Professional Documents
Culture Documents
Lise Fontaine Fullthesis Postomodification
Lise Fontaine Fullthesis Postomodification
Cardiff University
2008
DECLARATION
This work has not previously been accepted in substance for any degree and is not
concurrently submitted in candidature for any degree.
STATEMENT 1
This thesis is being submitted in partial fulfilment of the requirements for the degree of
PhD.
STATEMENT 2
STATEMENT 3
I hereby give consent for my thesis, if accepted, to be available for photocopying and
for inter-library loan, and for the title and summary to be made available to outside
organisations.
I hereby give consent for my thesis, if accepted, to be available for photocopying and
for inter-library loans after expiry of a bar on access previously approved by the
Graduate Development Committee.
This thesis explores the relevant systems which model the choices speakers
make when referring to objects. Referring expressions have received
relatively little attention in Systemic Functional Linguistics, although from a
purely structural perspective some work has attempted to account for
postmodification in the nominal group. The main goal of this thesis is to
produce a theoretical and analytical approach to referring expressions
including complex referring expressions in particular. This requires a shift in
perspective from structural (‘nominal group’) to functional (‘referring
expression’).
In addition the thesis has produced an XML database of over 3,000 fully
analysed referring expressions and a second database including detailed
coding of complex referring expressions.
i
Acknowledgements
When I first set out to do a PhD, I could never have imagined the journey it would
take me on. Having nearly reached the end of this long, fascinating, difficult,
challenging, impossible journey, I am finally able to express my gratitude to those who
were so important, instrumental and supportive throughout. If only I could represent
this journey as a system network of meaningful options that I encountered along the
way!
There can be no PhD without a supervisor and so first and foremost I’d especially
like to thank my supervisor Gordon Tucker for countless hours of support,
encouragement and suggestions. Of course any errors in this thesis are my
responsibility alone but the success of the thesis is the result of the many discussions
and exchanges I have had with Gordon whose input has been invaluable.
A special thank you to Tom Bartlett for his very useful comments and probing
questions on the draft of this thesis.
I would also like to thank Robin Fawcett. Thank you Robin for getting me started
on this fascinating topic and for being so generous with your materials, knowledge
and experience. You have been consistently supportive and I have greatly
appreciated this.
There were times when I thought I would never complete this thesis. I have my
parents to thank for the fact that I have. I don’t know how they managed it but they
somehow raised me to believe I could probably do anything I wanted to. They both
supported me and encouraged me throughout my life. Any strength, intelligence,
determination and kindness I have come directly from them. They are the reason why
someone like me from a small town in remote Northern Ontario, Canada somehow
managed to get to this place. Thank you mom for being such a great mom. If my dad
were alive today, I know he would be very proud. My only regret is not getting to this
point soon enough.
I’m sure there were times when my children wished they had never even heard of
a PhD. I’d like to thank them for being so understanding and not complaining about
having a mother whose most frequent response has been ‘no, sorry, I don’t have time
right now’, especially when ‘now’ has lasted for years. Thank you Steven and Rowan
for being such amazing people and thank you Ashwyn for coming along just in time to
be part of the light at the end of the tunnel.
Anyone who has lived with someone in the final stages of their PhD knows what
challenges and sacrifices it brings. I’d like to thank my partner Kevan who has been a
constant source of support and encouragement, showing patience deserving of
sainthood. He has been picking up the pieces of my fragmented life for the last
several years and I’m sure that he is almost as relieved as I am that we are near the
end.
I would also like to thank Yves for believing in me and for being my intellectual
rock.
In addition to my family I have been very lucky throughout my life to have made
friends with really great people. Without naming individual names, it is important for
me to thank my close friends who have always told me I could do this. Their support
has been so important to me and I am really grateful.
Last but certainly not least, I would like to thank my colleagues in the Centre for
Language and Communication for their support and encouragement. There have
been many times when I had to pass on opportunities because of my commitment to
the PhD and everyone has been patient and understanding. I would like everyone to
know that I really appreciate the support and faith they have shown me.
ii
Abstract.........................................................................................................................i
Acknowledgements.....................................................................................................ii
iii
3.4 Transitivity....................................................................................................58
4.1 An overview of the system networks for the nominal group ........................65
4.1.1 Halliday 1964 .......................................................................................65
4.1.2 Halliday and Hasan 1976.....................................................................67
4.1.3 Halliday 1977 .......................................................................................70
4.1.4 Halliday and Matthiessen 2004............................................................76
v
8.3.1.2 Scheme Mode................................................................................208
8.3.1.3 Coding Mode..................................................................................210
8.3.1.4 Review Mode .................................................................................211
8.3.1.5 Statistics Mode...............................................................................212
8.3.2 The coding scheme............................................................................214
8.3.3 Coding complex referring expressions...............................................224
vi
10.3.2 Results for nominal groups coded with [by relation with other thing] .292
10.3.2.1 Prepositions introducing a postmodifying prepositional group ...294
10.3.2.2 Apposition ..................................................................................297
10.3.3 Results for nominal groups coded with ‘postmodifier’........................298
10.3.4 Multiple qualifiers ...............................................................................299
References ...............................................................................................................325
Appendix ..................................................................................................................334
vii
Dans le monde qui nous est donné, la science
implante des oasis d’ordre, de calcul, d’efficacité ;
mais existent, demeurent et deviennent d’immenses
forêts d’ignorance, d’erreur, de passion, de
souffrance, d’inquiétude.
(Fourastié)
Chapter 1: A linguistic perspective on referring expressions
The situation, as realised in (1), has two participants: ‘my friend’ and ‘a new
coat’. What we are interested in here is the construction or composition of the
expression(s) produced by the speaker to refer to the object(s). The shift in
perspective is then on the referring expression rather than on the nominal
group.
1
Chapter 1: A linguistic perspective on referring expressions
2
Chapter 1: A linguistic perspective on referring expressions
a term is required for the linguistic units that serve to identify (or
refer to) whatever we are talking about when we make a
statement about something. The term that is most commonly
used in philosophical semantics for this purpose is ‘referring
expression’. We also want a term for the linguistic units that are
employed in order to ascribe particular properties to whatever is
being referred to.
Clearly, Lyons proposes that linguists adopt the term referring expression
in order to be able to talk about the linguistic units used by speakers to identify
whatever they are talking about. Although the term referring expression
originates in philosophical studies of reference and referring, it is currently
most commonly used in psycholinguistic research, philosophy and natural
language processing (usually in language generation). Linguistic researchers
tend not to use this term and instead focus on its related structural or syntactic
units like the noun phrase or nominal group. Oddly this structural classification
is also true for functional linguists where one might expect to find a more
functional approach (cf. Keizer, 1992).
The use of the term nominal group in place of the more standard noun
phrase has a rather long standing tradition of its own, although a more limited
scope. Halliday has consistently used the term Nominal Group, since as early
as his first major publications (1956), and Systemic Functional linguists have
followed suit in the use of this term. As Halliday and Hasan (1976:39) explain:
For the moment we will not concern ourselves with the distinction between
‘group’ and ‘phrase’ in Halliday’s model of Systemic Functional Linguistics.
3
Chapter 1: A linguistic perspective on referring expressions
4
Chapter 1: A linguistic perspective on referring expressions
This may or may not be the case; we cannot be certain. However, most
research in referring expressions would suggest that it is in fact the other way
around; that the noun phrase is simply one possible structural realisation of a
referring expression. The speaker’s intention to refer to some object (i.e. to
build a referring expression) would precede the building of a noun phrase. We
will return to this in the following chapters when we consider this in more detail.
If we think back to example (2), it seems quite obvious that the speaker would
have the intention to refer before having an available noun phrase since the
speaker might have preferred to express the situation differently with different
referring expressions, even if each ‘REFERENT THING’ remained the same as in
(3) and (4).
5
Chapter 1: A linguistic perspective on referring expressions
Mithun’s interpretation of what she calls descriptive labels is that they are
all “surface morphological verbs ... (but) they clearly function in the same way
as formal nouns syntactically” (Mithun, 1976:31, cited in Rijkhoff, 2002:13).
This should illustrate that it is perhaps unreasonable to assume that all
languages have nouns and that all referring expressions are realized
linguistically as noun phrases. This is clearly an extreme position. It is beyond
the scope of this thesis to explore this area any further. It is sufficient evidence
however to suggest that the structural realisation of a referring expression is
language dependent and this should not be forgotten. This is especially
important since we will be analysing the English language. It is easy to rely on
dominant languages as a model for other language systems. However it does
seem clear that the correlation between a referring expression and its
realisation as a noun phrase is quite strong in English, although this is not
always the case as will be discussed later.
6
Chapter 1: A linguistic perspective on referring expressions
The entities (NPs) refer to are not entities in the external physical
world. Referents of NPs are rather mental representations of
entities as they are created, stored, and retrieved in the minds of
the speech participants. ... Thus a distinction must be made
between a referent (mental construct) and its ontological
counterpart in the physical world, if it exists.
We will return to this in further detail in Chapter 2 when considering work done
on referring expressions in philosophy and semantics.
At this point then, we can define a referring expression as a linguistic
expression that a speaker produces in order to transmit his or her
representation of a given referent. Any attempt at a complete list would be
futile as it would near infinity, but for the purposes of illustration, examples of
referring expressions include: I, you, he, they, it, houses, the tall boy, a quick
trip, two apples, five of those oranges, a cup of coffee, etc. These referring
expressions can be said to be simple in the sense that they represent one and
only one referring expression. However, a referring expression may have
other referring expressions embedded in it as in, a car with a sunroof, people
who eat meat, etc. In doing so, the speaker creates a temporary ‘situation’
which relates the referents (this will be explained in detail in Chapter 7). This
relationship of embedded referring expressions is shown in (5) and (6):
7
Chapter 1: A linguistic perspective on referring expressions
Describing the aims of this thesis is more difficult than one might imagine
because of the paradox of what was said above: this is not a study of the
nominal group and yet it is. In this chapter, I have argued for a linguistic
perspective to the study of referring expressions. In English at least, this
means also studying the nominal group as it is the primary linguistic resource
for referring to an object (cf. ‘entity’ in Lyons, 1977 and in Keizer, 1992).
Although linguists generally disagree on the functions of constituent elements
of the nominal group, there is much agreement within specific theories as to its
syntax. With respect to referring expressions however the main difficulty is due
to the fact that they seem to relate directly to what is happening within the
brain of the speaker where the referent is some cognitive representation or
mental construct. This thesis does not intend to develop a cognitive model of
referring expressions per se, but rather to move as closely as possible in this
direction so that a fully functional account of referring expressions can be
developed.
There are two main areas of concern when dealing with referring
expressions from a systemic functional perspective. There is the
(simultaneous) relation between function and form. In formal terms (i.e. the
syntax), there is little more to do in describing the nominal group (provided we
ignore for the moment somewhat problematic expressions such as the top of
the mountain, or the side of the house, see Fawcett (2007) for a detailed
account and the discussion here in Chapter 4, Section 4.3.2 and Chapter 7,
Section 7.3). Fawcett (2000, 2007, and forthcoming b) provides a very
comprehensive account of the functional syntax of the nominal group.
However in terms of function, with respect to complex referring expressions,
there remains a great deal of work to do. As Fawcett (2004) explains, the
8
Chapter 1: A linguistic perspective on referring expressions
9
Chapter 1: A linguistic perspective on referring expressions
The preceding sections have attempted to set the stage, as it were, for the
presentation of the thesis. If the aims are not perfectly clear, this should be
understandable since the argument developed throughout this research –
although simple in thought – is complex and difficult to talk about. The main
problem is perhaps due to the fact that there is no functional or semantic term
for expressions which are realized as a nominal group. Butler (2003a:251)
calls them “nominal expressions which designate entities”. It is almost
10
Chapter 1: A linguistic perspective on referring expressions
11
Chapter 1: A linguistic perspective on referring expressions
12
Chapter 1: A linguistic perspective on referring expressions
13
Chapter 1: A linguistic perspective on referring expressions
In closing this section, we will review the main aims of this thesis. They
are:
• to develop a methodology for analysing referring expressions following
the system networks as closely as possible.
• to contribute to the systemic functional theoretical model of language in
the area of referring expressions by presenting a detailed description of
referring expressions.
• to develop a new theory of complex referring expressions which is
based on systemic choice rather than on structural or lexical
classification.
The next section will complete this chapter with an overview of the thesis.
14
Chapter 1: A linguistic perspective on referring expressions
concerning the system networks and it identifies the areas in the framework
that deal implicitly with referring expressions, namely participant roles in the
transitivity system. A particularly important contribution in this chapter is the
discussion of transitivity as this will form the basis of the theoretical
developments in Chapter 7.
Chapter 4 describes the nominal group within the framework of Systemic
Functional Linguistics. Its main goal is to explain the grammatical description
of the nominal group following Halliday’s description of the nominal group from
1964 to 2004. It also includes important discussions of complex nominal
groups which have direct implications for the view of complex referring
expressions presented in this thesis. In addition, Halliday’s ‘measure’
nominals are considered in terms of the challenges they create for the analyst.
Finally Martin’s (1992) work on the nominal group from his developments on
participants in discourse is also briefly considered.
In Chapter 5, the focus is on the model developed within the Cardiff
Grammar, which is the model the current work is based on. It provides the
framework for functional syntax used in the analysis. This chapter is essential
for understanding the terminology used in the Cardiff Grammar as well as the
basic relations in functional syntax such as conflation, componence, filling and
exponence. More detail is given here on the role of the system networks in
this model.
Chapter 6 offers an overview of the treatment of referring expressions in
the Cardiff Grammar from the generation perspective and from within the
Cardiff model of language. This is essential since the relevant system
network, which is the ‘thing’ network, is discussed in relative detail. This
includes a presentation of the relevant algorithm for referring to ‘thing’. As the
system network for ‘thing’ is referred to throughout the remainder of the thesis,
it provides the necessary background both for understanding the theoretical
developments in this thesis and the analytical approach to referring
expressions.
In Chapter 7 the main arguments are presented for why a new theoretical
approach is needed in the study of complex referring expressions.
Consideration is given here to the more standard or traditional approach to
postmodification in the nominal group. The main contribution of this chapter is
15
Chapter 1: A linguistic perspective on referring expressions
16
Chapter 2: An overview of selected approaches to referring expressions
Whales do not really belong with the fish they superficially resemble, since
the similarity of form and behaviour conceals radical differences of
structure and function. (Gareth Evans, 1982:1)
2.1 Introduction
17
Chapter 2: An overview of selected approaches to referring expressions
18
Chapter 2: An overview of selected approaches to referring expressions
2.2.1 Frege
The earliest studies of referring expressions (even if the term itself shifts in
definition depending on the field) stem from philosophy and logic. It was then
developed outward and into linguistics via two principal paths: one is through
semantics (e.g. Lyons, 1977) while the other is through computational
linguistics (e.g. Appelt, 1985). Searle’s work on reference has led indirectly
from philosophy to studies of discourse and pragmatics (e.g. Searle, 1969).
For the philosopher, the central notion is that of unique reference: the use
of an expression to uniquely identify some object. Philosophers working on
this topic seem to have focused almost exclusively on naming and what
constitutes a unique definite description. Evans (1982:1) summarises the
philosophical view of referring expressions as follows: “the class of referring
expressions has traditionally been taken to include proper names, definite
descriptions (‘the tallest man in the world’); demonstrative terms (‘this man’)
and some pronouns.”
Even among philosophers, there has been much dispute as to what
constitutes a referring expression. Part of the problem lies in interpretations of
the non-existence and/or truth value of the expression in question. The main
explorations attempted to resolve puzzles such as whether or not an
expression can be said to be referring if it has no reference and whether a
sentence containing an expression referring to something that does not exist
can be said to have reference.
The first real treatment of reference is attributed to Frege (1892), although
according to Searle (1969:77), it can be traced back as far as Plato. For Frege
a referring expression is an expression that refers to a unique object; in other
words, a definite description. In his view, a referring expression is a type of
sign and by sign Frege (1892/1993:24) means “any designation representing a
proper name, which thus has as its reference a definite object (this word taken
19
Chapter 2: An overview of selected approaches to referring expressions
Frege’s claim was that each expression in (7) has a different sense but that
they both share the same reference. Reference, for Frege, is the actual object
being designated; in other words, the thing that is being referred to. According
to Frege (1892/1993:26) “the reference of a sign is an object perceivable by
the senses … an internal image”. Sense, however, is part of the meaning of
the sign but separate from reference. Frege defines sense as the mode of
presentation of the sign; the way in which the expression is present. As Frege
explains (1892/1993:24), “the sense of a proper name is grasped by
everybody who is sufficiently familiar with the language or totality of
designations to which it belongs”. It must be kept in mind that in Frege’s
terminology, sign and proper name mean the same thing as referring
expression.
In addition to this, he also claimed that an expression may have a sense
but no reference as in example (8) below (from Frege, 1892/1993:25). In this
example, the sense may be grasped but there is no unique object to which this
expression refers and therefore it cannot be said to have reference. As a
result, this expression would not be classed as a proper name (referring
expression) in Frege’s terminology. The opposite is not possible according to
Frege, it is impossible for a sign to not have a sense (i.e. it is impossible for a
sign to have a reference but no sense), since as he explains (1892/1993:25),
“to every expression belonging to a complete totality of signs, there should
certainly correspond a definite sense”.
20
Chapter 2: An overview of selected approaches to referring expressions
2.2.2 Russell
Whereas Frege was concerned with the distinction between sense and
reference, Russell focused on the distinction between referring and describing,
where expressions which he calls descriptions are not referring expressions.
A description, according to Russell (1905:479), “consists of several words,
whose meanings are already fixed, and from which results whatever is to be
taken as the ‘meaning’ of the description”. This type of expression does not
constitute a referring expression for Russell. There are two types of
descriptions: definite and indefinite (or what Russell calls ‘ambiguous’).
Indefinite descriptions are of the form ‘a so-and-so’, e.g. a man, and definite
descriptions are of the form ‘the so-and-so’, e.g. the man (Russell,
1919/1993:46).
Russell (1905) used his theory of denoting to attempt to solve the problem
of the expression of entities which do not exist, e.g. the present king of France.
He presents the law of the excluded middle as a puzzle that can be solved by
his theory of denoting. The law of the excluded middle states that any
proposition can be determined to be true or false, such that either the
proposition or its converse must be true, i.e. “either ‘A is B’ or ‘A is not B’ must
be true” (1905:485). The problem with this for propositions containing non-
21
Chapter 2: An overview of selected approaches to referring expressions
existing entities (i.e. the present king of France, as in the present king of
France is bald), is that the law will not hold. Clearly neither ‘the present king of
France is x’ nor ‘the present king of France is not x’ can be said to be true.
Russell solves this by distinguishing between primary and secondary types of
denoting phrases (ibid.:489).
2.2.3 Strawson
In (9), Strawson explains (1950:325) that the context of the utterance of the
expression the king of France is important since if uttered during the reign of
22
Chapter 2: An overview of selected approaches to referring expressions
Louis XIV, some would say the sentence is true, while if during the reign of
Louis XV, many would say it was false. Similarly the expression the greatest
French soldier, in example (10), may be or may not be uniquely referring
depending on the function of its use. If uttered in the sentence Napoleon was
the greatest French soldier, then the expression is not referring since it is not
being used to refer to a particular individual, instead it is being used to say
something about a particular individual (Napoleon). However, if the same
expression is uttered in the sentence the greatest French soldier died in exile,
then it is used to refer to an individual (Napoleon).
It is through this realization that Strawson sees referring, not as something
that an expression does, but rather as something that can be used by a
speaker in order for the speaker to refer to some object. Therefore truth
values and the real world existence of objects become irrelevant in his
approach to the theory of referring (cf. Searle, 1969).
Linguistic reference has inherited a great deal from philosophical studies of
reference. Traces of its influences are present in most if not all linguistic
theories and this is most notably relevant in the use of concepts such as
definiteness, meaning, sense, and reference. Although much of what we glean
from philosophical studies of language is important, some of it carries with it
concerns and definitions that are not relevant for linguistic studies (e.g. the
arguments concerning definiteness, cf. Fawcett 1980). The use and meaning
of the term ‘reference’, for example, is inconsistent both in philosophy and in
linguistics (see Lyons, 1977:199 and Evans, 1982:7). Nevertheless, any study
of referring expressions must understand the roots of its development and very
clearly this lies in the earliest philosophical studies of linguistic reference. In
the next section, contributions to the study of referring expressions in
computational linguistics are considered.
23
Chapter 2: An overview of selected approaches to referring expressions
24
Chapter 2: An overview of selected approaches to referring expressions
We can build models and try to figure it out, but in the end, even with a
computer system that is trying to model human interaction, we cannot be sure
that this is in fact what humans do. Nevertheless, computational models are
the main means that we have at this point for testing models of language.
Computational linguistics is not always concerned with models of language.
There are cases of particular task-based programs that deal with an isolated
part of language. In language generation however, models of language are
needed in order to produce language in a way that resembles the output of a
human speaker (more or less accurately).
So while we would certainly not like to claim that computational systems
model what the brain is actually doing with language, we should not ignore
developments and discoveries from their implementation. In terms of referring
expressions, it does seem significant that certain types of expressions need to
be handled differently.
