3121 (documentation entities with function or relation spec
ified) terms. Figure 6.4 summarises the category shift patterns (the speci
fications starting from RE122, RE3121, RE32, RE42 and RE43 are omit
ted). These category shifts are expected from the discussion of the nature
of conceptual categorisation in Chapter 4, where two dividing criteria —
simple/complex and general/subject-specific — were recognised to be im
portant; all the relevant concepts are simple units, of different specificity.
Figure 6.4. Patterns of category shifts among representational entity terms.
In general, the types of category shift can be explained by the differ

ences in the factors of term formation, i.e. conceptual specification patterns,
choice of nuclei and categories of determinants. In a few cases, however,
the same specification patterns produce terms of different categories, e.g.
FFUN (functional link) terms with RE11 nuclei can become RE121 or
RE3121. Although this cannot be properly explained by means of con-
Table 6.13. Formation patterns of AEll and AE12 complex terms.
ceptual categories and specification patterns, the individual linguistic items

involved are mutually exclusive. For instance, in FFUN terms with RE11
nuclei representing RE121 and RE3121 concepts, or in FUSE (use) terms
with RE121 nuclei representing RE121 and RE3121 concepts, there are few
common nuclei.
6.1.3 Abstract entity (AE) terms

6.1.3.1 Information entities (AE1)
There are 27 (8 simple and 19 two-item) terms in AE11 (basic information
entities), 26 (3 simple and 23 two-item) terms in AE12 (linguistic entities)
and 7 terms (all simple) in AE13 (operational information entities). The for
mation patterns of complex terms belonging to AE11 and AE12 are listed
in Table 6.13. AE13 terms are not treated here as they only accommodate
simple terms.
INAT (nature), which is a dustbin category, is dominant in AE11, and
FUSE (use) is dominant in AE12. Although we cannot see positive ten
dencies in these terms with respect to conceptual specification patterns, the
absence of the specification of functional aspects is notable for  term
formation. This is because, in documentation, AE11 concepts (information,
meaning, concepts, etc) are regarded as abstract theoretical objects. An
other interesting phenomenon is that AE11 terms with PCON (constituent
elements) or PPAR (whole or affiliations) take CE (classificatory entity) and
RL1 (labels of relation) elements as nuclei and AE11 elements as determi
nants. This reveals the role of some of the classificatory entity elements in
term formation, as well as the affinity between RL1 and abstract or classi
ficatory entities. The roles of CE in term formation will be discussed later.
AE12 terms can be divided into three groups, i.e. those related to natu
ral language, to documentation languages (for indexing and retrieval), and
to programming languages. Though we cannot observe any consistent char
acteristics of this category as a whole, it is noted that specification of func
tional aspects (namely FUSE) is used to form documentation or program
ming languages only.
6.1.3.2 Systems of knowledge (AE2)

6.1.3.2.1 Subject fields (AE21)
Only 3 terms (1 simple and 2 complex) belong to the category of general
subject fields (AE211); they are not treated here as there are too few for any
meaningful generalisation. The category of specific subject fields (AE212)
accommodates 22 (4 simple, 17 two-item and 1 three-item) terms. The
formation patterns of 18 AE212 complex terms are listed in Table 6.14.
With only two exceptions, the patterns are fixed, i.e. AE211 (or
AE212) nuclei, mostly with entity determinants specified by FFUN (func
tional link). The two non-FFUN terms have complex nuclei constructed
by the FFUN specification. This shows that the first step of naming a spe
cific subject field is to specify the object of the study, starting from the
expressions that represent general systems of knowledge, such as "theory"
or "field".
Table 6.14. Formation patterns of AΕ212 complex terms.
6.1.3.2.2 Methodologies (AE22)

There are no terms representing AE221 (methodologies), while 109 terms
(1 simple, 36 two-item, 70 three-item and 2 four-item) belong to specific
methodologies (AE222). Table 6.15 lists the formation patterns of the 108
AE222 complex terms.
The dominant pattern is the specification of use (FUSE), found in 105
of the total 108 complex terms. This clearly shows that the formation of the
terms in this category follows a fixed pattern that reflects the nature of the
conceptual category. The nuclei are AE221 (general methodology: 100),
CE122 (complex classificatory entity seen as composed: 7) and AE222
(specific methodology: 1). It is notable that 92 of the 100 AE221 nuclei are
(method).
The variations of intra-term relations within FUSE patterns can be
classified into three groups corresponding to different levels of specificity:
(i) those with the intra-term relation USE (use), AFO (affected object) or
PRO (product) only; (ii) those with the relations MAN (manner) or MEA
(means) only, MAN and USE, MEA and USE, MAN and PRO, or MEA
and PRO; and (iii) those with AFO, MAN and PRO. The terms in the first
group belong to the same level because (a) USE, AFO and PRO do not co-
occur in terms of three or more items and (b) the USE determinants imply
types of objects or products, while the AFO or PRO determinants imply
types of activities. For roughly the same reason, the six patterns in (ii) are
considered to belong to the same level of specificity. The two terms in (iii)
Table 6.15. Formation patterns of AE222 complex terms.

Table 6.15. Formation patterns of AΕ222 complex terms (contd.).
are clearly more specific than the terms in (ii). Figure 6.5 illustrates this
situation.
We can also observe some tendencies of the determinants. Many de
terminants of USE belong to the activity concepts which imply types of
products or objects, e.g. representational entity production (AE232) or ma
terial or representational entity state change (AE242). Most determinants
of AFO or PRO, on the other hand, belong to material or representational
document-related entities (ME21 or RE31). This corresponds to the expla
nation for the coherence of the groups (i) and (ii) above1.
1
FUSE is expected to be the main specification pattern in the formation of methodology
terms irrespective of the domain, and the formation pattern of this category specific to the
field of documentation may reside more in the types of determinant categories. To confirm
Figure 6.5. Formation of AE222 terms by specification of use.
Table 6.16. Formation patterns of AE3 complex terms.
6.1.3.3 Subjective entities (AE3)

Only 5 terms (all two-item) belong to the category of roles/evaluations
(AE31), 6 terms (3 simple and 3 two-item) to relations (AE32) and 3
(all two-item) to psychological entities (AE33). Table 6.16 lists forma
tion patterns of AE3 terms. All the AE31 terms take FUSE (use), and 2
of three complex terms in both AE32 and AE33 take FDES (destination);
AE3 terms are characterised by the predominance of specification of pas-
this fully, a comparative study is necessary.

sive functional aspects. This contrasts with such categories as AE11, where
the lack of specification of functional aspects is noteworthy.
6.1.4 Classificatory entity (CE) terms

At the bottommost level of classificatory entities, independent and depen
dent units are distinguished. For example, CE111 are independent and
CE112 are dependent. The difference was established for facilitating the
classification of morphemes. We collapse this distinction, as we examine
CE terms and not morphemes.
Table 6.17. Formation patterns of CE complex terms.
CE1 (single/simple items as perceived by formal complexity) contains

two terms (1 simple and 1 three-item), CE121 (complex items seen as one)
contains 16 terms (7 simple and 9 two-item), CE122 (complex items seen
as composed) contains 24 terms (3 simple and 21 two-item), and CE2 (clas-

sificatory entity as perceived by relation or role) contains 8 terms (5 simple
and 3 two-item). Table 6.17 shows the formation patterns of the complex
terms in these categories.
We observe the absence of formation by specification of functional as
pects. This is because the specification of function implies a positive role
in the conceptual dynamics of the field, which leads to the positive char
acterisation of the concepts. Thus, as long as the terms are classified as
classificatory entities, they remain essentially passive. In fact, it is possible
to regard the classificatory entity terms as subject-specific abstract entities
(abstract documentation entities), the specific subclass of which could not
be accommodated for in the concept categories of abstract entities estab
lished in Chapter 4. This explains the affinity of the formation patterns
of information entity concepts (AE1 terms) with the CE terms; both lack
specification by positive functional aspects2.
6.1.5 Quality (QL) terms

Only QL12 (redefined quality concepts for documentation) and QL13
(qualitative/quantitative measurements) accommodate independent terms.
There are 36 (6 simple and 30 two-item) terms in QL12, and 21 (5 simple
and 16 two-item) terms in QL13. Table 6.18 lists the formation patterns of
complex terms for these two categories.
It can be observed that 22 of the 30 complex QL12 terms and all of the
QL13 terms take the conceptual specification pattern of IACO (attributed
concept), though the conceptual categories of determinants vary. The nuclei
of the other specification patterns of QL12 terms are all (characteris
tic), which is itself an independent term in the field of documentation. This
indicates that the quality concepts can only become independent terms as
labels of qualities attributed to some other concepts.
Closer inspection of IACO terms reveals the difference of the levels
of concepts between those which take ME (material entity), RE (repre
sentational entity), AE (abstract entity), CE (classificatory entity) or AC
(activity) determinants and those which take RL (relation) or QL (quality)
determinants. In the former case, the quality concepts are generally at-
2
Linguistic entity terms (AE12) have FUSE and FDES formation, but not FFUN nor
FROL, which shows they do not have positive functions themselves.
Table 6.18. Fonnation patterns of QL complex terms.
tributed to specific documentation-related concepts. In the latter case, very

general labels of qualities or quantities are attributed to specific aspects of
qualities or relations (or values of qualities are elevated to the quality la
bels or types of qualities, by taking the labels of qualities as nuclei; also
note that most of the nuclei that take QL or RL as determinants are QL11,
i.e. general labels of qualities). The formation of terms of the former type
is more specific than that of the latter. This is also supported by the fact
that some of the nuclei of the terms taking the former type of determinant
Figure 6.6. Formation of quality concept terms by specification of attributed concepts.
are the result of the latter type of formation, as in -IACO-

-IACO- ] ([expression=method=of]-IACO-[acceptance-IACO-nature]).
Figure 6.6 illustrates this situation.
6.1.6 Relation (RL) terms

RL12 (redefined labels of relations) accommodates 20 (2 simple and 18
two-item) terms, RL13 (relational measurements) accommodates 29 (6
simple and 23 two-item) terms, and RL2 (various values of relations) ac
commodates 26 (6 simple and 20 two-item) terms. Unlike QL (quality),
there are terms that represent values of relations.
Table 6.19 lists the formation patterns of 18 RL12 and 23 RL13 com
plex terms. As in QL1, IACO (attributed concept) terms are the majority.
The other specification patterns in RL12 terms are mostly tied up with a
specific nucleus of (relation), which is itself a term. The INAT (na
ture) terms of category RL13 are either specific combinations, such as
(logical sum) and (logical product), or terms with complex nuclei
formed by specification of IACO. So the different levels of term formation
illustrated in Figure 6.6 for terms representing quality labels are also visible
in the formation of terms representing relation labels.
Table 6.20 lists the formation patterns of 20 RL2 complex terms. Of
these 20 terms, 18 have (relation) as the nucleus, with the specification
Table 6.19. Formation patterns of RL1 complex terms.
pattern of IMAN; with only two exceptions, the terms of this category rep
resent various subconcepts of (relation), which is itself a term in the
field of documentation. Like CE terms, these terms might also be regarded
as being very close to subtypes of abstract entity concepts, together with
terms representing quality concepts.
6.1.7 Activity (AC) terms

6.1.7.1 Units of activity/activity indicators (AC1)
Table 6.21 lists 20 (18 two-item and 2 three-item) terms belonging to the
category of units of activity (AC1). There are no simple terms. The applica-
Table 6.20. Formation patterns of RL2 complex terms.
Table 6.21. Formation patterns of AC1 complex terms.
tions of specifications of attributed concept (IACO: 13) and of complemen

tary elements of functions (FCOM: 7) to generic AC1 concepts make up the
basic formation patterns of this category. These IACO terms are formed in
a similar way to IACO terms representing quality (QL) or relation (RL)
concepts, the nuclei that represent units of activities being "labels" of ac
tivities. FCOM terms, on the other hand, have nuclei whose "active" nature
is stronger. All the FCOM terms take AFO or PRO determinants.
6.1.7.2 Action(AC21)
The conceptual category of action (AC21) accommodates 97 (14 simple,
80 two-item and 3 three-item) terms. Table 6.22 lists the formation patterns
of complex AC21 terms. The majority (69) of action terms take FCOM
(complementary elements of functions) as the specification pattern, the ac
tual linguistic representations of which vary: affected object (AFO: 40),
manner (MAN: 23), means (MEA: 3), MAN + AFO (2) and MEA + AFO
(1). The lack of PRO is basically due to the definition of the subdivisions of
the category AC, where a concept category AC23 (production) is separately
defined.
The terms with other than FCOM specifications tend to take indepen
dent terms as nuclei. Though the nuclei of many FCOM terms are them
selves terms, we can recognise two levels of term formation; the general
non-terminological action concepts are first restricted by the specification
of complementary elements of the action concepts, while other specifica
tion patterns are mostly applied to already restricted concepts.
6.1.7.3 Transference (AC22)

The category of transference (AC22) accommodates 22 (6 simple, 14 two-
item and 2 three-item) terms. The formation patterns of AC22 complex
terms are listed in Table 6.23. The domain-dependent subcategorisations
are maintained in the table.
The dominant conceptual specification pattern of this category is
FCOM (complementary elements of functions), which is used in 12 out of
the total of 16 complex terms. The determinants of the FCOM terms repre
senting representational entity transfer (AC222) and material entity transfer
(AC223) are information or document entities — mainly representational
entities for AE222 and material entities for AC223 — though the types of
determinants are not rigid; the combinations of the nuclei and determinants
define the final category of concepts.
6.1.7.4 Production (AC23)

There are 21 (10 simple and 11 two-item) terms in the category of produc
tion (AC23). The number of simple terms is relatively high. Table 6.24
shows the formation patterns of the complex terms of this category. FCOM
(complementary elements of functions) is the only conceptual specification

Table 6.22. Formation patterns of AC21 complex terms (contd.).
pattern. The intra-term relations are product (PRO: 4), manner (MAN: 5)
and means (MEA: 2). AC231 nuclei tend to require PRO, while AC232 or
AC233 nuclei take MAN or MEA. This can be explained by means of the
nature of the nuclei, i.e. the nuclei which are not themselves terms tend to
take PRO. It is interesting to compare AC232 and AC233 terms with AC222
and AC223 terms, where the types of objects are implied but are actually
represented linguistically. In AC22 terms, the specific subtypes of objects
seem to be important (perhaps partly because many AC22 morphemes im
ply types of means or manner within the domain), while in AC23, not the
specific subtypes of the products but means or manner are regarded as im
portant for differentiation.
6.1.7.5 State change (AC24)

State change (AC24) accommodates 26 (12 simple and 14 two-item) terms.
As in AC23, the ratio of simple terms is high. Table 6.25 lists the formation
patterns of 14 complex terms. Among them, 13 terms take FCOM (com
plementary elements of functions) as the conceptual specification pattern.
All but one of the nuclei represent AC24. The exception is special in that
the nucleus functions as "categoriser", an equivalent to classificatory enti
ties used to construct other concepts. The other FCOM terms have some
variations with respect to their intra-term relations, i.e. with affected object
(AFO: 5), manner (MAN: 5) or means (MEA: 2). The nuclei of MAN or
MEA tend to be terms themselves, analogous to formations of AC23 terms.
6.1.7.6 Summary of term formation of specific activities (AC2)

All the subcategories of AC2 take FCOM (complementary elements of
functions) as their central conceptual specification pattern; this is very nat
ural. As we have briefly observed before, the role of FCOM can be divided
into two: (i) that of forming specific terms to represent concepts, based on
general AC concepts which are not independent; and (ii) that of naming
narrower concepts of the independent AC concepts in documentation. On
the other hand, the other specification patterns are in general only applied
to the AC concepts which are independent in documentation.
6.2 Cross-categorial examinations
In the previous section, we found typical or motivated term formation pat

terns through the examination of the data ordered on the basis of conceptual
categories of terms. In the process, we found a few characteristics that can
be better clarified from the cross-categorial point of view, which we exam
ine here.
6.2.1 Affinities among conceptual categories

We noticed in 6.1 similar term formation patterns among different concep
tual categories. Some of these reveal the status of the conceptual categories
within the overall conceptual structures and operations in the field of docu
mentation.
6.2.1.1 "Active" concepts

Disregarding the detailed characteristics, we can recognise that ME111
(people), ME12 (organisations), ME22 (machines) and RE4 (software),
which are understood to have the common trait of "being active", are dom
inated by the specification of functions (FFUN). The formation patterns of
these classes of terms reflect the basic status of these categories in the field
of documentation, which is mainly determined by the trait "being active".
Interestingly, RE1 (broad representational entities) and RE312 (simple
documentation entities) also take FFUN as dominant specification patterns,
though these concepts do not by themselves have the conceptual trait of
activeness. They reflect the specific position of these conceptual categories
in documentation, i.e. some representational entities are conceptually ma-
nipulated from the point of view of the function they take. It is interest
ing to contrast these with the conceptual categories characterised by the
dominance of FUSE (where functional aspects are noted only passively),
i.e. ME212 (non-document information carriers), AE12 (linguistic entities)
and AE2 (subjective entities). These are regarded as taking a passive role
in documentation.
6.2.1.2 Documentation-related entities

ME2112 (documents by content or structure), RE311 (complex documen
tation entities) and RE312 (simple documentation entities) share a common
dominant pattern, i.e. specification of information content and representa
tion (PICR). Irrespective of their being material or representational, they
share the common conceptual trait that they can contain information (con
tents or representation). We can understand why PICR is also notable in
linguistic entities (RE2); in this case, the meaning of the linguistic symbol
is regarded as a content. In the field of documentation, these categories
are perceived as being similar, with the perspective based on the distinction
similar to that of "content" and "carrier".
6.2.1.3 Units of activity, quality and relation

The last major cross-categorial affinity is recognised among QL12 (labels
of qualities), QL13 (qualitative and quantitative measurements), RL12 (la
bels of relations), RL13 (relational measurements) and AC1 (units of ac
tivities). They are all characterised by the dominance of the specification
of attributed concepts. For qualities, relations or units of activities to be
come technical terms (and, thus, domain-specific concepts), they must be
attributed to some other more specific concepts. This corresponds to some
extent to the nature of the categories established in 4.2.1.
These shared characteristics, together with the motivated term forma
tion patterns for each conceptual category, explain the quantitative relation
ship between conceptual specification patterns and the broad conceptual
categories briefly outlined in 5.3.2.
6.2.2 The status of classificatory entities (CE)
The status of classificatory entity needs independent and special examina

tion from very different points of view, not because of the particular nature
Table 6.26. Classificatory entity items used in the terms representing other categories.
of terms that belong to this class but because of the behaviour of CE mor
phemes used as nuclei, which appear in other broad categories.
Table 6.26 lists the linguistic items representing CE concepts used as
the nuclei for constructing the terms of other categories. It shows that the
category shifts are triggered by only a limited number of conceptual speci
fication patterns, i.e. FFUN (functional link), FUSE (use), PPAR (whole or
affiliations), PCON (constituent elements), RORI (origin), IFOA (formal
attributes) and a single case of PICR (information content and representa
tion).
The category shifts can be divided into three types:
a) Combinations whose categories are systematically determined by the
categories of the determinants: PPAR, PCON or PICR terms belong to
this group. Examples are (code element) and (con-
cept system). Though the examples in the corpus are limited, this pattern
of category shift is transparent and seems to have wider applicability.
b) Combinations whose categories are not systematically determined from
the point of view of nuclei and categories of determinants: They consti
tute the central patterns of the resultant categories, and terms are formed
basically in the same way as the terms whose nuclei belong to the same
categories as the resultant concepts. Examples are FFUN terms repre
senting ME22, e.g. (reception body; receptor), and RORI terms
representing ME21111, e.g. (publication thing; publication).
c) Combinations whose categories are not systematically determined and
which do not belong to the central patterns of term formation of the
resultant categories: From the point of view of the nuclei, this type is
similar to b), but from the point of view of the resultant categories, this
type is more idiosyncratic, because the formation patterns do not belong
to the central patterns of the categories. Examples are FUSE and IFOA
terms with the nucleus (element), e.g. 2 (binary element).
In the field of documentation, a) and b) seem to be productive, while c)
is not. In correspondence with the a) and b) productive patterns of category
shifts, the use of CE nuclei in term formation can be described as follows:
a) CE are used as secondary concepts with respect to the categories of
resultant concepts, specifying (from the reverse point of view) config
urations, structures, units, etc. The CE concepts, not being nuclei of
substance themselves, do not occupy the generic position in the set of
resultant concepts. Thus, these CE concepts are used as "helpers" for
the formation of terms in the relevant conceptual categories.
b) CE concepts are used in the same way as the generic concepts of the
target categories. In this case, the CE concepts share at least one charac
teristic with the generic concepts of the target categories, e.g. "system"
(CE) and "device" (ME). In this case, as nuclei, these CE concepts can
be seen as representing generic categories of special hierarchies which
cut across different broad categories.
In addition, of course, CE concepts are used as the nuclei of indepen
dent CE term formations, in which case the category of CE is very close to
AE.
6.3 Summary and reflections
6.3.1 On the description of term formation patterns
So far, our observations have revealed a number of typical tendencies and

conceptually-motivated formation patterns of documentation terms. What
is presented in this chapter, however, does not fully exhaust the useful anal
yses made possible by the descriptive devices introduced in Chapters 4 and
5. Using these three descriptive devices, i.e. conceptual categories, concep
tual specification patterns and intra-term relations, it is possible to proceed
to a more detailed account of conceptual patterns of term formation. For
instance, we can also analyse the following aspects.
1. The restrictions or preferences of conceptual specification patterns vis-
à-vis the conceptual categories of determinants and nuclei.
2. The correlation tendencies between the conceptual categories of deter
minants and nuclei.
3. The status of embedded combinations within the overall patterns of term
formation for each conceptual category, and the relations between sec
ondary combinations and term-forming combinations.
4. The distinction between the tendencies attributed to the conceptual cat
egories or specification patterns in general and the tendencies that may
be better attributed to individual language items (or the specific concepts
they represent).
These points were not systematically analysed but only touched upon
when they were thought to be important. By exploiting these aspects, it
will be possible to further clarify the previously observed tendencies, such
as: The more the overall concepts represented by terms are stable and cen
tral to the domain, the more indirect specifications or intra-term relations
are used, e.g. RSCO (specification of scope) in ME21112 (documents) or
MEA (means) of FUSE (specification of use) in ME212 (machines and im
plements). It will also be possible to clarify more systematically whether
the realisation of different intra-term relations within the same conceptual
specification pattern shows different levels of the same specification or dif
ferent types motivated by different conceptual characteristics. Some of
these questions are dealt with in Kageura (1993).
In addition, the status of the regularities of formation patterns in some
conceptual categories would further be clarified by extending the analyses.
For instance, while the formation patterns in such domain-dependent cate

gories as ME2111 are attributable exclusively to the terminology of docu
mentation, the formation patterns observed in AE21 terms seem to be valid
in other domains as well. This would be clarified by further observations
coupled with the terminological data of different domains; we consider the
comparative viewpoint to be important now that the overall term formation
patterns of a domain are clarified.
The description of conceptual patterns presented in this chapter, there
fore, can be augmented in many ways. The principal aim of the present
book, however, is to show the relevance and potential of the methodological
framework which we defined in Part I for the study of the dynamics of ter
minology. From that point of view, the concrete analysis of term formation
patterns has been carried out to the level that was required for the theoretical
study of terminology, i.e. to illustrate the major patterns within the concep
tual system of the field of documentation reflected in its terminology3.
6.3.2 Reflections on methodological devices and procedures

In 3.2.2, two main methodological claims were presented in relation to
the description of conceptual patterns of term formation: (i) that concep
tual categories attributed to terms and their constituent elements should be
placed within the overall conceptual system of the domain to which the
terms in question belong; and (ii) that intra-term relations should be reor
ganised as conceptual specification patterns, defining the position of de
terminants vis-à-vis the nuclei. These are the concrete devices to describe
term formation patterns in accordance with the perceptions of terminology
introduced in Chapter 2, namely, that some systematic factors in the exist
ing terminology of a domain determine the formation of new terms, which
is in turn attributed to the conceptual system of the domain. Through the
actual descriptions of conceptual patterns of term formation presented in
6.1 and 6.2, the validity of these claims was shown, especially with respect
to the conceptual system and to the conceptual specification patterns.
3
Recall that in Chapter 2 it was argued that the basic target of the theoretical study of
terminology should be the terminology of each domain; the concrete descriptions in Chap
ters 4 to 6 were the theoretical condition for showing the relevance of the methodological
framework, even if, as in the present book, the detailed description of formation patterns of
individual documentation terms is not a primary goal.
6.3.2.1 Conceptual system

The introduction of the overall conceptual system corresponding to the ter
minology of documentation helped clarify the motivated and major con
ceptual patterns of term formation for the terminology of documentation.
This is as distinguished from the description of individualised restrictions
of combinations in complex terms which happen to belong to the terminol
ogy of documentation.
As for the individual conceptual categories established in Chapter 4,
most of them, including the broadest categories, proved useful as method
ological devices or guides for observing conceptual patterns of term forma
tion, though some of them accommodate only a small number of terms for
which no meaningful generalisations were possible. In addition, some, e.g.
ME11 (people) and major document-related categories, proved to consti
tute a coherent category in which motivated and consistent patterns of term
formation could be observed.
At a more specific level, the basic viewpoints (introduced in 4.2.2)
from which smaller categories were defined — i.e. the distinctions of gen
erality and specificity of concepts, of simple and complex concepts, of
basic and functional or relational aspects, of general and subject-specific
concepts — and the categories established according to these viewpoints
proved to be very useful. These distinctive viewpoints helped capture the
general phenomena shown, for instance, in Figures 6.2 and 6.3. At a con
ceptual level, what is shown in Figure 6.3 is tautological; that general repre
sentational entity concepts (RE11) become specific representational entity
concepts (RE12) by some conceptual specification is an analytical truth. In
the same manner, that general representational entity concepts (RE11) be
come specific representational entity concepts whose function or relations
are specified by the specification of functional (e.g. FFUN or FUSE) or re
lational (e.g. PPAR) aspects is also analytically true. What is important,
however, is the observation that terms representing broad representational
entity concepts with specific functions or relations are actually formed by
the application of the specification of functional or relational aspects to the
items representing general representational entities in a systematic way.
This could only be revealed by using these conceptual categories as a means
of description.
6.3.2.2 Conceptual specification patterns

The introduction of conceptual specification patterns has two implications.
Firstly, it fixes the viewpoint from which relations among constituent ele
ments are observed; the conceptual specification patterns define the status
of the determinant with respect to the nucleus. This contributed to the sys
tematic capture of the formation patterns of terms within the overall con
ceptual system to which the terminology is attributed. The understanding of
the formation patterns of, for instance, RE121 (broad representational en
tities with function or relation specified) or RE122 (broad representational
entities with representational structure or type specified) became possible
with the introduction of conceptual specification patterns. Secondly, it gen
eralised the relations between constituent elements by grouping together
the different intra-term relations into the different realisations of the same
specification pattern, as typically shown in FFUN (specification of func
tional link) that accommodate intra-term relations AFO (affected object),
PRO (product), MEA (means) and MAN (manner). This contributed to
capturing the basic conceptual motivation behind different linguistic reali
sations of complex terms.
6.3.2.3 Remaining problems

A few concrete problems remain to be examined and studied further if one
is to pursue the full-scale description of dynamics of terminology.
Firstly, the conceptual categories and conceptual specification patterns
need further refinement, both by themselves and in relation to the descrip
tive framework. For instance, some categories are too small and so no
motivated patterns were observed. A different definition of conceptual cat
egories might reveal systematic patterns which are hidden in the current
framework. Also, the category of classificatory entity, when applied to
terms, can be interpreted as being close to that of abstract entity, while
when classificatory entity is applied to nuclei, the nuclei are revealed to be
have differently than when representing abstract entities. This implies that
it would be better in some cases to distinguish the conceptual system for
morphemes from the conceptual system for terms.
The second problem is concerned with the allocation of linguistic items
(terms and constituent elements) to conceptual categories and the assign
ment of intra-term relations to pairs of items. If examined individually,
some controversial cases will be found. For instance, (mass

media), which is classified as ME212 (non-document information carriers),
is very different from other terms in this category; this sort of treatment was
inevitable, partly because of the conflict between the granularity of concep
tual categories and the number of terms accommodated in each category.
Though no concrete problem arose in the analyses, different assignments
might have been more useful in capturing some patterns.
Thirdly, and from a different point of view, it might be the case that
some of the phenomena described in 6.1 and 6.2 are more profitably treated
from different points of view. For instance, some phenomena of category
shifts can be explained more adequately by metaphorical transformation of
meaning, now that the conceptual patterns of category shifts are revealed
(cf. Way 1991).
Like all linguistic analyses involving meaning or concepts, there is no
instant solution to these problems. The step-by-step refinement of categori
sations and assignments through trial and error for a wider variety of data
as well as the theoretical examination and clarification of the criteria and
procedures are perhaps the only ways to overcome the first two problems.
The last point will be complemented by the analysis based on a flexible
viewpoint of the aspects of terms. The analyses carried out in Chapters 4
to 6, therefore, from the point of view of the concrete study of the forma
tion patterns of terms in the field of documentation, are only a first step;
but they distinctly show the validity of the theoretical and methodological
framework sketched in Chapter 3.
6.4 From term formation patterns to the dynamics of terminology
Chapters 4, 5 and 6 described the conceptual patterns of term formation

observed in the existing terminological data. As our objective is to reveal
the dynamics of terminology, it is necessary to interpret the descriptions so
far from the point of view of "dynamics of terminology". Recall that, in
Part I, it was mentioned that we are concerned with the dynamics of ter
minology as observed in the synchronic slice of the internal structure of
existing terminology. In saying that, we adopted the assumption that this
dynamics itself does not change in the creation of new terms (see Figure
2.2). Under this assumption, all the observations concerning the term for-
mation patterns (presented in 6.1) can be straightforwardly interpreted as

the prediction of the possible formation patterns of new terms.
For instance, we observed that existing terms representing people
(ME111) are formed by the specification of either functional link (FFUN)
or whole or affiliation (PPAR). So we can extend this observation to fore
cast possible formation patterns of new terms representing "people" in the
field of documentation. That is, we can reasonably conclude as follows:
A new ME111 complex term will be formed by either specifi

cation of FFUN or PPAR, starting from the ME111 nuclei. The
likelihood that the specification of FFUN will occur is roughly
three to four times higher than the likelihood that PPAR is used
when a new term is produced as a complex term.
It is also possible to forecast the combinations of intra-term relations

that will be used in the formation of new terms. For instance, we saw that
the methodology (AE222) concepts specified by USE (use) or PRO (prod
uct) can be further specified by MEA (means) or MAN (manner). From the
quantitative tendencies of these terms we can reasonably forecast the form
of possible new terms.
In this sense, what we have observed about the formation patterns of
terms is expected to be valid for possible new terms and, thus, for the ex
tended range of the terminology of documentation expected in the future.
However, this is still too coarse and insufficient as a theory of dynamics of
terminology.
To proceed from here, there are two choices. One is to discard the
assumption adopted so far and regard the creation of new terms and the
growth of terminology as a unique historical event. Following this route,
we would most probably end up excluding the possibility of incorporating
assumptions about potential developments of terminology into the study
(cf. Kageura 2000), or incorporating various real-world factors which are
manageable only for a small number of individual terms but not with re
spect to a coherent set of terminology. The alternative is to ask whether
there are ways of exploring further the description of the dynamics of ter
minology and, thus, refining the expectations of new terms, while maintain
ing the basic assumption which was broadly called "structural" in Chapter
2. The question to be answered if taking the latter approach would be, for
instance, "Is it possible to forecast how many unseen conceptual specifi

cation patterns would be used for the formation of new terms of a given
category?" As briefly mentioned in the Introduction and in Chapter 2, it is
to this direction that we will proceed in Part III.
Part III
Quantitative Patterns of
Terminological Growth
Chapter 7
Quantitative Analysis of the Dynamics

of Terminology: A Basic Framework
In Part II, the conceptual patterns of term formation were observed. The
level of description remained necessarily broad to leave room for the de
scription of general tendencies of term formation. A quantitative approach
can be used to complement the conceptual description and explore finer
tendencies of term formation or, more precisely, the growth patterns of ter
minology. This is a field so far virtually unexploited in the research field
of terminology and even in the field of lexicology in general. By exploring
this, we will be able to shed light on aspects of the dynamics of terminology
that have been unexposed so far. This chapter first explains what aspects of
terminological growth can be captured by the quantitative approach, then
presents the quantitative method. Lastly, the conditions and assumptions
necessary for applying the quantitative method will be examined.
7.1 Patterns of terminological growth
In Chapter 6, we observed, for instance, that terms representing complex

documentation units as representational entities (RE311) are formed mainly
by specifications of information content and representations (PICR) and of
formal aspects (IFOA) (see 6.1.2.3a and Table 6.9). We also saw that most
of the nuclei represent complex documentation units (RE311). On the basis
of these observations, it became possible to forecast the possible patterns
of the formation of new terms, e.g. which specification patterns are likely
to be used and how, etc.
Descriptions of conceptual patterns of term formation

Putative (grown-up) situation of terminology
Quantitative descriptions of terminological growth
Figure 7.1. The position of the descriptions of terminological growth.
Following these descriptions, it is natural to ask questions such as:

1. When a new term with an RE311 nucleus is created, what is the likeli
hood that one of the existing RE311 nuclei will be used? Alternatively,
what is the likelihood that a new linguistic item (and thus most prob
ably a new concept) will be used as the nucleus? More generally, how
are morphemes representing different conceptual categories used in con
structing the terminology of documentation as a whole?
2. Turning our eyes to the conceptual specification patterns, we observed
that, in total, 12 specification patterns were used in the formation of
complex terms that represent RE311. When the terminology of docu
mentation grows and becomes twice the size of the observed data, for
instance, we can ask: Will only these 12 specification patterns of the
original corpus be used in the RE311 complex terms? Or is there a pos
sibility that new specification patterns will appear when the terminology
grows?
The answers to these questions constitute an extension of and a comple
ment to the description of the conceptual patterns of term formation. With
them, we would then be able to forecast the dynamic developments of the
Quantitative Framework 167
terminology when the size of the terminology changes. Figure 7.1 gives a
rough image of the position of the quantitative description of terminologi
cal growth in relation to the descriptions of the conceptual patterns of term
formation.
In order to pursue the characterisation of patterns of terminological
growth as exemplified above, it is necessary to clarify the mathematical
methods and the relation between the mathematical methods and the phe
nomenon of the dynamics of terminology. Let us turn to these tasks now.
7.2 The mathematical model
For the sake of convenience, let us assume a situation where we are in
terested in the increase in the number of the constituent elements or mor
phemes when the size of a terminology is increased. In other words, we
are concerned with knowing, on the basis of given terminological data,
how many different morphemes would be used if the size of a terminol
ogy became 1.5 times, twice, three times, etc. the size of the given data. By
assuming the existence of the same terminology in these various sizes, we
may regard the original terminological data as a sample, derived from the
population which has the "essential" probabilistic structure of the termi
nology construction. In the following, starting from the basic explanation
of the binomial model of morpheme distribution in terminology, we intro
duce binomial interpolation and extrapolation, which, if properly applied,
fit very elegantly the concept of terminological growth. In this section, the
basic terminology of probability and statistics will be used without expla
nation; see DeGroot (1984) for the basic background of probability and
statistics, and Baayen (2001) for a detailed explanation of the quantitative
model adopted here. The notation used in the present study was taken from
Baayen (2001).
7.2.1 The binomial model

Let us start with defining the binomial model — a derivation of the urn
model — which compares the construction of terminology with the extrac
tion of numbered balls from an urn. The urn model is the most commonly
adopted framework for establishing a probabilistic model of distributions
of words in texts (Yule 1944; Chitashvili & Baayen 1993; Ogino 1998) and
of morphemes in terminologies (Kageura 1998a).
In the present case, the mathematical model is constructed over the
distribution of morphemes in a given terminology. The model assumes that
there exist S element types1 in the urn. This equals to assuming that the
terminology population consists in total of S different morphemes. It is also
assumed that r token numbers of each morpheme type are included in the
urn, such that the relative token frequency of each morpheme type in the urn
equals the population probability of the morpheme type. This constitutes a
basic model of the population from which the actual terminological data is
assumed to be obtained.
The sampling of the terminological data corresponds to selecting mor
phemes from the urn, one after another. We assume that we are only con
cerned with the selection of morphemes in the terminology as a whole, and
not with the construction of individual terms. Sampling with replacement
corresponds to the binomial model, while sampling without replacement
corresponds to the hypergeometric model. These two are asymptotically
identical, and either can be used in the present situation. In any case, the
sample terminological data consisting of TV morpheme tokens is regarded
as equivalent to the set of N morpheme tokens chosen randomly from the
urn. So the basic model is constructed on the basis of morphemes and their
distribution in the terminology, and the level of individual terms is disre
garded in the model.
Formally, the situation can be defined as follows. First, assume that
there are S different types or events, w1, w2..., ws, in the population:
Definition 1 S : Population number of types or events.
Definition 2 w1,w2, ..., wi, ..., ws : Types or events in the population. In
the present case, an event corresponds to a morpheme type.
To each of these types, the population probability pi is assigned:
Definition 3 p1, P2, ..., Pi, ..., ps: Population probabilities of the events
W1,W2, ..., Wu ..., WS.
This is equivalent to what is represented by the urn.
1
In this and the following chapters, we use the terms "type", "token", "different" and
"running" in the standard sense of quantitative linguistics. For instance, the meanings of
"morpheme type" and "different morpheme" are identical; so are the meanings of "mor
pheme token" and "running morpheme". We keep using both pairs of terms because one or
the other fits some contexts better.
When a sample of N tokens is taken from the population, the proba

bility that a morpheme wi appears exactly m times in the sample can be
expressed as follows, under the binomial model:
where X indicates a random variable and:

Definition 4 N : Sample size in token number of morphemes.
Let ƒ (, TV) express the number of times that wi occurs in the sample
of size N:
Definition 5 ƒ (, İV) : The frequency of occurrence of wi in a sample of
size N.
The expectation of ƒ (, Ν) is given by:
where E indicates the expectation. In the derivation of the last line, we

use the fact that we obtain the probability 1 by summing up all the possible
occurrences of an event, which in this case is:
Let V{m, N) represent the number of different elements that occur

exactly m times in a sample of size TV:
Definition 6 V(m, N) : The number of different morphemes or morpheme

types which occur m times in a sample of size N.
The expectation of V(m, N) is:
Finally, let V(N) indicate the total number of different morphemes in

a sample of size N:
Definition 7 V(N) : The number of different morphemes or morpheme
types in the sample of size N.
The expectation of V(N) can be expressed as the sum of all the ele
ment types that occur at least once, thus:
Alternatively, E[V(N)] can be defined as the total number of events

in the population, S, minus the number of events that do not occur in the
sample of size N. Thus, using the probability that an event wi does not
occur in the sample of size TV, i.e.
= (1-Pi)N,
E[V(N)] can be expressed as follows:
Ideally, if we knew or could estimate from the data the number of S

types in the population and the probability p 1 , p2 ... ps of each element
type w1,w2,... ws, a distribution model of morphemes could be completed.
We would then be able to estimate the number of morpheme types in the
data of arbitrary size N. This is the basic framework, the binomial model,
for the distribution and growth of morphemes in terminology in particular
and for words in language in general.
7.2.2 Incompleteness of the sample

The parameter to be estimated in the above framework is the population
probability pi of each event (morpheme) wi, if we take the ordinary as
sumption that we know the population number of events; we can obtain the
distribution model of terminology population if we estimate pi for all wi
(i = 1,2, ..., S).
According to ordinary statistics, the probability pi for the event wi
in binomial distribution can be estimated from the data using the sample
relative frequency:
which constitutes the unbiased estimator for the binomial model (Minoya
1994).
For language data, however, it is widely known that this estimation is
not optimal (Baayen 2001; Chitashvili & Baayen 1993; Kita 1999; Man
ning & Schütze 1999). This problem arises because it is nearly always
expected that there are events (or morphemes in the present case) which do
not appear in a given sample. As long as we see the data as a sample of the
population, a view which we naturally have come to adopt in dealing with
the concept of dynamics, the data is incomplete in the sense that there are
events that exist or may come to exist but do not appear in the given sample.
In order to assess the degree of the incompleteness of data, Chitashvili
& Baayen (1993) introduces a measure called coefficient of loss (CL). This
measure calculates the ratio of the number of events which are lost by esti
mating the number of events in a sample equal in size to the original, using
the sample relative frequencies as estimations of the population probabili
ties, based on the binomial model2 :
Kageura (1998a) has shown that the coefficient of loss of the English and
Japanese terminological data of the four domains, i.e. computer science,
2
The reason why the view adopted in CL, i.e. collapsing the distinction of the events
and focusing only on the number of events, can be justified is examined in Kageura (2000).
agriculture, psychology and physics, are all greater than 20 per cent. This
indicates that the terminological data is statistically incomplete and, thus,
the estimation of population probabilities by sample relative frequencies is
not valid. Note that, from the point of view of the dynamics of terminol
ogy, this incompleteness is an opportunity, not a deficiency, because the
completeness of a given terminological sample means that the dynamics of
terminology is already exhausted and that the terminology is dead.
7.2.3 Binomial interpolation and extrapolation

To overcome the problem of parameter estimation, Good (1953) proposes a
method of modifying the probability allocations in order to leave room for
the unseen events to have non-zero probabilities. This is what is commonly
called Good-Turing estimates (Baayen 2001; Church & Gale 1991; Gale
& Sampson 1995; Kita 1999). However, the method cannot be used in the
present case, because there is no way of knowing how many unseen events
there are in the population.
Good & Toulmin (1956) proposes a method of observing how, given
the original data, the number of events increases or decreases when the
sample size is changed, i.e. the method of interpolation and extrapolation,
conditioned on a given sample. The method fits excellently the present
situation, because the interpretation of the methodological framework in
the application to terminology corresponds exactly to the basic problem we
are concerned with, which was formulated in 7.1.
Let us introduce binomial interpolation and extrapolation formally.
First, let TV be the size of the original sample. In the case of termino
logical data, this is given by the token number of morphemes in the data,
as defined above. Let us think of data, conditioned on the original (which
roughly means that they share the same probabilistic structure), whose size
is λN for the arbitrary real number λ(> 0).
The expectation of the number of different morphemes that occur m
times in the second sample of size λN, E[V(m, λN)], is defined as:
where E[V(m + k,N)] indicates the expectation of the number of mor

pheme types that occur m +  times in the original sample of size TV.
The expectation of the number of all the different morphemes,

E[V(λN)], on the other hand, is defined as follows:
If we allow ourselves to estimate E[V{k, N)} and E[V(N)] by
Ê[V(k,:N)] = V(k,N)
and
Ê[V(N)] = V(N)
respectively, it is possible to calculate the number of morphemes and the

number of morphemes that occur m times in the sample of size XN.
On the basis of the binomial model and under the assumption that the
population is known, equations (7.3) and (7.4) can be derived from equation
(7.1) applied to the sample of size XN (Good & Toulmin 1956):
If we only use the range m +  ≤ N and  ≤ (λ —1)N for actual calcula

tion, the term for combinatorics in the last line can be reduced to:
which leads to equation (7.3). Equation (7.4) can be derived immediately

from equation (7.3).
Using formulae (7.3) and (7.4), we can calculate the number of differ
ent morphemes for data of any size, by using the quantitative structure of
the morphemes in the given sample (see Chapter 8 for concrete examples
of application). We can also calculate the number of different conceptual
specification patterns within a certain category when the number of terms is
increased on the basis of the quantitative structure of the conceptual specifi
cation patterns of that category. So binomial interpolation and extrapolation
offers a powerful means to observe the dynamics of terminological growth.
Due to a technical problem concerning numerical computation, the ac
tual estimation does not work for a size of more than twice the original
sample size; this is because of the terms (λ — 1)i in equations (7.3) and
(7.4). Sometimes, when the distribution of items in the given data is sparse,
the value even starts to fluctuate at around 1.5 times the original data size.
In Chapter 9, where the growth patterns of terms for each conceptual cate
gory are described, we will observe this problem in a few conceptual cat
egories. This problem can be avoided by using further assumptions on the
distributional nature of events3. These will not, however, be adopted here,
as the basic interpretative framework does not change even if we use these
methods.
7.2.4 The growth rate

Equations (7.3) and (7.4) give the growth of the events that occur m times
and as a whole, respectively. It is convenient here, for later discussion, to
introduce the concept of growth rate, i.e. the rate at which the number of
3
Baayen (2001) gives an accessible explanation of these methods; for the application of
these methods to terminology, see Kageura (1998a).
events (i.e. number of morpheme types in the present situation) increases,

after sampling N tokens. The growth rate is mathematically given as the
first derivative of E[V(N)]. The explanation below is based on Baayen
(2001).
To derive the formula for the growth rate, we must make two prepa
rations. Firstly, let us define structural type distribution, which gives the
number of population events whose probabilities are greater than or equal
to a given probability p:
As before, S indicates the number of population events. Also,
The function G{p) is a step function whose value jumps at  such that at
least one event takes the population probability p. Let pj be the population
probability that is assigned to at least one event in the population, where the
subscript is assigned in ascending order of probability, from the smallest
value to the largest. Then the value of the jump at the probability pj is
given by:
AG(pj)=G(pj)-G(pj+1).
Secondly, we rewrite the equation given in (7.2) using the Poisson ap
proximation to the binomial distribution:
where e is the base of the natural logarithm,

Noting that
when m = 0, E[V(N)] can be rewritten as follows:

Rewriting the equation (7.6) by using (7.5), we obtain the integral form
for the expression of E[V(N)]:
where dG(p) = ∆G(p) around pj, and 0 otherwise.

As E[V(N)] defines the value of V(N) with respect to N, i.e. the
growth of events or morphemes, the first derivative of E[V(N)] gives the
growth rate P(N):
The penultimate step uses the fact that E[V(1, N)] = Npe-Np according
to the Poisson distribution.
In the actual application, further clarification of the concept of growth
rate is needed; this will be given in 8.2.3.1. It is interesting to observe that
the growth rate of the morphemes at sample size N is given by the estimated
number of morphemes that occur only once divided by the sample size
N. If we approximate E[V(1,N)] by V(1,N), the growth rate can be
calculated on the basis of the given data. Incidentally, the growth rate is
equal to the probability mass assigned to the unseen events by the Good-
Turing estimates (Good 1953).
7.3 General conditions for the application
In the process of defining the mathematical model of binomial interpreta

tion and extrapolation, a few conditions were assumed. For the model to
be appropriate for the framework of interpretation, these conditions should
be examined in relation to the characteristics of the phenomena to which
the model is applied. Though the individual matters to be considered will
be discussed in the following chapters, where the concrete analyses are car
ried out, it is useful to give general illustrations of how these conditions are
related to the terminological phenomena.
7.3.1 Application to the growth of morphemes

Suppose that we are interested in estimating the pattern of use of new mor
phemes in forming terms when a terminology grows, e.g. when the size
of a terminology becomes twice as big as the original sample data, how
many new morphemes will be used in the terminology? This is the same as
the general question described in 7.1. Technically, binomial interpolation
and extrapolation can be straightforwardly applied to give an answer to this
question. But what kind of conditions does the model impose on the per
ception of the terminological phenomena, as a result of the mathematical
conditions adopted in the establishment of the model?
There are two important aspects to be considered here. Firstly, the
mathematical model is constructed over the distribution of morphemes, by
passing the level of individual terms. This is clearly shown by the fact that
the size of a terminology is defined by the token number of morphemes and
not by the number of terms, as we saw in the previous section. In other
words, a terminology is regarded as a bag of morphemes. Secondly, the
occurrence of morphemes is assumed to be random.
The simplest way to overcome the first problem and recover the level
of "term" is to transform the size of the data given by the token number
of morphemes by means of the average length of terms, as calculated by
the number of morphemes per term in the original data. Applying this idea
means that we accept that the average length of terms as counted by the
number of morphemes remains constant even if the size of the terminology
changes. Formally, it means that, in the relation between the number of
terms  and the token number of morphemes N, i.e.
N = α,
the condition that α, the average length of a term as counted by the number
of morphemes, is constant, irrespective of the size of the terminology. The
actual value of α is obtained by N/T of the original data.
This effectively assumes that the ratio of terms of each length does not
change. Therefore, for instance, if our original terminological sample con
sists of 20 terms, of which 10 are simple terms, 5 are terms with two con
stituent morphemes, 3 are terms with three morphemes and 2 are terms with
four morphemes, then we assume that, when the sample becomes twice as
big and contains 40 terms, it will consist of 20 simple terms, 10 terms with
two morphemes, 6 terms with three morphemes and 4 terms with four mor
phemes.
This is an obvious oversimplification of the actual terminological phe
nomena. It is intuitively more natural to assume that the ratio of longer
terms increases when the terminology grows (cf. Baayen 2001). It is, how
ever, not a critical oversimplification with respect to the structural assump
tion that was adopted in Part I (see especially 2.2.2)4.
Note that this simplification only becomes apparent because a rigid
mathematical model is adopted, while in Part II (and, in fact, in many
qualitative studies of terminology) this problem simply remains unnoticed,
though it has always existed. In that respect, the simplification introduced
by assuming α to be constant is no worse than many of the implicit and
unnoticed simplifications adopted in existing studies.
The second problem, the randomness assumption, is a well-known
problem in lexical statistics in general (Baayen 2001). In the present case,
assuming the randomness of the occurrence of morphemes means ignoring
two factors in terminology. The first is related to the distinction between a
core and a non-core set of terms or morphemes within the terminology of a
4
As will be shown in Chapter 9, the actual analysis sometimes leads to results that con
tradict the assumption that the distribution of the length of terms does not change. Within
the overall assumption adopted in the present study, it is technically possible to introduce
another model which traces the pattern of changes of the average length of terms and the
ratio of simple and complex terms according to the changes in sample size. In that sense,
the assumption adopted here is still a simplification even under the structural assumption.
We nevertheless do not incorporate the more complex model because the distribution of
word-length is a separate research matter that needs thorough examination as a separate
research topic.
domain and within each conceptual category. The second is related to the
dependency of morphemes within individual terms.
The distinction between core and non-core terms or morphemes has
not been taken into account in the present study. We have assumed, not
unreasonably, that the data constitutes a representative sample of a termi
nology. In that sense, as long as we do not try to draw any conclusions
about the centrality and/or non-centrality of terms or morphemes, omitting
the distinction of core and non-core terms and morphemes does not cause
any trouble concerning the consistency of the present study5.
The dependency of morphemes in individual terms is an aspect that has
so far been maintained and taken into account by means of the distinction
between nucleus and determinant, as well as the distinction between term-
forming and embedded combinations. The binomial model ignores this.
Can this be justified? Based on an experiment, Kageura (1998a) showed
that disregarding the intra-term dependency of morphemes does not affect
the accuracy of the quantitative model. His experiment — which is based
on the same idea as Baayen (1996), in which the lexical dependency in texts
is examined — compares the growth patterns of morphemes calculated by
the random permutation of terms in a terminology and by the random per
mutation of morphemes disregarding the unit of terms. No statistically sig
nificant difference was observed for both the English and Japanese termi
nologies of the four domains, i.e. computer science, agriculture, physics
and psychology6. In general, therefore, as long as we can disregard the
distinction or grade of central and non-central sets of terms in terminology,
it is safe to disregard the intra-term coherence of morphemes.
7.3.2 Application to conceptual specification patterns

Let us now suppose that we apply binomial interpolation and extrapolation
to conceptual specification patterns. An intuitively obvious example was
5
In the study and practice of terminology in general, this distinction seems to have been
dealt with individually, when, for instance, the editors of a terminological dictionary have
to choose a limited number of terms to be listed in the dictionary. The very meaning of the
concept of the centrality and non-centrality of terms, though frequently referred to, has not
yet been well examined.
6
Recall that the morphemes defined for Japanese terms are somewhat closer to words
in English. In the experiment for English terms, Kageura (1998a) used lemmatised words
instead of "morphemes".
given in 7.1. In Chapter 6, we observed that a total of 12 specification

patterns were used in the formation of complex terms representing RE311
(complex documentation units as representational entities). When the data
becomes twice the size of the original, for instance, is there any possibility
that new specification patterns are used?
In this case, the mathematical model is applied to the distribution of
the conceptual specification patterns, i.e. N represents the token number of
specification patterns in the category (thus equalling the number of complex
terms), while V(N) represents the number of types of conceptual specifi
cation patterns. Here the randomness of the occurrence of conceptual spec
ification patterns is assumed.
This randomness assumption is related to two aspects. Firstly, the ran
domness assumption disregards the distinction between centrality and non-
centrality among the complex terms involved. This is on a parallel with
the distinction of the centrality and non-centrality of terms or morphemes
discussed above. We do not take any external factors into account that may
lead us to conclude that a certain type of complex term is more central than
another in the sense that terms that belong to this class grow faster than
others7. This simplification is in accordance with the overall assumption
we set out in 2.2.
Secondly, depending on how the model is applied, the randomness as
sumption may disregard the distinction of the level of specification within
the overall conceptual system. If, for instance, the model is straightfor
wardly applied to the specification patterns of a broad conceptual category
in which the same specification patterns are used at two different levels,
then the application involves this type of simplification. Note this second
problem exists also in the analysis of morphemes in relation to their posi
tion (such as nucleus or determinant) in terms.
It is theoretically possible to avoid this by partitioning the data further
so that the same specification at different levels can be distinguished. This
kind of operation, however, may cause a technical problem by reducing
the size of the data. In general, in applying the quantitative model, it is
necessary to consider its statistical reliability as well as the reliability of
its underlying assumptions. Though statistical reliability is not discussed
in the present study, it is important to emphasise that, in general, we need
7
On the contrary, we claim that some types of complex terms are central among existing
terms; the very observation pursued in Chapter 6.
data of reasonable size for the quantitative approach to work properly. The
actual analyses to be carried out in the following two chapters are subject
to this constraint as well.
Chapter 8
Growth Patterns of Morphemes in

the Terminology of Documentation
As a preliminary step to complementing the observation of conceptual pat

terns of term formation, we examine the morphological-conceptual dynam
ics as manifested by the growth patterns of morphemes representing certain
conceptual categories in the terminology of documentation. As such, this
chapter corresponds to and complements Chapter 4. In the process, this
chapter also introduces some concepts which will be used in Chapter 9, the
chapter complementing Chapter 6 and devoted to the quantitative observa
tion of the dynamics of terminology. Due to the problem of size, it it not
possible to carry out quantitative analyses of morphemes for all the concep
tual categories identified in Chapter 4. Here, instead, we first observe the
growth patterns of morphemes according to the broad conceptual categories
and then the growth patterns of morphemes for selective subcategories. We
start by examining the basic premises for applying the quantitative model
to the terminology of documentation.
8.1 Some premises of the analysis
8.1.1 Concepts and individual morphemes

In this chapter, the basic unit of analysis is the individual morpheme used
in the terminology. It is useful first to explain how the analysis of individ
ual morphemes complements the description of conceptual patterns of term
formation. This is because sometimes it seems to be assumed that analyses
based on individual linguistic elements fall short of conceptual analyses.
This attitude is most typically held by the studies in which concepts are
claimed to precede terms in terminology (Felber 1984; Lara 1998/1999).
It is in fact straightforward to relate individual morphemes to concepts;
simply declaring that individual morphemes represent corresponding indi
vidual concepts is sufficient and theoretically valid. Formally, by indicating
concepts with < > and lexical representations with italics, we can define
two one-to-one functions, one constituting the reverse of the other, between
a concept and a morpheme. This is illustrated in the following (with "infor
mation" used as an example):
< information > = fc (information)
information — fr(< information >)
where fc is the function that maps a lexical representation to a concept and
fr is the function that maps a concept to a lexical representation.
Theoretically, this one-to-one mapping appears at the final stage of
the detailed conceptual analysis of linguistic items. If one tries to obtain
the conceptual category that can distinguish every different linguistic phe
nomenon, then the concepts indicated by different linguistic representations
should all be distinguished, simply because they take different linguistic
representations. The concepts thus established can be referred to straight
forwardly by the corresponding linguistic representations using the func
tion fc. It is only when some sort of generalisation is important, as in the
analyses carried out in Part II, that concepts or conceptual categories that
do not correspond one-to-one to linguistic items are required. Because we
aim here to carry out an analysis of the detailed patterns of terminologi
cal growth which complement the generalised descriptions of conceptual
patterns, it is natural to carry out the analyses at the most detailed level,
where individual linguistic items have a one-to-one correspondence with
concepts. Thus, the morphological dynamics is assumed to represent the
conceptual dynamics at the finest level.
The only problems are synonymy and polysemy. The problem of pol
ysemy can be avoided by proper conceptual analysis, and we have already
clarified the morphemes to which more than one conceptual category is al
located in Chapter 4. In the present analysis, which uses the conceptual
categories established in Chapter 4, therefore, polysemy does not cause a
problem. The issue of synonymy has not been properly addressed in the
present study. It might be possible to bypass this problem by simply claim-
Growth Patterns of Morphemes 185
ing that there is in reality no complete or absolute synonymy (Ullmann

1962; Lyons 1977; Sparck Jones 1986; Cruse 1986) and to insist that dif
ferent linguistic representations represent different concepts. However, this
claim requires further clarification concerning what type of concepts we are
talking about. Here, instead of making this claim, we assume that the num
ber of synonymous morphemes is very small and that the effect caused by
synonym can be regarded as statistically insignificant. This is a reasonable
assumption as terminology tends towards clear representation of concepts.
8.1.2 The quantitative status of the data

In the Introduction, we gave a quantitative summary of the data of 1228
documentation terms in Wersig & Neveling (1984). The ratio of simple
terms, terms with two morphemes, etc. were given in Table 1 on page 5.
Table 2 on page 5 gave the basic number of morphemes. As the number
of morpheme tokens in the data was 2,668, the average length of a term, as
counted by the number of morphemes, is:
By allocating conceptual categories to each morpheme in Chapter 4, we

obtain a total of 845 different morphemes (see 4.3.1), among which 10
morpheme types are conceptually indeterminate. The number of morpheme
types and tokens for each broad category were given in Tables 4.10 to 4.13.
Using the notation introduced in the previous chapter, the quantities of
the morphemes in the data are expressed as follows:
N = 2668,
V(N) = 845.
Table 8.1 shows V(m. N), the number of morphemes that occur m
times in the data. More than half of the morpheme types occur only once.
The growth rate P(N), calculated by letting E[V(1,N)] be approximated
by V(l,N),is:
Table 8.1. V(m, N) and m in the data of documentation terms.

m V(m,N) m V(m,N) m V(m,N) m V(m,N)
1 All 10 6 19 4 33
2 164 11 2 20 2 39
3 54 12 3 23 3 44
4 41 13 8 25 1 48
5 18 14 2 26 շ 58
6 11 15 2 27 1 99
7 13 16 2 30 1
8 8 17 4 31 2
9 11 18 1 32 1
This means that a new morpheme token used to construct a new term can
be a new type with the probability of 0.177. The coefficient of loss of the
data is:
CL = 0.236.
The values of V(N) and CL clearly indicate that the data does not exhaust
the potentiality of terminology as seen from the angle of the distribution of
morphemes, i.e. there is room for the terminology to grow with the incor
poration of new morphemes.
8.1.3 The mathematical assumptions and the data

Let us turn now to the examination of the validity of the randomness as
sumption. As mentioned in the previous chapter, we assume that there is no
difference among terms in the data with respect to their centrality in the ter
minology. What should be empirically validated is whether the intra-term
dependency of morphemes significantly affects the randomness assumption
of the occurrence of morphemes.
To observe this, 1,000 random permutations were applied both at the
level of terms and at the level of morphemes, and, for each, the mean num
ber of different morphemes was calculated for 20 equally-spaced intervals
in the original data, i.e. at 1/20, 2/20, etc. of the original data size. In
this way, as mentioned in the previous chapter, the growth curves with and
without the effect of the intra-term coherence of morphemes were obtained.
The left panel of Figure 8.1 shows the growth curve calculated on the
basis of 1,000 permutations of terms and morphemes. The dots indicate
Figure 8.1. The effect of the randomness assumption.
the result of term-level permutation, and the solid line indicates the result
of morpheme-level permutation. No noticeable difference can be observed
between these two. The difference is well within the 95 per cent inter
val indicated by the dashed lines. The right panel of Figure 8.1 shows
the discrepancy of the numbers of morpheme types between the two for
20 equally-spaced intervals, i.e. the number of morphemes calculated by
morpheme-level permutations subtracted by the number of morphemes cal
culated by term-level permutations. The difference is at most one mor
pheme type. Thus, we can conclude that the randomness assumption of the
distribution of morphemes will not affect the conclusions drawn from the
mathematical model.
As mentioned in 7.3, we assume that the average length of a term, i.e.
α = N/T = 2.17, is constant, irrespective of the size N. The ratio of
simple terms, terms with two morphemes, etc. is assumed to be constant as
well. Thus, the observation made on the basis of TV can always be trans
ferred to the observation on the basis of the number of terms. In addition,
in the following observations, we assume that the ratio of morpheme tokens
representing each conceptual category is constant irrespective of N.
8.2 The dynamics of morphemes by broad conceptual categories
Table 8.2 shows the quantitative nature of morphemes belonging to the

seven broad conceptual categories, i.e. material entity (ME), representa
tional entity (RE), abstract entity (AE), classificatory entity (CE), quality
(QL), relation (RE) and activity (AC). There are 10 types and 11 tokens of
conceptually-indeterminate morphemes, so the total of morphemes belong
ing to the seven conceptual categories is smaller than the total number of
the morphemes. The category of classificatory entities has the lowest val
ues for both P(N) and CL, which implies that it is relatively less likely to
encounter a new morpheme representing classificatory entity concepts.
Table 8.2. Quantitative nature of morphemes for broad categories.