Referring expressions, as a domain of interest, have received the greatest
attention from the field of Natural Language Processing (NLP). The field of
NLP is where artificial intelligence, computer science and linguistics merge in
order to develop computational systems that can generate and/or understand
language. NLP covers a variety of computational linguistic tasks such as
machine translation, speech recognition, text-proofing, automated
summarisation, information retrieval and extraction, question answering,
natural language understanding and natural language generation. To one
extent or another, all NLP tasks are part of either natural language
understanding (NLU) or natural language generation (NLG). Machine
translation is an example of how these two areas need to work together since
to translate, the system must first understand in one language and then
generate language in another language.
In language generation, many approaches have been developed by
computational linguists to solve the problem of generating referring
expressions. For example, it has been approached from a Pragmatic
perspective (e.g. Appelt; 1985; Appelt and Kronfeld, 1987; and Dale and
Reiter, 1995), from a discourse perspective (e.g. Paris and McKeown, 1987),
from a syntactic perspective (e.g. Power, 1999), or indeed from the
perspective of the computational architecture itself, i.e. what the computer
25
Chapter 2: An overview of selected approaches to referring expressions
26
Chapter 2: An overview of selected approaches to referring expressions
addressee from being able to identify the pen that the speaker has in mind. In
this way, the entities in the set of domain entities play a role in generating
referring expressions to the extent that the expression generated must
eliminate the distractors and identify the target entity.
So in many cases, although clearly not all, referring expression generation
seeks to identify one unique object from a set of other possible referents.
There seems to be no question that this expression will be realized as a noun
phrase: referring expression generation is then a procedure for putting
together a noun phrase that will identify (and discriminate) one particular entity
(Reiter and Dale, 1997:68). Specifically, “the generation of referring
expressions involves finding a noun phrase that identifies an entity ... to the
hearer in the current discourse context” (ibid.:79).
The other concept that plays an important role in referring expression
generation is disambiguation or the process whereby the addressee is able to
identify the target referent unambiguously (Reiter and Dale, 1997:81). This is
one area where the computational system clearly differs from human language
generation as every speaker can think of instances where during the course of
a conversation one speaker generates a referring expression that is not
unambiguously identified (see below on Grice’s Brevity maxim). Otherwise we
would not likely ever hear the question which one. Nevertheless, the
computational system has no specific need to replicate human performance
and in fact it is far more likely that it needs to be efficient. Therefore
disambiguation is built into the algorithm for generating referring expression to
a degree that is probably not the case for human speakers.
The main developments have been in terms of computational efficiency
and in terms of what types of referring expressions could be generated,
including the extent of complexity of the resultant noun phrase. In Appelt’s
early work (1985:72); the only referring expressions that his system could
generate were “singular, specific, definite noun phrases that do not contain
explicit quantifiers.” It was not long before his system was developed to go
beyond simple discriminating descriptions. Dale (1989) developed EPICURE,
an NLG system that concentrated on referring expressions. This system
became a benchmark for the development of referring expression algorithms.
Heeman and Hirst (1995:6) took the representation of entities as “not simply
27
Chapter 2: An overview of selected approaches to referring expressions
data structures, but (as) mental objects that agents have beliefs about.” This
was an important development following on views of referring that were
developing around that time concerning what beliefs speakers and addressees
may hold. This provided researchers with a link to psycholinguistic work done
on how speakers collaborate on referring expressions (we will discuss this
further in Section 2.4). Based on this, Heeman and Hirst (1995) were able to
develop a computational model of collaboration on referring expressions (see
Section 2.4.2).
28
Chapter 2: An overview of selected approaches to referring expressions
They describe the type of referring expression used in the study as follows
(Dale and Reiter, 1995:235):
goal.
Here the speaker would have the additional communicative goal of warning
the addressee not to touch the table as they sat down.
29
Chapter 2: An overview of selected approaches to referring expressions
They include one other constraint which deals with lexical preference for
the head of the nominal group, for example the selection of dog, a basic-level
lexical item, as compared to poodle, a non-basic-level lexical item, in referring
to this animal:
30
Chapter 2: An overview of selected approaches to referring expressions
there are no hard and fast rules we can use to determine which
tasks belong within microplanning and which more properly
belong in the domain of document planning or surface
realization: it is clear that many research groups are still
exploring the issues here ... document planning can be viewed
as primarily relying on domain knowledge, whereas surface
realization relies primarily on linguistic knowledge; microplanning
is then concerned with those tasks that require attention being
paid to interactions between domain knowledge and linguistic
knowledge.
32
Chapter 2: An overview of selected approaches to referring expressions
33
Chapter 2: An overview of selected approaches to referring expressions
34
Chapter 2: An overview of selected approaches to referring expressions
35
Chapter 2: An overview of selected approaches to referring expressions
• The speaker uses the noun phrase intending the addressee to be able
to identify the referent uniquely against their common ground;
• The speaker satisfies her intention simply by the issuing of that noun
phrase;
• The course of the process is controlled by the speaker alone.
36
Chapter 2: An overview of selected approaches to referring expressions
37
Chapter 2: An overview of selected approaches to referring expressions
accept RE
Initial RE repair RE
reject RE expand RE
replace RE
The process of building a referring expression will always begin with what
Clark and Wilkes-Gibb call an initial referring expression. This may then be
either accepted or rejected by the addressee, or even rejected by the speaker.
Example (14) illustrates an initial referring expression which is accepted by
the addressee.
38
Chapter 2: An overview of selected approaches to referring expressions
(18) A. Okay, and the next one is the person that looks like
they’re carrying something and it’s sticking out to the left.
It looks like a hat that’s upside down. (Initial RE)
(19) B. The guy that’s pointing to the left again? (Reject and
Replace RE)
39
Chapter 2: An overview of selected approaches to referring expressions
As we can see then, initial referring expressions are always tentative; they
are seen as complete or successful only through the participation of the
addressee. This model accounts nicely for interactions which are in nature
collaborative. Some initial work done on collaborating on referring expressions
in non-collaborative settings suggests that the system presented in Clark
Wilkes-Gibb needs to be developed further. In a study of Safety Hearings
(Fontaine, 2004a), speech partners seem to develop different strategies
following a rejected referring expression depending on whether they are seen
as collaborating or not on the task at hand. The results in the particular case
studied show that the same witness will employ different refashioning
strategies depending on whether the referring expression was initiated during
an exchange with his lawyer or during cross-examination.
The issues concerning referring expressions are rather complex involving
intention, beliefs, cooperation, collaboration and power relations. For our
purposes here, we will not discuss this in any more detail. However, the point
needs to be made that building a referring expression is indeed a process
involving more than simply the speaker’s intentions.
We will turn now to a different type of collaboration among speakers with
respect to referring expressions. This concerns convergence.
2.4.3 Convergence
40
Chapter 2: An overview of selected approaches to referring expressions
41
Chapter 2: An overview of selected approaches to referring expressions
42
Chapter 2: An overview of selected approaches to referring expressions
43
Chapter 2: An overview of selected approaches to referring expressions
One of the main contributions of corpus linguistics has been in its ability to
help us discover recurrent patterns. In the study of referring expressions, this
type of research can be of value for a number of reasons. Patterns found in
very large corpora may reflect a kind of cultural norm of convergence for
certain types of referring expressions. They may also be indicators of
multiword or formulaic sequences that suggest the sequence functions rather
as a unit, stored and retrieved whole, rather than being generated by the
grammar. Both cases have significant implications for the notion of speaker
choice in language production. In this section we will consider some
contributions from the study of such multiword expressions to the study of
referring expressions.
Stubbs (2004) did not set out to study referring expressions but his work on
very frequent phrases has identified multi-word patterns that form part of
particular referring expressions. He introduces the term phrase-frame to
account for a string of words (n-gram) with one variable slot. The example
given by Stubbs is plays a * part in where * marks the variable slot.
Expressions such as it plays a significant part in do not necessarily affect
referring expressions although what is expected to follow in would be a
referring expression: e.g. it plays a significant part in the overall picture or it
plays a significant part in people’s lives (both examples taken from a simple
Google search). A related type of string can also be taken from a corpus
analysis. This concerns focussing on the parts of speech or the word class of
words. Strings identified using parts of speech are called POS-grams.
Searching for POS-grams can also be done in combination with lexical items
as in the following: PREPOSITION + DETERMINER + singular-NOUN + of +
DETERMINER (e.g. ‘at the end of the’ and ‘in the middle of the’). This pattern
was identified by Stubbs as “one of the most frequent 5-POS-grams in the
BNC (British National Corpus).” The top three 5-phrase POS-grams identified
by Stubbs were as follows with examples to illustrate each.
44
Chapter 2: An overview of selected approaches to referring expressions
As has been shown in several corpus studies, different text types can vary
considerably in terms of grammatical realisation (Biber et al, 1999, Biber et al,
1998, Quirk et al, 1985). According to Biber et al (1998), “texts from different
registers often differ dramatically in the use of referring expressions”. This
specific element tends to be reduced to studies of nominal groups (or noun
phrases) in texts.
In this section, we will compare results from two very large corpus studies
of the noun phrase, illustrating the difficulty in researching this area of the
grammar.
Quirk et al (1985) studied approximately 17 000 noun phrases from the
Survey of English Usage. They made a two-way distinction in classifying noun
phrases. The first class is the simple noun phrase, where the head is not
modified; this would include cases of determiner + noun and noun only noun
phrases. As a subclass of the simple noun phrase, noun phrases realised by
either a name or a pronoun are further classified. The second class contains
examples of the complex noun phrase; this includes all other types of noun
phrases. There was a further distinction in this class for multiple modification,
where a noun phrase was found to have more than one type of modification
(e.g. more than one adjective, or an adjective and a postmodifier).
45
Chapter 2: An overview of selected approaches to referring expressions
Table 2-1 Noun phrase structure and distribution for the entire sample (Quirk et al,
1985:1351)
46
Chapter 2: An overview of selected approaches to referring expressions
Table 2-2: Noun phrase structure and distribution for the individual corpora (Quirk et
al, 1985:1351)
Categories Total Simple Noun Phrases Complex Noun Phrases
All Names + All Multiple
Pronouns Modification
Informal Subject *2984 2064 1941 148 62
Speech (2212)
(4192) Not *2220 1169 677 811 327
subject (1980)
Fiction Subject *4466 2220 1943 211 92
(5234) (2431)
Not *3991 1682 754 1121 434
subject (2803)
Serious Talk Subject *3693 1745 1478 343 127
(4599) (2088)
Not *3602 1273 599 1238 492
subject (2511)
Scientific Subject *1831 720 459 447 175
writing (1167)
(2936) Not *2456 629 163 1140 524
subject (1769)
The main observations made from these results are as follows (ibid.:1350-1).
• Less than 1/3 of all noun phrases in the entire sample are complex,
therefore approximately 2/3 of all noun phrases are without any
modification.
• Scientific writing differs most significantly from other text types due to a
much higher frequency of complex noun phrases and specifically a
much higher frequency of noun phrases with multiple modification.
• Almost 50% of all noun phrases consist of pronouns or names.
47
Chapter 2: An overview of selected approaches to referring expressions
48
Chapter 2: An overview of selected approaches to referring expressions
2.6 Summary
49
Chapter 2: An overview of selected approaches to referring expressions
language system works when viewed across very large collections of texts.
Certain texts seem to have canonical patterns, the most obvious of which are
found in academic prose as we have seen. Results from studies in corpus
linguistics would seem to indicate that there are external factors, other than
speaker intention, acting on the generation of referring expressions.
This chapter has presented the vast and complex nature of the study of
referring expressions from the most important contributors in the field. In the
next chapter, an overview of Systemic Functional Linguistics (SFL) is given
which explains the foundations of the approach taken here to referring
expressions. The most important concept drawn on from SFL is that of
transitivity and consequently part of Chapter 3 is devoted to a discussion of
Halliday’s views on transitivity. Therefore many key concepts in SFL are
explained in the next chapter and these are necessary as they contribute to
the theoretical approach to referring expressions presented in this thesis.
50
Chapter 3: An overview of Systemic Functional Linguistics
51
Chapter 3: An overview of Systemic Functional Linguistics
53
Chapter 3: An overview of Systemic Functional Linguistics
referring to entities in the world and the ways in which those entities act on or
relate to each other”. Consequently, we will only briefly cover the interpersonal
and textual metafunctions for the sake of completion and focus instead on a
more detailed account of the experiential metafunction.
54
Chapter 3: An overview of Systemic Functional Linguistics
Figure 3-2: Three-strand analysis of the clause (Halliday and Matthiessen, 2004:182)
55
Chapter 3: An overview of Systemic Functional Linguistics
Figure 3-3: Disjoint Subject, Theme, and Actor (adapted from Halliday and
Matthiessen, 2004:182)
56
Chapter 3: An overview of Systemic Functional Linguistics
57
Chapter 3: An overview of Systemic Functional Linguistics
3.4 Transitivity
58
Chapter 3: An overview of Systemic Functional Linguistics
59
Chapter 3: An overview of Systemic Functional Linguistics
the associated role of the potential participants for that particular type of
process.
This system (Figure 3-5) only shows the process types. As stated earlier,
this is the system that represents the choices available to the speaker in terms
of experiential meaning. Experiential meaning, as we have already seen,
covers the range of meanings available to the speaker in representing his or
her experience of the world. The role of the system network is to represent the
options available in the grammar. Therefore, this system network represents
the choices available to the speaker, in terms of how he or she can represent
his or her own experience. Each process type is explained briefly in Table 3-1,
along with its respective participants.
For example, as a spectator at a football match, a speaker wanting to
comment on a particular move during the game will choose one of these six
types of meaning to represent the experience that they want to express
(however unconsciously). This might be through a material process such as
John kicked the ball, or a mental process such as John saw the ball, or a
relational process such as John is a great footballer. In each example, the
participants included are represented as having different functional roles. John
in each case is either represented, by the speaker, as Actor, Senser or Carrier,
respectively. However it is important to make clear that these relations are not
arbitrary, each process construes a different type of experience.
60
Chapter 3: An overview of Systemic Functional Linguistics
The process and participants represented in the clause reveal only part of
the transitivity picture. If we refer back to the description of transitivity given by
Halliday (1970:30), transitivity also includes ‘the attendant circumstances’.
The circumstances are seen as peripheral to the core of the clause. They
describe the process or situation in some way. In standard SFL, there are nine
main types of Circumstance. These are listed in Table 3-2, where each type is
illustrated with examples. If we reconsider the examples above, each one is a
representation of experience and in each case, additional meaning could have
been included in the form of an attendant circumstance. Considering each of
the examples again in turn, the speaker may have chosen to express the
following representations, where circumstances are underscored, processes
are in bold and participants are in italics: John kicked the ball very quickly; At
that point, John saw the ball; For over three years, John was a great
footballer.
Language is instantiated through the system networks. In the examples we
have see in this discussion of transitivity, the system network representing the
experiential metafunction, the systemic choices selected lead to the realization
of an instance of language (in these cases, the clause). The categories we
have discussed represent the functions or types of meanings that can be
identified in the clause but nevertheless result from the selections made by the
61
Chapter 3: An overview of Systemic Functional Linguistics
Table 3-2: The Nine Basic Types of Circumstantial Elements (adapted from
Halliday and Matthiessen, 2004:262)
Type question answered example
Extent distance how far? He ran three miles[Circ.:Extent:distance].
duration how long? He ran for three days[Circ.:Extent:duration]
frequency How frequently? He ran every day[Circ.:Extent:frequency]
Location place where? He ran in Toronto[Circ.: location: place]
time when? He ran last year[Circ.: location: time]
Manner means by what means? He saved her with a rope[Circ.: manner:
quality how? means]
comparison like what? She saved him quickly[Circ.: manner:
degree How much? quality]
She ran like the wind[Circ.: manner:
comparison]
She loved him more than anyone[Circ.:
manner:degree]
Cause reason why? She ran because she loved to[Circ.:
purpose for what purpose? cause:reason]
behalf on whose behalf? She ran to raise money[Circ.:
cause:purpose]
She ran for her sister[Circ.: cause:behalf]
Contingency condition under what In the event of fire[Circ.: contingency:condition]
conditions? leave the building
3.5 Summary
63
Chapter 4: Halliday’s Nominal Group
64
Chapter 4: Halliday’s Nominal Group
65
Chapter 4: Halliday’s Nominal Group
Figure 4-1: partial view of Halliday’s principal systems for nominal group
(1964/1976:131)
66
Chapter 4: Halliday’s Nominal Group
For Halliday, all nominal groups must have a head element, so the difference
made here reflects the difference found in examples (25) and (26), where the
head element is underscored. In (26), the head is non-nominal as it is a
number (called Numerative in later work). This is in contrast to other
approaches that would treat the head as being left empty (ellipted) as in (27)
for example.
In (25), the head of the nominal group is apples (nominal) but in (26) it is
two (non-nominal). Therefore following the options in the network, the option of
‘noun’ is selected for example (25), i.e. apples and ‘non-nominal’ is the option
selected in producing example (26), i.e. two.
The sketches of the systems presented are lacking in detail to the point
where we cannot reasonably discuss them further except to compare the
treatment of person reference presented in Halliday and Hasan (1976), shown
in Figure 4-3 below.
Although these early descriptions have since changed form substantially
from Halliday (1964), we can see the beginnings of the types of meanings
expressed with the nominal group. To continue our discussion of the relevant
system networks, we will turn to work done by Halliday and Hasan (1976) on
person reference. It should be noted that their work focussed on cohesion and
so it approaches reference from this perspective. In other words, they
developed reference in terms of its textual cohesion rather than considering
any potential textual meaning within the referring expression itself.
from theory building to how to apply the theory (cf. Halliday 1994 and Halliday
and Matthiessen, 2004).
Reference, in the sense used by Halliday and Hasan (1976:31) is to be
distinguished from lexical items having a semantic interpretation “in their own
right” as opposed to those items which are “directives indicating that
information is to be retrieved from elsewhere”. We can see this distinction
clearly in example (28) taken from the corpora of this thesis, where some
referring expressions have a semantic interpretation and others require this
information to be retrieved from elsewhere. Not every use of ‘he’ in this
example has the same referent, yet the speaker and addressee have no
trouble in understanding who is being referred to (coreference is marked with
the index i or j, to mark the referent for ‘he’ in each case).
(28) Johni went to see the doctorj and hej did some tests and
hei?j? also said hei had a sinus infection. I’m glad hei finally
went.
68
Chapter 4: Halliday’s Nominal Group
Reference
↓
Exophoric Endophoric
[situational] [textual]
↓
anaphora cataphora
[to preceding text] [to following text]
The term ‘referring expression’ covers Halliday and Hasan’s ‘reference’ but
it is much broader, extending beyond it since we would want to include those
expressions having a semantic interpretation of their own (e.g. lexical items
such as nouns). Reference, seen as a cohesive device, as is the case for
Halliday and Hasan, does not include, for example, initial referring
expressions; those expressions which are interpretable in their own right. For
a reference to be anaphoric or cataphoric, there must be a referring expression
in the text with which it is co-referential (e.g. the doctor and John in example
(28) above). In a sense, this type of reference (i.e. naming and exophoric
reference), must precede endophoric reference.
Figure 4-3 illustrates Halliday and Hasan’s (1976:44) system for person
reference. Although it is presented more as a taxonomy than a system
network, we clearly get a sense of the division between two major categories
that person reference can have. In referring to a person, there is a selection to
be made between the person having a speech role or some other role. Within
speech roles, the person may have the role of speaker or addressee. In other
words, the reference is to the speaker or to the addressee. In the case of
other roles, further distinctions can be made between singular and plural and
human and non-human.
69
Chapter 4: Halliday’s Nominal Group
Figure 4-3: System network for person reference (Halliday and Hasan, 1976:44),
with examples in italics
In order to reconcile the 1964 and 1976 versions of the nominal group’s
systems, the person system would seem to replace the ‘personal’ option in the
1964 CLASS_AT_HEAD system (cf. Figure 4-1 above). The 1976 version is
clearly more functional, capturing functional distinctions in terms of person
reference, even if ‘person’ as a system name must be seen as very broad and
rather closer to the notion of ‘thing’. As Halliday and Hasan (1976) was
devoted to cohesion and not to the nominal group, we need to turn to Halliday
(1977) to begin to complete the picture of the systemic options of the nominal
group.
The system networks for the nominal group in Halliday (1977) offer more
detail but the system networks remain unannotated. Using an example (see
example (29) below) taken from Halliday (1977:208), we begin to see
Halliday’s first complete theoretical vision of the nominal group. Table 4-2 on
page 76 shows Halliday’s analysis of the nominal group; interpreting it from
each of the three semantic components: ideational (experiential and logical),
interpersonal, and textual. It should be noted that the example given by
Halliday is a full clause. However since the unit we are focussed on is the
nominal group, we will ignore, for our purposes, the treatment of the other
elements of the clause.
70
Chapter 4: Halliday’s Nominal Group
As for the clause, Halliday also considers that the nominal group expresses
three types of meaning (see also Halliday and Matthiessen, 2004:328-329).
The first of these is the Ideational component which is further divided between
the logical and the experiential component. According to Halliday (1977:177),
logical meaning only applies to complexes (containing paratactic or hypotactic
structures): “complexes are univariate (recursive) structures formed by
paratactic or hypotactic combinations … at the rank in question … all other
structures are multivariate (non-recursive)”. The nominal group, like the
clause, can be a complex (univariate) or a simplex (multivariate). The main
difference in terms of the logical component is that simplexes have no logical
structure. As Halliday states: “the logical component is distinct from the other
three (components) in that all logical meanings, and only logical meanings, are
expressed through the structure of ‘unit complexes’: clause complex group
complex and so on” (Halliday, 1977:178). Therefore, for the nominal group we
are currently considering, a pair of unoiled garden shears, there is no logical
structure since this nominal group is a simplex. However, at the level of word,
Halliday states that there is a complex: “groups consisting of more than one
element are simultaneously structured both as word constructions
(multivariate) and as word complexes (univariate)” (ibid.:210) and therefore
there is a logical analysis to be made at the level of word complex for the word
complex formed by unoiled garden shears (see example Figure 4-5 below).
We will return to complex nominal groups in Section 4.3.1 below when we
consider the Qualifier element of the nominal group and the grammatical
structure of nominal complexes.
The experiential component includes the various types of participants as
well as other semantic distinctions. The system network for nominal group
with respect to experiential meaning is shown in Figure 4-4.
71
Chapter 4: Halliday’s Nominal Group
Figure 4-4: Simplified system network for nominal group with respect to the
Experiential Component (from Halliday, 1977:213)
As can be seen from Figure 4-4, the entry point to the system network for
the nominal group is (+Thing). Consequently, ‘thing’ is also at least one of the
entry conditions for the system network for Participant Type. It is impossible to
discuss this system network any further as there is insufficient information. We
will nevertheless discuss the analysis following the explanation given by
Halliday (1977). Figure 4-5 illustrates the logical and experiential component
for the word complex.
Figure 4-5: logical and experiential components for unoiled garden shears
72
Chapter 4: Halliday’s Nominal Group
Table 4-1: Analysis of the logical and experiential component of a pair of unoiled
garden shears
a pair of unoiled garden shears
Logical β : α : Head
Modifier
α : Head β : Modifier
γ:Modifier β: Modifier α : Head
Experiential Numerative: Epithet Classifier Thing:
Measure Common
One difficulty with the above analysis is that Halliday classifies this nominal
group as a simplex and with regard to a pair of unoiled garden shears, he
states that it “contains no logical structures” (Halliday, 1977:209). However,
his analysis does provide for a logical structure for the nominal group, in
addition to the logical structure of the word complex (unoiled garden shears).
The only possible interpretation is that the logical structure is in fact meant to
account for the entire nominal group, a pair of unoiled garden shears; as a
word complex.
What is of particular interest in the presentation of this example is that at
one point Halliday hyphenates ‘pair-of’. In the analysis of the experiential
functions of this nominal group, he labels ‘pair-of’ as Numerative.
Figure 4-6: Simplified system network for the prepositional group with respect to
the Experiential Component (Halliday, 1977:215)
73
Chapter 4: Halliday’s Nominal Group
74
Chapter 4: Halliday’s Nominal Group
Figure 4-7: Simplified system network for the nominal group with respect to the
Interpersonal Component (Halliday, 1977:217)
Finally, the nominal group is also analysed from the perspective of the
textual component. With respect to the textual component, as Halliday points
out (1977:224), the order of the elements within the group is fixed, “but it is the
thematic principle that determines this fixed sequence and explains why in the
verbal and nominal group, the element that has deictic value comes first: this is
the element that relates to the ‘here and now’”. A simplified view of the system
network representing the semantic options for the textual component of the
nominal group is given in Figure 4-8. What is illustrated here is the system
network for ‘deixis’ and determiners. The combined description of the three
components can be seen in Table 4-2.
75
Chapter 4: Halliday’s Nominal Group
Figure 4-8: Simplified system network for the nominal group with respect to the
Textual Component (Halliday, 1977:219)
Table 4-2: Full analysis for the nominal group a pair of unoiled garden shears
(Halliday, 1977:221)
a pair of unoiled garden shears
Logical β: α : Head
Modifier
α : Head Β : Modifier
γ:Modifier β : Modifier α : Head
Experiential Numerative: Epithet Classifier Thing:
Measure Common
Interpersonal Connotation Person:
other
Textual Deictic
76
Chapter 4: Halliday’s Nominal Group
description of the nominal group itself in terms of the experiential structure has
not changed considerably since then. The one area that has changed
significantly concerns the logical structure of the nominal group but this is an
area that will be discussed in the next section.
Halliday and Matthiessen (2004) present a partial view of the system
network for the nominal group and the only view we get is of the
DETERMINATION system. In a sense, this is the view of the nominal group
system relative to textual meaning. The Deictic, as an element of the
experiential structure of the nominal group, has, according to Halliday and
Matthiessen (ibid.:322), “the greatest specifying potential” and is therefore
what comes first in the structure of the nominal group. There is no
presentation of the system networks for the nominal group concerning
experiential or interpersonal meaning. We will discuss the experiential
structure in the next section.
The DETERMINATION system is illustrated in Figure 4-9, the output of which
is an expression which would function as Deictic in the structure of the nominal
group (see Section 4.2.1 below). This system has changed in terms of
organisation from the 1977 version presented above. The restructuring
includes the ‘non-specific’ and ‘specific’ options, especially the much improved
‘person’ option, which reflects the person system described in Figure 4-3 and
Figure 4-4 above. It is not clear how this system would produce embedded
possessive nominal groups such as my father’s (example from Halliday and
Matthiessen, 2004:314) since there is no option for this shown in the system
network. Due to lack of space, the system network shown here does not
provide the full set of options for the ‘non-interactant’ option. The missing sub-
networks are as follows:
‘non-interactant’ → {‘one only’ OR ‘more than one’};
‘one only’ → {‘non-conscious’ OR ‘conscious’};
‘non-conscious’ is realised as Deictic:its;
‘conscious’ → {‘female’ OR ‘male’};
‘female’ is realised as Deictic:her;
‘male’ is realised as Deictic:his;
‘more than one’ is realised as Deictic:their.
77
Chapter 4: Halliday’s Nominal Group
Figure 4-9: System network for Deixis in the nominal group (Halliday and Matthiessen, 2004)
We can reasonably assume that similar system networks exist for the
remainder of the functions of the nominal group. This would represent the
options available for the selection of Thing, Numerative, Epithet, Classifier and
Qualifier (see below for the full description of the nominal group). In the
absence of such a description from Halliday and Matthiessen 2004, we can
only refer to Halliday 1977 as the most complete presentation of the system
networks for the nominal group, especially when the various sub-networks he
presents are seen together. We have no way of knowing what options are
available for completing the picture of the nominal group in systemic terms.
We will now turn to the description of the structure of the nominal group in
order to better understand how the elements all fit together.
78
Chapter 4: Halliday’s Nominal Group
The move away from the more traditional Noun Phrase was not without
good reasons initially. From his earliest writings, Halliday has preferred the
term Nominal Group in place of Noun Phrase for two main reasons. The first
is that he followed other linguists using nominal group (for example Allen,
1951). In what is considered to be the first publication of his theory of grammar
(Halliday, 1956), Halliday uses the term nominal group and has consistently
done so since. More importantly, however, there were terminological issues in
the wider model as compared to other popular linguistic theories at the time.
The distinction between group and phrase is at the centre of these issues. As
we saw in Chapter 1, Halliday’s use of nominal group rather than noun phrase
reflects the considerable differences between his verbal group and Chomsky’s
verb phrase, rather than any fundamental differences between the nominal
group and the noun phrase (Halliday and Hasan, 1976: 39). Although
Halliday’s nominal group shares enough features and structure with
Chomsky’s noun phrase to be considered as a relatively similar unit, there is
no verb phrase in Halliday’s model. Instead, Halliday considers the verbal
units of the clause to form a group, which he naturally labels the verbal group.
Wanting to avoid any problems that might arise due to misunderstanding the
terminology, nominal group, which is clearly differentiated from noun phrase,
was preferred. Currently, Halliday’s followers have all adopted the term
nominal group.
There is however there is one type of phrase in SFL which makes the
distinction between group and phrase theoretically significant. For Halliday
(2003:19), the terms are not equivalent: “words expand to form groups ...
clauses contract to form prepositional phrases ... clauses and phrases get
embedded inside (nominal groups)”. The preposition acts as a reduced
process and the embedded nominal group is the complement of this process.
The parallels between the prepositional phrase as reduced clause and a
clause is shown in examples (30) and (31), both from Halliday and
Matthiessen (2004:360). The experiential meaning represented in prepositional
phrases is therefore Process and Participant, where the Participant, i.e. the
embedded nominal group, functions as Range (Halliday and Matthiessen,
79
Chapter 4: Halliday’s Nominal Group
The nominal group is seen basically as having a head element and the
potential for modification, both pre- and postmodification. The elements of
Halliday’s nominal group are essentially as follows (Halliday and Matthiessen
2004):
80
Chapter 4: Halliday’s Nominal Group
There is a strong relation between ‘thing’ and the head of the nominal
group in English. In SFL, we expect that the Head of the nominal group will be
conflated with Thing, as this is the most usual instance (Bloor and Bloor,
2004:150 and Halliday and Matthiessen, 2004:331). This combination makes
sense somehow in that the semantic core of the referring expression is what
will occur in the role of head. In other words, given that there is a denotational
relationship between the ‘thing’ (i.e. the noun) selected and a class of thing,
the head itself does not refer to a particular thing. It is the expression in its
entirety that is used by the speaker to refer to a particular object. It will be this
element that can be modified as in example (32), modified from Halliday and
Matthiessen (2004:332).
In (32) it is clear that the class of object being referred to is a cup and
specifically, the speaker is referring to a particular white cup.
The main distinction between Thing and head for Halliday and Matthiessen
(2004:325) is that Thing is an element of experiential structure and Head is an
element of logical structure. Furthermore, “(Thing) may be common noun,
proper noun or (personal) pronoun” but there could be other elements as head,
for example Deictic or Numerative. In other words, the head and thing are not
necessarily conflated in the nominal group (see example (33)). When the head
is a Deictic or Numerative, the expression is called a ‘measure’ nominal
(Halliday, 1994:195 and Halliday and Matthiessen, 2004:332), as shown in
Figure 4-10.
81
Chapter 4: Halliday’s Nominal Group
82
Chapter 4: Halliday’s Nominal Group
The nominal group is a rich resource for expressing meaning. It has the
potential for considerable complexity. This is described in the next section
where the nominal group is considered in further detail, focussing on the
functional grammar of more complex nominal groups. A firm understanding of
83
Chapter 4: Halliday’s Nominal Group
nominal group
the children in blue hats
Logical structure Premodifier Head Postmodifier
Experiential Deictic Thing Qualifier
structure
prepositional phrase
process Range
Logical structure pre-modifier head
Experiential structure Epithet Thing
84
Chapter 4: Halliday’s Nominal Group
85
Chapter 4: Halliday’s Nominal Group
86
Chapter 4: Halliday’s Nominal Group
This function of further describing seems to fit well with the purpose of referring
expressions. Examples (37) and (38) above are also examples of this type of
expansion as they further specify the referent.
The second type of expansion is called extension which has the function of
adding some new element, “an addition, or else a replacement, or an
alternative” (Halliday and Matthiessen, 2004:405). This is perhaps the most
difficult type of expansion to accept as accounting for the relation between the
Qualifier and the rest of the nominal group. As Halliday points out “the only
sense of extension which produces embedded clauses is that of possession
introduced by whose, of which/which or a ‘contact’ relative ending with of” (ibid.
432). An example of this is given in (40), from (Halliday and Matthiessen,
2004:433).
87
Chapter 4: Halliday’s Nominal Group
With three pages devoted to elaboration in the nominal group, half a page
to extension and five and a half to enhancement, it seems clear that the
function of expansion does not readily lend itself to explaining the function and
relation of the qualifier to the rest of the nominal group. It seems more a case
of trying to apply an existing framework for the clause to the nominal group
rather than attempting to account for embedded clauses or groups which
function as postmodifiers. With the current syntactic description of
postmodification being a case of embedding, we clearly do not have a case of
expansion in the sense of ‘tacking on’ as we find in paratactic or hypotactic
clause expansion. It is the result of recursion in producing a referring
expression (see Chapter 6). There is a loop and a re-entry into the system to
complete a satisfactory expression within the expression of the first level
expression. Expansion, on the other hand, is quite different when considered
systemically as, although there is a re-entry into the system, it is at the
completion of a unit. Recursion is a process that is taking place
88
Chapter 4: Halliday’s Nominal Group
There is one final area that we need to consider before ending this brief
presentation of Halliday’s treatment of postmodifying structures in the nominal
group. This concerns what we will call for the moment ‘of’ structures. We have
already seen an example of how problematic these structures are in example
(29) above, a pair of unoiled garden shears. Other examples of this type of
structure include the top of X, the tallest of X, several of X, etc. How these
forms are treated directly impacts on the overall structure of the nominal group
and in particular whether or not a nominal group of this type is classed as
having a Qualifier or not. For example, in a pair of unoiled garden shears, if
the Head element is taken as pair, then of unoiled garden shears would
constitute a Qualifier, postmodifying the head. Alternatively, if shears is the
Head element, then there is no Qualifier as an element of the nominal group.
In our discussion above, we saw that there are some problems with Halliday’s
account of these types of expressions. This is an area where there is
considerable difficulty in capturing both the function and syntax of these
expressions.
For Halliday and Matthiessen (2004), these particular ‘of’ structures are
extended numeratives, expressing meanings such as measure (a pack of
cards), facet (the front of the house) or variety (a kind of owl) among others
(examples from Halliday and Matthiessen, 2004:333). In such cases, Halliday
states that the Thing and Head elements are disassociated (i.e. not conflated).
In terms of syntax, this is somewhat of a neat solution as it avoids the
complications we saw earlier in example (29), a pair of unoiled garden shears.
However this can be problematic from the analytical perspective as it is difficult
to motivate how or why the Thing would be embedded (i.e. recursively
89
Chapter 4: Halliday’s Nominal Group
Figure 4-14: Analysis of the tops of the hills from Halliday and Matthiessen (2004:335)
90
Chapter 4: Halliday’s Nominal Group
Figure 4-15: Comparison of the experiential and logical analysis for example (20)
Martin’s interest lies more in discourse than the clause and his interest in
referring expressions is in relation to tracking participants in text or what he
refers to as “participant identification”. For example, he would consider any
embedded nominal groups within the nominal group as separate ‘participants’
for the purpose of phoricity, e.g. anaphora and cataphora, (1992:98) and
reference chains (ibid.:140). Participant identification, according to Martin
(ibid.:95), “refers to the strategies language use to get people, places and
things into a text and refer to them once there”. The way this is done in
English is through the nominal group (ibid.:97). As Martin points out
(ibid.:129), “all participants are realised through nominal groups but not all
nominal groups realise participants and some nominal groups realise more
than one”. His definition of participant states that a participant is “a person,
place, or thing, abstract or concrete, capable of functioning as Agent or
Medium in TRANSITIVITY” (Martin, 1992:129). Since it is traditional in SFL to
capitalise functional labels, ‘Participant’ refers to a function of the experiential
metafunction and ‘participant’ in Martin’s sense refers to a particular function of
a nominal group within a text, since a participant is a reference to a thing (in its
broadest sense).
Although Martin (1992) does not describe the nominal group in full detail,
he does provide a good description of its various pre-modifier elements. These
are shown in Table 4-4, including Martin’s four ‘pre-’ elements. Each of these
constitutes for Martin a type of nominal group embedding. The embedding is
illustrated in Figure 4-16 where we see the tree structure of example (46),
taken from Martin (1992:133). In this case, the nominal group has a Pre-
Deictic which has the function of ‘partitive’.
Table 4-4: Examples of the ‘pre-’ elements of the nominal group following Martin
(1992)
element of the nominal group example in bold
Pre-Deictic the top of the mountain
Pre-Numerative a pair of boots
Pre-Epithet the tallest of the mountains
Pre-Classifier that kind of gear
92
Chapter 4: Halliday’s Nominal Group
93
Chapter 4: Halliday’s Nominal Group
Figure 4-17: Analysis of a pair of unoiled garden shears following Martin (1992)
There is one problem with this view and that is that it does not seem to be
the case that Martin (1992) allows for Determiners to intervene between
Numerative and Epithet as would be needed if we consider an invented but
similar example as in (47) and (48).
94
Chapter 4: Halliday’s Nominal Group
4.4 Summary
This chapter has outlined the description of the nominal group in SFL in
terms of the system networks representing the systemic options available to
the speaker in its expression and in terms of its structural description. It has
also pointed out some of the challenging areas to be aware of when analysing
this group. The earlier descriptions of the nominal group are informative as
they assist in the understanding of later developments, as well as offering a
contribution to ‘filling in the blanks’ in the current state of the theory. It is
important to present these descriptions here as they are not adequately
presented elsewhere. The nominal group, both in terms of structure and
function, has been largely overlooked.
Although this presentation has been relatively brief, we can nevertheless
draw some inferences. The first perhaps is the parallel between the nominal
group and the clause. Halliday is clear on the multifunctional nature of the
nominal group, which he has paralleled with the clause such that the nominal
group expresses ideational, interpersonal and textual meaning. However the
experiential meaning is more obviously present, if not inherent to the nominal
group, and it is this metafunction more so than the other two that suggests
parallels with the clause. Secondly, the three-strand analysis would seem to
call into question the interpretation in the logical component. It is very difficult
to see how elements that have a textual function such as Deictic can be
labelled as having a modifier function in the logical structure.
Martin’s view of the nominal group as ‘participant’ points out an area that is
identified as a need in this thesis; there is insufficient terminology to discuss,
functionally, the expressions we are interested in. Calling them nominal
groups forces a focus on structure and in fact as Martin makes clear, it is not
an adequate term to cover the reference to ‘things’ since not all nominal
groups are ‘participants’. The term ‘participant’ is still problematic as it is the
same term as is used for the participants in the situation (associated to
processes). The use of ‘referring expression’, as used here, is not without
difficulties, especially since it carries with it considerable history and diversity
in meaning. Nevertheless, we do need a term, as Lyons argued (1977:23),
that will allow us to classify and refer to these expressions. The problem is
95
Chapter 4: Halliday’s Nominal Group
clearly that there are different levels of referring and therefore naming the term
appropriately and placing this term within the model is problematic. As is often
the case, we inherit terminological problems from other fields. This problem
has largely been ignored. Typically the focus is given to the structural
realisation of these expressions without identifying any status to the semantics
of the expressions. The position taken in this thesis is that it seems a
potentially backwards approach, especially for semantically- and systemically-
minded functional linguists.
In the next chapter, the Cardiff model of language is presented in order to
situate the work developed here within this model. Important terminology is
explained in Chapter 5 including the basic relations in functional syntax such
as conflation, componence, filling and exponence. More detail is given here
on the role of the system networks in the Cardiff Grammar. The Cardiff
Grammar offers important contributions to the study of referring expressions
and is therefore essential to the development of this thesis.
96
Chapter 5: The Cardiff Grammar
Figure 5-1: The main components of a systemic functional grammar (from Fawcett,
2000:36 and Fawcett, Tucker and Lin, 1993:121)
One rather significant area in which Cardiff SFG differs from Sydney SFL is
in the overall framework in terms of how language, or the theory of language,
is viewed. For Halliday, language is very much a social semiotic system;
speakers use language to interact with others in culturally determined ways. In
the Cardiff grammar, although this view is shared, it is developed within a
99
Chapter 5: The Cardiff Grammar
100
Chapter 5: The Cardiff Grammar
The role of the system networks in the Cardiff grammar also support the
primacy of production over comprehension as they reflect, as we will see, the
semantic options available to the performer (in the broadened sense of
speaker).
Although Lamb’s work on stratificational grammar greatly influenced the
early development of SFL, Lamb himself argues the reverse and claims that
language understanding (‘recognition’) is the primary process from which
language generation is derived. Lamb offers four main reasons for this position
(1999:131):
101
Chapter 5: The Cardiff Grammar
This raises some very important questions as one of the central concepts
in SFL is the notion of choice. The notion of choice drives the theory and
because of this, the theoretical framework of SFL is oriented towards language
production. Cardiff SFG explicitly takes the position that language production is
of prime importance, although it must be acknowledged that Lamb makes
some strong arguments for the contrary. Even though the concept of choice
plays an integral and important role in language production in systemic
linguistics (i.e. systemic options in Sydney SFL and semantic options in Cardiff
SFG), the same cannot be said for its role in language understanding.
Although the Cardiff grammar is developing its model of language
understanding, we will not pursue it here.
102
Chapter 5: The Cardiff Grammar
some way attempting to model human cognition, it does not claim to provide
an accurate description of cognitive behaviour.
The model we see in Figure 5-2 has been described in differing degrees of
detail in the following publications: Fawcett (1980, 1993, and 2000) and
Fawcett, Tucker and Lin (1993). The diagram contains a slightly simplified
view of the main components, leaving out any detail concerning the language
understanding components. What we see here is the relationship among the
various components for language production; inputs and outputs and the flow
of information. Although we are mostly concerned with the lexicogrammar, we
will present an overview of the full model in as brief a manner as possible.
The topmost components of the model concern the cognitive and
sociolinguistic processes: the processes required to plan what to say and to
guide the selection of lexical and grammatical units so that the output (e.g.
sentence or clause) matches the speaker’s goals and intentions. The lower
components, specifically the sentence planner, relate most directly to
traditional linguistics (e.g. syntax and morphology). With input from the higher
components, the sentence planner produces the best formal and semantic
representation. We will now look at each component in more detail.
The overall planner, which is the first and highest component, is a kind of
‘overseer’ which directs discourse planning and guides the system as a whole.