Category V(N) N V(1,N) P(N) cL
ME 143 487 89 0.183 0.254
RE 117 456 62 0.136 0.219
AE 72 331 38 0.115 0.230
CE 35 158 14 0.089 0.177
QL 146 314 92 0.293 0.255
RL 132 322 62 0.193 0.238
AC 190 589 97 0.165 0.223
8.2.1 The growth of morphemes

Assuming that both the ratio of the number of morpheme tokens repre
senting each conceptual category and the length of terms remain constant,
we can compare the changes in the number of morpheme types in accor
dance with the growth of a terminology. Applying binomial interpolation
and extrapolation to the distribution of morphemes for each broad concep
tual category, and plotting them as a function of the number of terms 
(= N/2.17), we obtain Figures 8.2 and 8.3. In Figure 8.2, the solid vertical
line shows the point of the original data (N = 2, 668;  = 1, 228) and the
dashed vertical line shows the point where  = 614, i.e. half the original
data size. The data was extrapolated to 1.5 times the original. Figure 8.3
magnifies the growth of morphemes up to half the original sample size in
Figure 8.2.
Figure 8.2. The growth curves of morphemes for broad conceptual categories.
Figure 8.3. The growth curves of morphemes for broad conceptual categories (part).
Table 8.3. The ratios of morphemes for broad conceptual categories

0.25T 0.5T 0.75T  1.25T 1.5T
ME 16.6 16.6 16.8 17.1 17.4 17.8
RE 15.3 14.5 14.2 14.0 14.0 14.0
AE 8.7 8.7 8.7 8.6 8.6 8.5
CE 5.0 4.7 4.4 4.2 4.1 4.0
QL 16.1 16.9 17.3 17.5 17.8 18.3
RL 14.9 15.5 15.8 15.8 15.7 15.4
AC 23.5 23.2 23.0 22.8 22.5 22.1
At the beginning, the number of material entity (ME) morphemes is

bigger than that of quality (QL) morphemes, but this reverses when the size
of the terminology sample is around 400. At the very beginning, the num
ber of representational entity (RE) morphemes is greater than that of both
QL and relation (RL) morphemes, but soon the number of RE morphemes
becomes smaller. At the right hand side of Figure 8.2, where the size of the
terminology becomes bigger, the growth curve of classificatory entity (CE)
morphemes flattens. It is also observed that the curves of activity (AC), re
lation (RL) and abstract entity (AE) morphemes become gentler compared
to the curves of QL, ME and RE morphemes.
Table 8.3 shows the change in the ratio of the numbers of morphemes
for the seven conceptual categories at six observation points, i.e. 0.25,
0.5, 0.75T, , 1.25 and 1.5T1. Though subtle, the ratios of RE (repre
sentational entity), AE (abstract entity), CE (classificatory entity) and AC
(activity) morphemes become smaller as the terminology grows, while the
ratios of ME (material entity) and QL (quality) morphemes grow steadily.
The ratio of RL morphemes first increases and then starts decreasing.
8.2.2 The average use of morphemes

We have observed the differences in the growth patterns of morphemes for
different conceptual categories. However, the number of morpheme tokens
belonging to these conceptual categories differs from category to category.
Thus, it is important to complement the above observation with some mea
sures which take into account the number of tokens. The average number
1
As the ratio was calculated by binomial interpolation with random permutation, the
ratios of categories at the data point  are not exactly the same as those in Table 4.10.
Figure 8.4. The average use of morphemes for broad conceptual categories.
of use N/V(N) is one such measure; it reveals the degree of the repeated
use of morphemes.
Figure 8.4 illustrates the changes in the average use of morphemes
for each broad conceptual category, plotted against T. The average use
of morphemes representing AE (abstract entities) is the highest when the
size of the terminology is small. When the size becomes bigger, the aver
age use of CE (classificatory entity) morphemes rises above that of abstract
entity morphemes. Through the repeated use of morphemes, abstract en
tity and classificatory entity concepts contribute to establishing connections
among terms in the terminology. This observation also matches our general
understanding of the nature of these conceptual categories: The stores of
highly-abstract concepts such as abstract entities and classificatory entities
are inherently small, and this small number of concepts can be used, in
combination with other concepts, in a variety of situations.
On the other hand, the average number of the use of both QL (qual
ity) and RL (relation) morphemes remains low. This corresponds to the
interpretation that these concepts are used for the differentiation of related
concepts in term formation. The categories of RE (representational en

tity), ME (material entity) and AC (activity) are located between these two
groups. This is also intuitively reasonable, though in a negative way, as
these categories are rather more substantive than abstract and classificatory
entities, and they do not seem to have differentiation as their main role.
8.2.3 The growth rate

8.2.3.1 Two definitions of growth rate
Let us now turn to observe the growth rate, P(N), of morphemes for the
broad conceptual categories. Before the analyses, two different notions of
growth rate should be defined. The first is the growth rate of morphemes as
observed in the morpheme depository of each conceptual category. Letting
NCAT refer to the number of morpheme tokens in the broad conceptual
category CAT (e.g. ME, RE,...), we define the growth rate as:
For instance, the growth rate of ME (material entity) morphemes calculated

according to the definition (8.1) is (see Table 8.2):
Pm(N CAT ) measures the conditional probability, i.e. the probability that
the token is a new type of CAT when a new token is CAT. From the
point of view of constructing terminology, therefore, P m {N C A T ) can be
interpreted as indicating the degree of expectation of a new morpheme type
when the particular category CAT is called upon. Note that P(N) shown
in Table 8.2 is in fact P m (N C A T )·
Alternatively, we can define the growth rate of morphemes of individ
ual categories vis-à-vis the size of the terminology:
The growth rate of ME morphemes according to this definition is:

Figure 8.5. The growth rate of morphemes for broad conceptual categories.
Pt(N CAT ) shows the unconditional probability; it measures the probabil

ity that the new morpheme token represents CAT and is a new type. Thus,
the value of P t ( N M E ) , i.e.0.033, represents the probability that a new mor
pheme token employed to construct a new term is a new type and represents
a material entity concept. P t (N CAT ) can also be understood as the contri
bution of conceptual categories to the new types.
In the present context, where the focus is on the dynamics of mor
phemes belonging to various conceptual categories within the terminology
of documentation, it is natural to assume that the use of morphemes is es
sentially regulated by the structure of the terminology. Thus the values of
Pm(NCAT) and Pt(NCAT) for different categories should be compared on
the basis of the overall size of the terminology.
8.2.3.2 The growth rate of morphemes

Figure 8.5 shows the values of P m (N) and P t (N), respectively, extrapo
lated up to 1.5 times the original data size for the seven broad conceptual
categories. What was implied in relation to Figures 8.2 and 8.3 can be con
firmed more precisely from the right panel of Figure 8.5. When the size of
the terminology is small, morphemes representing AC (activity) concepts
grow faster than the morphemes representing other concepts. Activity con
cepts are followed by morphemes representing ME (material entities) and
RE (representational entities). As the terminology grows, the growth rate

of morphemes representing AC rapidly decreases, becoming smaller than
the growth rates of ME and QL (quality). When the terminology grows
even further, the growth rate of AC morphemes is also expected to become
smaller than the growth rate of RE morphemes.
The growth rate of morphemes representing relations (RL) also de
creases fast, and does not stop decreasing when the terminology grows. To
wards the end of the range of observation, the decrease in the growth rates
of morphemes representing qualities (QL), material entities (ME), repre
sentational entities (RL), classificatory entities (CE) and (though to a lesser
extent) abstract entities (AE) slows down, while the growth rate of mor
phemes representing activities (AC) and relations (RL) keeps decreasing.
This implies that, if the size of the terminology becomes much bigger, the
number of different AC morphemes will become smaller than the number
of QL and ME morphemes, and perhaps that of RE morphemes as well.
The left panel of Figure 8.5 shows a pattern similar to that of the right
panel, with a few important differences. Firstly, the growth rate of mor
phemes representing qualities (QL) is much higher than that of material en
tity (ME) morphemes. This indicates that, when quality concepts are called
on by the terminology, new concepts are much in demand, whereas when
material entity concepts are required, the likelihood that new concepts are
needed by the terminology is much lower.
Secondly, the growth rate of relation (RL) morphemes is consistently
higher than that of activity (AC) concepts. In fact, the growth rate of re
lation morphemes is similar to that of quality (QL) morphemes when the
size of the terminology is small; these two categories have a similar role in
the earlier-stage terminology construction. When a terminology becomes
larger, however, the role of relation concepts becomes more like that of ab
stract entity (AE) or classificatory entity (CE) concepts. This corresponds
well to the nature of the conceptual category "relation" examined in Chap
ter 4.
8.3 The dynamics of morphemes by subcategories
The seven broad categories have considerable internal diversity, as can be

seen from the subcategories defined in Chapter 4. It is therefore necessary
to observe the growth patterns at the level of subcategories in order to clar-
ify the characteristic dynamics of morphemes. In the growth patterns of

morphemes at the level of subcategories, we only observe the average use
of a morpheme and the growth rate Pm(NCAT) for subcategories. This is
because the number of morpheme types and Pt(NCAT) depend much on
the number of tokens and, thus, it is not very revealing to compare these
among categories whose token sizes differ greatly.
As a compromise for the factor of quantity and the conceptual coher
ence of the subcategories, we have chosen the following middle-level con
ceptual subcategories for observation. Material entity concepts (ME) are
divided into three, i.e. animate entities (ME1), information carriers (ME21)
and others (ME2x). Representational entity concepts (RE) are divided into
four, i.e. broad representational entities (RE1), linguistic entities (RE2),
documentation entities (RE3) and software entities (RE4). Abstract entity
concepts (AE) are divided into three: information entities (AE1), systems
of knowledge (AE2) and subjective entities (AE3). Classificatory entity
concepts (CE), due to their size and nature, are not divided. Quality con
cepts (QL) and relation concepts (RL) are each divided into "labels" (QL1
and RL1) and "values" (QL2 and RL2). Activity concepts (AC) are divided
into units of activity (AC1), action (AC21), transference (AC22), produc
tion (AC23) and state change (AC24).
Table 8.4 shows the basic quantities of each subcategory. The size in
terms of tokens as well as the growth rate and the coefficient of loss differ
greatly among the subcategories even within the same broad categories. A
few subcategories, i.e. RE4, AE3 and AC1, may be too small for quantita
tive analyses to be properly carried out. Thus, care should be taken when
these categories are observed.
Figure 8.6 shows the average use and the growth rate Pm(NCAT) of the
subcategories of material entity, representational entity, abstract entity and
classificatory entity concepts, plotted up to 1.5 times the original data size
(for convenience, classificatory entity morphemes are plotted in the same
panel as abstract entities). To help visual observation, the same absolute
scale is adopted for the -axis.
The three subcategories of material entity concepts show considerable
variance. The average use of morphemes representing animate concepts
(ME1) is consistently higher for all ranges of observation. The growth
rate is correspondingly consistently lower. The values of average use and
growth rate for the morphemes representing information carriers keep the
Table 8.4. Basic quantities of subcategories.

Subcategory V(N) N V(1,N) V(N) cL
ME1 26 107 16 0.150 0.248
ME21 80 284 50 0.176 0.252
ME2x 37 96 23 0.240 0.259
RE1 25 98 13 0.133 0.213
RE2 20 82 10 0.122 0.208
RE3 59 247 32 0.130 0.219
RE4 13 29 7 0.241 0.234
ΑΕ1 23 127 11 0.087 0.214
AE2 28 168 16 0.095 0.233
 21 36 11 0.306 0.236
CE 35 158 14 0.089 0.177
QLl 36 100 22 0.293 0.237
QL2 110 214 70 0.293 0.261
RL1 30 102 18 0.193 0.245
RL2 102 220 53 0.193 0.235
AC1 14 31 8 0.165 0.238
AC21 65 256 31 0.121 0.203
AC22 25 61 15 0.246 0.249
AC23 36 115 14 0.122 0.181
AC24 50 126 29 0.230 0.255
middle position. Material entity concepts other than these two show low av
erage use and a high growth rate. Towards the right-hand end of the obser
vation range, we see that the growth rate of information carriers is expected
to become larger than that of the morphemes representing non-information
carriers. This implies that, among material entity concepts, the bigger the
terminology becomes, the higher is the demand for new morphemes for
information carriers.
Three of the four subcategories of representational entity concepts, i.e.
broad representational entities (RE1), linguistic entities (RE2) and docu
mentation entities (RE3), are very close to each other, both in average use
and in growth rate. Despite the different token sizes (RE3 is twice to three
times bigger), the degree of repetition and of new morphemes required in
the construction of the terminology is similar. The morphemes representing
software entities (RE4) show a somewhat unstable transition, perhaps be
cause of the size of the data (but we can still rely on the overall tendencies).
It is notable that when the terminology becomes bigger, the decrease in
the growth rate of documentation entities (RE3) becomes much slower than
Figure 8.6. Average use and growth rate of subcategories for ME, RE, AE and CE.
that of the others. This corresponds to what we observed in the transition of

the growth rate of information carriers (ME21), which also becomes flatter
compared to other categories when the size of the terminology becomes
larger. This may be because these two are "core" concept classes in the field
of documentation and they maintain room for incorporating new concepts
represented by new morphemes, even when the size of the terminology
becomes very large.
Among the subcategories of abstract entities and classificatory entities,
information entities (AE1), system of knowledge (AE2) and classificatory
entities (CE) show similar patterns. Their average use is very high (in fact
much higher than those of all the other subcategories of all the other broad
categories) and their growth rates are very low (lower than those of most
subcategories). These three categories share the characteristics of abstract
entity and classificatory entity concepts observed in 8.2.1. Only the mor
phemes representing subjective entities (AE3), whose transitions are unsta
ble perhaps due to the size of the data, show a different pattern.
Figure 8.7. Average use and growth rate of subcategories for QL, RL and AC.
Figure 8.7 shows the average use and the growth rate Pm(NCAT) of the
subcategories of the quality, relation and activity concepts. As in Figure
8.6, they are plotted up to 1.5 times the original data size, with the same
absolute scale.
Morphemes representing quality and relation concepts show very sim
ilar internal patterns, though their absolute values differ. The average use of
"labels" (QL1 and RL1) is consistently higher than that of "values" (QL2
and RL2). The growth rates of "values" are higher at the beginning, but
they quickly decrease and come closer to, or even lower than, the growth
rates of "labels". The patterns confirm that the qualitative difference be
tween labels and values is reflected in the quantitative patterns of the use of
the morphemes in terminology construction.
The difference between quality concepts and relation concepts also be
comes clear, especially in the right-hand end of the transition patterns of
the growth rates. In the case of quality concepts, the decrease in the growth
rates slows down for both QL1 and QL2. In fact, the curve of the growth
rate of QL1 is nearly flat towards the end of the observation range. In
the case of RL, on the other hand, the decrease in the growth rate of RL1
does not seem to slow down at all, while that of RL2 does slow down.
Thus the phenomenon observed in 8.2.3.2 seems to come mainly from RL1
morphemes. This corresponds to our qualitative observation of the charac
teristics of relations, i.e. some relation concepts, especially in RL1, have a
similarity with abstract entity concepts.
The subcategories of activity concepts show a certain degree of diver
sity. Action (AC21) and production (AC23) concepts are somewhat closer
in terms of their patterns of average use, while the other three subcate
gories show similar transition patterns. This also holds for the patterns of
the growth rates when the size of the terminology is not very big. Un
fortunately, we cannot readily see a qualitative explanation behind these
patterns.
Chapter 9
Quantitative Dynamics in Term Formation
In this chapter, the growth patterns of various elements of term formation

are quantitatively analysed and described. These observations complement
the descriptions of conceptual patterns presented in Chapter 6 in two ways:
(i) They extend the general descriptions of conceptual patterns by taking
into account the distribution of individual concepts represented by mor
phemes, and (ii) they explore the dynamic aspects of term formation pat
terns by introducing dynamic quantitative analyses. In the following, the
basic setup of the analyses is first clarified; then the results of the analyses
are presented. Like Chapter 6, this chapter contains detailed descriptions
of term formation specific to the terminology of documentation.
9.1 Basic setup of the analyses
9.1.1 Conceptual categories under observation
First of all, it is necessary to clarify the level of conceptual categories at

which the quantitative analyses are carried out. Table 9.1 shows the cat
egories adopted in this chapter. Because the quantitative analyses aim at
complementing Chapter 6, the same level of conceptual categories used in
6.1 should be adopted for the framework of observation here. This is also
a theoretical requirement because the categories should be coherent in the
sense that they constitute the basic units in which regularities of term for
mation patterns manifest themselves. Meanwhile, it is also necessary that
the categories have equivalent status with respect to the overall concep-
Table 9.1. Conceptual categories used as a basis of observation.

Category Terms Simple Determinants
111 People (types) 26 3(11.5%) 27
ME12 Organisations 38 1 (2.6%) 39
ME2111 Documents (types) 136 34 (25.0%) 112
ME212 Non-documents 54 8 (14.8%) 47
ME22 Machines and implements 59 7(11.9%) 64
REl Broad representational entities 51 15 (29.4%) 38
RE2 Linguistic entities 45 16 (35.6%) 30
RE311 Documentation entities (complex) 113 17 (15.0%) 97
RE312 Documentation entities (simple) 60 8(13.3%) 55
RE4 Software entities 40 12 (30.0%) 32
AE1 Information entities 60 18(30.0%) 42
AE21 Subject fields 25 5 (20.0%) 21
AE22 Methodologies 109 1 (0.9%) 182
CE Classificatory entities 50 16 (32.0%) 35
QL1 Labels of types of qualities 57 11 (19.4%) 46
RL1 Labels of types of relations 49 8(16.3%) 41
RL2 Values of relations 26 6(23.1%) 20
AC1 Units of activities 20 0 (0.0%) 22
AC21 Action 97 14 (14.4%) 86
AC22 Transference 22 6(27.3%) 18
AC23 Production 21 10 (47.6%) 11
AC24 State change 26 12 (46.2%) 14
tual structure of the domain . The analyses and discussions in Chapter 6

showed that the level of the categories adopted there was basically valid in
that sense.
Due to the problem of size, however, a few modifications to the level of
the categories were admitted. In principle, the distinction between general
and specific concepts was collapsed. For instance, we put together RE11
(broad representational entity - general) and RE12 (broad representational
entity - specific), and AE211 (general subject field) and AE212 (specific
1
Suppose, for instance, we observed one characteristic in the terms representing one
conceptual category and another, different characteristic in the terms representing another
conceptual category. In this situation, unless the two conditions for the conceptual cate
gories here are satisfied, it can always be argued that the difference in the characteristics
may be due to badly-defined categories and may not reveal anything about the character
istics of term formation. Although we are not concerned with direct comparisons of the
formation patterns of different categories, they are still important to guarantee the overall
significance of the descriptions.
Quantitative Dynamics in Term Formation 203
subject field). A certain number of categories are not observed here because
they are too small for quantitative analysis and cannot be collapsed into
neighbouring categories. These are: document parts (ME2112: 3 terms),
places/locations (ME24: 4 terms), parts of information entities (ME25: 5
terms), parts of documents (RE32: 18 terms) and subjective entities (AE3:
14 terms). Some categories, i.e. RL2 (values of relations), AC1 (units of ac
tivities), AC22 (transference), AC23 (production) and AC24 (state change),
may be too small for quantitative analysis, so the results of quantitative
analysis applied to these categories will have to be interpreted with care.
9.1.2 Elements of observation

For each category in Table 9.1, the following elements are investigated.
9.1.2.1 The nucleus

For the nucleus, individual morphemes are used as the basic elements of
quantitative analysis. This is because the conceptual categories in the nu
cleus position, even taken at the finest level, are highly limited in each
category, i.e. there are 6 at most (in RE311 and AE1). The quantitative
method is therefore not useful for analysing them. In addition, categorial
variations in the nucleus position were already examined in the qualitative
observation in Chapter 6.
The basic questions related to the quantitative dynamics of the nucleus
are thus: When a new term that represents a particular conceptual category
is created, what is the likelihood that a new morpheme is used as the nu
cleus of the term, and what is the likelihood that morphemes already used
as nuclei are used again? How will the likelihood change when the termi
nology grows?
9.1.2.2 The determinant

The analysis of the determinant is based on the bottom-level conceptual cat
egories (henceforth "determinant categories") and individual morphemes
(henceforth simply "determinants"). Unlike with the analysis of the nuclei,
the diversity at the level of conceptual categories made qualitative gener
alisation difficult. For this reason, an in-depth observation of determinant
categories was not carried out in Chapter 6. The quantitative analysis of the
bottom-level conceptual categories will compensate for the lack of exami

nation of determinants in Chapter 6.
Positional differences among the determinants, which occur in terms
with more than three constituent elements, are not taken into account in the
analysis. This is a compromise adopted for quantitative reasons; the sample
is too small for the quantitative method to work with this distinction.
The basic questions concerning determinants are: When a new term
representing a particular category is created, what is the likelihood that a
new morpheme or a new conceptual category is used for the determinant(s),
and what is the likelihood that morphemes or conceptual categories already
occurring in the determinant position are reused? Also, will the likelihood
change, and, if so, how?
9.1.2.3 The conceptual specification pattern

For each conceptual category, only term-forming conceptual specification
patterns are analysed. Although the analysis of secondary specification pat
terns is in theory interesting and important, we do not pursue it because
it would require a distinction of the levels of term formation within each
conceptual category; the number of secondary specification patterns would
then be too small for quantitative analysis.
The basic questions concerning conceptual specification patterns are
thus: When a new term representing a particular conceptual category is
created, what is the likelihood that a new conceptual specification pattern is
introduced? What is the likelihood that existing patterns are reused? Also,
will the likelihood change, and, if so, how?
9.1.2.4 The dynamics of term formation

For each category in Table 9.1, the above four elements — i.e. nuclei, de
terminants, determinant categories and conceptual specification patterns —
are analysed, in sample sizes up to 1.5 times that of the original, by ap
plying binomial interpolation and extrapolation. By setting the observation
range relative to the scale of the original data (instead of, say, extrapolating
up to 200 terms irrespective of the original size of the category), we ob
serve the dynamics in the overall terminology of documentation, instead of
comparing the dynamics of term formation for each category.
The above elements of observation are not mutually independent. For

instance, if only a very limited number of morphemes are used as the nuclei
of terms in a category, then the determinants of the terms of the category
might well vary so as to compensate for the limitation of the nuclei. There
fore, the relationships between the four elements of observation, namely,
the nuclei, the determinants and their categories, and the conceptual spec
ification patterns, will also be investigated, in addition to the dynamics of
each element.
9.1.3 The measures and the framework of interpretation

9.1.3.1 The three measures for observation
In order to observe the dynamics of term formation, we use three measures
for each of the above four elements of observation.
Firstly, we observe the change in the number of types (of nuclei, de
terminants, etc.) according to changes in the sample size. As mentioned
above, we observe changes that occur when the sample size is decreased or
increased by up to 1.5 times that of the original using binomial interpolation
and extrapolation. The number of types is calculated by applying formula
(7.4) in Chapter 7 for λ = 0 to 1.5.
Secondly, in order to see clearly the change in the number of types
according to changes in the sample size, we observe the growth rate, i.e.
the first derivative of the developmental curve of the number of types, as
defined by equation (7.8). More specifically, for each category we observe:
for λ = 0 to 1.5.
Though we stated above that the relationships among the growth pat
terns of the four elements are taken into account, these measures cannot be
used for direct comparison. For instance, suppose that the number of nu
clei increases much faster than the number of specification patterns. This
comparison does not mean anything in itself because the nuclei and the
specification patterns are qualitatively different in the first place.
To solve this problem, referring to the growth and growth rate of all
the terms in the data as the baseline, we introduce a third measure, i.e. the
relative number of types. The relative number of types is defined as the
ratio of the number of elements in the observed category to the number of

elements in all terms in the data at the same sample size2. Formally, the
relative number of types is defined by:
VCAT(ΛN)
VALL(λN)
for λ = 0 to 1.5. In the above formula, N is the token number of elements,

VCAT(N) is the number of elements (types) in the observed category, and
VALL(N) is the number of elements (types) in all the terms at the sample
size N3.
The three measures have different characteristics when different cate
gories are compared. When using the number of types or the growth rate,
the values calculated at the same number of terms should be compared. On
the other hand, the relative number of types can be compared at the same
relative observation point, e.g. at the original sample size, at a size 1.5 times
larger than that of the original, etc., even if the numbers of terms are differ
ent. This is because the relative number of types is normalised with respect
to the absolute size by means of the baseline, i.e. the growth patterns of all
the terms.
9.1.3.2 Framework of interpretation