This is the component that plans the propositional content in consultation with
the belief system, in other words, this is where the earliest decisions are made
in terms of what is going to be said or what is going to be talked about for a
given proposition. As we can see from the diagram, there is a two-way flow of
information between the two. There is a kind of consultancy relationship
between the overall planner and the belief system. The main difference
between a planning component and the belief system is that the latter as we
will see shortly is static (even though the items in it will change); it contains
objects. Planning components have a role to play in the entire process of
language generation; in a sense, they have jobs to do. The planners consult
103
Chapter 5: The Cardiff Grammar
the belief system for various reasons to assist in the planning process, a kind
of decision making.
Figure 5-2: Partial view of the main generative components of the COMMUNAL
computer model
The belief system is, in a sense, at the heart of both language generation
and understanding. Although this is not shown here, various components of
language understanding also need to consult the belief system. We get a
sense of its importance from the number of components that need to have
access to it at different stages of the process. The belief system itself is object
oriented “in the sense that it consists of a vast number of specific objects and
generic objects” (Fawcett, 1994:78). It is not a system in the dynamic sense of
something that is operational. However, it is assumed that the objects it
contains are organised systemically. As Fawcett explains (1990:164), the
belief system includes “general and specific beliefs about (‘knowledge of)
situations and things in some domain; specific beliefs about the content of
104
Chapter 5: The Cardiff Grammar
105
Chapter 5: The Cardiff Grammar
clause and produces the formal representation. The role of the system
networks is to integrate the various decisions or selections made in the various
higher components, for example the microplanner. It is not a decision making
component but rather an integration and production mechanism.
There are two stages in this component. The first is the ‘traversal’ of the
system networks, i.e. a complete pass through the networks. This may in fact
involve more than one pass. Since the system networks represent a potential,
their traversal leads to the identification of the realization rules which will then
apply to the realization of the instance at hand. The second stage is then to
apply the realization rules thereby producing the formal representation. An
example of how this works is given below.
The system networks in the Cardiff Grammar have a great deal in common
with those discussed in Chapter 3. Both the Sydney and Cardiff system
networks rely on the same basic notation and presentation. Therefore in this
section, there is no need to review what a system or a system network looks
like. Nevertheless, there are some important differences and these will be
discussed here. In the Cardiff Grammar, the system networks represent the full
potential of the semantic options available to an individual speaker. The
systemic features are completely semantic in nature (Fawcett, 2005:7n). The
concepts of choice and dependency are central as Fawcett explains (1980:19):
107
Chapter 5: The Cardiff Grammar
1. rule number
2. network feature(s)
3. any conditions on the rules
4. rule operations.
For example, in the small system network presented in Figure 5-3 below, if the
feature [situation] is selected (as it would need to be in order to generate a
clause), then as Fawcett (2008a:100) explains, the corresponding realization
rule is to insert a clause and within the clause to insert a main verb (see
Section 5.3 below for a description of the Cardiff theory of functional syntax). If
the feature [information giver] is selected then the realization rule specifies that
the Subject must be positioned before the Operator.
Realization rules may be very simple or very complex. For example, the
realization rule for the feature [thing] in the system network for thing (Fawcett,
1998) states:
60 : thing:
if congruent_thing then ngp,
if minor_relationship_with_thing then pgp.
This example shows the rule number (60), the system feature ([thing]), and
the operation (insert unit, i.e. ‘unit insertion rule’). This rule handles the
difference between examples such as the woman and to the man as in the
woman gave the ticket to the man. In the first case, the woman, rule 60 would
insert a nominal group. In the second case, it would insert a prepositional
group for to the man. A more complex rule will have conditions such as in the
108
Chapter 5: The Cardiff Grammar
As stated in rule (5) above, the rule applies when either [prediction] or
[future time reference point] has been selected. It also describes the
conditions of realization dependent on whether or not the system feature
[negative] has also been selected. If it has not been selected, the Operator
will be expounded by will and if it has, the Operator will be expounded by wo.
This will be combined with the realization rule for [negative], which is as follows
(and also applies when the feature [confirmation seeker] is selected:
The final realization then for [prediction] and [negative] will be that the
Operator will be expounded by won’t, as in He won’t phone. It is therefore the
features of the system network and the realization rules which create the
formal representation. As Fawcett explains (2000:149),
As the examples above show, the realization component has the function
of integrating the selected semantic features and they control the realization of
the form. There is no need to consider more complex rules as an explanation
of their application would be beyond the boundaries of this thesis.
It is also possible for a realization rule to specify another semantic feature
that must be selected on a future pass or on the same pass through the
network. In the former, a preference is given for a particular feature on a future
109
Chapter 5: The Cardiff Grammar
pass. This is one way in which preselection occurs, where a realization rule
preselects other features. According to Tucker (1998:46), “the preselection of
one feature from a system is equivalent to stating that in the relevant context
there is no systemic choice, and consequently no choice in meaning or in
structural realization”.
The first pass through the networks deals with the clause ([situation]),
whereas the second pass typically will serve to fill an element of the clause
such as Subject. Subject is most commonly filled by a nominal group ([thing])
and the selection of [thing] would have been preselected from a realization rule
determined by the selection of certain semantic features on the first pass.
Preselection may also occur if the planner has passed down an instruction to
select a particular feature (e.g. validity assessment). Same pass rules cover
relationships between features which can be thought of as a kind of co-
dependency. For example certain process types are far more likely to prefer
progressive aspect and others are far more likely to disprefer progressive
aspect. In other words, action type processes in the present tense are more
likely to be progressive than are mental type processes (cf. John is bouncing
the ball versus John bounces the ball and John sees the bird versus John is
seeing the bird).
M validity unassessed
no circumstance
manner . . A
circumstance
time position A
Key:
= enter all systems to the right = choose between the features to the right
Figure 5-3: A very small systemic functional grammar for the English clause
(Fawcett, 2008a:93)
110
Chapter 5: The Cardiff Grammar
5.2.3.2 Probabilities
111
Chapter 5: The Cardiff Grammar
Figure 5-4: A simplified system network for the ‘information’ sub-network of MOOD
(Fawcett, 2008a:157)
Presenting the full system networks for generating a clause is far beyond
the limits of space afforded here. However from the very limited examples
shown above in Figure 5-3 and Figure 5-4, the concept of the system network
should be clear. The selection of a given feature may lead to another system
or to a set of systems, or it may lead to a realization rule. Systems may be
entered in parallel, as is the case shown in Figure 5-3 where there are four
systems in parallel (shown by ‘{‘ ). In theory, parallel systems are meant to be
entered simultaneously.
The short examples of system networks shown in Figure 5-3 and Figure
5-3 are not detailed enough to illustrate the process of traversing the network.
The full system network is far too large to present here. Instead, as a brief
example, we will consider some of the semantic features selected in order to
generate a simple clause such as I drove the car (invented example), using the
system networks in the computational implementation of the Cardiff Grammar
as presented in Fawcett, Tucker and Lin, 1993. The first selection (which is in
fact preselected) is the feature [situation] (other features are possible, such as
[thing], but [situation] is the semantic feature for the clause). This then leads to
112
Chapter 5: The Cardiff Grammar
∑|Cl → S/Theme/Agent
M < “drive”
113
Chapter 5: The Cardiff Grammar
C/Affected
This notation states that a sentence is generated and filled with a clause.
The clause has three elements: Subject (S), Main Verb (M) and Complement
(C). The Subject is conflated with Theme and Agent functions and the
complement is conflated with the role of Affected. It also states that the Main
Verb is expounded by the item ‘drive’. This terminology will be explained in the
next section. The realization rules have been omitted here to save space but
they would be similar in nature to the examples shown above. Furthermore,
there would be a realization rule associated to the verb ‘drive’ which would
ensure the correct morphology of the verb given the semantic options selected
in the system network. It should be clear from the results of the first pass that
the process is incomplete. Further passes are necessary in order to complete
the intended clause (I drove the car). This involves re-entering the system
network to generate the two participants (I and the car). This will be done
following a realization rule from the TRANSITIVITY system which would state that
the network must be re-entered preferring the [thing] feature. This preselection
leads to the generation of a referring expression.
What this short demonstration shows is that the process of traversing the
system networks is complex and very detailed. The example has barely
scratched the surface of the system network. The point here is not to describe
the traversal process in detail but to provide a sufficiently simple example in
order to give the reader an idea of the steps involved and the relationship
between the system network and realization. In the next chapter, a more
detailed account of traversal will be given in the discussion of the treatment of
referring expressions in the Cardiff Grammar. To complete this overview of the
Cardiff Grammar, the next section describes the Cardiff theory of functional
syntax.
114
Chapter 5: The Cardiff Grammar
a sound theory”. It has been claimed by Fawcett (2000:1) that current systemic
functional theory has no theory of syntax with the exception of his book.
The statement that SFL has no theory of syntax must be considered in
context. As Fawcett explains, Halliday’s “Categories of the theory of grammar”
(1961) was indeed a theory of syntax. However as developments in systemic
functional linguistics progressed, the focus was shifted towards meaning at the
expense of any similar developments in the theory of formal representation.
According to Butler (2002:61), the proposals presented in Fawcett (2000) are
most welcome since “(the book) not only sets out in detail the approach to the
syntactic part of the grammar taken by the Cardiff group, but also offers
detailed and explicit arguments for these views, in relation to the proposals
made by Halliday”.
In this section we will consider the syntactic framework within the Cardiff
Grammar. According to Fawcett (2007:8), there are four key concepts which
contribute to the syntax of any unit in the Cardiff grammar. These are:
1. the system network that defines the language’s meaning
potential,
2. the selection expression of features chosen from it on any
one traversal of it,
3. the realization rules that these trigger, and
4. the structure that is their output in any one instance,
consisting of syntax and items.
115
Chapter 5: The Cardiff Grammar
elements, and items. The term item refers to “word base forms and inflectional
morphemes” (Tucker, 1998:48n). These form the basic categories of the
functional syntax in the Cardiff grammar. They are illustrated in Figure 5-7.
The relationships among the various components are shown in Figure 5-8.
The relationship amongst them is explained by Fawcett (2008a:76) as follows:
Figure 5-5: Basic syntax of the Figure 5-6: Basic grammar of the clause in the
clause in the Cardiff Sydney Grammar
grammar
116
Chapter 5: The Cardiff Grammar
117
Chapter 5: The Cardiff Grammar
118
Chapter 5: The Cardiff Grammar
Table 5-1: The main elements of the clause in the Cardiff Grammar
Element Example
Subject (S) John, in John drove the car
Operator (O) does, in John does drive the car
Auxiliary Verb, several types (X) is, in John is driving the car
Infinitive Element (I) to, in John wants to drive the car
Main Verb (M) drove, in John drove the car
Main Verb Extension (MEx) asleep , in John fell asleep
Complement (C) the car, in John drove the car
Adjunct, many types (A) yesterday, in John drove the car yesterday
Linker (&) but, in But John drove the car
Binder (B) because, in because John drove the car
Formulaic Element (F) thanks
Let Element (L) let’s
Negator (N) not, in John did not drive the car
119
Chapter 5: The Cardiff Grammar
Figure 5-10: The relationship between Meaning and Form for situation and clause
(from Fawcett, 2008a:47)
In addition to the unit of the clause, the following remaining units will now
be considered in turn:
• nominal group (ngp)
120
Chapter 5: The Cardiff Grammar
Table 5-2: Examples (Fawcett, 2007) of individual elements of the nominal group,
element being sampled is in italics, head is underscored, each example
may be composed of more than one element.
Element of the nominal group: Example:
representational determiner (rd) a recording of her voice
selector (v) (always ‘of’), in various positions a recording of her voice
partitive determiner (pd) the back of the house,
fractionative determiner (fd) half the population
quantifying determiner (qd) two cultures
superlative determiner (sd) the fastest of the runners
ordinative determiner (od) the first of the runners
qualifier-introducing determiner (qid) those of her family
typic determiner (td) a different brand of oil
deictic determiner (dd) the castle
modifier (m) the old castle
head (h) the old castle
qualifier (q) the old castle in the centre of the city
ngp > (&) (rd) (v) (pd) (qd) (v) (sd or od) (v) (dd) (m)* h (q)*,
121
Chapter 5: The Cardiff Grammar
Figure 5-11: A common configuration of the nominal group in the Cardiff grammar
122
Chapter 5: The Cardiff Grammar
The prepositional group is another important unit with respect to this study
as it also functions as an expression that refers to a referent (either referent
thing or referent situation).
The prepositional group (pgp) is a unit having a preposition as its head
together with two other elements: completive (cv), which is an obligatory
element, and an optional prepositional temperer (pt). The completive element
is a significant element for the study of referring expressions as it is most
frequently filled by a nominal group (cf. Chapter 9)
The completive is always filled with either a nominal group or a clause. In
the diagram above it is shown filled with a nominal group. The preposition (as
the head of the group) is typically expounded by an item (e.g. ‘of’). An
example of a fully labelled tree diagram of the prepositional group can be seen
in Figure 5-14.
Key
pgp: Prepositional Group
tp: prepositional temperer
p: preposition
cv: completive
123
Chapter 5: The Cardiff Grammar
The quality group (qlgp) is perhaps the most developed unit in the Cardiff
grammar, having been described extensively in Tucker (1998), and arguably
more developed as a unit in the Cardiff grammar than in any other account of
syntax. This group expresses the syntax and functions associated to both
adjectives and adverbs. It is called the quality group as it covers the range of
qualities used to describe referent things and referent situations.
Consequently it is most commonly found filling the modifier element in the
nominal group (for adjectives), as seen in Figure 5-15, and filling the adjunct
element in the clause (for adverbs). The pivotal element of the quality group is
the apex, which is always expounded by an adjective or an adverb depending
on whether the quality group is functioning relative to a referent thing or a
referent situation.
Figure 5-15: Full tree diagram showing a nominal group, a quality group and a
prepositional group
124
Chapter 5: The Cardiff Grammar
Key:
ex = extent; d = determiners (quality deictic determiner, quality quantifying determiner); t =
temperers (te = emphasizing temperer; ta = adjunctival temperer); td = degree temperer; a =
apex; sc = scope elements (may be repeated [sc1] [sc2]; f = finisher (can occur either before or
after the scope elements.
Table 5-3: Examples of the quality group with an adjective expounding the apex
(from Tucker, 1998)
Ex td ta a td f sc
Much angrier than ever with him
so clever
politically stupid
strong enough
Table 5-4: Examples of the quality group with an adverb expounding the apex
Ex td ta A td f sc
very carefully
More carefully than ever
125
Chapter 5: The Cardiff Grammar
these elements may be filled by another unit (often a nominal group in the
case of the quantity group, as in numbers) or expounded by an item. Examples
of the realization of quantity group are illustrated Table 5-5.
Key:
ad = adjustor; am = amount; qtf = quantity finisher
Key:
po = possessor; g = genitive element;
o = own element
Figure 5-18: Diagram showing the generic structure of the deictic determiner
126
Chapter 5: The Cardiff Grammar
Table 5-6: Examples of the realization of the genitive cluster (from Fawcett, 2000:307)
po g o
my
the man ‘s
his own
his very own
Table 5-7: Main Processes and Participant Roles in the Cardiff grammar
Process ([Pro]) Participant Roles Example
Action Agent [Ag] I [Ag] bought [Pro] a new game [Af]
Affected [Af]
relational Carrier [Ca] That [Ca] is [Pro] a great game [At]
Attribute [At] I [Ca] have [Pro] two dogs [Pos]
Possessed [Pos] My mother [Ca] is [Pro] in Canada [Loc]
Location [Loc]
mental Perceiver [Perc] John [Cog] knows [Pro] the answer [Ph]
Cognizant [Cog] John [Perc] saw [Pro] the moose [Ph]
Emoter [Em] John [Em] loves his new job [Ph]
Phenomenon [Ph]
environmental (no Participant Roles, only It rained [Pro] today
Process)
127
Chapter 5: The Cardiff Grammar
(49) The war [S/Ag] made [M] Ivan [C/Af-Ca] very rich [C/At]
(from Fawcett, forthcoming a)
5.4 Summary
129
Chapter 5: The Cardiff Grammar
130
Chapter 6: The place of referring expressions in the Cardiff model
132
Chapter 6: The place of referring expressions in the Cardiff model
Figure 6-1: The relationship between referent situation and referent thing (from
Fawcett, forthcoming a)
Although noun senses may seem at first as though they are classified in an
ontological taxonomy, they are represented within the relevant system
133
Chapter 6: The place of referring expressions in the Cardiff model
networks. The system networks for noun lexis are system networks, with entry
conditions (often complex entry conditions) and they have realization rules
attached to them. Nouns in English denote classes of objects and relative to
the discussion in the previous chapter concerning the relationship between a
conceptual object and the actual object that is being referred to by the
speaker, the head noun of a nominal group will have the function of identifying
the class of thing for the object (Fawcett, forthcoming a). This classification
system for nouns is called the ‘cultural classification’ system, as individual
languages may have different arrangements of noun senses. As we will see
below, when the cultural classification is not deemed sufficient (for whatever
reason) for specifying the referent, it will be further sub-classified with an ad
hoc description for the purposes of the current referent. The expression may
also be further defined through systems such as particularization and
quantification for example.
134
Chapter 6: The place of referring expressions in the Cardiff model
she takes are working (see example (50) above). However in the earliest
planning stages, the basic logical form (Lin, 1993) would resemble something
similar to what is presented in (51):
135
Chapter 6: The place of referring expressions in the Cardiff model
136
Chapter 6: The place of referring expressions in the Cardiff model
The first thing to note in the algorithm is that a distinction is made between
the referent (R) and the referring expression (RE). The referent (R) can be
seen as the mental construct of Referent Thing (see above) while the referring
expression (RE) is the target expression that has yet to be determined. The
algorithm first names the referent X in order to identify it since it has no
linguistic expression at this point and it is simply a concept in the mind of the
speaker.
The main concepts needed here are: recoverability, distractors,
disambiguation, namability and novelty. Recoverability concerns whether or
not the speaker believes the addressee will be able to identify the referent
through some type of reference, in other words whether or not the addressee
can recover the referent’s meaning (Cook, 1991:25). This affects the selection
of the referring expression because if the speaker knows that the addressee
will easily recover what the referent is then a pronoun, for example, can be
used by the speaker to refer to the referent. The lower the ambiguity, the more
strongly recoverable the referent is. A weakly recoverable referent needs
some kind of qualifier (e.g. ‘the one who/that ...’ or ‘the Noun that ...’). If a
referent is totally unrecoverable, then it is considered novel and the referring
expression will name the referent or an expression will be produced that allows
the addressee to identify the referent, typically through the cultural
classification system.
The first question asked by the algorithm is whether or not X (the referent)
has been mentioned already in the clause (e.g. John cut himself). If yes, then
X is considered to be recoverable (it was already mentioned) but surprising,
since it is not expected. This will generate a reflexive pronoun. If no, then the
next question asks whether X is the same referent as any other referring
expression in working memory. If the answer is yes, then we can follow the
algorithm through and the output will be that X is either strongly recoverable,
recoverable by naming or weakly recoverable (which leads to another
algorithm). As an example of these, the output would produce respectively, it,
Alex, or the one called Alex (this last example taken from:
http://www.digitalspy.co.uk/forums/showthread.php?p=17698163).
137
Chapter 6: The place of referring expressions in the Cardiff model
Figure 6-2: the basic algorithm for generating a referring expression (Cook, 1991:34)
If the answer to the second question is no, then we move on through the
algorithm to determine whether X is the referent of a referring expression in a
subordinate clause. If it is, then the possible outputs are the same as those
above (e.g. I went home after I saw it/Alex/the one called Alex). The next
question determines whether or not X was a referent at any point earlier in the
discourse. If yes, then X is either recoverable by naming or weakly
recoverable (which leads to another algorithm).
Part of the role of the algorithm for referring expressions is to determine
whether or not there are any potential distractors. Distractors are other
referents that could potentially be mistaken for the current referent X (c.f.
Chapter 2, Section 2.3.2). For example, if there are two books on a table, a
speaker cannot simply say ‘please pass me the book on the table’. The
presence of another book may cause a distraction and so in order to be certain
that the addressee will be able to correctly identify the referent, this has to be
138
Chapter 6: The place of referring expressions in the Cardiff model
considered and this is done in terms of consideration for the addressee; a kind
of assessment of whether or not the referent is identifiable from the
addressee’s perspective. If during the course of the algorithm one or more
distractors are present, then there is a loop introduced until the algorithm is
satisfied that the addressee will be able to identify X. Note that if the speaker
does not wish take into account consideration for the addressee, then this step
can be bypassed; however the risk is that this will lead either to confusion or to
a kind of collaborative exchange to determine the referent.
If we now run the algorithm for the referring expression we want to
generate (i.e. the chemo pills she takes), we find we can follow the path
through the algorithm seen above (Figure 6-2) which would give the pathway
shown in Figure 6-3, resulting in our referring expression being deemed a
‘novel referent’.
139
Chapter 6: The place of referring expressions in the Cardiff model
speaker can be fairly confident that the addressee will be able to identify what
is being referred to even if this particular type of pills is unknown to the
addressee as the speaker will be able to believe that the addressee will be
able to adequately identify the object being referred to by the speaker. As a
result certain features will have been selected as the output of the algorithm –
a selection expression. This will determine the features selected in the system
network for referent thing. The system networks are the most important
component of the grammar and we will now discuss the selections
encountered through the network. It is however important to realise that it is
the algorithm that determines which selections are chosen. Although this
algorithm was designed for an implementation in a computational linguistics
system, it nevertheless is also attempting to model one way in which the
human language processor may process the selection process. In other
words, there must be some cognitive mechanism or procedure for deciding
how to refer to an object. Basing the model on what works computationally is
reasonable since it is impossible to test the functionality of such a mechanism
in a human. For the time being then we can only assume that there is such a
mechanism and that the above algorithm is a reasonable attempt to model it.
Whether it represents or not an actual human decision making process is not
very important for our purposes, it simply needs to be recognised that
something like this governs the way in which referring expressions are
generated whether by a human or a computer.
140
Chapter 6: The place of referring expressions in the Cardiff model
to have access to the full and complete array of semantic features. The main
purpose here is to illustrate the process and an example will suffice.
The partial view of the system network for referent thing presented here is
shown in Figure 6-4 through to Figure 6-11. It should be noted that this
representation is a somewhat simplified version from that presented in version
5 of the thing network (Fawcett, 1998) and a more recent update (Fawcett,
2004). An earlier version is also available in Fawcett (1980) and Cook (1991).
The network and its realization rules are very complex and detailed and as a
result, the presentation here will be more of an overview, sampling the
network, rather than a comprehensive exploration of the system networks and
their realization rules.
We will assume, as a result of the generation process to this point, that the
feature [congruent thing] has been preselected as part of a realization rule
from an earlier pass (i.e. the stages of generation that we have not covered in
detail). We also know, following the algorithm for referring expressions, that
we are generating an expression for a novel referent. Therefore as we first
enter the system network for thing in Figure 6-4, the system for [congruent
thing] is selected and this leads us to Figure 6-5.
Figure 6-4: Partial and simplified view of the system for referent thing
In this system, there are only two semantic options: [stereotypical thing]
and [recoverable location]. Since the referent is novel, the next system entered
is [stereotypical thing] (see Figure 6-5). This system contains two semantic
options: [interactant] and [outsider]. The [interactant] system seen here is
simplified substantially. However the full network extends that presented here
141
Chapter 6: The place of referring expressions in the Cardiff model
in order to account for referring expressions such as myself, you who are ...,
each other, etc. The [interactant] option may in fact lead to other systems
outside the INTERACTANT system. As Fawcett explains (Fawcett, 2004:2),
“some of the dependent features lead in turn to further systems, which are also
entered from later systems entered from [outsider], e.g. those that generate
expressions such as ‘two of you’, ‘a group of us’, ‘each other’ and ‘we who
support democracy’”. The fact that our referent is novel also implies that it
cannot be an interactant (i.e. it is not the speaker/performer or the addressee);
however this would have been determined in the microplanner as we saw
above. Consequently, the [outsider] option is selected and we now enter the
[outsider] system in Figure 6-6, which is considerably more complex than
those we have just passed through. The full potential of referring is achieved
through the OUTSIDER system. Fawcett claims (Fawcett, 2004:2) that “it is the
feature [outsider] - a ‘non-interactant’ being an ‘outsider’ - that leads to the full
richness of choices in meanings that are realized in the nominal group”.
Figure 6-5: Partial and simplified view of the system for congruent thing (as
continued from Figure 6-4)
142
Chapter 6: The place of referring expressions in the Cardiff model
Even though we did not discuss this in the section of the algorithm that we
did discuss, it also determined that the referent is not a thing being exclaimed
at (e.g. what great chemo pills as in what great chemo pills they are!), nor is
the referent unknown ([seeking specification of thing]), as in what are
working?. Note that both the [cultural classification potential] option and the
[thing exclaimed at] option both lead to the same systems in this network as
we can see in Figure 6-8. The difference between the two would be expressed
in the realization rules where a different formal representation would result
from having the feature [thing exclaimed at] present in the selection expression
(as opposed to [cultural classification potential]), and this accounts for the
difference between what great chemo pills and great chemo pills, for example.
The last semantic option in the [outsider] system is [thing relating out], which
represents expressions like which/that are working and this is not the case for
the referent in question. Figure 6-9 represents the system networks for
[seeking specification of thing] and [thing relating out].
Figure 6-6: Partial and simplified view of the system for outsider (as continued from
Figure 6-5)
The system for [recoverable thing] is extensive and only shown here, as in
Figure 6-7, in a simplified and reduced form. Unfortunately space does not
143
Chapter 6: The place of referring expressions in the Cardiff model
144
Chapter 6: The place of referring expressions in the Cardiff model
Figure 6-7: Partial and simplified view of the system for recoverable thing (as
continued from Figure 6-6)
The [recoverable thing] system (Figure 6-7) will also realise expressions
such as it when the feature [by token classification recoverable thing] is
selected; that when [by location relative to performer recoverable thing] is
selected; and those that were here by the selection of [by description
recoverable thing].
Figure 6-8: Partial and simplified view of the system for cultural classification
potential and thing exclaimed at (as continued from Figure 6-6)
145
Chapter 6: The place of referring expressions in the Cardiff model
Figure 6-9: Partial and simplified view of the system for seeking specification of
thing and thing relating out (as continued from Figure 6-6)
Figure 6-10: Partial and simplified view of the system for cultural classification and
recoverable cultural classification (as continued from Figure 6-8)
From Figure 6-11on page 149, we can see that there are only two options
in the [cultural classification potential options] system: [ad hoc description] and
146
Chapter 6: The place of referring expressions in the Cardiff model
[no ad hoc description]. The important distinction here is whether or not the
cultural classification of the referent is sufficient for unambiguous identification
by the addressee. In this case, we know that the cultural classification
selected by the speaker is pills, which is a regular count noun and therefore we
infer that the option [count cultural classification] has been selected in the
[cultural classification] system, followed by the selection of [plural cultural
classification] and [particularized plural] (see Figure 6-10). The selection of the
feature [particularized plural] reflects the meaning realized by the in the chemo
pills she takes since what is being referred to are particular pills (cf. discussion
of definite descriptions in Chapter 2).
Recall that two systems were entered in parallel, so we will now turn to the
second of the two, which contains two options: [ad hoc description] and [no ad
hoc description]. As discussed above, if as a result of the planning and
consultation processes which take place at a higher level of planning and
generation, it was determined that the cultural classification is sufficient to
specify the referent, then no further description is necessary and none will be
generated. If however this is not the case and the output from the
microplanner determined that further sub-classification is necessary then [ad
hoc description] is selected and this leads to seven parallel systems as we can
see in Figure 6-11 (we will ignore the problem in this network where no
description could be selected within the system of AD HOC DESCRIPTION as this
is discussed in Fawcett, 2004). These systems cover the range of description
possible: AGE, EPITHET, AFFECTIVE ATTITUDE, RELATIONSHIP WITH ANOTHER THING,
PLACE. Examples of the realizations of each option are given in Table 6-1.
147
Chapter 6: The place of referring expressions in the Cardiff model
Key:
1 = ad hoc description ; 2 = age; 3 = affective attitude; 4 = general epithet;
5 = minor relationship with thing; 6 = role in other situation; 7 = alternative specification;
8 = deictic place.
148
Chapter 6: The place of referring expressions in the Cardiff model
Figure 6-11: Partial and simplified view of the system for ad hoc description and no ad
hoc description (as continued from Figure 6-8)
There is one further system attached to each of the describing end features
in the network shown in Figure 6-11 (i.e. excluding those that are ‘not’
describing, e.g. [not by deictic place]) but as they all have the same role, they
are not included in the diagram to save space. Instead these last options for
consideration will be discussed without visual representation. Each of these
types of qualifier can be further described as being one of two types:
‘classifying’ or ‘depicting’. The classifying function serves to sub-classify the
class of thing (which is similar to the traditional label of restrictive modifiers),
whereas the depicting function (cf. non-restrictive modifiers) serves to offer
more detail in terms of description or information. As Fawcett (2004:5)
explains, the main difference with the depicting function of qualifiers is that it is
often marked: “when such a unit is ‘depicting’, there is a rich set of variants in
written language, as between (i) ‘unmarked’, realized by commas, (ii)
‘downgraded’ realized by brackets, and (iii) ‘afterthought’ (which may be simply
‘presented as if it is an afterthought’), realized by dashes”.
149
Chapter 6: The place of referring expressions in the Cardiff model
As stated above, for each of the final describing features, there is a further
distinction to be made in terms of whether or not the type of description is
classifying or depicting. In the current example, it is classifying. Therefore the
list of features selected from the AD HOC DESCRIPTION system is: [not by age],
[by general epithet], [not by affective attitude], [not by minor relationship with
thing], [by role in other situation], and [classifying description of thing by role in
situation]. As a result the full list of semantic features selected thus far is:
It should be noted that the options [not by age], [not by affective attitude],
[not by minor relationship with thing], etc. must be selected, as these are
considered as semantic options; they are included in the selection expression
to signal that these systems are not entered. Consequently they must be
included in any statement of selected features (selection expression).
As stated earlier, this network is not complete. It does not end here and
many of the branches (systems) shown are not end points but continue and
result in realization rules. These are not included here due to space
constraints but they must not be forgotten. Having selected the feature
[cultural classification], the cultural classification system would be entered in
order to generate the head of the nominal group. Since we know that the head
that was selected by the speaker is pills, the ultimate selection in the cultural
classification system would lead to the selection of pill (h < “pills”, since it is
known that the cultural classification is plural).
Two of the semantic features selected will lead to re-entry rules. These are
[by general epithet] and [by role in other situation]. The former will lead to a
modifier being introduced into the nominal group and re-entry will be to fill this
150
Chapter 6: The place of referring expressions in the Cardiff model
element with chemo. The latter will lead to a qualifier being introduced in the
nominal group. Re-entry for this condition will be to fill the qualifier with a
clause (she takes). The process will continue until all items have been
expounded and the chemo pills she takes is realized. The details of the
various passes through the system networks will not be discussed here as
they are not directly relevant to the main purpose of this chapter.
Two important components of the system networks in the Cardiff Grammar
have been left out of this presentation. The first is the realization rules and the
second is the probabilities associated to the semantic options in the individual
systems. The main purpose here is to gain a basic understanding of the place
of referring expressions in the Cardiff Grammar. There is no need to discuss
the realization rules or the probabilities. With respect to the probabilities, they
will become a relevant point of discussion in Chapters 9 and 10 where the
results of the current study are presented.
Having completed the description of the semantic options expressed in the
chemo pills she takes, the discussion will now turn to the formal representation
of referring expressions. As just stated, the realization rules themselves will not
be discussed. Instead the functional syntax of the current example will be
presented.
The previous section illustrated a sampling of the system network for thing
by following one example and retracing the semantic options expressed in the
example. Due to time and space constraints, some of the detail of the full
network has been omitted. In many cases then, the end points of the systems
shown here in Figure 6-4 through to Figure 6-11 are not the final end points of
the network; each would either be extended into further systems going into
further semantic options, lead to re-entry points to previous nodes in the
network (typically specifying a preference for selection) or lead to realization
rules. For example, it may be clear at this point that when [by role in other
situation] is selected, it will lead to a re-entry rule that will include an instruction
to enter the system networks to generate a referent situation (i.e. a clause).
The details of re-entry will not be discussed here.
151
Chapter 6: The place of referring expressions in the Cardiff model
The goal of this section is to relate the semantic features to the formal
representation as concerns referring expressions. It is useful to review the
main components of the Cardiff Grammar, which were discussed in the
previous chapter. These components are illustrated here again in Figure 6-12.
The system network of semantic features for thing has been presented as well
as the relevant higher planning component, the microplanner, which
determines many of the features selected in the network. From Figure 6-12 we
can see that the output from the semantic component (the system network) is
a selection expression of semantic features. Concerning the referring
expression in the current example, the selection expression has been given
as:
{[thing], [congruent thing], [stereotypical thing], [outsider], [cultural
This selection expression then feeds into the formal component of the
lexicogrammar, the realization rules and potential structures. The potential
structures were discussed in detail in the previous chapter. Here we will only
consider those involved in the production of the chemo pills she takes, as our
current example. As already stated the realization rules will not be presented
but instead discussed informally as the tree structure for the current example is
built up.
152
Chapter 6: The place of referring expressions in the Cardiff model
Figure 6-12: The main components of a systemic functional grammar (from Fawcett,
2000:36 and Fawcett, Tucker and Lin, 1993:121)
Before the system network for thing is even considered, the basic structure
of the clause is generated from the first pass through the system networks. In
the current example, the formal representation would be as in Figure 6-13
where the main components of the clause are included and it is already
determined that the Subject is conflated with Agent and Theme. One of the
outcomes of this first pass is a rule leading to a re-entry into the system
networks in order to fill the Subject/Theme/Agent element with the feature
[thing] pre-selected.
Figure 6-13: Labelled tree diagram for the clause the chemo pills she takes are
working on the first pass of the system network
153
Chapter 6: The place of referring expressions in the Cardiff model
Figure 6-14: Output from the first pass through the first pass through system network
for THING
154
Chapter 6: The place of referring expressions in the Cardiff model
Figure 6-15: The final labelled tree diagram for the chemo pills she takes are working
This presentation of the formal representation will end here although there
is much more to say. As this thesis is not contributing directly to the
computational model, there is no point in detailing the various traversals of the
system network or the realizations rules. Furthermore the description so far
has provided sufficient detail so that the approach in the Cardiff Grammar to
the overall process of referring should be clear in general. The probabilities
associated to each semantic option have also been omitted from this
discussion as they are not particularly relevant to understanding the process of
referring.
6.5 Summary
The aim of this chapter was to describe the place of referring expressions
in the Cardiff model of language. Much of what has been presented here is
not published material and so is not readily available to researchers. This
chapter offers a sufficiently detailed account of linguistic reference in order for
the overall process to be understood. There is insufficient space here for a full
and complete account which would ideally include a full discussion of the
system network for thing and the realization rules. What this chapter has done
155
Chapter 6: The place of referring expressions in the Cardiff model
is explain the approach taken in the Cardiff model since no other linguistic
model has attempted to provide as much detail concerning this particular area
of research. First, the conceptual and (pre)semantic representations of
referring expressions were described. This area which attempts to explain
what might be considered a pre-linguistic stage of referring. Next the relevant
parts of the micro-planner were described which illustrated some of the
considerations a speaker may go through in planning a referring expression.
Following this, the system network for ‘referent thing’ was presented in as
much detail as possible; however the full network is vast and complex and it
would be impossible to discuss each and every semantic option and the
accompanying realization rules in this thesis. Finally, the functional syntax of
the nominal group was discussed. The Cardiff Grammar has a very detailed
account of the function and structure of the nominal group. Understanding this
approach to nominal group syntax is very important to the current research as
it is this approach that is adopted as the analytical approach to the study of
referring expressions in this thesis.
The next chapter will present the theoretical contribution made by this
thesis by formulating a new approach to the linguistic study of complex
referring expressions. It will do so by drawing on the previous chapters
including the current one and also by giving consideration to the more
standard or traditional approach to postmodification in the nominal group. Most
importantly, this chapter develops the theory of transitivity for the study of
complex referring expressions. It is possibly the most important chapter in the
thesis.
156
Chapter 7: Towards a functional grammar of complex referring expressions
7.1 Introduction
referring expression:
situation.
157
Chapter 7: Towards a functional grammar of complex referring expressions
158
Chapter 7: Towards a functional grammar of complex referring expressions
159
Chapter 7: Towards a functional grammar of complex referring expressions
then the nominal complement is there to fulfil this role. This would further
suggest that the nominal complement is not related to any role it may have in
the referring expression. Furthermore it also implies that the lexical head of
the nominal group must be ‘in place’ before any modification. If this were not
true then there could be no notion of complement in the analysis of the noun
phrase. Regardless of whether we consider that head nouns are not modified
(cf. Fawcett, 2000) or whether we consider that they are, the systemic
functional view of these expressions is that the noun senses are represented
in the system networks (see Chapter 6). Therefore it is not possible in current
descriptions for the head to be selected before any other systemic features
(with the exception of preselection rules). The approach taken in the Cardiff
Grammar suggests, as was explained in the previous chapter, that the head
element of the nominal group is generated first from the realization rules
before any modifiers are but this does not suggest that the selection of any ad
hoc description is made after the selection of the cultural classification or the
item expounding the head. As has been argued in this thesis, the referent, at
the point of preparing to produce the referring expression, is a mental
construct or conceptual object, in the mind of the speaker; it is this to which the
speaker wishes to refer. We have also seen that in the Cardiff grammar, the
head noun represents the cultural classification of the referent; therefore in a
sense, it ‘stands in’ for the referent. It is itself at times a description, when
there is a choice to be made among different classifications.
160
Chapter 7: Towards a functional grammar of complex referring expressions
There are two main researchers who have examined the tests and criteria
for the distinction between complement and adjunct postmodifiers. Fries
(1999) and Keizer (2004) both explored the main arguments for these
161
Chapter 7: Towards a functional grammar of complex referring expressions
162
Chapter 7: Towards a functional grammar of complex referring expressions
163
Chapter 7: Towards a functional grammar of complex referring expressions
For Fawcett (forthcoming, b) the head of the nominal group in example (56)
is tree and there is no qualifier postmodifying it. His analysis is illustrated in
Figure 7-1 below. In his view, the prepositional group, from a point p on the
ground, is functioning as a qualifier but it is embedded within the nominal
group filling the element of representational determiner. Therefore with this
interpretation, the referring expression would be classed as a simple referring
expression. This illustrates the need for criteria in determining the head of the
nominal group when attempting to account for the referring expression whether
from the production or analysis perspective.
The account provided by Fawcett (forthcoming, b) is not the only approach
to this nominal group. It is perfectly reasonable to consider that the head is
‘angle’ (i.e. the speaker is talking about an elevation angle that is measured
from a tree top to a ground point and he or she is not talking about a tree).
The head element of the nominal group (as the cultural classification of the
referent) can been considered as representing the referent being referred to by
the speaker. Therefore whether through its meaning as culturally classifying
the referent or through some pronominal reference, the head can be argued to
have been selected by the speaker to represent the referent. The remainder
of the elements of the nominal group then serve to further specify the referent
in some way (and there are various ways in which this can be done).
164
Chapter 7: Towards a functional grammar of complex referring expressions
Figure 7-1: Tree diagram for the angle of elevation of the top of a tree from a point
P on the ground, adapted from Fawcett (forthcoming, b)
One of the instances found is given in example (57) below (taken from:
www.boardofstudies.nsw.edu.au/hsc_exams/hsc2000exams/hsc00_maths/96MATIS.pdf).
165
Chapter 7: Towards a functional grammar of complex referring expressions
utterance would want to refer to Object1 (cf. Chapter 6 for the treatment of
referring expressions in the Cardiff model), which, in the speaker’s mind, must
be a defined space or angle since in the planning stages, the speaker would
know that they want to identify this as 30°. It is very difficult to see how tree
could express the cultural classification of the referent.
Re-assigning the head element from tree to angle offers a number of
possibilities for exploring the functions of the remaining items. Three structural
options have been identified in terms of the elements of the nominal group
where angle is expounding the head element. These are listed below in
examples where each qualifier is indicated in square brackets but the internal
structure is not marked.
Table 7-1: Options for the structure of the angle of elevation of the top of a tree
from a point P on the ground
Possible arrangement of elements Example marked for qualifier(s)
in the nominal group
1. det head [q] the angle [of elevation of the top of a tree from a point P
on the ground]
2. det head [q] [q] the angle [of elevation of the top of a tree] [from a point
P on the ground]
3. det head [q] [q] [q] the angle [of elevation] [of the top of a tree ] [from a point
P on the ground]
There would seem to be no dispute that the top of a tree and a point P on
the ground express the unit of nominal group, with tree as head element in the
first instance and point as the head in the second one. The difficulty is in
identifying the relation amongst angle, elevation, tree, and point. Option 1 in
Table 7-1 proposes that the head element is angle and that elevation modifies
angle, and that the top of the tree modifies elevation as does from a point P on
the ground. This is reasonable as the nominal group is not different from an
example such as the solar elevation angle is the elevation angle of the sun,
where it is clear that solar and elevation modify angle. The difficulty is with
from a point P on the ground. This element is somewhat different since it can
be preposed as in from a point P on the ground, the angle of elevation of the
top of a 10 m. tall building is 30°. However in this case it has a circumstance
function rather than qualifier. This suggests that it is not inherently linked to
elevation as a modifier.
166
Chapter 7: Towards a functional grammar of complex referring expressions
The second option includes two qualifiers, both of which expand the
referring expression by modifying angle. The main difference between options
1 and 2 is that in 2, from a point P on the ground is modifying angle whereas
the top of a tree is modifying elevation. In other words this would be
expressing something similar to this is an angle which is from a point on the
ground and it is an elevation angle and the type of elevation is that of the top of
a tree. Alternatively this could be expressed as one nominal group as follows:
a tree-top elevation angle which is from a point P on the ground. This makes
no sense, or rather it makes no mathematical sense. It should be clear that
there can be no angle for a single point and that the concept of elevation
requires two points of reference. It would be impossible to argue in favour of
option 2 therefore since it would be necessary to see both post nominal
expressions modifying either angle or elevation, which is clearly not the case in
option 2. Consequently option 2 can be dismissed without further discussion.
Option 3 includes three qualifiers, all of which are modifying angle. If we focus
solely for a moment on the relationship between the head angle and of a top of
a tree, it becomes clear that this approach is not desirable as it will
immediately feel as though something is missing for similar arguments such as
those presented against option 2. A discussion of the lexical and cognitive
constraints involved here are beyond the scope of this research but it is clear
that they play an important role. The preferred option therefore is option 1
where there is only one qualifier in the nominal group and the complexity of
this expression is maintained within the qualifier through multiple embedded
qualifiers. The tree diagram representing option 1 is given in Figure 7-2.