In order to set the interpretative framework, we adopted the following as
sumptions concerning term formation:
1. The form of a new term in a given conceptual category is decided in
accordance with the forms of the existing terms in the category. If the
new term is a simple term, then only a nucleus will be used. If the new
term is complex, then (a) determinant(s) and a conceptual specification
pattern will also be used.
2
Direct comparison in different categories is not desirable for two reasons. Firstly, as
categories differ in size, the comparison does not make sense. For instance, comparing the
number of morphemes in 50 nuclei in one category and the number in 100 nuclei in another
category is not valid. Secondly, we are concerned not with the comparative characterisation
of the formation patterns of terms in different conceptual categories, but with the dynamics
of term formation for each conceptual category itself.
The actual value of N depends on the size of the observed categ ory. So if the size of
the category is 100 by token number of elements, it is necessary to calculate VALL100) by
interpolation.
2. The occurrence of actual items in the nucleus or determinant position is

regulated by the structure of the existing items in the nucleus or determi
nant position, respectively. Likewise, the occurrence of the specification
patterns is regulated by the structure of existing specification patterns.
Under these assumptions, which are in fact restatements of the assump
tions adopted so far in the present study, the dynamics of the elements in
each conceptual category (with simple terms also taken into account) can be
most clearly observed by plotting the number of types or the growth rates
calculated for each element on the axis of the sample size normalised into
the number of terms. This enables us to observe the developmental curves
of the three measures for the four elements of observation.
9.1.3.3 The status of simple terms

Because we are interested in the formation and growth of terms within a
terminology, as opposed to the formation of complex terms, we take sim
ple terms into account. The inclusion or omission of simple terms affects
the different elements of observation differently. Since simple terms are
regarded as consisting of the nucleus only, their omission would distort the
quantitative evaluation and thus the prediction of the growth of the nuclei.
On the other hand, as the determinants and the specification patterns
can only be defined in complex terms, the inclusion or exclusion of simple
terms in the analysis does not change their values for term creation.
It is necessary, however, to take into account the ratio of simple terms
(or, from the other point of view, the ratio of the determinants or specifica
tion patterns, which generally correlates negatively with the ratio of simple
terms) when the growth of the nuclei, the determinants, etc. is interpreted
within the growth of terms in which simple terms are included. This is be
cause it is intuitively obvious that, in general, the more simple terms are
contained in the category, the more rapidly the number of nuclei increases
and the more slowly the number of determinants or specification patterns
tends to increase against the number of terms.
9.2 Preparation for the analyses
Let us observe here the developmental patterns for all 1,228 terms in the
data and then look at the general tendencies of the relative number of types
of each element by conceptual category. This provides a basic reference for

interpreting the developmental curves of the measures in each conceptual
category.
9.2.1 The growth patterns of all the terms in the data

The basic quantities of the four elements, i.e. the nuclei, the determinants,
the determinant categories and the specification patterns, for all the data are
listed in Table 9.2. The notation introduced in Chapter 7 is used in the table:
TV refers to the sample size in token number of elements; V(N) refers to
the total number of different items in the sample of size TV; V(1, TV) refers
to the number of different hapax legomena in the sample of size TV; V{N)
refers to the growth rate; and L refers to the coefficient of loss. The
token number of nuclei is greater by one than the number of terms. This
is because there is one complex term with a dual nucleus (i.e.
[reader-printer]) which belongs to the category of "machines
and implements" (ME22). The token number of conceptual specification
patterns is the same as the number of complex terms, i.e. 981, as we only
deal with term-forming specification patterns.
Table 9.2. Basic quantities of nuclei, determinants, determinant categories and specification
patterns.
N V(N) V(1,N) V(N) cL
Nuclei 1229 443 284 0.231 0.259
Determinants 1104 632 448 0.406 0.286
Determinant categories 1104 90 9 0.008 0.053
Specification patterns 981 20 1 0.001 0.019
Figure 9.1 shows the developmental patterns of the numbers of types of

nuclei, determinants, determinant categories and conceptual specification
patterns for all the data (vertical axis). The numbers of items were obtained
by binomial interpolation and plotted against the number of terms from 0
to 1,228 (horizontal axis).
Throughout the range of observation, the number of determinants is
largest, followed in order by the number of nuclei, the number of determi
nant categories and the number of specification patterns. The number of
nuclei and determinants keeps increasing until the number of terms reaches
Figure 9.1. The growth patterns of nuclei, determinants, determinant categories and specifi
cation patterns f or all the data.
1,228 (and is expected to continue increasing thereafter). The number of

conceptual categories of determinants and the number of conceptual spec
ification patterns, on the other hand, converge; this is natural because their
types are limited. These tendencies can be confirmed by the values of P(N)
and L shown in Table 9.2.
9.2.2 The overall tendencies of relative growth

In order to make the descriptions below clearer, it is convenient to have
a rough idea of how much the relative number of elements in individual
categories varies. Table 9.3 shows the values of the relative number of the
four elements of observation — i.e. the nuclei (NUC), the determinants
(DET), the determinant categories (CDET) and the specification patterns
(SPEC) — calculated on the basis of an extrapolated sample 1.5 times the
size of the original. The maximum and minimum values for each of the
four elements of observation are underlined4.
4
Recall that the inter-categorial comparison of the values of the relative number of types
is justified because the size factor is abstracted in the relative number of types.
Table 9.3. The relative number of types of the four elements of observation for each con
ceptual category as calculated at a sample size 1.5 times that of the original.
Category NUC DET CDET SPEC
ME111 People (types) 0.33 0.91 0.55 0.16
ME12 Organisations 0.28 0.91 0.68 0.54
ME2111 Documents (types) 0.53 0.96 0.92 0.76
ME212 Non-documents 0.37 0.79 0.55 0.74
ME22 Machines and implements 0.38 0.84 0.67 0.58
REl Broad representational entities 0.45 0.89 0.73 0.80
RE2 Linguistic entities 0.50 1.01 0.85 0.82
RE311 Documentation entities (complex) 0.33 0.98 0.81 0.75
RE312 Documen tation entities (simple) 0.32 0.99 0.89 0.78
RE4 Software entities 0.71 0.75 0.81 0.45
AE1 Information entities 0.50 0.95 0.61 0.84
AE21 Subject fields 0.54 0.74 0.72 0.40
AE22 Methodologies 0.14 0.72 0.67 0.31
CE Classificatory entities 0.48 0.87 0.61 0.88
QL1 Labels of types of qualities 0.58 0.92 0.81 0.27
RL1 Labels of types of relation s 0.47 0.99 0.79 0.49
RL2 Values of relations 0.39 1.05 0.61 0.34
AC1 Units of activities 0.35 0.82 0.55 0.17
AC21 Action 0.38 0.82 0.71 0.45
AC22 Transference 0.41 0.71 0.63 0.65
AC23 Production 0.61 0.88 0.54 0.11
AC24 State change 0.81 1.03 1.00 0.25
Average 0.45 0.89 0.71 0.52
In general, the increase in the number of elements in each conceptual

category is expected to be slower than the increase in the number of el
ements for all the terms, because the terms in a conceptual category are
expected to be more coherent. So it is natural that the values in Table 9.3
tend to be less than 1. The minimum relative number of types of nuclei is
0.14 (AE22), and the maximum is 0.81 (AC24). The average is 0.45. The
minimum relative number of types of determinants is 0.71 (AC22), and the
maximum is 1.05 (RL2). The average is 0.89. The minimum relative num
ber of types of determinant categories is 0.54 (AC23), and the maximum is
1.00 (AC24). The average is 0.71. The minimum relative number of types
of specification patterns is 0.11 (AC23), and the maximum is 0.88 (CE).
The average is 0.52. It is clear that the nucleus is in general much more co
herent than the determinant. Only the determinants in three categories and
the determinant categories in one category take a value equal to or larger

than 1.
From these we can see, for instance, that the relative number of types
of nuclei for ME2111, 0.53, does not mean that the nuclei of this category
are particularly coherent, even though they are much more coherent than
the nuclei of all the terms. On the other hand, the relative number of types
of determinants for AE22, 0.72, shows that this category is comparatively
coherent with respect to the use of determinants. Though Table 9.3 is insuf
ficient for inter-categorial comparisons because it only takes into account
the point values and not the change in values according to changes in the
sample size, it is sufficient for our immediate purpose of setting a rough
baseline for the observation of the dynamics of term formation for each
conceptual category.
9.3 The observation of dynamics for each conceptual category
In this section, we describe the dynamics of the four elements of observa

tion and their relationships for each conceptual category listed in Table 9.1.
The number of types, the growth rate and the relative number of types are
observed in sample sizes up to 1.5 times that of the original for each cat
egory. This gives a forecast of growth patterns in sample sizes larger than
that of the original.
9.3.1 Material entity (ME) terms

9.3.1.1 Animate -people - types (111)
Table 9.4 shows the basic quantities of the four elements for 26 111
terms, i.e. N; V(N); V(1,N); V(N); and CL. In Figure 9.2, panel (a)
shows the developmental curve of the number of types, panel (b) shows the
developmental curve of the growth rate, and panel (c) shows the develop
mental curve of the relative number of types of nuclei (solid line), determi
nants (dotted-and-dashed line), determinant categories (short-dashed line)
and specification patterns (long-dashed line). The vertical line in each panel
shows the original sample size. As stated, the data is extrapolated to sizes
up to 1.5 times that of the original, so in Figure 9.2 the range of observation
is from 0 to 40 terms.
T 9.4. Basic quantities of nuclei, determinants, determinant categories and specification
patterns in ME111.
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 26 8 6 9.17 0.231 0.271
Determinants 27 24 22 35.28 0.815 0.338
Determinant categories 27 13 5 14.97 0.185 0.187
Specification patterns 23 2 0 2.04 0.000 0.002
Figure 9.2. The number of types, the growth rate and the relative number of types of nuclei,
determinants, determinant categories and specification patterns for ME111
terms.
Panel (a) shows that the value of the determinants is consistently higher
than the values of the other elements. This is followed, from highest to
lowest, by the values of the determinant categories, the nuclei and the spec
ification patterns5. The difference between the number of determinants and
the number of nuclei is rather large. The number of specification patterns
converges at a very early stage of the graph, and the increase in the number
of determinant categories slows down towards the end. The numbers of
5
As mentioned above, the comparison of the number of these elements is in itself not
meaningful, as the four elements consist of different classes of items (other than the deter
minants and the nuclei). The number of determinant categories is by definition not larger
than the number of determinants, and the total number of conceptual specification patterns
is only 21. However, by comparing the order of the elements' values and the discrepancies
among them with those of all the terms in Figure 9.1, we can intuitively observe the char
acteristics of the growth of terms in this category. This is why we refer to the order of the
values of the four elements in panel (a). In the following, we will refer to the order of the
numbers of nuclei, determinants and determinant categories in the same spirit.
determinants and nuclei keep increasing.

This is confirmed by the developmental curves of the growth rates in
panel (b). The growth rate of specification patterns converges to zero, which
indicates there will be no new specification patterns other than the specifi
cation of functions (FFUN) and the specification of the whole or affiliations
(PPAR). The growth rate of nuclei converges to about 0.25, thus a new nu
cleus is expected to keep appearing for every four new nucleus tokens, ir
respective of the overall number of terms. The growth rate of determinants
also seems to converge to about 0.8, i.e. the number of determinants keeps
increasing quickly and steadily. In contrast, the growth rate of determinant
categories continues decreasing in the graph, which is reflected in the flat
tening pattern of the number of determinant categories in panel (a). As the
growth rate of determinant categories becomes smaller than that of the nu
clei, the number of nuclei will eventually become greater than the number
of determinant categories6.
Panel (c) shows that the increase in the number of nuclei for 111
terms is much smaller than the increase in the number of nuclei for all the
terms. It levels off at about 35 per cent in the middle of the graph, well
below the average for the relative number of conceptual categories at 1.5
times the original sample size. In contrast, almost the same number of de
terminants is used as for all terms (about the average of the categories). The
number of determinant categories, however, is limited and becomes smaller
in relative terms when the number of terms becomes larger, well below the
average of the categories. Only a very limited number of specification pat
terns are used, which is also far less than the average of the categories.
The terms representing 111 concepts, therefore, are constructed
from a relatively small number of nuclei and a very limited number of con
ceptual specification patterns. In contrast, a fair variety of determinants is
used, though with a limited range of conceptual categories. In the case of
growth, the number of nuclei is expected to increase steadily though slowly,
and the number of determinants is expected to increase rapidly, though the
conceptual categories they represent would remain somewhat limited. It
is unlikely that new conceptual specification patterns would be introduced
even if the number of ME111 terms should become larger.
6
Theoretically, this is a matter of course when N → ∞, because the number of deter
minant categories is limited. What is important here is that we can forecast how this will
happen in this category on the basis of the quantitative investigation.
patterns in ME12.
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 38 10 6 12.88 0.158 0.235
Determinants 39 34 31 49.30 0.795 0.339
terms.
9.3.1.2 Animate - organisations (ME12)

Table 9.5 shows the basic quantities of the four elements for 38 ME12
terms, and Figure 9.3 shows their growth patterns for up to 57 terms. Panel
(a) shows that the value of the determinants is the highest among the four
elements, and the difference between the number of determinants and the
number of nuclei is rather large. This partly reflects the fact that the ratio
of simple terms in this category is very low.
Among the four elements, the growth rates of nuclei and determinant
categories converge, as shown in panel (b). Thus, the number of these ele
ments is expected to keep increasing steadily. The growth rate of determi
nants is high, but it is still decreasing at the end of the graph, implying that
the growth of the number of determinants will slow down. The growth rate
of specification patterns is approaching zero, so the number of specification
patterns is not expected to increase much.
Panel () shows that the increase in the number of nuclei is much
smaller (even smaller than that of the 111 terms) than the increase in
the number of nuclei for all the terms, becoming less than one-third at
the end of the graph. This can be attributed partly to the use of
(book=house; library) as the nucleus in many terms (see 6.1.1.2). The
increase in the number of determinants is about the same as that of all the
terms, while the relative number of determinant categories becomes smaller
when the number of terms becomes greater. The graph also shows that
about 60 per cent of all the specification patterns are used in the terms of
this category.
The terms representing ME12 concepts are constructed from a small
number of nuclei and a limited number of conceptual specification pat
terns, with a wide variety of determinants whose conceptual categories are
slightly limited. In the case of growth, the number of new nuclei should in
crease steadily, though at a slow rate, while new determinants are expected
to occur at a high rate, though the rate itself will decrease slowly. New
conceptual specification patterns are expected, but they will not be many.
9.3.1.3 Inanimate - information carriers - documents (ME2111)

terms, and Figure 9.4 shows their growth patterns. The value of the num
ber of determinants is the highest, followed by that of nuclei and then by
that of determinant categories. The difference between the numbers of de
terminants and nuclei is relatively small, reflecting the fact that the ratio of
simple terms is high in this category. At the beginning, the number of deter
minant categories is larger than the number of nuclei7. The developmental
curve of the number of specification patterns flattens out.
Panel (b) shows that the growth rates of determinants and nuclei keep
decreasing even at the end of the graph. Thus, the ratio of new determinants
and nuclei will decrease as the number of terms increases. The growth
rate of determinant categories becomes almost flat, and the growth rate of
specification patterns becomes zero at the end of the graph. Thus, there will
7
Recall that we argued that the number of nuclei will surpass that of determinant cat
egories for ME111 terms. The ME2111 terms, whose number is much larger than that of
ME111 terms, somehow realise that forecast, although, of course, the structure is different.
patterns in ME2111.
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 136 51 37 68.45 0.272 0.278
Determinants 112 95 84 135.09 0.750 0.336
terms.
be no more than 14 specification patterns in total after the number of terms

becomes more than 1.5 times that of the original.
From panel (c), it can be observed that the developmental curve of the
relative number of nuclei flattens out at about half that of all the terms.
The number of specification patterns in this category is about 75 per cent
of that of all the terms. The relative number of determinants and their cat
egories is more than 0.9, indicating that the number of determinants and
their categories increases at a pace similar to those of all the terms. The
comparatively high relative numbers of nuclei, specification patterns and
determinant categories are notable.
The terms representing ME2111 concepts are constructed from a rel
atively wide variety of nuclei and a comparatively wide range of specifi
cation patterns, with a reasonable range of determinants, whose categories
vary as widely as the corresponding categories of all the terms. In the case
of growth, the number of nuclei is expected to keep increasing, though the
pace of this increase will slow down gradually. The number of determinants
is expected to increase rapidly, but the pace of this increase, too, will slow
down. The number of conceptual specification patterns will not increase
after 13 or 14 patterns occur.
9.3.1.4 Inanimate - information carriers - non-documents (ME212)

terms, and Figure 9.5 shows their growth patterns. The number of de
terminants is consistently higher than the number of other elements. The
difference between the number of determinants and the number of nuclei
becomes greater towards the end of the graph, as the increase in the number
of nuclei begins levelling off. The increase in the number of determinant
categories and specification patterns also begins to level off.
Panel (b) confirms these points. The growth rates of these three el
ements approach zero. So the number of nuclei in this category will be
absolutely, not relatively, limited8. The growth rate of determinants is high
but it also keeps decreasing in the graph.
Panel (c) shows that only the developmental curve of the relative num
ber of specification patterns flattens out, and it does this at a value above
the average (see Table 9.3). The relative number of the other three elements
keeps decreasing.
The terms representing ME212 concepts are constructed from a lim
ited set of nuclei, with a reasonable (but comparatively narrow) variety of
determinants, whose categories are limited. A fair number of specification
patterns is used. In the case of growth, few new nuclei, determinant cat
egories and specification patterns are expected to occur. Thus, the basic
concepts of this category are fixed. It is only the determinants that add
conceptual variations.
8
The ratio of simple terms is about 15 per cent for ME212 terms, so new nuclei are
expected to keep occurring under the assumption that the ratio of simple and complex terms
is constant irrespective of the size of the data. Here, however, the number of nuclei is
expected to be finite. This implies that at a certain stage no new nuclei will occur, which
in turn means that no new simple terms will occur. This contradiction always exists for
categories whose number of nuclei is expected to be limited but whose ratio of simple terms
is not zero. This makes invalid the assumption that the ratio of simple and complex terms as
well as the ratio of complex terms with lengths of 2, 3, etc. are constant. This, however, is a
problem to be addressed in the next stage of research and is put aside in the present study.
patterns in ME212.
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 54 20 9 23.19 0.167 0.208
Determinants 47 36 31 50.96 0.660 0.325
terms.
9.3.1.5 Inanimate - machines and implements (ME22)

terms. Because one term, (reader printer), has a dual
nucleus, the number of nuclei is 60. Figure 9.6 shows their growth pat
terns. The overall pattern of growth for the four elements is closer to the
patterns observed in 111 and ME12 terms in that the number of nuclei
is consistently smaller than the number of determinant categories.
Panel (b) shows that the growth rates of determinants and nuclei keep
decreasing steadily; the growth rate of nuclei is expected to converge to
zero if extrapolated to a larger size. The growth rates of determinant cate
gories and specification patterns level off at small but non-zero values. This
shows that the number of these two elements keeps increasing slowly when
the number of terms increases9. Note that the growth rate of specification
9
As the total number of conceptual categories and specification patterns is finite, their
patterns in ME22.
N V(N) V(1,N) E[V(1.5N)] P(N) CL
Nuclei 60 21 12 26.00 0.200 0.237
Determinants 64 51 44 71.50 0.688 0.330
terms.
patterns shows unstable transitions, which may be due to a numerical cal

culation error.
Panel (c) shows that the increase in the number of nuclei, determinants
and determinant categories is comparatively smaller than that for all the
terms. The developmental curve of the relative number of determinant cat
egories flattens out towards the end of the graph. The relative number of
specification patterns goes down at the beginning and then increases, re
flecting the fact that the number of specification patterns keeps increasing
slowly but steadily.
The ME22 terms are constructed from a comparatively small number
of nuclei, with a reasonable but less-than-average variety of determinants
and their categories, and a comparatively wide variety of specification pat
terns. In the case of growth, the nuclei may well be limited to a finite
growth rates should converge to zero in the long ran. What is shown here is that this would
only occur when the number of terms increases greatly.
number in the long run, while the number of specification patterns and de
terminant categories is expected to grow slowly but steadily.
9.3.2 Representational entity (RE) terms

9.3.2.1 Broad representational entities (REl)
Table 9.9 shows the basic quantities of the four elements for 51 REl terms.
Figure 9.7 shows their growth patterns. The number of determinants is
consistently higher than the number of other elements, but the difference
between the number of determinants and the number of nuclei is not large.
This is partly because the ratio of simple terms is high (about 30 per cent,
see Table 9.1). Panel (b) shows that the growth rates of all four elements
keep decreasing, thus it is expected that the developmental curves of the
numbers of all four elements will become flatter when the number of terms
increases, though at different rates.
Panel (c) shows that the relative number of specification patterns con
sistently takes a high value (about 0.80; the average is 0.52), while the
relative number of all the other elements is roughly average (see Table 9.3).
The REl terms are constructed from a reasonable number of nuclei, de
terminants and determinant categories, with a comparatively wider variety
of conceptual specification patterns. This tendency is expected to continue
in the case of growth.
9.3.2.2 Linguistic entities (RE2)

Table 9.10 shows the basic quantities of the four elements for 45 RE2 terms.
higher than the number of other elements, but the difference between the
number of determinants and the number of nuclei is not large. This is partly
due to the high ratio of simple terms in this category, which results in a
token number of nuclei 1.5 times larger than that of determinants.
From panel (b), it can be observed that the growth rates of nuclei and
specification patterns basically converge, the former to about 0.3 and the
latter to zero. Thus, no more specification patterns are expected, while
about one-third of the nuclei are expected to be new when the number of
terms increases, irrespective of the size of the data. The growth rates of
determinants and determinant categories keep decreasing, though the pace
is very slow.
patterns in REL
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 51 21 13 26.78 0.255 0.245
Determinants 38 33 29 46.88 0.763 0.332
determinants, determinant categories and specification patterns for REl
terms.
Panel (c) shows that the relative number of types for all four elements
is higher than the average (see Table 9.3) and that that of determinants is
even increasing.
The RE2 terms are constructed from a fair number of nuclei, with a
rich variety of determinants whose conceptual categories vary widely. A
limited but reasonable number of conceptual specification patterns is used.
In the case of growth, the number of nuclei is expected to keep increasing
steadily. The number of determinants and determinant categories will also
increase, but the pace of increase will slow down. No new specification
patterns are expected to occur.
Table 9.10. Basic quantities of nuclei, determinants, determinant categories and specifica
tion patterns in RE2.
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 45 20 14 26.64 0.311 0.279
Determinants 30 29 28 42.75 0.933 0.354
determinants, determinant categories and specification patterns for RE2 terms.
9.3.2.3 Documentation entities (RE3)

9.3.2.3.1 Documentation entities - complex units (RE311)
Documentation entities (RE3) is one of the central categories in the field of
documentation. Table 9.11 shows the basic quantities of the four elements
for 113 RE311 terms. Figure 9.9 shows their growth patterns. Panel (a)
shows that the number of determinants is consistently larger than the num
ber of other elements. The difference between the number of determinants
and the number of nuclei is rather large, even though the ratio of simple
terms is not low (and the token number of nuclei is larger than that of deter
minants). The developmental curve of the number of specification patterns
flattens out at an early stage. The curve of the number of nuclei also flattens
out towards the end of the graph.
Panel (b) confirms these tendencies. The growth rate of specification
patterns converges to zero in the middle of the graph, so few new specifi
cation patterns are expected to occur. The growth rate of nuclei approaches
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 113 32 14 37.39 0.124 0.194
Determinants 97 86 75 121.08 0.773 0.336
determinants, determinant categories and specification patterns for RE311
terms.
zero towards the end of the graph, so the increase in nuclei in panel (a)
will level off if extrapolated further. The growth rate of determinants keeps
decreasing, while that of determinant categories becomes flat.
Panel (c), together with Table 9.3, shows that the developmental curves
of the relative number of determinants, their categories and specification
patterns flatten out above the average. The relative number of nuclei is
below average and keeps decreasing.
The RE311 terms are therefore constructed from a comparatively small
number of nuclei, with a reasonably rich variety of determinants and deter
minant categories. A fair number of conceptual specification patterns are
used. In the case of growth, the number of nuclei is expected to converge,
and no new specification patterns are expected to occur. The number of
determinants is expected to keep increasing, but the pace will slow down,
while the number of determinant categories is expected to keep increasing
slowly, with the pace slowing down slightly.
9.3.2.3.2 Documentation entities - simple units (RE312)

Table 9.12 shows the basic quantities of the four elements for 60 RE312
terms. Figure 9.10 shows their growth patterns. Panel (a) shows that the
number of determinants is consistently the highest. The difference between
the number of determinants and the number of nuclei is rather large. The
developmental curves of determinant categories and specification patterns
become flatter towards the end of the graph.
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 60 18 9 21.92 0.150 0.207
Determinants 55 51 47 73.89 0.855 0.346
determinants, determinant categories and specification patterns for RE312
terms.
Panel (b) shows that the growth rate of specification patterns becomes
almost zero at the end of the graph. The decrease in the growth rate of
the nuclei levels off. The growth rate of determinants keeps decreasing
steadily. The growth rate of determinant categories also keeps decreasing,
though less steeply towards the end of the graph.
Panel (c) shows that the developmental curves of the relative number
of determinants, determinant categories and specification patterns become
stable, above the average values (see Table 9.3). It is only the nuclei whose
relative number of types is below the average; it keeps decreasing.
The RE312 terms are constructed from a comparatively small number
of nuclei, with a reasonably rich variety of determinants and determinant
categories. A fair number of conceptual specification patterns are used.
In the case of growth, the number of nuclei is expected to grow slowly
but steadily. Recall that in constructing RE312 terms, nuclei representing
different categories are used (see Figure 6.4). The steady increase in the
number of nuclei may correspond to this qualitative feature. The number of
determinants is expected to keep increasing, but the pace will slow down.
The number of determinant categories will also keep increasing, but at a
slower pace which will become progressively slower. Few new concep
tual specification patterns are expected to occur when the number of terms
increases further.
The overall growth pattern of RE312 terms is similar to the pattern of
RE311 terms. However, in the RE311 terms the number of nuclei is most
probably limited, while in the RE312 terms it is expected that new nuclei
will keep occurring, though at a slow rate.
9.3.2.4 Software entities (RE4)

Table 9.13 shows the basic quantities of the four elements for 40 RE312
number of nuclei is consistently larger than the number of other elements.
Though this is partly due to the high ratio of simple terms in this category,
it is still noteworthy (compared, for instance, with ME2111, RE2 or AE1
terms). From the graph, it is expected that determinants will outnumber
nuclei if the number of terms increases further.
Panel (b) shows that the growth rate of specification patterns converges
to zero immediately. The number of specification patterns is therefore ex
pected to remain at six. The growth rate of determinants almost converges
above 0.6, thus the number of determinants will increase steadily. On the
other hand, the growth rate of nuclei keeps decreasing, which shows that
the increase in the number of nuclei is expected to slow down. The steep
downward curve of the growth rate of nuclei and the upward curve of the
growth rate of determinant categories at the end of the graph can be a result
of instability in the numerical calculation.
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 40 26 19 34.20 0.475 0.291
Determinants 32 24 20 33.66 0.625 0.314
determinants, determinant categories and specification patterns for RE4 terms.
Panel (c), together with Table 9.3, shows that the relative number of
determinants and specification patterns is smaller than the average; they
keep decreasing slowly. The relative number of nuclei is much higher than
the average, though it also keeps decreasing. The relative number of de
terminant categories is higher than the average, and it is not expected to
decrease.
The RE4 terms are constructed from a very wide variety of nuclei, with
a comparatively small number of determinants whose categories, however,
vary. A small number of conceptual specification patterns is used. In the
case of growth, the determinants and their categories are expected to keep
increasing steadily, while the increase in the number of nuclei is expected
to slow down. No new specification patterns will occur.
The diversity of the nuclei in this category may be a reflection of the
fact that this category consists of three related but different types of subcat
egories. It may also be because the terms of this category are peripheral to
the field of documentation, and only those representing core concepts are
recognised as terms of this field.
9.3.3 Abstract entity (AE) terms