167
Chapter 7: Towards a functional grammar of complex referring expressions
Figure 7-2: Tree diagram for the angle of elevation of the top of the tree from a
point P on the ground, following option 1 for the structure of the nominal
group elements.
There may be other alternative analyses for this particular example but the
point being made here is that the head element is central to the analysis of
referring expressions. In the Cardiff Grammar, as was shown in Chapter 5, the
approach to referring expressions does not first identify the head element of
the nominal group. It is the result of the entire process of traversing the
system network for THING. The head is thus produced simultaneously as a
result of the selection of semantic options in the system network. Traditionally
when analysing the nominal group (noun phrase), analysts assume the head is
there and that all other elements are present to modify the head element.
However if we consider the nature of how referring expressions are produced
– in other words, how nominal groups are generated – the picture shifts and
we must rather ask how are these other elements contributing to the referring
expression. Why does the speaker feel that the cultural classification is in
sufficient or undesirable? Clearly these types of questions are of a more
cognitive nature. The answers are only in the mind of the speaker. Both Fries
168
Chapter 7: Towards a functional grammar of complex referring expressions
(1999) and Keizer (2004), after extensive consideration of the problem of post
nominal modification, have concluded that a more cognitive approach is
needed to explain why these elements are present. The position taken here is
that the nominal group must first be considered in the context of building a
referring expression.
The next section provides an overview of how transitivity has already been
applied to the nominal group even though it has never been described fully in
these terms. One could argue that transitivity is a cognitive approach. In this
thesis it will be argued that transitivity is a concept that applies primarily to
things rather than processes and as such it is a good place to begin a new
perspective on researching the nominal group.
nominal group
the children in blue hats
Logical structure premodifier head postmodifier
Experiential Deictic Thing Qualifier
structure
prepositional phrase
Process Range
premodifier head
Logical structure
Epithet Thing
Experiential structure
Figure 7-3: Analysis of the children in blue hats Halliday and Matthiessen (2004:324)
170
Chapter 7: Towards a functional grammar of complex referring expressions
view of prepositional phrases, we are not given any sense of the variety of
processes that may be expressed by ‘minor processes’ nor are we able to
understand how the Participant (cf. participating entity) included in the minor
process is related, if at all, to Thing. In Section 7.5 below a new theory is
presented which attempts to account for this relationship. Before beginning
the presentation of the new theory, we will consider the treatment of transitivity
in the nominal group as developed in the Cardiff Grammar.
7.4.2 Nominalization
171
Chapter 7: Towards a functional grammar of complex referring expressions
Figure 7-4: Incongruence in the relations of two units of logical form to language
(Fawcett, forthcoming a)
173
Chapter 7: Towards a functional grammar of complex referring expressions
174
Chapter 7: Towards a functional grammar of complex referring expressions
nominal group), rather than as Range, i.e. what was being practised. The
speaker, if desired, has available the option of referring to the process of
practising and alternatively, she could have said practising their routine. This
is not the case and the speaker seems to have chosen to refer to a thing, i.e.
practice as an object.
175
Chapter 7: Towards a functional grammar of complex referring expressions
The problem here is that with the nominalization analysis, practice is seen
as having a role of Process. The analysis of example (62) where practice is
seen as a nominalization is given in Figure 7-9. If lexical items such as practice
are consistently viewed within this approach, without consideration of the type
of qualifier, then this would have implications for the overall model in terms of
how these types of words are processed. If we go back to the simplified view
of the model, the speaker has a conceptual object in mind and through several
planning stages, he or she generates an expression to refer to the object in
question. In order for practice to be seen as a process, the speaker would
have in mind a situation rather than an object. Since we do not know how
these types of lexical items are treated (stored or processed) in the brain, it is
very difficult to argue for one representation over another. However, in keeping
with what we feel we do know about the process of generating referring
expressions and the concepts of object and thing, then practice is, at least for
this speaker, a thing (referent) that he or she wants to refer to. In the Cardiff
Grammar, a decision is made in the planning stages to describe the referent
on an ad hoc basis and this is how the qualifier is generated as part of the
referring expression. Following the ideas presented above, when this
happens, this creates an additional situation in which the item which classifies
the referent in cultural terms (i.e. the head of the nominal group) has an
additional role with respect to this additional situation.
The additional role assigned to the referent is necessarily related to the
situation represented by the qualifier (this will be explained in more detail
below). This implies that, for a practice of their routine, the speaker decided
that practice (the cultural classification of the referent thing) was insufficient for
the current purposes and an ad hoc description is generated. As a result, a
practice and their routine are two referent things that are included in some ad
hoc (temporary) situation for the purposes of referring. There is thus a relation
between the two, at least in the speaker’s mind. This is even clearer with a
practice that we watched last week as shown in Figure 7-8.
176
Chapter 7: Towards a functional grammar of complex referring expressions
Figure 7-8: a practice we watched last week showing role in other situation
7.5.1 Transitivity
177
Chapter 7: Towards a functional grammar of complex referring expressions
If this is applied to a clause, we see quickly that the relation will not hold
(not in a strict sense at least). For example, in (63) and (64) it is possible to
argue that there is a transfer that takes place between the man and the ball,
yet it is not perfectly clear that there is a transfer between the boy and the girl.
If we tried to apply the mathematical relation, we find the following for (63):
a=‘the man’ ; b=‘hit’ ; c=‘the ball’. If we are able to define a relation, R,
between a and b, it is not the same as the one between b and c and therefore
there is no transfer in the mathematical sense. However, there is no need for
us to attempt to do this since it does not invalidate the use of transitivity. It
does mean that we need to be careful and consistent in its definition and use.
Linguistically, the notion of transfer has followed along the lines of Agent –
Patient relations: i.e. with the man hit and the ball was hit, there is a transfer to
the ball, from the man. What is clear is that a is in relation with c, by some
indicator b.
In its purest sense then, with respect to transitivity, given a R b ∧ b R c ⇒
aRc, we know, as we saw above, when dealing with a clause, we cannot
obtain aRc. However given b R c implies a, in other words, the relation a ↔ b
178
Chapter 7: Towards a functional grammar of complex referring expressions
179
Chapter 7: Towards a functional grammar of complex referring expressions
referring potential. This is not the intention here. It was explained in Chapter 1
(cf. Rijkhoff, 2002: 28) that only nominal groups have referring potential;
elements of the nominal group do not. The challenge with complex referring
expressions is how to represent the additional role the referent has in the ad
hoc situation (μ). As will become clear in this section, the solution proposed is
to treat the head element of the nominal group as a ‘representative’ of the
referent. A full discussion of this area and potential counter proposals is
beyond the scope of the thesis. Therefore we will turn now to the presentation
of the theoretical approach to complex referring expressions as developed in
this thesis.
Axiom 1
In producing a complex referring expression a speaker will
construct a temporary conceptual situation, μ, for the purposes of
describing the target referent in a situation other than the one in
which the target referent is being represented.
Axiom 2
In any given complex referring expression the target referent,
represented in the nominal group as the head element, will have
a functional role in μ.
180
Chapter 7: Towards a functional grammar of complex referring expressions
Axiom 3
In any given complex referring expression the qualifier element of
the nominal group will be conflated with temporary conceptual
situation μ.
181
Chapter 7: Towards a functional grammar of complex referring expressions
Figure 7-11: the chemo pills she takes are working with transitivity analysis
This approach may seem obvious when the qualifier is filled by a clause.
However the position taken here is that it should apply regardless of the
structural nature of the qualifier. The consequence of Axiom 3 is that both [by
role in other situation] and [by minor relationship with thing], as semantic
options, would lead to conflating the qualifier with μ. This may suggest that the
organisation of this section of the system network should be reconsidered but
this is beyond the scope of the current presentation.
The semantic option [by minor relationship with thing] will lead to a
realization rule that will insert a qualifier in the nominal group which will be
filled by a prepositional group. In other words, this feature describes post
nominal prepositional groups. As was shown in Section 7.2 above, this area
has proved to be very challenging for linguists. There are three main
contributing factors to Axiom 3 which includes both post nominal clauses and
post nominal prepositional groups (or phrases). The first is the theoretical
approach to complex referring expressions developed here. The second is the
semantic option [by minor relationship with thing], which is describing a
relationship between ‘thing’ (i.e. the head of the nominal group) and the
completive element in the prepositional group. The third is Halliday’s theory
that the prepositional phrase is a reduced clause. Therefore, there seems to
be no obvious objections to the view that post nominal prepositional groups,
182
Chapter 7: Towards a functional grammar of complex referring expressions
together with the head of the nominal group (‘thing’), constitute part of the
temporary conceptual situation, μ.
The approach developed here can be illustrated with (65) below, a little
house with a yard, taken from the GAE corpus in this study. Axiom 1 states
that the speaker will construct a temporary conceptual situation within which
the ‘thing’ (referent) has a role. In this example, the speaker has constructed a
situation involving both house and yard, but the situation is considered implicit
as there is no linguistic element identifying the relation (process). This is
illustrated in Figure 7-12.
Axiom 2 states that the head of the nominal group, in this case house, will
have a role in the situation, μ. Consequently, the head will be conflated with
some Participant or Circumstance role. Finally, axiom 3 states the qualifier will
be conflated with μ.
In the speaker’s mind, during some stage of planning, there must be a
cognitively constructed (even if only symbolically) situation in which X is a
member of the situation, S. For example, with a little house with a yard, there
must be, in the speaker’s mind, a situation in which house and yard, as two
referents, co-occur, even if the situation is simply to describe the existence of
the two referents in a shared space. In order for the speaker to include other
referents in the referring expression for the purpose of referring to the desired
referent, there has to be a defined referential space, i.e. situation in which the
two (or more) referents occur and through which they are related by a relation
that is either explicitly or implicitly expressed. Since this situation is
constructed only for the purposes of referring to the referent, it is considered
here as temporary.
183
Chapter 7: Towards a functional grammar of complex referring expressions
The only problem left is identifying the role that the head element has. The
theory asserted here simply states that the head element will always have a
functional role with respect to the situation expressed by the head and qualifier
together. Full criteria have been developed for the application of this theory.
The criteria are presented and discussed in Section 7.6.1 below. However to
conclude this section, a solution is proposed by simply rephrasing the
expression. It should be reasonably clear that the situation constructed for the
referring expression is one in which house and yard are related in some
relationship such as a house which has a yard. Therefore the role to be
conflated with the head is that of Possessor and the completive is conflated
with the role of Possessed, as shown in Figure 7-13.
In this section the theory behind the approach developed in this thesis has
been presented. An example was used to illustrate how it can be applied to
184
Chapter 7: Towards a functional grammar of complex referring expressions
The criteria given here have been developed following Neale (2002). Her
criteria were developed for analysing transitivity in the clause. The criteria are
not the same but the approach taken here is very similar and adapts her
criteria.
As has already been stated, in this theory, the head of the nominal group
for a complex referring expression will always carry an additional role (see
Axioms 1 and 3). The role conflated with the head may be a Participant type
role or a Circumstance type role. In the discussion above it was made clear
that the relation may be explicit or implicit. A clearly explicit expression of the
situation and the additional role assigned to the referent is typically found when
the qualifier is filled by what is traditionally called a relative clause. An implicit
expression of the situation is found when the relation is not explicitly identified.
A common example of this is when the qualifier is filled by a prepositional
group. In this case, the relation is deduced based on the role it is deemed to
have with respect to the qualifier and the use of the criteria presented here
becomes essential. There are some cases where the relation is deemed to be
covert and the relation can only be assumed, as will be explained below.
The presentation of the set of criteria for analysing complex referring
expressions is divided into three sections. The first concerns the case where
the head is conflated with a Participant Role. The second represents cases
where the head is conflated with a Circumstance Role. The last division of the
186
Chapter 7: Towards a functional grammar of complex referring expressions
criteria discusses the covert relations where it is very difficult if not impossible
to feel confident about the relation between the head element and the qualifier.
All examples cited here have been taken from the corpora used in this study
unless otherwise stated.
The head of a nominal group that also has a qualifier element will have an
additional role which reflects its function in the situation created by the
qualifier. When q is filled by a clause, either h will have a role within the
embedded clause, i.e. it is also a participant of the embedded clause as in ‘a
parent who is on maternity leave’, or it will be related to the qualifier through
some covert (not expressed) relational situation, as in ‘plans to come home’.
The head is conflated with Actor when the qualifier answers the question:
“what did h do?”
example: the person going abroad
test: what did the person do? the person went abroad
The head is conflated with Goal when the qualifier answers the question “what
was done/what happened to h?”/”who did something to h?”
example: a nice long message I typed to you
test: what was done to the message? the message was typed.
The head is conflated with Beneficiary when the qualifier answers the
question “who gave something to h?”/”what was given to h?”
example: the man that John gave the ball to (invented example)
test: who gave something to the man? John.
What was given to the man? The ball.
187
Chapter 7: Towards a functional grammar of complex referring expressions
The head is conflated with Senser when the qualifier answers the question
“who perceived/etc. something?”
example: the woman who wants the apartment (invented example)
test: who wants the apartment? the woman wants the apartment.
The head is conflated with Phenomenon when the qualifier answers the
question “what was perceived/etc.?”
example: the building she wants
test: what does she want? she wants the building
188
Chapter 7: Towards a functional grammar of complex referring expressions
Identifying relations
The head is conflated with Identified when the qualifier answers the question
“what is x?”/”what is the name of h?/”which x?” etc.
example: a game called Junior Trivia
test: rephrase for equivalence - what is Junior Trivia? Junior
Trivia is a game.
Possessive relations
The head is conflated with Possessor when the qualifier answers the question
“what does h possess/own/have?”
example: a little house with a yard
test: rephrase for equivalence - what does the little house have?
the little house has a yard.
The head is conflated with Possessed when the qualifier answers the
question “who/what possesses/owns/has h?”
example: the role of the Aboriginal community
test: rephrase for equivalence - who has a role? The Aboriginal
community has a role.
The head is conflated with Sayer when the qualifier answers the question
“what did h say?”
example: the man who said no (invented example)
test: what did the man say? the man said no.
The head is conflated with Verbiage when the qualifier answers the question
“who said h?”/”what was said?”
example: something I said
test: who said something? I said something.
189
Chapter 7: Towards a functional grammar of complex referring expressions
The head is conflated with Recipient when the qualifier answers the question
“what was said to h?”
example: the man to whom John told the secret (invented example)
test: what was said to the man? the secret was told to the man.
There were no examples found in the corpora which expressed any other
type of relation. In other words, Behavioural and Existential types of relation
did not occur. This does not suggest that this is not possible but rather reflects
perhaps the fundamental meaning of the participants involved in these process
types.
Manner where the head expresses the role of specifying manner. (where q
answers what situation is modified by h? i.e. what situation is done in manner
h?)
190
Chapter 7: Towards a functional grammar of complex referring expressions
example: the way the different partners have integrated their own
interests
test: how have the partners integrated their own interests? the
way
Cause where the head expresses the role of specifying cause (q answers
what situation is modified by h? i.e. what situation is done for cause h? h
expresses a purpose or cause)
Extent where the head expresses the role of specifying extent (q answers
what situation is modified by h? i.e. what situation is done to the extent
expressed by h? h expresses an extent or duration)
example: the last day working in the old building
test: how long were you working in the old building? we were
working in the old building for the last day.
The cases referred to as covert relations are difficult since the situation
created by reference to the referent’s cultural classification and the qualifier is
not overtly expressed. In other words there is nothing that categorizes the
type of process and/or participants. This is, in many ways, similar to Halliday’s
view of the prepositional phrase as a reduced clause. However covert
relations do not only exist when the qualifier is filled with a prepositional group.
191
Chapter 7: Towards a functional grammar of complex referring expressions
We find covert relations when the qualifier is filled with a nominal group,
prepositional group, quality group and also in some cases when it is filled with
a clause (usually a non-finite clause). Examples of these are given in (66) to
(69) below.
The most frequent realization of this type of relation is when the qualifier is
filled with a prepositional group as is shown in examples (70) and (71).
In all of these cases, the relation governing all the referents in the situation
is not expressed and therefore it cannot be readily identified. In (70) above for
example, there is no obvious way to associate problem and school. However
in the example used for the head conflated with a role of ‘Possessor’ as seen
above in a little house with a yard, it is possible to relate house and yard in a
relational situation where yard is something the house should have. With
problem and school, there is no such association possible. It might be
tempting to consider that as a ‘thing’ problem developed or took place at
school which could lead to seeing the additional situation as a material type
but a more conservative approach has been developed here for these cases.
The approach taken here is to proceed on the theoretical assumption that a
situation does exist for the speaker, as discussed above, and that the speaker
has selected not to express it. In other words, the speaker has opted for a
192
Chapter 7: Towards a functional grammar of complex referring expressions
In (72), friends is conflated with the role of Carrier and the completive in the
prepositional group, across the street, is conflated with the role of location in
space since the referent is being described in relation to a location.
Example (75) is one that is perhaps most open to debate as both the
intended syntax and functional relation could be called into question. The
treatment of this nominal group here is to consider that the trouble is caused
by the nail and therefore the completive, that nail, is best described as
expressing the role of why. Therefore it follows the same analysis as in (74)
above where the completive is conflated with a role of cause.
194
Chapter 7: Towards a functional grammar of complex referring expressions
7.7 Summary
The goal of this chapter was to motivate and present a new theoretical
framework for the analysis of complex referring expressions. The post head
elements of the nominal group have been identified by many linguistics as
being the most challenging area of nominal group function and syntax.
However as Keizer (2004) and Fries (1999) both conclude more research
needs to be done in this area. Both suggest that the key approach to making
some progress in understanding these forms can be found in taking a more
cognitive approach to the study of the nominal group. It seems clear that this
can only be an attempt to reflect the intentions of the speaker and therefore a
reasonable starting position is from a framework that considers how speakers
refer, i.e. what intentions contribute to the production of the nominal group.
As discussed in earlier sections of this chapter, transitivity has been used
to explain some functions in the nominal group. However, it has never been
used to account for the presence of the qualifier with the exception of the
195
Chapter 7: Towards a functional grammar of complex referring expressions
196
Chapter 7: Towards a functional grammar of complex referring expressions
197
Chapter 8: A methodology for the analysis of referring expressions
198
Chapter 8: A methodology for the analysis of referring expressions
would have been impossible to develop the ideas presented here to the extent
that they have been had the original framework been held rigidly.
The research presented here has its primary focus on referring
expressions. The main goal is to reach a better understanding of their
production within a systemic functional framework. Specifically, the
multifunctional nature of referring expressions will be investigated, with
emphasis on the expression of experiential meaning as realised in referring
expressions. This involves looking specifically at the options selected by
speakers and attempting to account for this theoretically.
In the remainder of this chapter, the method of data collection and analysis
will be explained. The texts selected for the three corpora involved in this
study will then be described. Finally, a large section will be devoted to the
development of the methodological approach to the analysis. This includes
the software used to code or ‘tag’ the texts for the specific features being
analysed and the theoretical issues involved. The major contribution here is in
the development of a methodology that attempts to adapt and apply the
theoretical system networks in a coding scheme that enables the researcher to
recover the choices made in producing referring expressions.
The output of the analytical coding of the texts is in the form of XML. This
output is included in electronic form in the Appendix of this thesis on a CD-
ROM. Two XML databases have been constructed from the codings: REx.xml
and CREx.xml. The first, REx.xml, includes all referring expressions analysed
in this research and this totals 3 411 referents (see Table 8-2 below). It
represents each referring expression in a tagged format. The statistical
representation of the analysis is given in Chapter 9. The second database,
CREx.xml, includes all complex referring expressions and each one is tagged
in detail using XML in a similar way as the database with the analysis of the
individual referring expressions. However the amount of detail and the tags for
CREx.xml are different to REx.xml and for space saving reasons and for
reduced complexity, the databases have not been fused into one single
database. This would ideally be a project for further research. The databases
will not be discussed further as time and space does not permit. It is perhaps
useful for the reader to know that all the expressions analysed are included
here and that they are available for consultation if desired.
199
Chapter 8: A methodology for the analysis of referring expressions
200
Chapter 8: A methodology for the analysis of referring expressions
201
Chapter 8: A methodology for the analysis of referring expressions
instance and the larger extract would then have to be analysed by hand in any
case. It would seem then that there is little to no advantage to using it at this
early stage in the research. Once a fuller description of this part of the
grammar is available, we can then make good use of large corpora to verify
patterns and probabilities (for example Neale, 2002).
The most appropriate option for source of data then is for the researcher to
construct a corpus from written, spoken or electronic language. There is no
particular advantage of one mode over another with the exception that with
spoken language some additional variables are introduced and it may be
desirable to avoid these. Spoken language is often face to face (requiring co-
presence of participants) and this often introduces additional modes such as
non-verbal communication (gesture for example). Working with spoken
language also requires an enormous investment on the part of the researcher
since it must be first transcribed before the analysis can begin. Although this
is not a scientific problem, it would more than double the time and work
invested in the study. Since this research requires human parsing, it would
inevitably lead to a reduction in the amount of text that could be analysed over
a reasonable period. The only real disadvantage to using written (published)
texts is not having sufficient information about the speaker or speakers,
including important information about the editing process and any restrictions
on word count for example which may lead to a higher clause density.