9.3.3.1 Information entities (ΑΕ1)
Table 9.14 shows the basic quantities of the four elements for 60 AE1 terms.
the highest. The difference between the number of determinants and the
number of nuclei is comparatively small. The increase in the number of
determinant categories levels off at the end of the graph, while the number
of determinants and the number of nuclei increase almost linearly.
tion patterns in AE1.
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 60 24 20 34.00 0.333 0.304
Determinants 42 38 35 55.12 0.833 0.342
determinants, determinant categories and specification patterns for AE1 terms.
Panel (b) confirms these observations. The growth rate of determinant

categories becomes zero at the end of the graph, thus no more determi
nant categories are expected. The growth rate of specification patterns is
expected to converge to zero when the data becomes larger. On the other
hand, the growth rate of nuclei converges to about 0.35, thus new nuclei
keep appearing at a fixed rate. Perhaps this corresponds to the fact that
this category consists of three rather hybrid subcategories. The growth rate
of determinants keeps decreasing, though this decrease is expected to level
off.
Panel (c) and Table 9.3 show that the developmental curves of the rel
ative number of determinants, specification patterns and nuclei are stable,
all of them above the average values. The relative number of specification
patterns is especially high. The relative number of determinant categories
keeps decreasing and is expected to decrease to well below the average.
The AE1 terms are constructed from a variety of nuclei and determi
nants whose categories, however, are limited, with a wide variety of spec
ification patterns. In the case of growth, the number of determinants and
nuclei is expected to keep increasing steadily, while the increase in the num
ber of specification patterns is expected to slow down and eventually level
off. Few new determinant categories are expected to occur.
9.3.3.2 Systems of knowledge (AE2)

9.3.3.2.1 Subject fields (AE21)
Table 9.15 shows the basic quantities of the four elements for 25 AE21
number of determinants is the highest. The difference between the number
of determinants and the number of nuclei is small. Towards the end of the
graph, the slopes of the developmental curves of the determinants and the
determinant categories become less steep, while the developmental curve
of the nuclei does not show the same tendency.
Panel (b) confirms these points. The growth rates of determinants and
determinant categories decrease steadily and rapidly. The developmental
curve of the growth rate of nuclei, on the other hand, flattens out at a non
zero value. The growth rate of nuclei is expected to become higher than
that of determinants if the number of terms increases further. The growth
rate of specification patterns approaches zero.
Panel (c), together with Table 9.3, shows that the relative number of
determinants and their categories, which is already smaller than the average
Table 9.15. Basic quantities of nuclei, determinants, deteiminant categories and specifica
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 25 13 9 17.41 0.360 0.261
Determinants 21 17 13 22.71 0.619 0.303
determinants, determinant categories and specification patterns for AE21
terms.
at the end of the observation range, keeps decreasing steadily. It is only the
nuclei whose relative number is above average at 1.5 times the sample size.
The AE21 terms are constructed from a variety of nuclei with a com
paratively limited number of determinants, determinant categories and
specification patterns. In the case of growth, new nuclei are expected to
occur at a fixed rate. The number of determinants and determinant cate
gories will also increase, though this pace will slow down. Only a few new
specification patterns will occur when the size of the data becomes larger.
These tendencies correspond to the qualitative nature of the terms in
this category, i.e. only those whose determinants belong to a certain con
ceptual category are identified as being relevant to the field of documenta
tion. The range of nuclei is also limited, but when more specific subject
fields are identified, complex nuclei are used, and these new (complex) nu
clei will keep occurring.
9.3.3.2.2 Methodologies (AE22)

Table 9.16 shows the basic quantities of the four elements for 109 AE22
number of determinants is by far the highest among the four elements.
Compared to determinants, nuclei and determinant categories are very
small in number. This is partly a reflection of the fact that the token number
of determinants is much higher than the number of nuclei.
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 109 11 8 14.81 0.073 0.279
Determinants 182 110 91 153.28 0.500 0.317
determinants, determinant categories and specification patterns for AE22
terms.
Panel (b) shows that the growth rates of nuclei, determinant categories
and specification patterns suddenly drop at the very beginning of the graph.
The developmental curve of the growth rate of nuclei seems to have flat
tened out with a small but non-zero value. The curve of the growth rate of
specification patterns also becomes almost flat, with a very small but still
non-zero value. The growth rate of determinant categories keeps decreas
ing.
From panel (c) and Table 9.3, it can be observed that the relative num
ber of all four elements is below the average, though the number of deter
minant categories and specification patterns increases towards the end of
the graph. The small relative number of nuclei is particularly notable.
The AE22 terms are thus constructed from a very small number of
nuclei with a comparatively limited number of determinants and their cat
egories, and fairly limited specification patterns. The conceptual variety is
almost exclusively supported by determinants. In the case of growth, the
number of determinants is expected to keep increasing rapidly compared
to the other elements, though the pace of this increase is expected to slow
down. New nuclei and specification patterns will occur at a very slow but
fixed rate.
9.3.4 Classificatory entity (CE) terms

Table 9.17 shows the basic quantities of the four elements for 50 CE terms.
Figure 9.15 shows their growth patterns. Panel (a) shows that the increase in
the number of nuclei is faster than that of determinants at the very beginning
(note the high ratio of simple terms in this category), though the relation
soon becomes reversed. The difference between the number of nuclei and
the number of determinants is not large, though it becomes greater and
greater. Towards the end of the graph, the curves of the number of nuclei
and determinant categories become flatter.
Panel (b) shows that the growth rates of determinant categories and of
nuclei decrease rather steeply, while the growth rate of determinants de
creases slowly. The decrease in the growth rate of specification patterns
almost levels off to a non-zero value. Thus, it is expected that new specifi
cation patterns will keep occurring.
Panel (c) and Table 9.3 show that the relative number of specification
patterns is the highest among all the categories; it is about the same as
that of all the terms. The relative number of nuclei is about average, while
the relative number of determinants and determinant categories is below
average and keeps decreasing.
The CE terms are constructed from a relatively large number of nuclei
which, however, may well be finite, with a slightly restricted variety of
determinants. Many specification patterns are used. In the case of growth,
new nuclei are expected to occur less and less, while new determinants
tion patterns in CE.
N V(N) V(1,N) E[V(1.5N)] P(N) CL
Nuclei 50 23 13 28.19 0.260 0.238
Determinants 35 30 26 42.71 0.743 0.328
determinants, determinant categories and specification patterns for CE terms.
will occur with a relatively high ratio. New specification patterns will keep
occurring steadily, though with a low ratio.
9.3.5 Quality (QL) terms

Table 9.18 shows the basic quantities of the four elements for 57 QL1 terms.
Figure 9.16 shows their growth patterns. The number of determinants is the
highest throughout. The ratio of the number of nuclei with respect to the
number of determinants is comparatively high (note that the ratio of simple
terms is relatively high), but the difference grows larger. The developmental
curve of the number of specification patterns flattens out at an early stage
of the graph. The developmental curve of determinant categories becomes
flatter.
Panel (b) shows that the developmental curves of the growth rates of
specification patterns and determinants flatten out at zero and at about 0.8,
respectively. Thus, no new specification patterns are expected, while new
tion patterns in QL1
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 57 29 20 37.76 0.351 0.280
Determinants 46 40 36 57.75 0.783 0.336
determinants, determinant categories and specification patterns for QL1 terms.
determinants are expected to occur steadily. The growth rate of determinant

categories decreases steeply and is expected to approach zero if the number
of terms becomes larger.
Panel (c) and Table 9.3 show that the relative number of specification
patterns is very small, while those of the other three elements are slightly
higher than the average.
The QL1 terms are thus constructed from a reasonable variety of nu
clei, determinants and determinant categories and with a very limited num
ber of specification patterns. The qualitative investigations in Chapter 6
show that the variety of nuclei is due to some extent to the introduction
of complex nuclei to construct more specific QL1 terms. In the case of
growth, no new specification patterns will occur, and few new determinant
categories will be introduced, i.e. the conceptual "form" of the terms of this
category is restricted.
9.3.6 Relation (RL) terms

9.3.6.1 Labels of types of relations (RL1)
Table 9.19 shows the basic quantities of the four elements for 49 RL1 terms.
Figure 9.17 shows their growth patterns. Panel (a) shows that the number
of determinants is the highest among the four elements. The difference be
tween the numbers of determinants and nuclei becomes greater towards the
end of the graph. The developmental curve of the number of specification
patterns flattens out. The curve of the number of determinant categories
becomes flatter as well.
tion patterns in RL1.
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 49 21 14 27.04 0.286 0.268
Determinants 41 38 36 56.31 0.878 0.349
determinants, determinant categories and specification patterns for RL1 terms.
specification patterns and determinants flatten out, the former at the value
of zero and the latter at around 0.9. Thus, no new specification patterns
are expected to occur, while new determinants are expected to keep occur
ring at a high, constant rate of around 0.9. The growth rates of nuclei and
determinant categories keep decreasing. The pace of the decrease of the de
terminant categories is noteworthy; it may converge to zero when the data
becomes larger.
Panel (c) and Table 9.3 show that the relative numbers of nuclei and
specification patterns are more or less average, while the relative numbers
of determinants and of determinant categories are above average.
The RL1 terms are thus constructed from a reasonable number of nu
clei, with a comparatively wider variety of determinants and their cate
gories. As in QL1 terms, the variety of nuclei may be attributable to the
introduction of complex nuclei for more specific terms. Conceptual speci
fication patterns are limited, but not so much when compared to other cat
egories. In the case of growth, the number of determinants is expected to
increase steadily, while no new specification patterns are expected to oc
cur. The number of nuclei and that of determinant categories will increase,
though the pace will slow down, especially for the determinant categories.
Although the relation between the number of determinant categories
and the number of nuclei is reversed between QL1 and RL1, the general
dynamic patterns of growth of the four elements are rather similar. This
reflects the qualitative affinity between these two categories.
9.3.6.2 Values of relations (RL2)

Table 9.20 shows the basic quantities of the four elements for 26 RL2 terms.
of determinants is the highest among the four elements. Despite the fact
that the ratio of simple terms is relatively high (about 23 per cent), the
difference between the number of determinants and the number of nuclei is
significant. The number of determinant categories is expected to become
smaller than that of nuclei when the data is further enlarged.
the determinants, nuclei and specification patterns are all almost flattened
out with non-zero values. So the numbers of these three elements keep
increasing steadily. The growth rate of determinants remains one, i.e. no
existing determinants will be reused.
Panel (c) and Table 9.3 show that, among all the conceptual categories
examined here, the relative number of determinants is the highest at 1.5
times the sample size and keeps increasing. On the other hand, the relative
tion patterns in RL2.
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 26 9 8 13.00 0.308 0.321
Determinants 20 20 20 30.00 1.000 0.358
determinants, determinant categories and specification patterns for RL2 tenns.
number of the other elements is smaller than the average, though the value
of specification patterns increases towards the end of the graph.
The RL2 terms are constructed from a comparatively small number of
nuclei, with a rich variety of determinants whose categories, however, are
comparatively limited. A comparatively small number of specification pat
terns is used. In the case of growth, almost all the determinant tokens used
are expected to be new determinant types. New nuclei and new specifica
tion patterns also keep occurring steadily, though at a much lower rate.
9.3.7 Activity (AC) terms

9.3.7.1 Units of activities (AC1)
Table 9.21 shows the basic quantities of the four elements for 20AC1terms.
of determinants is the highest. The difference between the number of de
terminants and the number of nuclei is rather large, reflecting the fact that
Quantitative Dynamics in Terni Formation 237
there are no simple terms in this category. The developmental curve of the
number of specification patterns completely flattens out at an early stage
of the graph. The increase in the number of determinant categories slows
down when the number of terms becomes larger.
tion patterns in AC1
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 20 7 4 9.25 0.200 0.216
Determinants 22 18 16 25.69 0.727 0.327
determinants, determinant categories and specification patterns for AC1 terms.
Panel (b) shows that the growth rate of specification patterns becomes
zero at a very early stage. The instability of the developmental curves of the
growth rates of the other three elements at the right-hand side of the graph
may be due to a problem in numerical calculations, caused by the small size
of the data. Still, we can observe rough tendencies, i.e. the growth rate of
determinant categories approaches zero, while the decrease in the growth
rates of determinants and nuclei are expected to level off.
From panel (c) together with Table 9.3, it can be observed that the rel
ative numbers of all four elements are below average and keep decreasing.
The small value of the relative number of specification patterns is particu
larly notable.
The AC1 terms are thus constructed from a comparatively small vari
ety of nuclei, determinants and determinant categories, with very limited
types of conceptual specification patterns. In the case of growth, no new
specification patterns are expected to occur, while fewer and fewer new
determinant categories will be observed. Determinants and nuclei are ex
pected to increase steadily.
9.3.7.2 Action (AC21)

Table 9.22 shows the basic quantities of the four elements for 97 AC21
terms. Figure 9.20 shows their growth patterns. The number of determi
nants is the highest among the four elements. The difference between the
number of determinants and the number of nuclei increases as the num
ber of terms becomes larger. The developmental curve of the specification
patterns flattens out.
tion patterns in AC21.
N V(N) V(1,N) E[V(1.5N)] P(N) CL
Nuclei 97 31 16 37.38 0.165 0.216

Determinants 86 66 54 91.31 0.628 0.317
determinants, determinant categories and specification patterns for AC21
terms.
The growth rate of specification patterns becomes zero at the end of

the graph in panel (b). The developmental curve of the growth rate of de
terminant categories also becomes flat (the upward curve at the tail is most
probably due to a computational error). The growth rates of nuclei and
determinants keep decreasing.
Panel (c) and Table 9.3 together show that the developmental curves
of the relative number of all four elements become almost flat or only keep
decreasing slowly towards the end of the graph. Their values at the end of
the observation are slightly below the average, except for the determinant
categories, whose value is equal to the average.
The AC21 terms are constructed from a comparatively limited number
of nuclei, determinants and specification patterns, and a reasonable variety
of determinant categories. In the case of growth, the determinant categories
may keep increasing steadily but slowly. The increase in nuclei and deter
minants will slow down little by little. No new specification patterns are
expected to occur.
9.3.7.3 Transference (AC22)

number of determinants is consistently higher than the number of the other
elements, though the difference between the number of determinants and
the number of nuclei is rather small, reflecting the high ratio of simple
terms in this category.
Panel (b) shows that the growth rate of specification patterns converges
to a non-zero value very early on; it is expected that new specification pat
terns will keep occurring when the number of terms increases. The decrease
in the growth rates of the other three elements also levels off at the end of
the graph with a non-zero value. Thus, these elements are also expected to
grow steadily and constantly.
Panel (c) shows that the relative numbers of determinants, determinant
categories and nuclei keep decreasing, while the relative number of speci
fication patterns first decreases but then starts to increase. From Table 9.3
it can be observed that only the relative growth of specification patterns is
higher than the average at 1.5 times the original data size.
The AC22 terms are constructed from comparatively smaller numbers
of nuclei, determinants and determinant categories while specification pat-
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 22 9 6 11.77 0.273 0.253
Determinants 18 13 11 18.28 0.611 0.312
terms.
terns vary rather widely. The wider variety of specification patterns is par
ticularly notable if compared with other AC2 terms; this may be because
many of the activity concepts represented by the AC23 nuclei are subject-
specific and thus can take a wider variety of modification. In the case of
growth, all four elements, including specification patterns, are expected to
increase more or less steadily.
9.3.7.4 Production (AC23)

number of nuclei is consistently higher than those of the other three ele
ments, reflecting the extremely high ratio of simple terms in this category.
But the difference between the number of nuclei and the number of de
terminants becomes smaller as the number of terms becomes larger. The
developmental curve of the number of specification patterns becomes flat
N V(N) V(1,N) E[V(1.5N)] V(N) cL
Nuclei 21 14 8 16.96 0.381 0.252
Determinants 11 10 9 14.61 0.818 0.326
terms.
at a very early stage. The number of determinant categories also seems to

converge at the end of the graph.
Panel (b) clearly shows these tendencies. The growth rate of the speci
fication patterns becomes zero at a very early stage, the growth rate of deter
minant categories reaches zero at the end of the graph, and the growth rate
of nuclei is also expected to become zero if the size of the data increases10.
It is only the determinants whose growth rate is reasonably high, although
this also decreases steadily.
Panel (c) shows that the relative numbers of nuclei and determinant
categories also keep decreasing. The curve of the relative number of speci
fication patterns flattens out at a very low value. In fact, both the specifica
tion patterns and the determinant categories of AC23 terms take the lowest
values of the relative number (see Table 9.3).
10
Thus, the AC23 terms are yet another clear example disproving the assumption that the
ratio of simple and complex terms is constant irrespective of the size of the data.
N V(N) V(1,N) E[V(1.5N)] P(N) cL
Nuclei 26 19 16 26.91 0.615 0.313
Determinants 14 14 14 21.00 1.000 0.354
terms.
The AC23 terms are constructed from a fair variety of nuclei and de
terminants, whose categories however do not vary much. Only a very small
number of specification patterns is used. In the case of growth, the increase
in nuclei is expected to slow down rapidly, perhaps to the stage where no
new nuclei will be observed. Few new determinant categories will be intro
duced, and no new specification patterns will be used.
9.3.7.5 State change (AC24)

terms. Figure 9.23 shows their growth patterns. The number of nuclei is
consistently higher than the numbers of the other elements, reflecting the
high ratio of simple terms. Unlike with AC23 terms, the number of nuclei
keeps increasing. In fact, other than the determinant categories, all the
elements seem to grow almost linearly.
Panel (b) confirms these points. The growth rates of nuclei, determi
nants and specification patterns converge to non-zero values (though that of
the specification patterns is very close to zero). Only the growth rate of the
determinant categories keeps decreasing.
Panel (c) and Table 9.3 show that the developmental curves of the rela
tive number of all four elements flatten out. The relative numbers of nuclei
and determinant categories are the highest among all the categories. The
relative number of determinants is also much higher than the average (the
second highest among all the categories). The relative number of specifica
tion patterns, on the other hand, is much lower than the average.
The AC24 terms are constructed from a very wide variety of nuclei and
determinants, whose categories also vary widely. In contrast, only a very
small number of specification patterns is used. In the case of growth, all
four elements are expected to keep increasing in number, though the growth
rates will vary. Only the increase in the number of determinant categories
is expected to slow down as the data become larger. The other elements
will keep increasing linearly. Given that the growth rate of determinants
converges to a higher value than that of nuclei, determinants are expected
to outnumber nuclei in the long ran.
9.4 Summary observations
In the previous section, we described the dynamic patterns of the growth of

nuclei, determinants, conceptual categories of determinants and conceptual
specification patterns for each conceptual category. Within the range of
observation and ignoring the actual values and detailed transition patterns
of the measures, a few general patterns of growth for each element can be
observed across the conceptual categories.
For nuclei and determinant categories, three major patterns can be
identified: (i) The number of elements keeps increasing linearly (the growth
rate converges to a non-zero value); (ii) the number of elements increases,
but the speed of increase slows down (the growth rate keeps decreasing
within the observation range); and (iii) the number of elements does not, or
is not expected to, increase further (the growth rate converges or is likely to
converge to zero). For determinants, the pattern (iii) was not observed. For
conceptual specification patterns, it is convenient to divide (iii) into two,
Table 9.26. General growth tendencies for the conceptual categories.
adding a fourth type: (iv) The number of elements does not increase at all
(the growth rate converges to zero at a very early stage of the observation).
Table 9.26 summarises the growth patterns on the basis of these gen
eral types, where ↑ indicates (i), indicates (ii), → indicates (iii) and ↓
indicates (iv). As in Table 9.3, NUC indicates the nuclei, DET the deter
minants, CDET the determinant categories and SPEC the conceptual spec
ification patterns. These tendencies are not rigid in the statistical sense but
show general trends. Table 9.26 and Table 9.3 give the overall tendencies
of the growth patterns of terms in each conceptual category in the field of
documentation. As a natural extension of the present study, examining the
conceptual motivations that lead to the dynamic tendencies described in
this chapter would be an important research topic for the next stage of the
research cycle.
9.5 Methodological reflections
In this chapter, we have shown that binomial interpolation and extrapola

tion provides a very useful and powerful tool for describing the patterns of
terminological growth, which can complement the description of the con
ceptual patterns of term formation. The descriptions in this chapter remain
rather general, without much reference to the absolute values of the mea
sures (e.g. how many nuclei are expected, etc.), mainly in order to avoid
the descriptions becoming unmanageably complicated. More detailed in
formation would be useful for contrasting the formation patterns of qualita
tively similar categories, such as QL1 (labels of qualities) and RL1 (labels
of relations) terms or RE311 (complex documentation entities) and RE312
(simple documentation entities) terms.
A few methodological problems remain. Firstly, categories such as
AE1, which are adopted in place of lower-level categories because of the
problem of size, may have impeded the randomness assumption which re
quired categories to be basically uniform. This problem can only be avoided
by enlarging the terminological sample. Another feature that may have con
tradicted the randomness assumption is term formation of multiple depth
(e.g. in ME12; see 6.1.1.2). This presents an important challenge as to how
qualitative and quantitative approaches should be combined.
Secondly, there are cases that contradict the assumption of constancy
in the ratio of simple and complex terms as well as among complex terms
of each length. As a first approximation, this can be solved technically by
incorporating the model of the distribution of the length of terms into the
distributional model. This, however, may also be related to the multiple
levels of term formation and may lead to the question of the combination
of qualitative and quantitative approaches.
Lastly, binomial interpolation and extrapolation becomes unreliable at
around twice the original data size. This is one reason why we observed
each category at only up to 1.5 times the original data size. As a result, we
could not derive any definite conclusions concerning growth patterns when
the growth rates do not converge within the observation range. More power
ful quantitative methods called LNRE models (Baayen 2001) are available
to overcome this problem.
Part IV
Conclusions
Chapter 10
Towards Modelling Term Formation

and Terminological Growth
Having carried out a concrete analyses of the patterns of term formation and
of terminological growth for the terminology of documentation, and having
shown how conceptual descriptions can be complemented by quantitative
descriptions within the structural framework, it is time to look back and
examine what has been achieved and to clarify what still remains to be
done.
We start by summarising and examining what has been done so far in
the present study. In the process, the technical problems to be addressed in
future work will also be clarified. After this, we will illustrate the wider per
spective of research in term formation and terminological growth in which
the present study can be situated.
10.1 Structural-synchronic modelling revisited
The present study makes two major contributions to the study of terminol
ogy. Firstly, we examined the requirements for a theory of terminology (in
Chapters 1 and 2) and provided a concrete theory of the structural dynamics
of terminology based on the example of the terminology of documentation
(in Chapters 6 and 9). Our principal theoretical concern was to establish a
theoretico-descriptive framework for the study of terminology and not just
a study about exemplar terms. Secondly, we consolidated both the concep
tual and the quantitative methodologies for the establishment of a theory of
the structural dynamics of terminology. Let us briefly review these points
in turn.
10.1.1 The theoretical framework

The present work started by examining the definitions of terms and re
lated notions, as well as the basic characteristics of terms and terminology.
Through this, we arrived at an essential contention for this study, namely
that the study of terminology should first and foremost target the terminol
ogy of a domain in its totality and should not just deal with individual terms
or arbitrarily selected examples of terms. This is a minimum requirement
for a theoretical study of terms — as distinct from a theory of something
applied to describing some exemplar terms — because a theory of terms or
terminology must reflect the essential nature not only of terms as empiri
cal objects but also of the very categories term and terminology. Closely
related to this is the contention that "concept" cannot be an essential con
solidating factor of terminology.
Based on this contention, the patterns of formation and growth for the
terminology of documentation were investigated and described, using a rep
resentative data of Japanese documentation terms. The major results of the
analyses were presented in Chapters 6 and 9. The descriptions in Chapters
6 and 9 together constitute a theory of the structural dynamics of the termi
nology of documentation. Let us emphasise that what constitutes a theory
of terminology is the description of the terminology of a domain as a whole.
As we distance ourselves from the precedence of concepts over terms, the
question arises as to whether what we described in Chapters 6 and 9 is
confined to Japanese terminology. This can only be answered by empirical
examination, because the theoretical standpoint of the present study does
not offer any answer to this question.
In this respect, the theoretical intention of the present study is very sim
ple. Instead of a study which states "these are the characteristics of some
terms (but which may also be relevant to other lexical items)", we wanted to
undertake a study which could state "these are the characteristics of the ter
minology of the domain we are currently interested in". This rather naive
shift of emphasis was essential for consolidating a study of terminology
(and more specifically of term formation and terminological growth) which
could be distinguished, de jure, from the study of word formation on the
one hand and from the study of lexicology on the other.
In the process of establishing the theoretical framework of the study,
we recognised a few basic characteristics of terms and terminology, and
adopted a few assumptions in relation to the study of terminology.
Towards Modelling Terminological Dynamics 251
Firstly, in relation to the distinction between langue and parole, we

found that terminology essentially belongs to the sphere of parole1. This
characterisation arises directly from the very definition of the concepts
"terms" and "terminology"; terms are defined with reference to a domain,
which is external to langue. The immediate consequence of this charac
terisation of terms and terminology to the present study was important but
rather subtle, i.e. the study of term formation and terminological growth
should address, so to speak, the realistic possibility of existence, instead of
such formal restrictions as the combinability of elements in complex terms.
This was accompanied by the theoretical/methodological contention which
affirms the importance of the overall tendencies of the formation patterns
which anchor the individual descriptions to the terminology of a domain.
Empirically, it was recognised that terminology is located somewhere
in the middle between the general natural language vocabulary and artificial
nomenclatures. While acknowledging that terminology, consisting of lexi
cal items, potentially has the same flexibility as other natural language phe
nomena, it was argued that terminology, as a matter of fact, tends towards
stronger systematisation of its structure, both in its function of representing
concepts and in its linguistic forms. As a result, the systematic aspect of
terminology was emphasised in the study. This contrasts with some current
studies and claims related to terminology, which emphasise the flexible and
the "natural" language aspect of terminology (e.g. Temmerman 2000; Za
wada & Swanepoel 1994), and allied the present study with the traditional
approach, in which the systematicity of concepts represented by terms is
assumed.
In setting up the target of the study, yet another important theoretical
distinction was made, i.e. the distinction between the system of terminol
ogy and the sphere of text and discourse. Based on this distinction, it was
hypothesised that the system of terminology, which gains independent sta
tus as a socio-linguistic phenomenon, is closely related to the inclination
of terminology towards rigidity and systematicity, while the sphere of dis-
1
Nevertheless, methodologically, the structural approach was adopted, assuming the ex
istence of the terminological sphere as distinct from the textual sphere or discourse. Al
though it was emphasised that the "structural" approach in the present study is a method
ological concept and does not contradict the perception that terminology belongs to the
sphere of parole, this methodological choice still limited the range of the investigation. The
consequence of this will be examined further in the next section.
course in which individual terms are used is linked with the flexible aspect
of terminology. From the point of view of individual terms, this distinction
corresponds to the dual, i.e. classificatory and descriptive, role of a term.
On the one hand, a term distinguishes a concept it represents from other
concepts represented by other terms and demonstrates the position and sta
tus of the concept within the overall conceptual structure represented by the
terminology. On the other hand, a term describes the concept it represents
within the descriptive structure of the discourse in which the term is used.
Based upon this distinction, the systematic patterns of term formation
and terminological growth, as observed in the overall system of terminol
ogy, were investigated, assuming that some systemic/systematic factors in
the existing terminology of a domain direct the formation of new terms and
the growth of terminology. As shown especially in Chapter 6, the actual
description of documentation terms demonstrated the conceptual system
aticky of term formation patterns in many conceptual categories. So the
assumption of systematicity of terminology has empirically proved to be
valid to a significant extent.
All in all, even though some of these assumptions are not completely
correct, they can be justified given the current state of terminological re
search and because the central objective of the present study is (i) to move
away from establishing a theory which only describes exemplar terms to a
genuine theory of terminology, and (ii) to show the possibility of describing
the formation and growth patterns of terminology, not just of some exem
plar terms, with due granularity. Both, we contend, are necessary for the
ories of terminology, irrespective of what kind of assumptions one adopts
with respect to the nature of terms and terminology.
10.1.2 Methodological aspects

In order to formulate a theory of the structural dynamics of the terminology
of documentation, the present study devoted a major part to establishing
a relevant methodological framework. There were two main interrelated
methodological challenges, i.e. (i) how to relate the description of term for
mation and terminological growth to the terminology of a target domain
properly, and (ii) how to attain a sufficient concreteness or granularity of
description while at the same time allowing for the dynamics to be inte
grated in the descriptions. These challenges were dealt with through the
combination of the two approaches, i.e. conceptual and quantitative. We

will summarise the methodological contributions of the present study and
the remaining problems below.
10.1.2.1 Conceptual approach