Electronically produced texts (namely email texts, webpages, blogs, or chat
transcripts for example) are not without problems but they seem to be a very
appropriate choice for investigating language choice in the lexicogrammar for a
variety of reasons. The use of electronic texts eliminates the occurrence of
gestural reference since speaker and addressee do not share a physical co-
presence. In addition, these texts are produced with no observer effects and
this leads to texts which are as naturally produced as possible.
The texts used in this study come from a large archive of personal email
messages. Messages were selected from two speakers, forming two relatively
small corpora. They were chosen as they were the only ones to have the
202
Chapter 8: A methodology for the analysis of referring expressions
following two text variables (see Fontaine 2004b for a discussion of text
variables in email texts). Firstly, they were composed online using HTTP-
based email software. This is important as it gives less opportunity for editing
as compared to POP-based email software. Secondly, both authors
composed linear texts, in other words their texts did not include any reported
or embedded text from previous messages. This is also important as it often
becomes difficult to readily separate speakers in recursively constructed (or
non-linear) email texts (Fontaine, 2004b). The advantage to using these types
of texts (i.e. personal emails) is that they are as close as possible to casual
conversation without imposing an artificial nature to the discourse, without any
observer effects, and without third-party editing or revision.
A third text was included, forming a third small corpus, in an attempt to
control for text type variables. As was explained above, the type of text is
nearly irrelevant for this study and therefore any distinctive type of text could
have been used. In light of this, a completely unrelated text, having a relatively
comparable word count was analysed so that we could come to some
understanding of what influences could be attributed to speaker differences,
text type differences and most importantly differences related uniquely to the
internal structure and function of the nominal group.
The corpora used in this study were constructed as follows. First the
archive of personal email texts was consulted to find speakers who would be
suitable for the study. The texts chosen were single speaker texts with no
reported or embedded texts from previous messages. Two speakers were
chosen to create two relatively small corpora of 4824 and 4391 words
respectively (see Table 8-1). The third text selected was intended to be
completely different from the personal email texts so that the results for the
analysis could be compared in several dimensions: texts written by one single
speaker; texts written by different speakers within a similar text type (personal
email texts); texts written by different speakers in different text types (personal
email texts and another type of text). The only restriction on this third small
corpus then was that it should be of a clearly different, unrelated type of text.
The text chosen was taken from an online educational brochure on the Model
Forest Network.
203
Chapter 8: A methodology for the analysis of referring expressions
Having now described how the corpora were constructed, the next section
will focus on how they were consulted for analysis.
This section focuses on the way in which the theory, which was presented
in the previous chapters, has been converted to application in a computer-
assisted analysis. It describes the software used to tag the data and how the
analysis was conducted. The amount of detail in the analysis makes coding
large amounts of data challenging.
204
Chapter 8: A methodology for the analysis of referring expressions
Figure 8-1: Tree diagram showing the functional roles filled by nominal group
(invented example)
205
Chapter 8: A methodology for the analysis of referring expressions
(77) the kind woman in the house on the next street (invented
example)
206
Chapter 8: A methodology for the analysis of referring expressions
The embedded nominal groups within the qualifier, the house on the next
street and the next street have not been analysed internally for the purposes of
the overall description of the nominal group. The main reason for this is, as we
will see below in the presentation of the software used to code the data, that
successfully analysing, coding and tracking multiple levels of recursivity in the
nominal group is especially challenging. It seems more important at this stage
to attempt an accurate description of these expressions at the first level of
reference, i.e. the expression referring to the conceptual object (i.e. the object
that the speaker has in mind). Complex referring expressions like (77) above
are analysed analogously to transitivity as we will see in Section 8.3.3 below
and as was described in Chapter 7.
Since the particular point of study here is referring expressions, any items
filling determiners and modifiers were not considered and only the structures
filling the role of qualifier were analysed in great detail. In other words,
determiners and modifiers were included without regard to the units filling
these elements. Future research should go beyond this first level of analysis
in order to detail multiple levels of referring and consider whether such use of
nominal groups can be seen as referring or not.
207
Chapter 8: A methodology for the analysis of referring expressions
This section will describe, as briefly as possible, the software used to code
the data analysed.
The data were analysed by hand and coded using the Systemic Coder
(O’Donnell, 1995; and available online: www.wagsoft.com/Coder/). This
software is excellent for coding data and it helps contribute to maintaining a
high degree of consistency in the analysis due to both the nature of the
software and the features it offers the user. The Coder has five main
interfaces or modes: Text; Scheme; Coding; Review; and Statistics.
The input to the software is the text to be analysed in plain text format (.txt
or unformatted text). In this mode the user determines the boundaries for the
units under analysis by inserting markers into the text. Figure 8-4 shows a
section of text being analysed with the boundaries in place around nominal
groups. The system then interprets these markers as boundaries and in
Coding mode, the system will prompt the user to identify the features found in
the unit. In this study, the unit of analysis is the nominal group. Identifying the
boundaries was labour intensive as it needs to be done carefully by hand and
for large texts, this is even more so. Sections of text that are not of interest to
the current analysis can simply be left as one unit as they can be left out of the
coding by using the ‘ignore’ function. Therefore, in fact, boundaries will also be
marking sections of text that are not of interest. As an example of this,
consider the first line of text in Figure 8-4: | I | ‘ll try to get | the book |. Clearly
‘ll try to get is not meaningful in the sense that this stretch of text is ignored in
the analysis. The boundaries are not suggesting that this is a unit.
Figure 8-4: Unit segmentation in Text Mode (personal names have been blocked)
Figure 8-5 shows the Scheme mode, but the scheme developed for this
study is very large and cannot fit in one window (see Sections 8.3.2 below and
8.3.3 below). Building the scheme was possibly the most challenging part of
the research. There were two main challenges. The first was in terms of
dealing with the affordances and constraints of the software. The Coder is not
able to handle embedded structures. This was a particular challenge given the
complexity of the nominal group and the relatively great number of
permutations and combinations of its elements. The second challenge was in
trying to build a scheme that would model, to a certain degree, the networks
for generating referring expressions, while still offering a way to make
209
Chapter 8: A methodology for the analysis of referring expressions
Before going into detail on the actual scheme used, we will complete the
overview of the Systemic Coder.
Once the scheme has been finalised, the text is ready for coding. In
Coding mode, the system simply prompts the user with one unit at a time
together with the features available from the Scheme. We can see this in
Figure 8-6 where the unit, the book is prompted and the features are made
available to the user on the right. The top window displays the text and partial
context. The window on the left lists the selected features. In Figure 8-6 the
210
Chapter 8: A methodology for the analysis of referring expressions
user must select from the available features for the system (the use of SMALL
FULL CAPS will be reserved for the names of features): POSITION (SUBJECT,
COMPLEMENT, CIRCUMSTANCE), REF-EX TYPE (THING, EVENT, DESCRIPTION) and
ROLE (PARTICIPANT, CIRCUMSTANCE), before the system can present any further
features in the scheme. Once the user has made a selection for all systems in
the network, the system will move on to the next unit and prompt the user and
the cycle begins again. The analysis is complete once all units (sections of
text surrounded by boundary markers) have been either coded with a selection
of features or ignored (see Figure 8-7).
The fourth mode allows the user to check the status of all units through a
Review interface, which can be seen in Figure 8-7. This is simply a means of
verifying that all units meant to be analysed were indeed analysed. It also
provides a shortcut to either ignoring individual units that are not desired in the
analysis or alternatively making ignored units available for whatever reason.
211
Chapter 8: A methodology for the analysis of referring expressions
Finally, once all coding is complete, the Statistics interface enables the
user to take advantage of the statistical tools that come packaged with the
system. The Coder has a very attractive statistical tool that allows for a variety
of comparisons and ways to view results. For example, results can be viewed
and analysed according to one particular system. In the above example, we
mentioned the POSITION system where each unit is analysed for the features
SUBJECT, COMPLEMENT, ADJUNCT and in Statistics mode, results can be made
comparatively among these features. The user may find it interesting to see if
his or her results are significantly different when the unit is in SUBJECT position
or not, for example. Comparative statistics can be obtained with the built in
statistics package where T-scores are calculated as a measure of statistical
significance. However, basic calculations such as the descriptive statistics
shown in Figure 8-8 can also be easily extracted. A full description of the
212
Chapter 8: A methodology for the analysis of referring expressions
213
Chapter 8: A methodology for the analysis of referring expressions
Figure 8-9 illustrates basic system notation. The simple system shown
here can be explained in words as follows. SYSTEM A contains two systems:
SYSTEM B and SYSTEM C. Upon entering the network, the user must enter
SYSTEM B AND SYSTEM C simultaneously, or in parallel (in practice this is of
course impossible but this means that the systems can be independent,
although related). For each system entered, one feature must be selected.
For example, upon entering SYSTEM A, the user will have to select from SYSTEM
B either FEATURE A OR FEATURE B AND from SYSTEM C either FEATURE C OR
FEATURE D. This is illustrated more specifically in Figure 8-10 which shows how
this notation is used in the coding scheme.
214
Chapter 8: A methodology for the analysis of referring expressions
Figure 8-10: Partial view of the coding scheme using system notation
215
Chapter 8: A methodology for the analysis of referring expressions
The Coder Scheme has the constraint that no two nodes in the scheme
can share the same name, which is both useful and very important for obvious
reasons. For example, once the term SUBJECT has been used for one node in
the scheme, it cannot be used elsewhere in the same scheme, hence the need
for some creativity in designing the scheme.
We will now begin the full description of the scheme. The scheme begins
with a system network for REFERRING_EXPRESSION. Upon entering the scheme
network, three systems are encountered which are entered in parallel as we
can see in Figure 8-11. Each system will be discussed in the following order:
POSITION, ROLE and REFERRING_EXPRESSION_TYPE. REFERRING_EXPRESSION-
TYPE is given last as it will require the most explanation and the first two will
only be described briefly.
The features in the POSITION system are in fact as were described above:
SUBJECT, COMPLEMENT, OR ADJUNCT. These features describe the position of
the referring expression. The features in the ROLE system relate to the role the
referring expression has in the transitivity analysis for the Experiential meaning
it has in its context of the clause. With these two feature coding systems, then,
the referring expression is being analysed within the context of the clause in
which it was used.
Figure 8-11: First stage of the scheme for analysing a referring expression
TYPE.
216
Chapter 8: A methodology for the analysis of referring expressions
The features available in this system are THING, EVENT, and DESCRIPTION.
217
Chapter 8: A methodology for the analysis of referring expressions
The examples shown in (78) to (83) illustrate the distinction among THING,
EVENT, and DESCRIPTION using examples from the corpora in this study.
(83) the girls who I used to work with at the company (coded
as THING)
218
Chapter 8: A methodology for the analysis of referring expressions
OUTSIDER and SOUGHT. SOUGHT is the name for the feature that refers to
referents that are not known (for example: who phoned? what did he see?).
219
Chapter 8: A methodology for the analysis of referring expressions
Figure 8-14: Reproduction of the system network for interactant reference (cf.
Chapter 6)
220
Chapter 8: A methodology for the analysis of referring expressions
221
Chapter 8: A methodology for the analysis of referring expressions
to be made between lexis and grammar (and this is not to say that there is), it
may very well be in this area.
Up to this point, the process of developing the scheme is relatively
straightforward since the elements of the nominal group can be identified as
being present or not in the expression being analysed. The detail for
HEAD_ONLY cases is desirable in order to make a distinction between those
expressions that are referring by naming, although this forces an unnatural
split between referring expressions with only a lexical head, as in flour, and
those with a determiner or modifier of some kind, as in organic flour. This
division is not really problematic in terms of coding as this information can be
retrieved very easily by grouping the results differently depending on what
information is being sought. For example, one could easily add the statistics
for LEXICAL_ITEM to any of the HEAD_PLUS statistics if that were desirable.
As we can see from the view of the coding scheme in Figure 8-15, there
are only three remaining features that lead on to further codings. These are
DETERMINER, PRE_HEAD and POST_HEAD. The determiner coding scheme
shown in Figure 8-16 illustrates the features analysed in the determiner-type
system. Here we attempt to code to a minimal degree of detail certain types of
determiner. This is one area in which the amount of detail present in the
theoretical system networks for generating referring expressions is significantly
reduced. The feature being targeted for the purposes of this study is
particularisation which refers to whether or not the referring expression refers
to a particular object or not.
222
Chapter 8: A methodology for the analysis of referring expressions
223
Chapter 8: A methodology for the analysis of referring expressions
{pre-head(quality(quality_1)), post-head(qualifier(no_qualifier))}.
As a second example, bad news of jimmy (from GAE), has the following
elements in the nominal group: ngp → m h q. The coding features selected for
this expression consist of:
{pre-head(quality(quality_1)), post-head(qualifier(qualifier...))}.
The details for the features for qualifier are given below in Section 8.3.3.
This completes the presentation of the analytical methodology for the first
round of analysis. In order to research the relationship between the head of
the nominal group and the qualifier, detailed codings were developed based on
the hypothesis that this relationship can be explained in terms of experiential
transitivity, in an approach that is analogous to transitivity in the clause (as a
system of experiential meaning). The method of coding these expressions will
be explained in the next section.
The coding scheme presented in this section covers the method of coding
all complex referring expressions. This scheme is significantly different from
the original scheme first designed for the task of analysing these expressions.
The final version was developed during the analysis and forms part of the
research developed in this thesis. It was not applied in a deductive manner,
but rather inductively where the goal was to develop a coding model that
worked with the data. It would be very informative to explore the changes that
shaped the current scheme but this is beyond the scope of the current work.
The theoretical motivation and argumentation concerning the development
of the coding scheme presented in this section is given in detail in Chapter 7.
Here we will simply explain this section of the coding in terms of how the
analytical methodology was carried out.
The scheme is shown in Figure 8-18 and represents the last section of the
coding analysis. It illustrates the main features that were developed for coding
complex referring expressions, in other words, all nominal groups with a
qualifier element.
224
Chapter 8: A methodology for the analysis of referring expressions
For each complex expression, two feature systems must be given values,
as can be seen in Figure 8-18: type of AD_HOC_DESCRIPTION and type of ROLE.
This is to say that each complex expression will be coded for having a role as
either a participant or a circumstance in the additional (or ad hoc) situation
(see discussion of μ in Chapter 7) and that each qualifier represents one of
three types of ad hoc description.
ad_hoc_situation
ad_hoc_relationship_with_other_thing
225
Chapter 8: A methodology for the analysis of referring expressions
post_modifier
226
Chapter 8: A methodology for the analysis of referring expressions
Head as Participant
227
Chapter 8: A methodology for the analysis of referring expressions
Head as Circumstance
(105) HES: more details about the guy from London (DESC-
MATTER)
228
Chapter 8: A methodology for the analysis of referring expressions
For the analyst then this is problematic in the sense that any objective
classification is nearly impossible. These cases have been treated as covert; in
other words, it is accepted that a relationship exists, that it exists in the mind of
the speaker, but that it has not been expressed in a clearly identifiable way
such that the analyst can feel confident in offering a distinct coding. The way
around this problem is to accept that the relationship is covert and consider it
to be a covert relational, where the head has the role of carrier (CARRIER_HEAD)
and the completive (i.e. the ‘other thing’) has a role as well. This is done
without specifying what process would be involved, although it is assumed to
be covertly relational.
Although this section has not given any detail concerning the theoretical
motivation behind the coding scheme (see Chapter 7), the coding scheme
itself has been presented in detail. It should be clear at this point how the data
was approached, how the methodology was carried out, and how the data
were coded for the features developed in the coding scheme.
8.4 Summary
amongst the various components of the nominal group. Given the importance
of probabilities to the Cardiff Grammar, Chapter 9 describes the frequency of
occurrence and interaction of the various functional components in referring
expressions and considers variation over individuals and text type. Chapter 9
will also discuss the relevant descriptive implications from this study.
230
Chapter 9: A functional analysis of simple referring expressions
consider the frequency and distribution of all referring expressions found in the
corpora, as analysed in this research. The main goal in this chapter is to
answer questions such as how much variation is there with respect to referring
expressions for a single speaker, how do two different speakers compare for
the same text type, and what are the main differences and similarities between
two very different text types. Basically, the goal is to describe, to the extent
possible, a detailed account of the functional syntax of referring expressions.
The organisation of this chapter is as follows. Before presenting the results
of the research on referring expression, we will consider some issues
concerning grammatical categories and how this relates to the work presented
here. In the presentation of the results, we will first consider the relationship
between referring expression, referent thing and nominal group. Next the
results for each referring expression realised as a nominal group are given in
detail. We will focus on the various components of these expressions, with
emphasis on ad hoc descriptions. In each case of presenting the relevant
results, the results from the three corpora analysed will be given in turn. First,
the results from the GAE corpus are given; representing one speaker’s use of
referring expressions at three time intervals over a five year period. Then we
will compare the results from this corpus to the other two corpora analysed in
this study. The chapter ends with a summary.
232
Chapter 9: A functional analysis of simple referring expressions
What different view might a linguist have if they were not bound to a
predetermined theoretical set or to its terminology. Admittedly, the research
conducted in this thesis is still very much dependent on structural categories
(e.g. nominal groups and clauses and other such grammatical units) despite
the fact that it argues for pushing towards a different perspective – that of the
choices made in producing referring expressions – and re-used terminology in
its use of referring expression as a linguistic unit.
Linguists have a definition ‘in intension’ (Fontaine and Kodratoff, 2002:7,
for the original use of the term see Arnauld and Nicole, 1662/1965) of each
grammatical category; in other words, they have defined each with specific
criteria, giving them the ability to classify and use terms in the fullest range
possible under the right conditions. For example, linguists define what a noun
is and what a nominal group is and how to recognise it. This is a definition in
intension as it is based on a priori criteria. Speakers also have some kind of
definition for the categories of linguistic items they use, although it may be
unconscious or implicit knowledge; this allows them to use language
appropriately. Functional linguists may benefit from applying a definition in
extension in order to account for all possible instances (in our case, instances
of referring). This would lead to a characterisation based on the instances
encountered.
In the case under study here, the instances we are considering are those
expressions that refer to a thing. Do we know its characteristic properties?
Can we identify the individuals that satisfy these properties? The answers are
not obvious. For the current purposes, a referring expression is recognised by
the ability to identify the referent by some mechanism, e.g. pointing or pronoun
reference.
Speakers are not only interested in representing physical objects.
Language allows speakers a very full range of expression. In addition to
abstract or conceptual objects (perhaps it is something imagined), speakers
may want to represent an activity. In principle, this is an abstract (conceptual)
object. However, in Hallidayan terms, the difference between representing an
object or an activity (event) lies in whether the speaker intends to represent his
or her experience of the concept involved in terms of a process or a
233
Chapter 9: A functional analysis of simple referring expressions
234
Chapter 9: A functional analysis of simple referring expressions
These are clearly invented examples and we know (intuitively) that our
language will let us get away with these expressions. We do not know whether
any speaker would really use these patterns or not or under what conditions or
for what purposes. The examples given in (109) to (111), found through a
Google search, show a similar pattern but the inverse relation, i.e. verb derived
from a noun.
Google is a search engine. It was first a proper name, but its use has
broadened and as a term, it has become very flexible. However it is
impossible to be certain about the process involved in its use. Does google
(noun) imply google (verb)? Is it an analogous use? Or is it, as many would
claim, a grammatical metaphor (i.e. a noun behaving like a verb: the verb
google is (metaphorically) like the noun google.
The position taken here is from that of the referring expression where
although the qualifier may include lexical relations to the head element, it is not
the motivating reason for the qualifier. In other words, we have not been
primarily concerned with the lexical semantics of the individual contents of
referring expressions but rather with the referring expression itself as a whole.
We will now turn to the results of their analysis.
235
Chapter 9: A functional analysis of simple referring expressions
Each of the above features will be considered in turn. The next section
describes the frequency of referring expressions, focussing specifically on
referent thing and the nominal group. In each case we will consider the
frequency and distribution of the coded feature(s) for one speaker (GAE) and
then compare that to the results from the other texts (HES and MFN).
Following this, the results for the features of position and role will be
presented.
236
Chapter 9: A functional analysis of simple referring expressions
Chapter 1, this is not true for all languages. In coding each referring
expression for type of referent (i.e. referent thing, referent situation, referent
description), we can measure to what extent this relationship is stable.
The relationship between referent thing and nominal group is represented
statistically in Table 9-1 and Table 9-2. Recall that the means of identifying a
referring expression initially was any unit having a role in transitivity as
Participant or Circumstance. Therefore the only possibilities for the type of
referring expression are a thing (nominal group), a situation (embedded
clause) or a description (quality group). Examples of each can be found in
examples (112) to (114).
(113) It was good that you weren’t going too fast [situation]
The difference is quite pronounced for the GAE corpus. The frequency of
occurrence for a referring expression to refer to a referent thing is stable at
89.2 per cent as an average. Where there is some variation, albeit very small,
it is with respect to situation and description. This variation is not statistically
significant and in fact the overall frequency of referent situation in this context
is so low that very little can be determined about its statistical value.
237
Chapter 9: A functional analysis of simple referring expressions
Table 9-2: Frequency of referring expressions by type for GAE, HES and MFN
GAE (set) HES MFN
total N 1673 1169 552
frequency (%) N frequency (%) N frequency (%) N
THING 89.2%• 1492 83.9% 981 84.4% 466
SITUATION 3.1%• 52 5.1%• 60 11.6%• 64
DESCRIPTION 7.7%• 129 10.9%• 128 4.0%• 22
Key for levels of statistical significance:
• 98% degree of confidence
† 95% degree of confidence
‡ 90% degree of confidence
The coded features for THING are shown in Figure 9-2. We will now
consider the results for these features. Note that the feature SOUGHT is not
considered any further. Also the only detail for INTERACTANT is as given in
Figure 9-2. There is considerable detail in the coding scheme for OUTSIDER
and the results for its features will be given following the general results for
THING.