In the conceptual description of term formation patterns, we took a rather
traditional view of concepts, effectively assuming that we could clearly de
fine conceptual categories, to which each lexical elements could essentially
be assigned unambiguously. This assumption, which goes against the cur
rent trend in the description of terminology emphasising the non-clearcut
nature of concepts (Cabré 1995; Temmerman 2000), was adopted mainly
for two reasons.
Firstly, we observed the tension between the systematicity and the flex
ibility of terminology, both in its function of representing conceptual struc
tures and in its linguistic representation. With respect to this tension, the
present study focused on investigating the systematic aspect of the dynam
ics of terminology vis-à-vis the overall structure of existing terms. The
view of concepts in the study is a direct result of this general theoretical
assumption.
Secondly, corresponding to the shift in the theoretical target from some
exemplar or individual terms to the terminology of a domain, it was nec
essary to introduce an overall conceptual system as opposed to isolated or
fragmentary conceptual categories or features. For this reason, we had to
adopt a traditional view of concepts and conceptual categories, both tech
nically and logically: technically because establishing and formulating a
conceptual system that accounts for not only systematicity but also flexi
bility or ambiguity is an independent research issue, and logically because
the flexible aspects of concepts such as prototypical effects or metaphoric
transformation assumes the existence of conceptual categories anyway (e.g.
Way 1991), if only for casting doubt on clear existence. In view of this, the
type of conceptual system established in the present study should be re
garded as the first starting point for any study of terminology that resorts to
"concept" as a descriptive or theoretical means.
Based on this view of concepts, we elaborated the form of describing
conceptual patterns of term formation, corresponding to the assumptions
and the aim of the study. In addition to introducing the conceptual sys
tem as a whole, we adopted as one of the major methodological devices
the conceptual specification patterns. Empirically they are little more than
ordinary intra-term relations, but theoretically they constitute an important
mechanism for incorporating a classificatory, as opposed to a descriptive,
aspect of term formation. Note that the classificatory aspect of term forma
tion is essentially linked with our assumption that the formation of terms is
systematic with respect to the structure of terminology, as opposed to the
use of terms in discourse.
The actual descriptions of the conceptual patterns of term formation,
carried out in Chapters 4 to 6, have shown that (i) these devices were useful
in revealing the systematic patterns of term formation vis-à-vis the total
ity of the terminology of a domain and that (ii) a considerable degree of
systematicity was actually observed in the formation patterns of documen
tation terms representing various types of concepts. It was therefore shown
that the basic assumptions of the present study as well as the conceptual
approach were useful as a first approximation to the theorisation of term
formation and terminological growth.
Under the current framework and assumptions, a few technical prob
lems remain nevertheless (see 6.3 as well for reflections on more technical
details). Firstly, we did not explicitly distinguish the general part and the
domain-dependent part of the conceptual system. Although this can be jus
tified for the immediate purpose of describing the formation patterns of the
terminology of documentation, it would be desirable to clarify the distinc
tion between general and domain-dependent parts of conceptual systems.
By doing this, it would become possible to use the general part of a con
ceptual system as a methodological resource for the analysis of terminolo
gies of other domains, which would in turn contribute to the clarification of
the conceptual factors of term formation of terminologies across different
domains.
The second problem is related to the simplicity of the conceptual sys
tem introduced in the present study. Formally, the concept system intro
duced in Chapter 4 has a simple hierarchical structure. However, as is
widely acknowledged, conceptual structures are essentially multidimen
sional (Bowker 1997; Kageura 1997b), which would lead to a lattice of cat
egories or would better be defined by the combination of features. In fact,
the multidimensional nature of concepts manifests itself in some of the de
scriptions in Chapter 6. For instance, the "horizontal" category shifts as ob
served in Figure 6.4 reveal that some cross-categorial conceptual character-
istics are involved. The affinity between formation patterns of quality terms
and relation terms also showed the existence of conceptual dimensions that
cut across these two categories. Although we referred to cross-categorial
relationships and the multidimensionality of the viewpoints of concept cat
egorisation wherever necessary, we did not fully explore this aspect. In this
respect, there remains further room for refining and re-formalising concep
tual systems and the descriptions of term formation patterns.
Lastly, we used the conceptual categories defined within the conceptual
system in two different ways. First, by attributing the categories to terms,
we used the conceptual categories as a conceptual "field" for observing the
systematicity of term formation. Second, by attributing the categories to
morphemes, we used the conceptual categories to observe the regularity of
combination patterns. For the latter purpose, it might have been better to
use the detailed conceptual features rather than to use the same concep
tual categories as defined for classifying terms. We did not elaborate this
because the main purpose of the current study is to describe the forma
tion patterns of terms rather than the combination restrictions or tendencies
of morphemes in terms. In the analyses in Chapters 6 and 9, we retained
the bottom-level categories assigned to morphemes, even though we did
not go into such a detailed level of analyses. This is because the bottom-
level categories will provide a good starting point for the exploitation of
the combination restrictions or tendencies of morphemes, which would be
better pursued in parallel with the type of study elaborated in this book.
10.1.2.2 Quantitative approach

As has been repeatedly stated, the conceptual descriptions of term forma
tion should remain necessarily broad in order to leave room for generalisa
tion; otherwise, it would ultimately become equivalent to a mere listing of
the existing terms in the data. In addition, although the conceptual descrip
tions can reveal the systematic aspect of term formation within the existing
terms, the relation between this systematicity and future prospects for term
formation is rather simplistic, i.e. the same systematicity prevails over fu
ture term formation.
These conditions produce a technical challenge, i.e. to increase the
granularity of the description while introducing the dynamic feature ex
plicitly in the description. It was in order to provide a solution to this
challenge that the quantitative analysis of terminological growth was in-
troduced. We adopted the method of binomial interpolation and extrapola

tion (Good & Toulmin 1956; Baayen 2001), an extension of Good-Turing
estimates widely known in the field of computational linguistics (e.g. Man
ning & Schütze 1999). By binomial interpolation and extrapolation, we
could estimate the change in the number of items relevant to terminology
construction beyond the given data size. In Chapter 9, we investigated
the growth patterns of constituent elements or morphemes as well as the
conceptual specification patterns of terms representing different conceptual
categories, revealing their dynamic tendencies expected in the growth of
terms. Generally speaking, the typical growth patterns were revealed by
the quantitative approach, which successfully made up for the insufficiency
of the conceptual approach.
Here also, however, a few technical problems remain. The first is
related to the randomness assumption of the distribution of morphemes
or specification patterns, an essential mathematical assumption for bino
mial interpolation and extrapolation. We have empirically proved (see also
Kageura [1998a]) that this assumption holds as long as we can maintain
the assumption that the set of terms is uniform and there is no distinction
between central and marginal terms within the terminological data. In its
actual application in Chapter 9, however, this condition was not fully main
tained. For instance, in the analysis of ME12 (organisation) terms in Chap
ter 9, we regarded the types of term formation as belonging to a single
uniform set, while the analysis in Chapter 6 revealed that there were two
different levels of term formation. The simplified treatment in this study
was mainly due to the quantitative constraint but needs further examination
in relation to the re-evaluation of the conceptual system.
The second problem is the assumption of the constancy of term length
distribution with respect to changes in the data size. Recall that the dy
namic model of terminology growth was constructed on the basis of the
distribution of morphemes, skipping the level of terms. It was in order to
interpret the morpheme-based model as the model of terminology growth
that we adopted the assumption of constancy of term length distribution,
letting the ratio of the number of terms against the token number of mor
phemes be constant irrespective of the size of the data. This is against our
intuition; our intuition tells us that the ratio of longer terms will become
larger when the size of a terminology becomes bigger2. More importantly,

the actual analysis revealed that this assumption is sometimes unsustainable
(see the note in 9.3.1.4): On the one hand, quantitative analysis implies that
new nuclei will be limited, which implies that the ratio of simple terms will
decrease, while on the other hand, under the current assumption, the ratio
of simple terms will not change. In order to account for this contradictory
phenomena, it is necessary to incorporate the dynamic model of term length
distribution. We leave this topic for further research because it constitutes
an independent and important research matter in itself; term length should
be taken into account in quantitative modelling of terminology growth in
the future.
The third and last problem is concerned with a simple technical de
ficiency of binomial extrapolation; the numerical calculation becomes un
stable when the data is extrapolated up to about twice the original sample
size. This imposed an obvious limitation upon the investigation. For in
stance, in Figure 9.15 in Chapter 9, we could not accurately estimate when
the growth rate of the conceptual specification patterns converges to zero
(note that it should become zero as the number of specification patterns is
limited). There are some models (LNRE models) that solve this problem
by incorporating the parametric distributions into the model of interpola
tion and extrapolation (Chitashvili & Baayen 1993; Baayen 2001). Though
Kageura (1998b) has shown that LNRE models were useful for the dynamic
analysis of terminological data, the small size of the data in the present
study prohibited the use of LNRE models.
10.1.3 The interpretation of the descriptions in structural dynamics

Finally, before going into the possible extension of the present study, it is
necessary to review the nature of the descriptions in Chapters 6 and 9 within
the framework of the study of structural dynamics. These descriptions can
best be interpreted as, so to speak, an accumulation/disaccumulation model,
i.e. the model assumes that when the number of terms increases (or de-
2
Baayen (2001) examines the effect of sample size to word length distribution, showing
that longer words tend to appear more when a sample is increased. Though what is exam
ined in Baayen (2001) does not directly support our intuition here, it still does contribute to
enforcing that intuition at a very general level, i.e. a larger number of longer words tends to
be observed when the size of the data is larger.
creases), no existing terms disappear (or no new terms appear). This also
means that no existing constituent elements or specification patterns are as
sumed to disappear when the number of terms increases. This contrasts
with what may be called a generation/decline model, which assumes that
some terms — and thus possibly morphemes and specification patterns —
may disappear even when the number of terms increases.
The fact that the descriptions in this study assume the accumula-
tion/disaccumulation model is explicit in the quantitative model3. This is
not so obvious in the conceptual descriptions, simply because the concep
tual descriptions are not sensitive enough to reveal this point explicitly. One
might think that replacing the randomness assumption of the distribution of
morphemes and specification patterns with a more complex model would
contribute to establishing a generation/decline model. This is not the case,
however, because the quantitative model is first and foremost constructed
over the distribution of morphemes, while the essential feature of a genera
tion/decline model should theoretically be attributed to the generation and
disappearance of terms.
In fact, as was clear from Chapters 6 and 9, the increase and decrease in
the terms is an explanatory parameter, not the phenomenon to be explained
in itself. What was to be explained was the systematic conceptual patterns
and quantitative structure which govern the formation of possible terms and
the growth of terminology when the terminology data size increases or de
creases. If we take a diachronic perspective, this corresponds to regarding
terminology growth as the addition of new terms to the existing terms, with
out being concerned with the disappearance of existing terms. However, it
should be recalled that, in the first place, we explicitly stated that we are
concerned with the dynamic potentiality (of creating new terms) observed
in the synchronic slice of the internal structure of terminology (see Chapter
2, especially Figure 2.2). It would therefore be misleading to interpret the
descriptions of the present study from a diachronic perspective.
How, then, can the contributions of the present study be located within
the overall perspective of term formation and terminological growth? Even
3
Technically, binomial interpolation leaves room to estimate an item occurring only
once in the original sample size will occur more than twice in a smaller sample size. This
is due to the fact that we used the binomial approximation to the hypergeometric model
(see Baayen [2001: 65-66] for the technical discussion concerning this point) and does not
mean that the binomial model incorporates the generation/decline type of dynamism.
if the study of terminology should first and foremost be concerned with the
terminologies of individual domains, is it not the case that actual termi
nologies of different domains sometimes interact with one another? How
can these be incorporated into the framework of the study of synchronic
dynamics? What is more, if terms belong to the sphere of parole, is it not
necessary after all to relate the structural dynamics of terminology to what
is actually happening to the terminology, i.e. evolution and change, in the
real world along a real time scale?
Now that we have obtained the descriptions of the dynamic potential
ity of the terminology of documentation as seen from the synchronic view
point, we can widen the perspective to what was illustrated in Figure 2.2.
Let us conclude the present study by elaborating this and suggesting further
directions of research in the dynamics of terminology.
10.2 Directions for further research: toward full exploitation of the

dynamics of terminology
It is best to conclude the work by formulating the further research directions

of the dynamics of terminology in relation to the research carried out so far.
Roughly speaking, there are two research directions to which the study of
dynamics can proceed. The first is to extend the range of the study within
the structural synchronic framework. The second is to incorporate the di-
achronic perspective and to model the change or evolution of terminology
as a real-world event within the actuality of language fact or parole.
10.2.1 Extension of the study of structural synchronic dynamics

Even if, as we contended, terminology is a concept that is first and foremost
consolidated as a vocabulary of a domain and thus crucially dependent on
individual domains, it is still important to pursue the formulation of a gen
eral theory of the dynamics of terminology. To this end, a comparative
study, based on concrete modelling of the dynamics of the terminologies
of individual domains (as was carried out in this book), is an essential first
step. A comparative study of the terminologies of different domains (or of
different languages) is an obvious extension of the study within the struc
tural synchronic framework.
In the study of the terminologies of more than one domain, there are
two aspects which belong to different levels that must be considered. The
first, a natural extension of the type of study we have carried out, is a com
parative analysis of the dynamics of terminologies of different domains.
The second is the study of the interaction of terminologies.
In the first type of study, at least two levels of analyses are possible:
1. A comparative study which takes into account the overlap of concepts
and lexical items represented by terms in different domains (or lan
guages). For instance, when Pugh (1984) compares the terminology
of information processing in English, French and Spanish, the interlin
gual overlap of concepts represented by terms was taken into account.
From the point of view of the study of the dynamics of terminology, it
is important to carry out the comparison not only between lexical items,
conceptual categories or conceptual specification patterns, but also be
tween the interactions of these aspects; the comparison of terminolog
ical data alone falls short of the comparison of the dynamics of termi
nologies. In that sense, it is essential to describe the patterns of term
formation and terminological growth for the terminologies of individual
domains. Though there are some comparative studies of terminology
across domains or languages (e.g. Miyajima 1981; Kageura 1994), what
they compare is the data rather than the dynamic characteristics of the
data.
2. A comparative study at the level of the formal structure of terminolo
gies, without taking into account the overlap of concepts, lexical items
or conceptual specification patterns. Let us explain this by means of a
simplified example. Take, for instance, the following three terminolo
gies, T(A), T(B) and T(C), of the domains A,  and C:
(A) = {a, ab, abc, bc, bd, acd}
() = {a, b, , e, . abce}
() = {p.pq.pqr,qr,qs,prs}
where elements separated by commas, such as a, ab, etc., in the curly
brackets indicate terms and individual lower-case letters, such as a, b,
etc., indicate constituent elements or morphemes of terms. If we take
into account the overlap of lexical items and concepts represented by
them, then  (A), the terminology of the domain Ճ, is much closer to
T(B) than to T(C) because there are many common lexical items, i.e.
α, b, c If, however, we focus on the formal constitution of the termi

nologies, then (A) is much closer to T(C). In fact, on the basis of the
mapping of the lexical items:
 ↔ 
b ↔ q
 ↔ r
d ↔ s
T{A) and T(C) are shown to be isomorphic and, thus, they have the
same formal structure, while the structure of T(B) is very different. The
comparative study of the dynamics of terminologies of different domains
at this level is essential for measuring the efficiency of terminology in
communication, etc. In fact, quantitative studies such as Ishii (1987b),
which compares the terminologies of different domains by using sum
mary statistics, can be interpreted as highly-simplified approximations
of this type of study; the structure of the terminology of each domain is
reduced to a summary statistic, and a static structure of the snapshot data
is taken into account. Though some studies that use summary statistics
for comparing different terminologies, such as Kageura, et. al. (1999),
can be regarded as a rough approximation, Kageura (2002) is the first
proper step towards this type of research.
While these comparative analyses presuppose a coherent structure of
the terminology of each domain at a starting point, the interaction of the
terminologies of different domains is essentially concerned with the ac
tual relations among domains which go beyond terminologies of individual
domains; the study of the interaction of terminologies thus naturally goes
beyond the range of the structural synchronic study. We will thus refer to
this aspect again in the next section, when we examine the incorporation of
various factors recognised in the actuality of parole.
10.2.2 Dynamics of terminology within the actuality of parole

The study carried out so far was anchored to the level of parole or language
facts, where the terminology is situated, basically by means of represen
tative samples of data. So far as they properly represent a snapshot of the
terminology of documentation, the data in our study reflects the actuality of
the terminology of documentation at the particular time when the data was
collected. For this reason, we have claimed that the structural dynamics
formulated was of the terminology of documentation de jure. Now, as ter

minology is a concept consolidated at the level of parole or language facts,
for the full exploration of the dynamics of terminology it will be necessary
to take into account the dynamics observed in relation to the actuality of
parole or language facts in which the terminology is located.
This, in a way, means moving from the study of structures to the study
of events. Pursuing the study of the theory of terminology as an event,
however, is logically doomed to fail and will result in a retrospective listing
and description of what has happened. This is related to the very nature
of "event" itself. By its definition, "event" refers to a unique and singular
happening in history. Linking some kind of expectation for the future to
the event, on the other hand, means that we recognise something repetitive
in the event. But recognising something repetitive in the event immediately
means that we are not looking at the event as a singular happening. The
study of the theory of terminology as an event is, if pursued consistently,
logically reduced to the retrospective confirmation of what has happened,
without any power of expectation. What Maeda (1989) stated about the
difficulty of the study of lexicology, i.e. that the phenomena that lexicol
ogists have to deal with are too concrete and real, equally applies to this
logical gap between the terminological (or, more generally, lexicological)
phenomena as unique historical events and the nature of scientific research
which requires predictability and repetitive characteristics4.
As we are concerned with the concrete modelling of the dynamics of
terminology, it is essential and most important to examine how the various
factors observed in the actuality of parole relevant to the dynamics of ter
minology — which nevertheless have not been taken into account so far —
can be incorporated into the model, rather than delving deep into philosoph
ical discussion of the relation between the event and the structure. Note that
the modelling of the dynamics should incorporate prediction, because the
very definition of the term "dynamics" implies prediction. In the study of
the dynamics of terminology, therefore, singular or unique phenomena of
the changes of terminology in the actuality of parole, which are by def
inition ungeneralisable, should be regarded as specific manifestations of
4
The discussion in this section, therefore, holds to some extent not only for the study of
terminology but also for the study of lexicology in general. After having tried to consolidate
the study of terminology as distinct from the study of lexicology in general, we now come
back to the problem shared by both lexicology and terminology.
some generalisable phenomena. The problem thus again becomes how to

characterise the specific phenomena of terminology in such a way that the
characterisation at the same time allows for the dynamics and the prediction
of the changes of terminology.
In order to clarify the task of refining the modelling of the dynamics of
terminology within the actuality of parole, therefore, it is necessary to ex
amine important factors relevant to the dynamics which would complement
and enrich the types of description carried out so far in this book. From the
discussions in Chapter 2, it is immediately clear that the diachronic perspec
tive should be incorporated and the factors related to the sphere of discourse
should be taken into account.
10.2.2.1 Validation of synchronic dynamics from the diachronic viewpoint

In 10.1.3, we discussed how the descriptions of the dynamics of terminol
ogy in Chapters 6 and 9 represent the dynamic potentiality of the terminol
ogy of documentation as seen in its synchronic state, and thus the possible
direction of the creation of terms predicted in the descriptions cannot be
straightforwardly mapped to the changes of terminology along the histor
ical or diachronic axis. On the other hand, the dynamic potentiality de
scribed in those chapters was assumed to constitute an important structural
factor which influences the actual diachronic changes of terminology, and
the potentiality only shows itself through the actual changes of terminology
along the diachronic axis. Let us start the discussion here, therefore, from
examining the relationship between the dynamic potentiality observed in
the synchronic state of terminology and the diachronic changes or evolu
tion of terminology.
If, for instance, the dynamic potentiality of terminology as described in
Chapters 6 and 9 was the only factor in the creation of terms, then the range
of terms newly created in the field of documentation would totally fall in
the range of patterns and tendencies predicted in Chapters 6 and 9, inas
much as the data used in the analyses properly represent the terminology of
documentation. Let us examine this point for a few conceptual categories
using the new data set. For that, we use JSLS (1997), which includes 1,511
terms in the field of library science. The data used so far was taken from
Wersig & Neveling (1984), whose original edition was published in 1976.
So JSLS (1997) is expected to reflect the changes in the terminology during
the 20 intervening years
Note that, although JSLS (1997) was the best choice for validation at
the time of writing this book, it has a different editing policy from Wersig
& Neveling (1984) and the field covered is not completely the same (library
science vs. documentation). Thus, the result of the observations cannot be
attributed solely to the gap between the synchronic dynamics and the actual
diachronic change of the terminology. Also, for some conceptual categories
established in Chapter 4, the number of terms in JSLS (1997) is smaller
than that in Wersig & Neveling (1984). This shows that regarding JSLS
(1997) as the result of monotonic accumulation from Wersig & Neveling
(1984) is not completely justified. This said, the comparison between Wer
sig & Neveling (1984) and JSLS (1997) would still be interesting for the
immediate purpose of obtaining the general relation between the dynamic
potentiality and the actual change of terminology.
As a rigid examination is not possible nor very useful given the differ
ences in the two sets of data, we only examine qualitatively the tendencies
of the conceptual specification patterns for the terms representing three cat
egories: material entity - animate - people - types (111); representa
tional entity - documentation entities - simple units (RE312); and quality
(QL). 111 was chosen because the analyses showed that the terms of
this category are expected to follow very coherent formation and growth
patterns. RE312 was chosen because it is one of the central conceptual cat
egories of the field of documentation and library science, and at the same
time the patterns show reasonable diversity. QL was chosen as a represen
tation of a non-entity category.
10.2.2.1.1 Material entity - animate -people - types (111)
In JSLS (1997), there are 31 (4 simple, 24 two-item and 3 three-item) terms
that belong to 111 (proper names were excluded). Table 10.1 shows
these terms with their conceptual specification patterns. The terms marked
with an asterisk (*) are also in Wersig & Neveling (1984)5. As observed in
6.1.1.1 (Table 6.1), specifications of functional link (FFUN) and of whole
or affiliations constitute the majority; the formation of 111 terms in
5
Tables 10.1, 10.2 and 10.3 show that the number of terms listed in Wersig & Neveling
(1984) is not large. This seems to reflect the difference in the field of coverage and editorial
policy rather than the creation and/or disappearance of terms. For this reason, we did not
make a quantitative analysis of the relationship between the terms in JSLS (1997) and the
terms in Wersig & Neveling (1984), nor did we merge these two to keep the data in line
with the monotonic accumulation model.
Table 10.1. Material entity - animate - people - types (111) terms in JSLS (1997).
general follows the patterns formulated in Chapter 6. However, the data

in JSLS (1997) contradicts the predictions, made in 9.3.1.1 (Figure 9.2),
that no new conceptual specification patterns were expected to occur. In
JSLS (1997), 5 new conceptual specification patterns appear, though the
total number of terms that take these specification patterns is small.
10.2.2.1.2 Representational entities - documentation entities - simple
units (RE312)
There are 55 (6 simple, 46 two-item and 3 three-item) terms that belong to
RE312. Table 10.2 shows these terms with their conceptual specification
patterns. The central specification patterns are the specification of func
tional link (FFUN: 11), followed by the specifications of information con
tent and representation (PICR: 10), of use (FUSE: 6), of formal attributes
(IFOA: 5), of scope (RSCO: 5) and of status (RSTA: 5). These are the same
as the major specification patterns observed in the RE312 terms in Wersig
& Neveling (1984) (see 6.1.2.3 and Table 6.10), though the exact ratios
of these patterns are different. JSLS (1997) also follows the tendencies
Table 10.2. Representational entities - documentation entities - simple units (RE312) terms
in JSLS (1997).
observed in 6.1.2.3, in that a few specification patterns, i.e. specifications

of destination, of whole or affiliations, of role, of nature and of contents,
occur in addition to the major patterns. A new specification pattern, i.e.
specification of origin, is observed in Table 10.2. This is consistent with
Table 10.3. Quality (QL) terms in JSLS (1997).
the prediction made in 9.3.2.3, where a few new patterns were predicted to
occur when new terms of this category are created.
10.2.2.1.3 Quality (QL)
There are 23 (2 simple and 21 two-item) terms that belong to QL. Table 10.3
shows these terms with their conceptual specification patterns. Here again,
the major pattern, the specification of attributed concept (IACO: 18 in total),
is the same as that observed in the QL terms in Wersig & Neveling (see
6.1.5 and Table 6.18). Other than IACO, only two specification patterns
are observed, of which specification of use (FUSE) was observed also in
Wersig & Neveling (1984). On the other hand, the quantitative analysis
in 9.3.5 predicted that few new specification patterns would occur, which
contradicts the data here; the specification or nature (INAT) occurs twice in
JSLS (1997). As in 111 terms, the general tendency of the specification
patterns observed in 6.1.5 holds for the terms in JSLS (1997), but not the
detailed predictions.
10.2.2.2 Synchronic dynamics and diachronic perspective

From the above examples, it can be seen that the model of the patterns of
term formation and terminological growth established on the basis of Wer-
sig & Neveling (1984) explain the tendencies observed in JSLS (1997),
though many terms are not listed in Wersig & Neveling (1984). This
proves that the basic starting hypothesis of the present study, formulated
in 2.2.1, is, in principle, valid. We can now formulate the following as an
empirically-valid observation.
Some systemic/systematic factors in the existing terminology

of a domain contribute to determining the formation of new
terms and the growth of terminology.
On the other hand, there are cases which violate the predictions of the
structural model. There are two possible reasons for this. Firstly, as men
tioned above, the editing policies and the fields of coverage of Wersig &
Neveling (1984) and JSLS (1997) are different. Secondly, even if Wersig &
Neveling (1984) and JSLS (1997) had the same editing policy and the same
coverage (and we thus would have been able to compare the equivalent data
for 1976 and for 1997), it is still expected that the model which is based on
idiosynchronic terminological data would not be able to predict completely
the actual diachronic changes of the terminology. This is because the for
mer only describes the realistic potentiality as expected from an existing
structure of terminology, while the actual changes of terminology are also
affected by other factors related to language facts or parole. The dynamic
potentiality observed in the existing terminology provides a necessary con
dition or a basic restriction in which actual changes of terminology take
place, but does not uniquely determine the formation of new terms.
In order to explore the study of the dynamics of terminology beyond
what has been carried out so far, it is technically necessary to overcome the
first obstacle of obtaining theoretically equivalent data at different times6.
Assuming that the problem of data collection is solved, it is necessary to
6
This causes a serious problem in the study of lexicology. While textual data is con
sidered to be a "natural" product, lexicological data can be consolidated only as a result of
secondary processing of the "natural" language data. Though theoretically speaking, the
lexicological sphere is not necessarily secondary to the textual sphere, lexicological data
is. One possible way of obtaining equivalent terminological data for different times is to
apply some automatic measures, as proposed in, for instance, Frantzi & Ananiadou (1999),
Hisamitsu, Niwa & Tsujii (2000) or Nakagawa (2001), to the textual data of different times
which are themselves obtained by consistent sampling with proper consideration of such
factors as register variation (Biber 1995).
examine the status of the structural synchronic model of dynamics vis-à-vis

the diachronic change of terminology.
Restating the point so far: (i) Inasmuch as it is claimed to have some
sort of prediction power, the model of the dynamics of terminology as for
mulated in this book should be validated in terms of actual data obtained at
different times, and (ii) the model will not be able to capture completely the
actual change of terminology. As the structural synchronic model aims at
grasping the dynamic potentiality in the structure of terminology, denying
the model just because it does not match the reality is not theoretically valid.
What is required, then, is to establish an overall framework for research into
the dynamics of term formation and terminological growth which takes ad
vantage of structural synchronic modelling and complements its limitation.
The first step in this direction is to establish structural synchronic mod
els for the terminological data of the same domain at different time slices,
and to trace the differences and changes in these models. This means re
garding history as a diachronic layer of synchronic slices (cf. Maruyama
1981). There is, however, an essential difference between this and a simple
comparison of the data at different times. In the latter, the structural dynam
ics behind the data was either unassumed or simply assumed to be constant,
without identifying and differentiating many factors that may affect the ac
tual change of the terminology; the direct comparison of the diachronic
data results either in a retrospective description of what has happened or
in coarse predictions in which different factors are not identified. Tracing
the changes in the models of the dynamics at different diachronic stages,
on the other hand, is based on the distinction of two factors, i.e. between
the structural synchronic dynamics, which regulates the potentiality of term
formation and terminological growth, and the diachronic change of the dy
namics. This contributes to modularising the factors that affect the actual
changes of terminology.
Figure 10.1 illustrates the basic schema of the study of the dynamics of
terminology which takes into account the diachronic perspective discussed
so far. This can be regarded as the elaboration of what was shown in Fig
ure 2.2. It also shows the difference between the diachronic study based
on the synchronic models of dynamics and the simple comparison of the
diachronic data.
Figure 10.1. Study of the dynamics of terminology from a diachronic perspective.
10.2.2.3 Dynamics in terminological space and dynamics in discourse

In 2.2.1, two spheres, i.e. the system of terminology and the sphere of dis
course, were identified as relevant to studying the dynamics of terminology
(see especially Figure 2.1). Of these two, only the system of terminology
has so far been taken into account. In order to explore fully the study of the
dynamics of terminology, it is crucial to incorporate relevant factors recog-
nised from the use of terms in discourse. These may include, among others,
the following:
1. How microscopic variations of individual terms in discourse affect the
actual change of terminology over time.
2. How the difference in terms across different register variations of text
and discourse affects the change of terminology over time.
3. How inter-domain interaction in discourse affects the structuring, for
mation and change of terms and terminology of one domain7.
Figure 10.2. The overall framework of the study of the dynamics of terminology.
At the moment, only a few studies deal with the dynamics of termi
nology in relation to the factors related to discourse. The pioneering work
by Tartier (2001) tries to explore the first aspect of the above three, tak
ing full advantage of the recent availability of corpora and the development
of computational tools. But the potential of her approach has yet to be
7
Note that the interaction of terminology mentioned at the end of 10.2.1 can be studied
at this stage.
fully exploited. Studies such as Shaw & Gaines (1993), Polanco, Grivel
& Royauté (1995) and Ibekwe-SanJuan (1998) touch upon the relation be
tween terms (and their variations) and variations of texts or interactions of
different domains. But their aim is to clarify the patterns of evolution or in
teractions of scientific domains through the observation of terminology; the
direction is opposite. Other studies such as Beaugrande (1996) or Ahmad
(1996) touch on some of these aspects, but they remain mostly speculative.
In addition, presently it is not at all clear how the factors in the sphere of
discourse and in the terminological space can be integrated into a single
modelling of the dynamics of terminology. This may perhaps become clear
only through concrete individual studies, such as Tarder (2001), that deal
with such aspects. Figure 10.2 illustrates the overall framework of the study
of the dynamics of terminology. At the moment, the schema necessarily re
mains rough, as the concrete integration of the different types of study is
yet to be established.
The extension of the study may go beyond the linguistically-bound
concept of discourse into the social and cultural sphere. This is the natural
consequence of the fact that terminology is a concept consolidated in the
sphere of parole. From the wider perspective of the study, therefore, the
vast space of the dynamics of terminology remains unexplored.
Looking back from here, it is clear that what is addressed in this book is
only a small step towards the full exploration of the complex world of term
formation and terminological growth. It was, however, an essential step, for
two reasons. Firstly, because the descriptions given in this study constitute
a core part of the overall study of the dynamics of terminology. Secondly,
because it tried to establish the study of the dynamics of terminology as
distinct from the study of the dynamics of data which happen to be a set
of terms. Though this sounds like a mere truism, it has so far largely been
overlooked in much of the otherwise interesting research in terminology.
Appendices
Appendix A
List of Conceptual Categories
Tag Conceptual Category

ME Material Entity
ME1 Animate
ME11 — People
111 Types
ME112 Parts
ME12 — Organisations
ME2 Inanimate
ME21 — Information Carriers
ME211 Documents
ME2111 Types
ME21111 (by mode of publication/formal structure)
ME21112 (by content/mode of publication)
ME21113 (by physical form)
ME2112 Parts
ME212 Non-Document Information Carriers
ME22 — Machines and Implements
ME23 — Matériák
ME24 — Places/Locations
ME25 — Parts of Inanimate Entities
ME251 Parts in General
ME252 Parts as Places/Locations
RE Representational Entity
RE1 Broad Representational Entities
RE11 — General
RE12 — Specific
RE121 (by function and relation)
RE122 (by representational structure and type)
RE2 Linguistic Entities
RE21 — Set Concepts
RE22 — Element Concepts
RE3 Documentation Entities
RE31 — Types
RE311 Complex Units
RE312 Simple Units
RE3121 (by function and relation)
276 Appendix A

RE3122 (by structure)
RE32 — Parts of Documents
RE4 Software Entities
RE41 — Programme Units
RE42 — Programme Parts
RE43 — Software Elements
AE Abstract Entity
ΑΕ1 Information Entities
11 — Basic
ΑΕ12 — Linguistic
ΑΕ13 — Operational
ΑΕ131 General
ΑΕ132 Specific
ΑΕ2 Systems of Knowledge
ΑΕ21 — Subject Fields
ΑΕ211 General
ΑΕ212 Specific
ΑΕ22 — Methodologies
ΑΕ221 General
ΑΕ222 Specific
 Subjective Entities
ΑΕ31 — Roles/Evaluations
ΑΕ32 — Relations
 — Psychological
CE Classificatory Entity
CE1 (by formal complexity)
11 — Single/Simple Items
111 Independent Concepts in Documentation
112 Non-independent Concepts in Documentation
12 — Complex Items
121 Seen as One
1211 Independent Concepts
1212 Non-independent Concepts
CE122 Seen as Composed
1221 Independent Concepts
CE1222 Non-independent Concepts
CE2 (only by relation or role)
CE21 Independent Concepts
CE21 Non-independent Concepts
QL Quality
QL1 Labels of Types of Qualities
QL11 — Undefined (General) Concepts
QL12 — Redefined Concepts for Documentation
QL13 — Qualitative/Quantitative Measurements
QL2 Values of Qualities
QL201 — Spatial Quality
QL202 — Temporal Quality
QL203 — Functional/Situational Quality
List of Conceptual Categories 277

QL204 — Original Situation
QL205 — Shape
QL206 — Colour
QL207 — Amount/Quantity
QL208 — Scale
QL209 — Generality/Abs traction
QL210 — Negative
QL211 — Naturalness
QL212 — Completeness/Detailedness
QL213 — Structure/Mode
QL214 — Representational Mode
QL215 — Manner/Means
RL Relation
RL1 Labels of Types of Relations
RL11 — Undefined (General) Concepts
RL12 — Redefined Concepts for Documentation
RL13 — Relational Measurements
RL2 Values of Relations
RL21 — General Values
RL22 — Spatial
RL23 — Temporal/Stage in Process
RL24 — Functional/Beneficial/Causal
RL25 — Other Mode of Relation
RL26 — Comparison
RL27 — Relative Status
AC Activity
AC1 Units of Activities/Activity Indicators
AC2 Specific Activities
AC21 — Action
AC211 Instant Activity
AC212 Static Durative Activity
AC213 Dynamic Durative Activity
AC22 — Transference
AC221 General Transfer
AC222 Representational Entity (Information) Transfer
AC223 Material Entity (Document) Transfer
AC23 — Production
AC231 General Production
AC2312 Static Durative Activity
AC232 Representational Entity Production
278 Appendix A

AC233 Material Entity (Document) Production
AC24 — State Change
AC241 Functional or Relational State Change
AC242 Material or Representational State Change
Appendix 
Lists of Intra-Term Relations and
Conceptual Specification Patterns
B.l List of Intra-Term Relations

Tag Intra-Term Relation
 DET (determinant) represents the COncept to which NUC (nucleus) is
Attributed
AFO DET represents the AFfected Object of NUC
CIN Connection INtroduction
CON DET represents the CONstituents of NUC
DES DET represents the DEStination of NUC
DTA DET is used as the Differentiation TAg of NUC
FOA DET represents the FOrmal Attributes of NUC
FUN DET represents the FUNction of NUC
ICR DET represents the Information Content & Representation of NUC
JXT JuXTaposition
LOC DET represents the LOCation of NUC
MAN DET represents the MANner/mode of NUC
MEA DET represents the MEAns of NUC
NAT DET represents the NATure of NUC
ORI DET represents the ORIgin of NUC
PAR DET represents the whole of which NUC is a PARt
PRO DET represents the PROduct of NUC
QUA DET represents the QUAntity of NUC
ROL DET represents the ROLe of the NUC
SCO DET represents the SCOpe of NUC
SSP DET SPecifies the Subject of NUC
STA DET represents the STAtus of NUC
І DET represents the TIMe of NUC
USE DET represents the USE of NUC
280 Appendix 
B.2 List of Conceptual Specification Patterns
Tag Conceptual Specification Pattern
EJXT Juxtaposition
FCOM DET specifies the complementary elements of NUC as a function, by AFO,
PRO, MEA and/or MAN
FDES DET specifies the use of NUC by DES
FFUN DET specifies the functional link of NUC, by FUN, AFO, PRO, MEA and/or
MAN
FROL DET specifies the role of NUC by ROL
FUSE DET specifies the use of NUC, by USE and/or MAN/MEA
IACO DET specifies the concept to which NUC is attributed by 
IFOA DET specifies the formal attributes of NUC by FOA
IMAN DET specifies the static manner of NUC by MAN
INAT DET specifies the nature of NUC by NAT
IQUA DET specifies the quantity of NUC by QUA
ODTA DET differentiates NUC by DTA
PCON DET specifies the constituent elements of NUC by CON
PICR DET specifies the information contents and representation of NUC by ICR
PPAR DET specifies the whole or affiliations of NUC by PAR
RLOC DET specifies the location of NUC by LOC
RORI DET specifies the origin of NUC by ORI and MAN or FOA
RSCO DET specifies the scope of NUC by SCO
RSTA DET specifies the status of NUC by STA
RSUB DET specifies the subject of NUC by SSP
RTIM DET specifies the time of NUC by TIM
Appendix 
List of Terms by Conceptual Categories
Each entry consists of (i) conceptual category tag, (ii) Japanese term, and
(iii) English term. "=" shows embedded combinations.
282 Appendix 
List of Terms by Categories 283
284 Appendix 
286 Appendix 
288 Appendix 
290 Appendix 
292 Appendix 
List of Terms by Categoríes 293
294 Appendix 
Appendix D
List of Morphemes by Conceptual Categories
Each entry consists of (i) conceptual category tag, (ii) Japanese morpheme
(* is added if the morpheme is itself a term), (iii) English translation and
(iv) occurrence in the corpus.
296 Appendix D
List of Morphemes by Categories 297
298 Appendix D
300 Appendix D
302 Appendix D
Bibliography
Achinstein, P. 1968. Concepts of Science —A Phylosophical Analysis. Baltimore:

Johns Hopkins University Press.
Adams, V. 1973. An Introduction to Modern English Word-Formation. London:
Longman.
Ahmad, K. 1996. "A terminology dynamic and the growth of knowledge: A case
study in nuclear physics and in the philosophy of science," In Galinski, . and
Schmitz, K-D. (eds.) Terminology and Knowledge Engineering '96. Frankfurt:
Indeks Verlag, p. 1-11.
Aitchison, J and Gilchrist, A. 1997. Thesaurus Construction and Use: A Practical
Manual (3rd ed). London: Aslib.
Akhmanova, O. 1974. "When is a word a 'term' and how do we distinguish
between terms and words of 'everyday' language?" In Akhmanova, O. and
Agapova, G. (eds.) Terminology: Theory and Method. Moscow: Moscow State
University, p. 22-30.
Aristotle, 1963. Categories and De Interpretatione. London: Clarendon Press.
[Ackrill, J. L. (trans.)]
Baayen, R. H. 1996. 'The effects of lexical specialization on the growth curve of
the vocabulary," Computational Linguistics 22(4), p. 455-480.
Baayen, R. H. 2001. Word Frequency Distributions. Dordrecht: Kluwer Aca
demic.
Bauer, L. 1983. English Word-Formation. Cambridge: Cambridge University
Press.
Beaugrande, R. de 1996. "LSP and terminology in a new science of text and dis
course," In Galinski, . and Schmitz, K-D. (eds.) Terminology and Knowledge
Engineering '96. Frankfurt: Indeks Verlag. p. 12-26.
Bessé, . de, Nkwenti-Azeh, . and Sager, J. C. 1997. "Glossary of terms used in
terminology," Terminology 4(1), p. 117-156.
Biber, D. 1995. Dimensions of Register Variation. Cambridge: Cambridge Uni
versity Press.
Bloomfield, Լ. [1933] 1984. Language. Chicago: University of Chicago Press.
Bowker, L. 1997. "Multidimensional classification of concepts and terms," In

Wright, S. E. and Budin, G. (eds.) Handbook of Terminology Management 1.
Amsterdam: John Benjamins. p. 133-143.
Budin, G. 1996. "Evolution of scientific terminology," In Galinski, . and
Schmitz, K-D. (eds.) Terminology and Knowledge Engineering '96. Frankfurt:
Indeks Verlag. p. 27-34.
Budin, G. and Oeser, E. 1995. "Controlled conceptual dynamics: From 'ordinary
language' to scientific terminology — and back," Terminology Science and Re
search 6(2), p. 3-17.
Cabré, M. T. 1993. La Terminologia: Teoria, Metodologia, Aplicaciones.
Barcelona: Editorial Antártida/Empúries.
Cabré, M. T. 1995. "On diversity and terminology," Terminology 2(1), p. 1-16.
Cabré, M. T. 2000. "Elements for a theory of terminology: Towards an alternative
paradigm," Terminology 6(1), p. 35-57.
Carlson, L. and Nirenburg, S. 1991. "World modeling for NLP," In EDR (ed.)
Proceedings of the International Workshop on Electronic Dictionaries. np.
Chan, L. M. 1994. Cataloging and Classification: An Introduction (2nd ed). New
York: McGraw-Hill.
Chitashvili, R. and Baayen, R. H. 1993. "Word Frequency Distributions," In
Altmann, G. and Hrebicek, L. (eds.) Quantitative Text Analysis. Trier: Wis
senschaftlicher Verlag. p. 54-135.
Chomsky, N. 1964. "Formal discussion: 'The development of grammar in child
language' by W. Miller and S. Ervin," In Bellugi, U. and Brown, R. (eds.) The
Acquisition of Language. Indiana: Purdue University Press. p. 35-39.
Church, K. W. and Gale, W. A. 1991. "A comparison of the enhanced Good-
Turing and deleted estimation methods for estimating probabilities of English
bigrams," Computer Speech and Language 5(1), p. 19-54.
Collins, A. and Quillian, M. R. 1969. "Retrieval time from semantic memory,"
Journal of Verbal Leaning and Verbal Behaviour 8(2), p. 240-247.
Cruse, D. A. 1986. Lexical Semantics. Cambridge: Cambridge University Press.
Crystal, D. 1992. An Encyclopedic Dictionary of Language and Linguistics. Ox
ford: Blackwell.
Culler, J. 1976. Saussure. London: Fontana.
Dahlberg, I. 1978. "A referent-oriented, analytical concept theory for INTER-
CONCEPT," International Classification 5(3), p. 142-151.
Daille, ., Habert, ., Jacquemin, . and Royauté, J. 1996. "Empirical observation
of term variations and principles for their description," Terminology 3(2), p.
197-257.
Datta, S. and Farradane, J. E. L. 1978. "A psychological basis for general classifi
cation," In Wojciechowski, J. A. (ed.) Conceptual Basis of the Classification of
Knowledge. New York: K. G. Saur. p. 319-331.
Bibliography 305
DeGroot, M. 1984. Probability and Statistics (2nd ed). Reading: Addison Wesley.
Deuleuze, G. 1973. "A Quoi Reconnait-on le Structuralisme?" In Chattet, F. (ed.)
La Philosophie au XXe Siècle. Paris: Hachette. [Nakamura, Y. (trans.) 1998.
"Kouzousyugi ha naze sou yobarerunoka," Nakamura, Y. (trans. & ed.) Ni-
jusseiki no Tetsugaku (Seiyou Tetsugaku no Chi VIII). Tokyo: Hakusuisya. p.
332-371.]
Desmet, I. and Boutayeb, S. 1994. "Terms and words: Propositions for terminol
ogy," Terminology 1(2), p. 303-325.
Downing, P. A. 1977. "On the creation and use of English compound nouns,"
Language 53(4), p. 810-842.
Ducrot, O. and Todorov, T. 1979. Encyclopedic Dictionary of the Sciences of
Language. Baltimore: Johns Hopkins University Press. [Porter, C. (trans.)]
EDR, 1989. Gainen Jisyo (TR-012). 2nd ed. Tokyo: Japan Electronic Dictionary
Research Institute.
Fabre,  1996. "Interpretation of nominal compounds: Combining domain-
independent and domain-specific information," COLING'96, p. 364-369.
Felber, H. 1984. Terminology Manual. Paris: Unesco and Infoterm.
Fellbaum, . (ed.) 1998a. WordNet: An Electronic Lexical Database. Cambridge:
MIT Press.
Fellbaum,  1998b. "A semantic network of English verbs," In Fellbaum, C. (ed.)
WordNet: An Electronic Lexical Database. Cambridge: MIT Press. p. 69-104.
Finin, T. 1980. The Semantic Interpretation of Compound Nominals (TR -96).
Illinois: University of Illinois at Urbana Champaign.
Foucault, M. 1968. "Sur l'archéologie des sciences: Réponse au cercle
d'epistemologie," Cahiers peur l'Analyse 9 (Génealogie des Sciences), p. 9-
40. [Ishida, H. (trans.) "Kagaku no koukogaku," Michel Foucault Shikou Syusei
3. Tokyo: Chikuma. p. 100-143.
Frantzi, T. K. and Ananiadou, S. 1999. "The C-value/NC-value method for ATR,"
Journal of Natural Language Processing 6(3), p. 145-179.
Fujii, T. 1976. " 'Doushi+teiru' no imi," In Kindaichi, H. (ed.) Nihongo Doushi
no Asupekuto. Tokyo: Mugisyobo. p. 97-116.
Fujita, K. 1984. Nihongo Meishi Renzoku Fukugougo no Kouzou Kaiseki. MSc
Thesis, Kyoto University.
Fujiwara, S. and Fujiwara, Y. 1987. Sougou Gakujutsu Yougosyu. Tokyo: Kinoku-
niya.
Gale, W. A. and Sampson, G. 1995. "Good-Turing frequency estimation without
tears," Journal of Quantitative Linguistics 2(3), p. 217-237.
Geeraerts, D. 1994. "Lexicology," In Asher, R. E. (ed.), The Encyclopedia of
Language and Linguistics 4. Oxford: Pergamon Press. p. 2189-2192.
Good, I. J. 1953. "The population frequencies of species and the estimation of
population parameters," Biometrika 40(3-4), p. 237-264.
Good, I. J. and Toulmin, G. H. 1956. "The number of new species, and the increase
in population coverage, when a sample is increased," Biometrika 43(1), p. 45-
63.
Greimas, A. J. 1966. Semantique Structurale: Recherche de Méthode. Paris:
Larousse. [Tajima, H. and Torii, M. (trans.) 1988. Kouzou Imiron: Houhou
 . Tokyo: Kinokuniya.]
Guda, R. V. and Lenat, D. . 1990. "Cyc: A mid-term report,"AI Magazine 18(3),
p. 32-59.
Hatcher, A. G. 1960. "An introduction to the analysis of English noun com
pounds," Word 16(3), p. 356-373.
Hayashi, C. 1993. Koudou Keiryougaku Josetsu. Tokyo: Asakura.
Hempel, C. G. 1952. Fundamentals of Concept Formation in Empirical Science.
Chicago: University of Chicago Press.
Herdan, G. 1960. Type-Token Mathematics: A Textbook of Mathematical Linguis
tics. 's-Gravenhage: Mouton.
Hisamitsu, T. Niwa, Y. and Tsujii, J. 2000. "A method of measuring term rep
resentativeness — baseline method using co-occurrence distribution," COL
ING'2000, p. 320-326.
Hockett, C. F. 1958. A Course in Modern Linguistics. New York: Macmillan.
Hoffmann, L. 1979. "Towards a theory of LSP," Fachsprache 1(1/2), p. 12-17.
Ibekwe-SanJuan, F. 1998. "Terminological variation, a means of identifying re
search topics from texts " COLING-ACL'98, p. 564-570.
Ishii, M. 1986. "Fukugou meishi no gokouzou bunseki ni tsuite no ichi kousatsu
— gakujutsu yougo wo rei ni —," Kokugogaku 144, p. 13-26.
Ishii, M. 1987a. "Fukugou meishi no kouzou to kinou," Mizutani, S., Tajima, K.,
Satake, H., Nomura, M., Ishii, M., and Kabashima, T. Moji-Hyouki to Gokousei
(Asakura Nihongo Shinkouza 1). Tokyo: Asakura. p. 145-173.
Ishii, M. 1987b. "Economy in Japanese scientific terminology," In Czap, Η. and
Galinski, C. (eds.) Terminology and Knowledge Engineering. Frankfurt: Indeks
Verlag. p. 123-136.
Ishikawa, T., Sakamoto, Y and Sato, M. 1986. "Mu projekuto ni okeru imi ma-ka
no gainen to taikei," Jouhou Syori Gakkai u Houkoku Shizengengo Syori
Kenkyukai Houkoku 55-1. np.
ISO 5127-1 1983. Documentation and information— Vocabulary — Part I: Basic
Concepts. Geneva: International Organization for Standardization.
ISO 704 1987. Principles and Methods of Terminology. Geneva: International
Organization for Standardization.
Jackendoff, R. 1983. Semantics and Cognition. Cambridge, Mass: MIT Press.
Jackendoff, R. 1987. Consciousness and the Computational Mind. Cambridge,
Mass: MIT Press.
Jackendoff, R. 1990. Semantic Structures Cambridge, Mass: MIT Press.
Bibliography 307
Jacquemin, . 2001. Spotting and Discovering Terms through Natural Language

Processing. Cambridge, Mass: MIT Press.
Jakobson, R. 1980. Framework of Language. Michigan: Michigan University.
Jespersen, . 1942. A Modern English Grammar on Historical Principles. Part
VI. Morphology. London: George Allen & Unwin.
JSLS: Japan Society of Library Science 1997. Tosyokan Jouhougaku Yougo Jiten.
Tokyo: Maruzen.
JIS 1989. Dokyumenteesyon Yougo (Kaisetsu, Sakuin). Tokyo: Nihon Kikaku
Kyoukai.
JIS X 0701 1989. Dokyumentesyon Yougo (Kihon Gainen). Tokyo: Nihon Kikaku
Kyoukai.
JIS X 0702 1989. Dokyumentesyon Yougo (Bunken). Tokyo: Nihon Kikaku Ky
oukai.
JIS X 0705 1989. Dokyumentesyon Yougo (Bunken oyobi Deta no Syusyu, Tokutei
oyobi Bunseki). Tokyo: Nihon Kikaku Kyoukai.
JIS X 0706 1989. Dokyumentesyon Yougo (Dokyumentesyon Gengo). Tokyo:
Nihon Kikaku Kyoukai.
Kageura, K. 1993. A Conceptual Analysis of Japanese Complex Noun Terms with
Special Reference to the Field of Documentation. PhD Thesis, University of
Manchester.
Kageura, K. 1994. "Differences in linguistic representations of concepts in
Japanese and English complex noun terms," Terminology 1(1), p. 103-119.
Kageura, K. 1995. 'Toward the theoretical study of terms," Terminology 2(2), p.
239-258.
Kageura, K. 1997a. 'On intra-term relations of complex terms in the description
of term formation patterns," In Lapierre, L., Oore, I. and Runte, H. R. (eds.)
Mélanges de Linguistique offerts à Rostislav Kocourek. Halifax: Les Presses
ď ALFA (a). p. 105-111.
Kageura, K. 1997b. "Multifaceted/multidimensional concept systems," In Wright,
S. E. and Budin, G. (eds.) Handbook of Terminology Management 1. Amster
dam: John Benjamins. p. 119-132.
Kageura, K. 1998a. "On the effect of intra-term morphological coherence on the
growth curve of morphemes in terminology," Mathematical Linguistics 21(7),
p. 311-323.
Kageura, K. 1998b. "A statistical analysis of morphemes in Japanese terminology,"
COLING-ACĽ98,p. 638-645.
Kageura, K. 2000. "The dynamics of phenomena and the dynamics of data: On
the relationship between events and structures in terminology," Mathematical
Linguistics 22(7), p. 281-302.
Kageura, K. 2002. "Measuring the distance of terminologycail structures: A pre
liminary study," Asian-Pacific Workshop on Terminology 2002, p. 69.
Kageura, K., Enosawa, Y. Santana-Rodriguez. M. and Yamilet-Santana. S. 1999.

"A Comparative Observation of English, Spanish and Japanese Terminological
Structure," In Sandrini, P. (ed.) Terminology and Knowledge Engineering '99.
Wien: TermNet. p. 726-737.
Kageyama, T. 1982. "Word formation in Japanese," Lingua 57, p. 215-258.
Kageyama, T. 1989. "Keitairon, gokouseiron," In Sakiyama, S. (ed.) Gengogaku
Yousetsu (1) (Kouza Nihongo to Nihongo Kyouiku 11). Tokyo: Meiji Syoin. p.
60-92.
Katz, J. J. 1972. Semantic Theory. New York: Harper & Row.
Keil, F. . 1979. Semantic and Conceptual Development. Cambridge, Mass: Har
vard University Press.
Kindaichi, H. 1976. "Kokugo doushi no ichibunrui," In Kindaichi, H. (ed.) Ni
hongo Doushi no Asupekuto. Tokyo: Mugisyobo. p. 5-26.
Kita, K. 1999. Kakuritsuteki Gengo Model. Tokyo: University of Tokyo Press.
Kokuritsu Kokugo Kenkyusho (ed.) 1961. Bunrui Goi
 ou. Tokyo: Syuei Syup-
pan.
Kuhn, T. 1962. The Structure of Scientific Revolutions. Chicago: University of
Chicago Press.
Lamberts, K. and Shanks, D. (eds.) 1997. Knowledge, Concepts and Categories.
Cambridge, Mass: MIT Press.
Langacker, R. W. 1987. Foundations of Cognitive Grammar I (Theoretical Prereq
uisites). Stanford: Stanford University Press.
Lara, F. 1998/1999. " 'Concepts' and term hierarchy," Terminology 5(1), p. 59-76.
Laurence, S. and Margolis, E. 1999. "Concepts and cognitive science," In Mar-
golis, E. and Laurence, S. (eds.) Concepts: Core Readings. Cambridge, Mass:
MIT Press. p. 3-81.
Lees, R. B. 1960. The Grammar of English Nominalizations. Bloomington: Indi
ana University Press.
Lenat, D. B. 1995. "Cyc: A large scale investment in knowledge infrastructure,"
Communications of the ACM 38(11), p. 38-45.
Lenat, D. ., Guha, R. V., Pittman, K., Pratt, D. and Shepherd, M. 1990. "CYC:
Toward programs with common sense," Communications of the ACM 33(8), p.
40-49.
Leonard, R. 1984. The Interpretation of English Noun Sequences on the Computer.
Amsterdam: North-Holland.
Levi, J. N. 1978. The Syntax and Semantics of Complex Nominals. New York:
Academic Press.
Lévi-Strauss, . 1958. Anthropologie Structurale. Paris: Librairie Plon.
[Arakawa, L, Ikumatsu, K., Kawada, J., Sasaki, A. and Tajima, S. (trans.) 1972.
Kouzou Jinruigaku. Tokyo: Misuzu.]
Lyons, J. 1968. Introduction to Theoretical Linguistics. Cambridge: Cambridge
University Press.
Bibliography 309
Lyons, J. 1977. Semantics. 2 vols. Cambridge: Cambridge University Press.

Lyons, J. 1981. Language and Linguistics: An Introduction. Cambridge: Cam
bridge University Press.
Lyons, J. 1995. Linguistic Semantics. Cambridge: Cambridge University Press.
Maeda, T. 1989. "Goi souron," In Tamamura, F. (ed.) Nihongo noGoi,Imi (Kouza
Nihongo to Nihongo Kyouiku 6). Tokyo: Meiji Syoin. p. 1-22.
Manning, C. D. and Schütze, H. 1999. Fundamentals of Statistical Natural Lan
guage Processing. Cambridge, Mass: MIT Press.
Marchand, H. 1969. The Categories and Types of Present-day English Word-
Formation. 2nd. ed. Munich: . . Beck.
Margolis, E. and Laurence, S. (eds.) 1999. Concepts: Core Readings. Cambridge,
Mass: MIT Press.
Maruyama, K. 1981. Saussure no Shisou. Tokyo: Iwanami.
Masterman, M. 1962. "Semantic message detection for machine translation, us
ing an interlingua," 1961 International Conference on Machine Translation of
Languages and Applied Language Analysis, p. 437-475.
McDonald, D. B. 1982. Understanding Noun Compounds. Ph.D. Dissertation,
Carnegie-Mellon University.
Mel' čuk, I. and Zholkovsky, A. 1988. "The explanatory combinatory dictionary,"
In Evens, M. W. (ed.) Relational Models of the Lexicon: Representing Knowl
edge in Semantic Networks. Cambridge: Cambridge University Press. p. 41-74.
Meyer, I. and Mackintosh, K. 2000. "When terms move into our everyday lives:
An overview of the de-terminologisation," Terminology 6(1), p. 111-138.
Miller, G. A. 1998. "Nouns in WordNet," In Fellbaum, . (ed.) WordNet: An
Electronic Lexical Database. Cambridge: MIT Press. p. 23-46.
Miller, G. A. and Johnson-Laird, P. N. 1976. Language and Perception. Cam
bridge: Cambridge University Press.
Minoya,  1994. Toukeigaku Nyumon. 2 vols. Tokyo: Tokyo Tosyo.
Miyajima, T. 1973. "Muimi keitaiso," In Kokuritsu Kokugo Kenkyusyo (ed.) Ko-
toba no u. Tokyo: Kokuritsu Kokugo Kenkyusyo. p. 15-30.
Miyajima, T. 1981. Senmongo no Syomondai. Tokyo: Syuei Syuppan.
Mombusyou (ed.) 1958. Gakujutsu Yougosyuu — Tosyokangaku Hen. Tokyo:
Dainihontosyo.
Mugdan, J. 1994. "Morphological units," In Asher, R. E. (ed.), The Encyclopedia
of Language and Linguistics 5. Oxford: Pergamon Press. p. 2543-2553
Murphy, G. L. and Lassaline, M. E. 1997. "Hierarchical Structure in Concepts and
Basic Level of Categorization," In Lamberts, K. and Shanks, D. (eds.) Knowl
edge, Concepts, and Categories. Cambridge, Mass: MIT Press. p. 93-131.
Nagashima, Y. 1980. "Gokousei no hikaku," In Kunihiro, T. (ed.) Onsei to Keitai
(Nichi-Eigo Hikaku Kouza 1). Tokyo: Taisyuukan. p. 235-285.
Nakagawa, H. 2001. "Experimental evaluation of ranking and selection methods
in term extraction," In Bourigault, D., Jacquemin, C. and L'Homme, M-C. (eds.)
Recent Advances in Computational Terminology. Amsterdam: John Benjamins.

p. 303-325.
Nishio, T. 1988. Gendai Goi no u. Tokyo: Meiji Syoin.
Nomura, M. 1973. "Fukuji ketsugougo no kouzou," In Kokuritsu Kokugo
Kenkyusyo (ed.) Densikeisanki ni yoru Kokugo u (V). Tokyo: Kokuritsu
Kokugo Kenkyusyo. p. 72-93.
Nomura, M. 1975. "Yoji kango no kouzou," In Kokuritsu Kokugo Kenkyusyo
(ed.) Densikeisanki ni yoru Kokugo u (VII). Tokyo: Kokuritsu Kokugo
Kenkyusyo. p. 36-80.
Nomura, M. and Ishii, M. 1988. Gakujutsu Yougo Goki Rensetsu
 ou. Tokyo:
Kokuritsu Kokugo Kenkyusyo.
NTT Communication Research Laboratory, 1997. Nihongo Goi Taikei. 5 vols.
Tokyo: Iwanami.
Ogino, T. 1987. "Nihongo no imi bunrui shian," Keiryou Kokugo Gakkai Dai 31
Kai Taikai Happy ou Shiryou. np.
Ogino, T. 1998. "Yourei ha ikutsu shirabereba yoika," Keiryou Kokugogaku 22(1),
p. 11-17.
Okuda, Y. 1983. "Wo kaku no meishi to doushi no kumiawase," In Gengogaku
Kenkyukai (ed.) Nihongo Bumpou — Rengoron (Shiryouhen). Tokyo: Mugisy-
obo. p. 21-149.
Okutsu, K. 1975. "Fukugoumeishi no seiseibumpou," Kokugogaku 101, p. 19-34.
Pearson, J. 1998. Terms in Context. Amsterdam: John Benjamins.
Piaget, J. 1969. Le Structuralisme. Paris: Presses Universitaires de France. [Tak-
izawa, T and Sasaki, A. (trans.) 1970. Kouzousyugi. Tokyo: Hakusuisya.]
Picht, H. and Draskau, J. 1985. Terminology: An Introduction. Surrey: University
of Surrey.
Piotrowski, T. G. 1968. Informacionnye Izmerenija jazyka. Leningrad: Nauka.
[Cited in Tuldava 1995]
Polanco, X., Grivel, L. and Royauté, J. 1995. "How to do things with terms in
informetrics: Terminological variation and stabilization as science watch indi
cators," Proceedings of the 5th International Conference on Scientometrics and
Infometrics, p. 435-444.
Pugh, J. M. 1984. A Contrastive Conceptual Analysis and Classification of Com
plex Noun Terms in English, French and Spanish with Special Reference to the
Field of Data Processing. PhD Thesis, University of Manchester.
Pustejovski, J. 1995. The Generative Lexicon. Cambridge, Mass: MIT Press.
Ranganathan, S. R. 1962. Elements of Library Classification (3rd ed). Bombay:
Asia Publishing House.
Rea, L. M. and Parker, R. Α. 1997. Designing and Conducting Survey Research:
A Comprehensive Guide (2nd ed). San Francisco: Jossey-Bass.
Rey, A. 1995. Essays on Terminology. Amsterdam: John Benjamins. [Sager, J. C.
(trans.)]
Bibliography 311
Rhyne, J. R. 1976. Lexical Rules and Structures in a Computer Model of Nominal

Compounding. Ph.D. Dissertation, University of Texas at Austin.
Rosch, E. 1978. "Principles of categorization," In Rosch, E. and Lloyd, B. B. (eds.)
Cognition and Categorization. Hillsdale, NJ: Lawrence Erlbaum Associates. p.
27-48.
Rosch, E., Mervis, . ., Gray, W. D., Johnson, D. M. and Boyesgraem, P. 1976.
"Basic objects in natural categories," Cognitive Psychology 8(3), p. 382-439.
Roth, I. and Frisby, J. P. 1986. Perception and Representation: A Cognitive Ap
proach. Milton Keynes: Open University Press.
Sager, J. C. 1990. A Practical Course in Terminology Processing. Amsterdam:
John Benjamins.
Sager, J.  1994. "Terminology: Custodian of knowledge and means of knowl
edge transfer," Terminology 1(1), p. 7-16.
Sager, J. C. 1998. "Terminology for translators, a new and different approach?"
Unpublished memo. np.
Sager, J. C. 1998/99. "In search of a foundation: Towards a theory of the term,"
Terminology 5(1), p. 41-57.
Sager, J. C, Dungworth, D. and McDonald, P. F. 1980. English Special Lan
guages — Principles and Practice in Science and Technology. Wiesbaden: Os
car Brandstetter.
Sager, J. C. and Kageura, K. 1995. "Concept classes and conceptual structures
— their role and necessity in terminology," Actes de Langue Française et de
Linguistique 7/8, p. 191-216
Saiga, H. 1957. "Gokousei no tokusitu," In Iwabuchi, E., Hayashi, O., Ohishi,
H., and Shibata, T. (eds.) Kouza Gendai Kokugogaku (II). Tokyo: Chikuma. p.
217-248.
Saito, R. 1981. "Keiyoushi gokan wo zenkou to suru fukugoumeishi no imi,"
Bungeikenkyuu 97',p. 52-63.
Sakakura, Α. 1966. Gokousei  u. Tokyo: Kadokawa.
Sakakura, Α. 1980. "Fukugougo," In Kokugo Gakkai (ed.) Kokugogaku Daijiten.
Tokyo: Tokyodo. p. 743-744.
Sapir, E. 1921. Language. San Diego, Harcourt Brace & Company.
Saussure, F. de. 1916. Cours de Linguistique Generale. Paris: Payot. [Kobayashi,
H. (trans.) 1940. Ippan Gengogaku Kougi. Tokyo: Iwanami.]
Schank, R. C. 1976. "Understanding natural language meaning and intention," In
Papp, F. and Szepe, G.(eds.) Papers in Computational Linguistics. The Hague:
Mouton. p. 259-292.
Seaborg, G. T. 1994. "Terminology of transuranium elements," Terminology 1(2),
p. 229-252.
Shaw, M. L. G. and Gaines, B. R. 1993. "A methodology for analyzing termino
logical and conceptual differences in language use across communities," In
Kohler, R. and Rieger, B. B. (eds.) Contributions to Quantitative Linguistics.

Dordrecht: Kluwer Academic, p. 91-138.
Shelov, S. D. 1982. "The linguistic nature of the term," Nauchno-Tekhnicheskaya
Infonnatsiya series 2, 16(9), p. 1-6.
Smith, E. E. and Medin, E. L. 1981. Categories and Concepts. Cambridge, Mass:
Harvard University Press.
Sonneveld, H. and Loening, K. L. (eds.) 1993. Terminology: Applications in
interdisciplinary communication. Amsterdam: John Benjamins.
Sowa, J. F. 1984. Conceptual Structures — Information Processing in Mind and
Machine. Reading, Mass: Addison-Wesley.
Sparck-Jones, K. 1986. Synonymy and Semantic Classification. Edinburgh: Edin
burgh University Press.
Takeuchi, K., Uchiyama, K., Yoshioka, M., Kageura, K. and Koyama, T. 2001.
'Analysis of relations between nouns and deverbal nouns in Japanese com
pounds based on lexical conceptual structure," PACLING 2001, p. 257-264.
Tamamura, F. 1985. Goi no u to Kyouiku 2. Tokyo: Kokuritsu Kokugo
Kenkyusyo.
Tanaka, H. (ed). 1988. Dictionary of Linguistics. Tokyo: Seibido.
Tartier, A. 2001. "Méthodes d'analyse automatique de ľ "evolution termi
nologique au travers des variations repérées dans les corpus diachroniques,"
Proceedings of the Conférence TIA-2001, p. 191-200.
Temmerman, R. 1995. "The process of revitalisation of old words: "splicing", a
case study in the extension of reference," Terminology 2(1), p. 107-128.
Temmerman, R. 1997. Terminology beyond Standardisation. Language and Cate
gorisation in the Life Sciences. PhD Thesis, University of Leuven.
Temmerman, R. 1998/99. "Why traditional terminology theory impedes a realistic
description of categories and terms in the life sciences," Terminology 5(1), p.
77-92.
Temmerman, R. 2000. Towards New Ways of Terminology Description: The So-
ciocognitive Approach. Amsterdam: John Benjamins.
Toyama, H. 1996. Daisuuteki Kouzou. Tokyo: Nihon Hyoronsya.
Tsuji, . 1998. "Senmon yougo ni okeru dougigo no kyougou to hakusuu/gosyu
no kakawari," Keiryou Kokugo gaku 21(5), p. 206-215.
Tuldava, J. 1995. Methods in Quantitative Linguistics. Trier: Wissenschaftlicher
Verlag.
Uchiyama, ., Takeuchi, ., Yoshioka, M., Kageura, . and Koyama, T. 2001.
"Senmon bunya ni okeru fukugou meishi kaiseki no tameno meishi bumpou
zokusei no bunrui ni tsuite," Keiryou Kokugo gaku 23(1), p. 1-24.
Ullmann, S. 1962. Semantics: An Introduction to the Science of meaning. Oxford:
Basil Blackwell & Mott.
Ungerer, F. and Schmid, H-J. 1996. An Introduction to Cognitive Linguistics. Lon
don: Longman.
Bibliography 313
Vachek, J. 1983. "On some less known aspects of the early Prague linguistic
school," In Vachek, J. and Duskova, L. (eds.) Praguiana : Some Basic and Less
Known Aspects of the Prague Linguistic School. Amsterdam: John Benjamins.
Warren, B. 1978. Semantic Patterns of Noun-Noun Compounds. Lund: Acta Uni
versitaris Gothoburgensis.
Way, E. C. 1991. Knowledge Representation and Metaphor. Dordrecht: Kluwer
Academic.
Wersig, G. and Neveling, U. (eds.) 1976. Terminology of Documentation. Paris:
Unesco.
Wersig, G. and Neveling, U. (eds.) 1984. UNESCO Jouhou Kanri Yougosyuu.
Tokyo: JICST and NIPDOC. [Japanese Version of Wersig, G. and Neveling, U.
(eds.) 1976)]
Wierzbicka, Α. 1996. Semantics: Primes and Universals. Oxford: Oxford Uni
versity Press.
Wüster, E. 1959/60. "Das Worten der Weld, schaubildlich und terminologisch
dargestellt," Sprachforum 3(3), p. 183-204.
Yoshikane, F., Tsuji, F., Kageura, K. and Jacquemin, C. 1999. "Detecting Japanese
term variation in textual corpus," The Fourth International Workshop on Infor
mation Retrieval with Asian Languages (IRAL'99), p. 97-108.
Young, Η. (ed.) 1983. ALA Glossary of Library and Information Science.
Chicago: American Library Association.
Young, H. (ed.) 1988. ALA Tosyokan Jouhougaku Jiten. Tokyo: Maruzen.
[Japanese Version of Young, H. (ed.) 1983)]
Yule, G. U. 1944. The Statistical Study of Literary Vocabulary. Cambridge: Cam
bridge University Press.
Yumoto, S. 1977. "Awasemeishi no imikijutsu wo megutte," Tokyo Gaikokugo
Daigaku Ronsyuu 27. p. 31-46.
Yumoto, S. 1979. 'Awasemeishi no kouzou — n + n taipu no wago meishi no baai
—," In Gengogaku Kenkyukai (ed.) Gengo no u. Tokyo: Mugisyobo. p.
367-395.
Zawada, . and Swanepoel, P. 1994. "On the empirical inadequacy of termino
logical concept theories: A case for prototype theory," Terminology 1(2), p.
253-275.
Index
A 172-174
conditions for application of
"abstract entity" concept 67
177-181
subcategories of 74
fluctuation of values of 174
"abstract entity" terms
binomial model 167-170
conceptual patterns of 136-142 Poisson approximation to 175
quantitative patterns of 227- Bloomfield 11, 40
231 Bowker 254
Achinstein 61 Budin 20, 40
"activity" concept 67
subcategories of 78-80 
"activity" terms
conceptual patterns of 146-152 Cabré 20, 253
quantitative patterns of 236- Carlson 63
243 Carnap 57
Adams 93 Chan 62
Ahmad 272 characteristic 10
Aitchson 62, 71 and semantic features 18
American structural linguistics 40 Chitashvili 37, 168, 171,257
Aristotle 55, 61 Chomsky 12
artificial language 15 Church 172
"classificatory entity" concept 67
average length of terms 178, 185
average use of morphemes 190-192, subcategories of 74-75
"classificatory entity" terms
194-199
conceptual patterns of 142-143
 quantitative patterns of 231-
232
Baayen 37, 168, 171, 172, 175, 178, classificatory regularity of terminol
179, 245, 256, 257 ogy 35, 252
basic-level nucleus 51-53 coefficient of loss 171, 186
Bauer 11,45 Collins 56, 57
Beaugrande 272 Colon Classification 62
Bessé 9, 24 comparative study of terminology
binomial interpolati on/extrapolation 260
competence 12 "representational entity" terms

complex terms 51-54 127-136,265-267
complex morphemes 88-89 concept system
concept 10, 17, 54-56 and concept categories 49, 56-
and morphemes 183-185 58
theoretical position of 21-24 classical view 57
concept categories 56-58, 64-80 for documentation terms 64-80
"abstract entity" 67, 74 hierarchy 57
"activity" 67, 78-80 in term formation 49, 54, 158
basic and functional 70 conceptual specification patterns
"classificatory entity" 67, 74- 58-60,106-113
75 "attributed concepts" 109
and concept system 49, 56-58 binomial interpolation/extrapo
"entity" 66 lation 179-181
general and subject specific 70 characteristics of 111-113
generality and specificity 69 "complementary elements of
and hierarchy 57 functions" 106
"constituent elements" 108
"material entity" 67, 71-72
core and non-core 180
and morphemes 80-86
"destination" 108
"quality" 67, 76-77
"differentiation tag" 110
"relation" 67, 76-77
"formal attributes" 108
"representational entity" 67, "functional link" 107
72-73 "information content and repre
separately-defined and depen sentation" 108
dent concepts 69 level of specificity 180
simple and complex 70 "location" 109
and terms 86-88 "nature" 109
concept patterns of term formation "origin" 109
49-51, 156-157 quantitative patterns of 204
"abstract entity" terms 136-142 "quantity" 109
"activity" terms 146-152 "role" 106
affinities of 152-153 "scope" 110
and "classificatory entity" 153- "static manner" 109
155 "status" 109
"classificatory entity" terms "subject" 110
142-143 in term formation 49, 159
"material entity" terms 117 "time 109
27, 264-265 "use" 107-108
problems of 159-160 "whole/affiliations" 108
"quality" terms 143-145, 267 conceptual structure 10, 57
"relation" terms 145-146 general and subject-specific 70
Index 317
multidimensional 254 dynamics of terminology 32-34

constituent elements 4 and discourse 33-34, 270-272
Cruse 58,185 and dynamics of terms 41-42
Cyc 63 and evolution of terminology
40-42
D and parole 261-263
potentiality 34
Dahlberg 62
structural 257-261
Daille 20 synchronic 34, 259-261
Datta 62 synchronic and diachronic
dependency structure 51-54
263-272
descriptive regularity of terminology
and term formation 160-162
35, 252
and terminological space 270-
determinant 51
272
quantitative patterns of 203- and terminology growth 165—
204
166
determinant categories
quantitative patterns of 203-
E
204
Deuleuze 39 EDR 63, 78
diachronic study of terminology 18, embedded combinations 88-89
263-272 "entity" concept 66
discourse 33-34, 270-272
documentation terms F
average length 185
Fabre 78
average use of morphemes
Farradane 62
190-192, 194-199
Felber 16-20,55, 184
coefficient of loss 186
Fellbaum 64, 78
conceptual categories 64-80
Finin 93
conceptual specification pat
Foucault 40
terns 106-113
Frege 57
growth of morphemes 188-190
Frisby 55, 56
growth patterns 208-209
Fujii 78
growth rate of morphemes 185,
193-199
G
intra-term relations 98-105
and randomness assumption Gaines 272
186-187 Gale 172
relative growth 209-211 gairaigo 5
domain 10, 30, 35 general theory of terminology 17
Downing 58, 93 Gilchrist 62, 71
Dungworth 12, 58, 93-94 Good 37, 172-173, 256
Good-Turing estimate 172, 176, 256 "product" 99

Greimas 39 qualitative nature of 102-104
Grivel 272 quantitative nature of 104-105
growth of morphemes 177-179, "quantity" 101
188-190 "role" 99
growth rate 174-176, 185, 192-199, "scope" 101
205 "status" 101
GTT 17 "subject specification" 101
in term formation 49
H "time" 101
Hatcher 93 "use" 99
Hayashi 28 Ishii 4,58, 93, 261
head 51 Ishikawa 63
Hempel 61 ISO704 55
hierarchy of concept system 57 isomorphism
Hockett 11 concepts and meaning 23-24
hypergeometric model 168 terms and words 23-24
I J
Ibekwe-SanJuan 272 Jackendoff 18, 54, 56, 57, 62, 78
intra-term relations 58-60, 98-105 Jacquemin 20
"affected object" 99 Jespersen 93
"attributed concept" 101 Johnson-Laird 57
as binary relation 60
comparison of 96-97 
"connection introduction" 98 Kageura 37, 48, 62, 168, 171, 179,
"constituents" 100 254, 256, 257, 260, 261
"destination" 99 kango 5
"differentiation tag" 102 Katz 57
form of 58-60 Keil 18,57
"formal attributes" 100 Kindaichi 78
"function" 99 Kita 171
"information content and repre
Kokuritsu Kokugo Kenkyusho 62
sentation" 100
Kuhn 35
"juxtaposition" 99
"location" 101
L
"manner/mode" 100
"means" 100 Lamberts 55
"nature" 101 language system 13
"origin" 101 langue 11, 251
"partitive" 100 Lara 184
Index 319
Lassaline 52, 57 in Japanese 4

Laurence 18, 55 Murphy 52, 57
Lees 93
Lenat 63 N
Leonard 93
natural language 15
Levi 93 Neveling 4, 185,263
lexical item 10 Nirenburg 63
lexical unit 9 Nkwenti-Azeh 9, 24
lexicalisation 12 Nomura 4
library classification system 61-62 normal science 35
LNRE models 245, 257 nucleus 51
Lyons 13, 18,39,56, 185 basic-level 51-53
quantitative patterns of 203
M
Mackintosh 20 
Maeda 262 Oeser 20
Manning 171,256 Ogino 78, 168
Marchand 93 Okuda 78
Margolis 18, 55
Maruyama 269 Ρ
Masterman 57, 63
"material entity" concept 67, Parker 28
subcategories of 71-72 parole 11, 251
"material entity" terms Pearson 20
conceptual patterns of 117 performance 12
27, 264-265 Piaget 39
quantitative patterns of 211- Piotrowski 38
220 PMEST 62
McDonald 12, 58, 93-94 Polanco 272
Meyer 20 polysemy 184
Miller 57, 64 population number of types 168
minimal element 4 population of terminology 168
Miyajima 10, 29,260 population probability 168
modifier 51 and sample relative frequency
morphemes 171
Pugh 48, 58, 62, 66, 94-95, 260
and concept 183-185
Pustejovsky 18
and concept categories 80-86
conceptually indeterminate 82
core and non-core 178-179 Q
growth of 189-190 "quality" concept 67
growth rate of 185, 193-199 subcategories of 76-77
"quality" terms "relation" terms

conceptual patterns of 143- conceptual patterns of 145-146
145, 267 quantitative patterns of 234-
quantitative patterns of 232- 236
233 relative number of types 205-206
quantitative patterns "representational entity" concept 67
"abstract entity" terms 227-231 subcategories of 72-73
"activity" terms 236-243 "representational entity" terms
"classifìcatory entity" terms conceptual patterns of 127-
231-232 136,265-267
concept categories 201-203, quantitative patterns of 220-
243-244 227
conceptual specification pat representative sample 38, 48
terns 204 Rey 35
determinant 203-204 Rhyne 93
determinant categories 203- Rosch 18,64
204 Roth 55, 56
growth rate 205 Royauté 272
interpretation 206-207
"material entity" terms 211- S
220
Sager 9, 12, 15, 20, 24, 51, 55, 58,
nucleus 203
62, 93-94
number of types 205
Saiga 58, 91
"quality" terms 232-233
Sakamoto 63
"relation" terms 234-236
sample 168
relative number of types 205-
incompleteness of 171
206
sample size 169
"representational entity" terms
Sampson 172
220-227
Sapir 11
and simple terms 207
Sato 63
Quillian 56, 57
Schank 55, 63
Quine 57
Schmid 56, 57
Schütze 171,256
R
Seaborg 20
randomness assumption 168,178,180 secondary combinations 88-89
in documentation terms 186- Shanks 55
187 Shaw 272
Ranganathan 62 Sowa 55, 63
Rea 28 Spack Jones 185
"relation" concept 67 structural approach 39-40, 250
subcategories of 76-77 structural dynamics 257-261
Index 321
accumulation/disaccumulation 160-162
model 257 and intra-term relations 49
generation/decline model 258 linguistic elements 51-54
structural semantics 39 "material entity" terms 117-
structural type distribution 175 27, 264-265
subject field 10 "quality" terms 143-145, 267
Swanepoel 20, 251 quantitative tendency of 48
synchronic dynamics 34, 259-261 "relation" terms 145-146
and diachronic perspective "representational entity" terms
263-272 127-136, 265-267
validation of 263-267 and word formation 46-51
synchronic study of terminology 18, terminography 28
263-272 terminologisation 12
synonymy 184 terminology 9
system of terminology 32-34 and artificial language 15
and discourse 33-34, 270-272 as an aspectual category 26-27
as a concept 26, 29-31
 conceptual approach 253-255
conceptual regularity 36-37
Takeuchi 78 diachronic study of 18, 263-
Tartier 20, 271,272 272
Temmerman 11, 20, 29, 39, 251, 253 dynamics of 32-34, 40-42,
term formation 160-162
"abstract entity" terms 136-142 as an empirical object 26, 29-
"activity" terms 146-152 31
affinities 152-153 evolution of 40-42
basic assumptions 45-46 flexibility of 15
building blocks of 46 general theory of 17-20
and "classificatory entity" 153— and lexicology 262
155 methodology of 35-39
"classificatory entity" terms and natural language 15
142-143 potentiality in 34, 268
classificatory regularity 50 quantitative approach 255-257
and conceptual categories 47- quantitative pattern 37-38
49, 56-58 recent trends in 20
conceptual elements 54-60 regularity of 35
and conceptual specification representative samples 38, 48
patterns 49 rigidity of 15
and conceptual system 49, 56- as a sample 168
58 standardisation 19
description of 51-60 synchronic study of 18, 263-
and dynamics of terminology 272
system of 32-34 U
theory of 21-24, 31-32 Ullmann 185
traditional theory of 16-20 Ungerer 56, 57
terminology growth urn model 167-168
binomial interpolation/extrapo
lation 172-174 V
binomial model 167-170
mathematical model 167-176 Vienna school 16
randomness assumption 168, vocabulary 9, 30
178, 180
urn model 167-168 W
terminology population 168 wago 4
terms 9 Warren 93
as an aspectual category 26-27 Way 57, 253
characteristics of 14-15 Wersig 4, 185,263
as a concept 26, 29-31 Woods 57
and conceptual categories 86- Wierzbicka 18
88 words 10-11
core and non-core 178-179
as a frame category 26-27
dynamics of 40-42
and terms 11-14
as an empirical object 26, 29-
word formation
31
semantic patterns of 47
evolution of 40-42
and term formation 46-51
as a functional class 12, 26
WordNet 64
and words 11-14
Wüster 16
theory of terminology 21-24, 31-32
conditions for 25-32
Y
range of generalisations of 27-
29 Yoshikane 20
structural approach to 39-40, Yule 168
250 Yumoto 58, 92-93
and terminography 28
thesaurus 61-62 Z
Toulmin 37, 172-173,256
Zawada 20, 251
Toyama 39
traditional theory of terminology
16-20
Tsuji 20
Tuldava 38

(Kyo Kageura) The Dynamics of Terminology A Desc PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Kyo Kageura) The Dynamics of Terminology A Desc PDF

Uploaded by

Copyright:

Available Formats

THE DYNAMICS OF TERMINOLOGY

TERMINOLOGY AND LEXICOGRAPHY

Terminology and Lexicography Research and Practice aims to provide

The Dynamics of Terminology.

JOHN BENJAMINS PUBLISHING COMPANY

can National Standard for Information Sciences — Permanence of Paper for

Library of Congress Cataloging-in-Publication Data

P305 .K34 2000

PART I: Theoretical Background 7

PART II: Conceptual Patterns of Term Formation 43

PART III: Quantitative Patterns of Terminological Growth 163

Chapter 9. Quantitative Dynamics in Term Formation 201

PART IV: Conclusions 247

University of Nantes. My research cooperation and many discussions with

hand, fosters application-oriented studies of the computational treatment of

The organisation of the book

Table 1. Number of terms by number of morphemes.

Table 2. Quantities of morphemes.

Japanese), kango (originated in Chinese) and gairaigo (originated in

ings (Kageura 1994). Table 1 shows a breakdown of the terms by number

Terminology: Basic Observations

1.1 Basic observations

1.1.1 Terms and related notions

It is unfortunate that, in the definition of "term", the authors use "word"

ogy [vol. 5, no. 1, 2000]). Any definition is controversial and provisional

1.1.2 Terms, words and parole

By defining a "terminology" as the set of terms of a subject field, terms

above definitions of "word" to suit the definitions of "term"; we can speak

Figure 1.1. The position of "term" in language.

terms in the same manner as studying words is theoretically valid (Kageura

1.1.3 Characteristics of terms

Figure 1.2. The position of terminology in natural and artificial language.

1.2 The traditional theory of terminology

1.2.1 Basic claims of the traditional theory

1.2.2 Restrictions of the traditional theory

First of all, the precedence of concepts over terms is dubious. If "the

In short, various aspects excluded from the traditional theory of termi­

1.3 Recent trends in terminology

A growing number of researchers have recognised the limitations of the

prototype theory. They emphasise the flexible relationship between con­

1.4 A missing piece for the theory of terminology

Despite the current development of research in terms and terminology, a

which makes it essentially different from the relationship between mean­

2. The descriptive structure of concepts in the traditional theory of termi­

The Theoretical Framework for the

If we search for a theory of terms or terminology, as distinct from mere de­

2.1 Conditions for the theory of terminology

Let us seriously consider the seemingly tautological claim that terminol­

"terminology" (in its nominalistic sense). We therefore start the discussion

2.1.1 Terminology as an aspectual category

the category "word". When we try to characterise a phenomenon which is

2.1.2 The range of generalisations

Studies of aspectual categories should have a scope of generalisation ap­

We should, for instance, recognise the difference between samples and

2.1.3 "What is a term/terminology ?" — revisited

discussion, because for a study about terms to be a study of terms de jure,

the study of terminology, therefore, it can be assumed that vocabulary and

2.1.4 Theories of terminology

The epistemological precedence of the concept terminology over term, indi­

2.2 The dynamics of terminology: target and method

2.2.1 The dynamic system of terminology

Figure 2.1. The difference between terms, terminology and discourse.

the domain independent of the individual situation of discourse. The lat­

In short, various aspects excluded from the traditional theory of termi

prototype theory. They emphasise the flexible relationship between con

which makes it essentially different from the relationship between mean

2. The descriptive structure of concepts in the traditional theory of termi

If we search for a theory of terms or terminology, as distinct from mere de

Let us seriously consider the seemingly tautological claim that terminol

Studies of aspectual categories should have a scope of generalisation ap

The epistemological precedence of the concept terminology over term, indi

the domain independent of the individual situation of discourse. The lat

Studying conceptual patterns is close to conventional structural seman

processing), (processing) is the nucleus and (data) is the deter

The academic study of concepts is generally considered to have origi

Concepts or conceptual categories ... are mental representa

According to the classical view of concept organisation, the representa

The next important theoretical consideration is whether binary concep

The subcategories of each conceptual category are best described sep