238
Chapter 9: A functional analysis of simple referring expressions
The features for THING make a distinction between referents having a role
in the exchange (i.e. an interactant role or a role outside the exchange) and
the expression itself being sought (e.g. ‘what (SOUGHT) did she give you?’).
The distribution of the three features for the GAE corpus is shown in Table 9-3
below. The occurrence of SOUGHT is negligible in this corpus, as well as in the
other two.
The data displayed in Table 9-3 is the frequency distribution for one
speaker over three time periods where 1 673 nominal groups were analysed in
total. The distribution of the referents is consistent in the first two periods with
outsider referents representing 71.4 per cent of the data in period one and
69.9 per cent in period two. We find a significantly different distribution in
period three where this figure rises to 80.9 per cent. Since we are not
attempting to account for this variation, we will not discuss any potential
reasons for this. However it would seem that there is room for more detailed
research here in order to establish whether a single speaker will vary in this
way or whether there will normally be a stable distribution in which case there
would be something aberrant in period three.
239
Chapter 9: A functional analysis of simple referring expressions
240
Chapter 9: A functional analysis of simple referring expressions
Table 9-4: Frequency of THING features across text types (GAE, HES, and MFN)
GAEset HES MFN
total N 1673 1169 552
frequency (%) N frequency (%) N frequency (%) N
OUTSIDER 73.9%• 1103 81.7%• 801 100.0% 466
INTERACTANT 25.8%• 385 17.9%• 176 0.0% 0
SOUGHT 0.3% 4 0.4% 4 0.0% 0
Key for levels of statistical significance:
• 98% degree of confidence
† 95% degree of confidence
‡ 90% degree of confidence
241
Chapter 9: A functional analysis of simple referring expressions
(N=299 for GAE) is calculated based on the total number of expressions found
having the feature INTERACTANT (N=385 for GAE), rather than considering the
rate of occurrence out of the total number of expressions analysed. The
second, called global counting, calculates the frequency as a percentage of
the total number available overall, i.e. all referring expressions analysed
(N=1 673 for GAE). In this case, it is perhaps more reasonable to focus on the
global counting as it considers the frequency of occurrence of performer and
addressee reference from the total of all referring expressions. Local counting
will always result in a sum of 100 per cent for a set of features whereas global
counting will not since the complete list of features is not given.
Regardless of which counting method we use, the differences found
between GAE and HES in terms of the frequency of occurrence of PERFORMER
242
Chapter 9: A functional analysis of simple referring expressions
attempting here to account for why this is so. We simply accept the
differences at this point and note that there are issues here for related types of
research, (register studies for example). However since it seems clear that the
speaker of the GAE corpus does not vary in her use of speaker reference, we
might be able to assume a similar stability for the speaker of the HES corpus.
If this assumption were tested and it held to be true, then the variation found in
Table 9-4 would be due to the differences in speaker. It might further suggest
that ten per cent variation is an indication of a threshold for text type with
respect to reference to speaker role. Clearly this is a figure which would need
to be tested in further research.
For the remaining features coded, the numbers are too small to determine
any degree of difference between the two speakers except for the Performer
Addressee features. However, the occurrence of EXCLUSIVE (‘exclusive we’)
seems to indicate some differences in use. The meaning of exclusive we is an
equivalent to the performer plural with the exclusion of the addressee and the
meaning of BOTH, ‘inclusive we’, is performer plural with the inclusion of the
addressee. If we combine these results, and consider ‘performer plural’, as
shown in Table 9-6, we find that the gap closes considerably.
243
Chapter 9: A functional analysis of simple referring expressions
The final feature for this portion of the coding scheme is OUTSIDER. As
explained in Chapter 8, in order to capture certain features using the coding
software, an initial split was made between those nominal groups coded as
OUTSIDER and not having any other elements in the group other than the head
(HEAD-ONLY) and those having other elements (HEAD-PLUS). In a moment we
will consider each one in detail, examining the features coded for both. First
though we will simply look at the division between these two features. Recall
once again that this division does not represent a division that exists in the
system network for referent thing. This is a classification developed in the
methodology of the analysis. It is important to note that although HEAD-ONLY is
a feature that includes pronouns (she, it, they, etc., it does not include
interactant reference (e.g. I, me, you, etc.), as this category refers only to
outsider reference.
The GAE corpus exhibits a fairly consistent distribution for local counting
as we can see in Table 9-7. However there is a slight amount of variation in
period one in comparison to periods two and three. However periods two and
three are not statistically different when their results are compared. When
global counting is used for the statistical analysis, there is considerable
variation across all three periods to greater or lesser degrees. Local counting
is the more reliable measure for this feature since it gives the distribution for all
outsider referents. The local statistics give a more accurate picture in the
sense that many other variables are held constant and only those expressions
that are known to have the feature OUTSIDER are included.
In Table 9-8, we find the same frequency distribution but this time
comparing the results from GAE with HES and MFN. Here we find that there is
considerable variation among all corpora and no particular patterns emerge. It
is clear that GAE and HES are most similar in this distribution. The difference
between the two is not quite ten per cent, which is a similar degree of variation
to the results found earlier in Table 9-4. The distinction being made concerns
the composition of the nominal group. Expressions coded as HEAD-PLUS
in addition to the head. We will gain a better picture of where the variation is
precisely by looking at the features for each of HEAD-PLUS and HEAD-ONLY in the
next sections.
Table 9-7: Frequency distribution for HEAD-PLUS and HEAD-ONLY for the GAE corpus
period 1 (GAE) period 2 (GAE) period 3 (GAE)
local counting N=257 N=306 N=381
(%) N (%) N (%) N
HEAD-PLUS 42.3%‡ 176 35.9% 110 36.2% 138
‡
HEAD-ONLY 57.7% 240 64.1% 196 63.8% 243
Table 9-8: Comparative frequency distribution for HEAD-PLUS and HEAD-ONLY for GAE, HES
and MFN
GAEset HES MFN
Global
Counting N=1673 N= 1169 N= 552
(%) N (%) N (%) N
HEAD-PLUS 25.3%• 424 33.8%• 395 74.5%• 411
HEAD-ONLY 40.6%• 679 34.7%• 406 10.0%• 55
Local
Counting N=1103 N=801 N=466
HEAD-PLUS 38.4%• 424 49.3%• 395 88.2%• 411
HEAD-ONLY 61.6%• 679 50.7%• 406 11.8%• 55
Key for levels of statistical significance:
• 98% degree of confidence
† 95% degree of confidence
‡ 90% degree of confidence
from the data certain results that lead to a better understanding of the choices
made by speakers.
Figure 9-3 recalls the features which were included in the analysis for
HEAD-ONLY nominal groups. If we compare this to Figure 9-4, which illustrates
the related parts of the system network for referent thing, we can see that the
coding scheme is a simplification in an array that is not meant to represent the
system networks. However, if the goal is to identify and count semantic
options (choices) such as [name of thing], [clock time], [recoverable thing],
etc., then the arbitrary relationship in the coding scheme is not entirely
relevant. Given the complexity of the system networks, mapping the relevant
features into a coding scheme in the software was a great challenge.
Figure 9-3: Coded features for nominal groups with feature HEAD-ONLY
246
Chapter 9: A functional analysis of simple referring expressions
Figure 9-4: Partial view of the system network for referent thing, illustrating semantic
options relevant to the HEAD-ONLY coding feature (seen in Figure 9-3)
The number of occurrences for the features DATE and TIME were so low (N
< 5) that they will not be included in the presentation of results below. Instead
the focus will be on the instances of nominal groups coded as having the
features PRONOUN, LEXICAL-ITEM or NAME. The results for the GAE corpus are
given in Table 9-9 where we can see the cross section of results over the three
periods. Given that the coded features here are grouped only by their single
element nature, the global counting method gives the most representative
picture of their frequency.
Irrespective of the counting method used, period one exhibits a statistically
significant difference in the distribution of HEAD-ONLY coded features. As
shown in Table 9-9, the frequency of pronoun use in period one is significantly
lower than the other two periods by nearly ten per cent. It is clear that the
large majority of these types of nominal groups have the head expounded by a
pronoun. Despite the low numbers of nominal groups having only a lexical
item expounding the head, there does seem to be a higher frequency in period
one as compared to the other two time periods (12.5 per cent as compared to
7.1 and 7.8 per cent respectively). The frequency of nominal groups coded for
NAME is reasonably constant for this speaker regardless of which counting
methodology is used. The slight differences in frequency are not statistically
247
Chapter 9: A functional analysis of simple referring expressions
significant. This result suggests that further research is required which may be
of particular relevance to language generation studies and the problems of
anaphor resolution.
We can now compare the results for GAE with HES and MFN as shown in
Table 9-10. Using local counting, GAE and HES show no statistically different
results, with pronouns representing nearly 60 per cent of the frequency and
NAME being the feature with the next highest frequency at nearly 30 per cent.
MFN however has a completely different distribution; here we find a much
lower frequency of pronoun use (41.8 per cent) and a much higher frequency
of the category LEXICAL-ITEM (36.4 per cent). The relative low frequency for
nominal groups with a name expounding head in the MFN corpus is
unsurprising given the nature of the text. However, it should be recognized that
there are some problems with identifying with certainty what a ‘name’ is in the
case of multi-word strings.
The frequency representation is slightly different when the global counting
method is used. As stated above, there is no particular reason why these
features should be seen to group together, except perhaps in terms of their
realization as a nominal group having only a head element and that this head
is expounded by a single lexical item.
248
Chapter 9: A functional analysis of simple referring expressions
Even though the differences between GAE and HES for pronoun frequency
are statistically significant, it is much less of a difference than that found in
comparison with the MFN corpus. GAE has a relative frequency of 24.3 per
cent while the frequency for HES is close at 20.4 per cent. The frequency for
the same feature in the MFN corpus is very low comparatively at only 4.2 per
cent (although the number of occurrences is very low, N < 30). Since MFN
has no interactant references, this means that the frequency of all pronouns as
a percentage of all nominal groups in the corpus is just over four per cent.
Names were grouped within the [head only] category; however they are
distinguished much earlier in the system network which was discussed in
Chapter 6. Table 9-10 compares the frequency of the use of names as
referring expressions across all three corpora. The probability of selecting the
semantic option of [name of thing] in the system network for thing is 5 per cent
(Fawcett, 2004). The frequency for [name] in the corpora studied here varies
from 1.3 per cent to 12.1 per cent, with the MFN corpus exhibiting the least
frequent use of naming. A more detailed discussion of this result is beyond the
scope of this thesis but clearly more research is needed in the area of naming
(cf. Vandelanotte and Willemse, 2002; Wray, 2002).
Table 9-10: Comparative distribution of head-only coded expressions for GAE, HES and MFN
GAEset HES MFN
(%) N (%) N (%) N
Local Counting
PRONOUN 59.9% 407 58.9% 239 41.8%• 23
LEXICAL-ITEM 9.3% 63 11.6% 47 36.4%• 20
†
NAME 29.9% 203 27.1% 110 12.7% 7
DATE 0.4% 3 1.0% 4 9.1% 5
TIME 0.4% 3 1.2% 5 0.0% 0
ADDRESS 0.0% 0 0.2% 1 0.0% 0
Global
N=1673 N=1169 N=552
Counting
PRONOUN 24.3% 407 20.4% 239 4.2%• 23
LEXICAL-ITEM 3.8% 63 4.0% 47 3.6% 20
†
NAME 12.1% 203 9.4% 110 1.3% 7
Key for levels of statistical significance:
• 98% degree of confidence
† 95% degree of confidence
‡ 90% degree of confidence
249
Chapter 9: A functional analysis of simple referring expressions
It is interesting to note that while the frequencies for PRONOUN and NAME
250
Chapter 9: A functional analysis of simple referring expressions
Local
period 1 (GAE) period 2 (GAE) period 3 (GAE)
Counting
ANAPHORIC 69.3% 88 77.9% 102 69.1 103
EMPTY 7.1% 9 8.4% 11 14.1 21
DEICTIC 17.3% 22 12.2% 16 12.1% 18
GENERAL 6.3% 8 1.5% 2 4.7% 7
Global
N=653 N=492 N=528
Counting
ANAPHORIC 13.5%• 88 20.7% 102 19.5% 103
EMPTY 1.4% 9 2.2% 11 4.0% 21
DEICTIC 3.4% 22 3.3% 16 3.4% 22
GENERAL 1.2% 8 0.4% 2 1.3% 7
Key for levels of statistical significance:
• 98% degree of confidence
† 95% degree of confidence
‡ 90% degree of confidence
Table 9-12: Comparative frequency of types of pronoun for GAE, HES and MFN
Local
GAEset HES MFN
Counting
anaphoric 72.0% 293 71.1% 170 52.2%‡ 12
empty 10.1% 41 10.0% 24 39.1%• 9
deictic 13.8% 56 16.7% 40 8.7% 2
general 4.2% 17 2.1% 5 0.0% 0
Global
N=1673 N=1169 N=552
Counting
anaphoric 17.5% 293 14.5% 170 2.2%† 12
•
empty 2.5% 41 2.1% 24 1.6% 9
deictic 3.3% 56 3.4% 40 0.4% 2
general 1.0% 17 0.4% 5 0.0% 0
Key for levels of statistical significance:
• 98% degree of confidence
† 95% degree of confidence
‡ 90% degree of confidence
251
Chapter 9: A functional analysis of simple referring expressions
It would be very difficult to relate the coded features for the HEAD-PLUS
category to the relevant system network for thing since they involve quite a
number of different areas which are not confined to one particular portion of
the network. If we were to try to identify the main systems involved in what
has been grouped in the coding feature HEAD-PLUS, then the main system
options would be recoverable_thing, cultural-classification_potential, and
thing_exclaimed_at, as shown in Figure 9-5 which is an excerpt from the
overview of the system network for thing seen in Chapter 6.
Figure 9-5: Partial view of the system network for referent thing
Without presenting the network again at this point, let us simply recall that
deeper into each of these systems, we find that what this coding feature
(shown in Figure 9-6) is trying to capture is the instances of any type of
‘addition’ to the cultural classification of a referent. This includes
particularizing, modifying or quantifying the referent.
252
Chapter 9: A functional analysis of simple referring expressions
253
Chapter 9: A functional analysis of simple referring expressions
9.4.1 Determiners
254
Chapter 9: A functional analysis of simple referring expressions
255
Chapter 9: A functional analysis of simple referring expressions
accounted for this degree of detail. However this was not the case as no such
instances were encountered. In fact as we shall see below, the occurrence of
any type of determiner with selector is extremely rare in the corpora analysed
here.
Using the figures gained from the local counting method on the GAE
corpus, we find that period one differs only with respect to period three and it is
not statistically different from the frequency distribution in period two. However
with the global counting method, the degree of confidence is less and it is
again period one that differs but this time only with respect to period two.
What seems to be consistent in period one is a different pattern to those found
in later periods. This is an area that requires further exploration for individual
speaker variation.
It is important to note that since the nominal groups counted here have the
feature HEAD-PLUS, there is at least one other element in addition to the head.
For the cases where there is no determiner (NO-DETERMINER), there must be
some other element. The only cases where there is no determiner and no
other elements in addition to the head are included under the feature HEAD-
256
Chapter 9: A functional analysis of simple referring expressions
One of the main differences between MFN and the other two corpora is the
relative high frequency of nominal groups coded as having the feature [head-
plus]. This is a good example of why it is useful to consider both sets of
statistics; in doing so, we gain a much fuller and more accurate picture of the
data. It is clear that regardless of the text or speaker, if a nominal group has
elements other than head, it is far more probable (frequency probability of
between 0.65 and 0.75) that it will also have the element of determiner as we
can see from Table 9-14. The distribution of determiners within HEAD_PLUS
expressions is the same for GAE and HES but MFN differs significantly (98 per
cent level) from both of the other corpora. The frequency probability covers a
greater range when considered as a frequency of all referring expressions (P =
0.19 to 0.48, with global counting). Although further research is required, there
is evidence to suggest that individual speakers do not vary greatly in the use of
determiner. Furthermore, speakers of the same text type do not seem to differ
greatly.
257
Chapter 9: A functional analysis of simple referring expressions
Table 9-14: Comparative frequency of DETERMINER for GAE, HES and MFN
Local
GAEset HES MFN
Counting
Total N N=424 N=395 N=411
DETERMINER 75.2% 319 70.4% 278 64.5%• 265
24.8% 105 29.6% 117 35.5%• 146
NO-DETERMINER
Global
N=1673 N=1169 N=552
Counting
DETERMINER 19.1%• 319 23.8%• 278 48.0%• 265
NO-DETERMINER 6.3%• 105 10.0%• 117 26.4%• 146
Key for levels of statistical significance:
• 98% degree of confidence
258
Chapter 9: A functional analysis of simple referring expressions
Some very interesting results are discovered when the results for
determination are compared amongst the three main corpora (see Table 9-16).
We must keep in mind that the frequency of genitive determiners is very low in
the MFN corpus and so the statistical analysis for this feature is not reliable.
However there is clearly a trend. There are no statistical differences for the
GAE and HES corpora and the results for the MFN corpus are significantly
different from both GAE and HES when considered as a percentage of all
determiners in this category (i.e. using local counting). However, as a
percentage of all the expressions analysed in each corpus, there are no
differences and the frequency occurrence of genitive determiners seems
stable regardless of speaker or text type. This result is very difficult to explain
as one would expect there to be differences at least in the MFN corpus.
Table 9-16: Comparative results for determiner for GAE, HES and MFN
GAEset HES MFN
Local Counting
Total N 319 278 265
GENITIVE 23.8% 76 20.9% 58 9.8%• 26
PARTICULARIZATION 76.2% 243 79.1% 220 90.2%• 239
Global Counting
Total N 1673 1169 552
GENITIVE 4.5% 76 5.0% 58 4.7% 26
PARTICULARIZATION 14.5%• 243 18.8%• 220 43.3%• 239
Key for levels of statistical significance:
• 98% degree of confidence
‡ 95% degree of confidence
† 90% degree of confidence
9.4.2 Particularization
259
Chapter 9: A functional analysis of simple referring expressions
called the indefinite determiners, we can use the results as we wish. In this
way, it makes it easy to compare with other frameworks which do not make the
same division and at the same time, the results here can be grouped if desired
with the quantifying determiners. In Table 9-17, we find again the results using
both counting methods and what we find is that regardless of counting method,
there are no statistical differences for any period in the GAE corpus. This
suggests an area of realization that is stable for a single speaker. We must
keep in mind that these figures relate only to deictic determiners and that there
are other realizations of particularized referring expressions; for example,
deictic determiners could be arguably classified as having a particularizing
function in the referring expression.
260
Chapter 9: A functional analysis of simple referring expressions
the speaker of the GAE corpus, Table 9-18 shows that the speakers of the
GAE and HES corpora differ significantly and both of these differ significantly
from the MFN corpus. This suggests then that the amount of ‘particularizing’,
at least where deictic determiners are concerned, may be a steady ratio for
one speaker and variable for different speakers and different texts. Reaching
any satisfactory conclusions on this point requires a far more detailed analysis
of particularization in general and the use of particularizing determiners for
individual speakers.
Table 9-18: Comparative results for PARTICULARIZATION for GAE, HES and MFN
GAEset HES MFN
Local Counting
Total N 239 220 239
particularized 59.4% 142 65.5% 144 68.2%• 163
not-
particularized 40.6% 97 34.5% 76 31.8%• 76
Global Counting
Total N 1673 1169 552
particularized 8.5%• 142 12.3%• 144 29.5%• 163
not-
particularized 5.8% 97 6.5% 76 13.8% 76
Key for levels of statistical significance:
• 98% degree of confidence
† 95% degree of confidence
‡ 90% degree of confidence
9.4.3 Selection
The analysis of the cases that fall into this area are perhaps the most
challenging of the nominal group. This is one of the areas where most
linguists struggle to account for the expressions encountered in the data as
they do not fit the notions of head and its relation to the other elements of the
nominal group. The Cardiff Grammar has developed an approach to
determiners called the selection principle which pushes the head element to
the far right of the nominal group in many cases (see also Chapter 7). By
introducing the notion of selection, many of the nominals encountered early in
the nominal group can be viewed as having a determiner function.
As a reminder of the type of determiner we are discussing here, let us
reconsider an example of the partitive determiner from Chapter 5, where the
261
Chapter 9: A functional analysis of simple referring expressions
nominal group, the back of the house (see Figure 9-9) has the following
elements: partitive determiner, selector, deictic determiner and head. The
Cardiff grammar has developed a very detailed account of determiners. The
order of determiners is as follows:
ngp > (&) (rd) (v) (pd) (qd) (v) (sd or od) (v) (dd) (m)* h (q),
Figure 9-9: Diagram of a nominal group illustrating the selector element for the
back of the house
262
Chapter 9: A functional analysis of simple referring expressions
Clearly the number of occurrences of the selector element is too low for
any conclusive statistical analysis. However there is one very interesting
pattern that emerges when the frequency of occurrence of this element is
viewed in its role in the clause. Where there are instances of selection, it is
always in either Complement or Adjunct position. No instances of selection
were found for Subject roles. In fact, a number of interesting results can be
found when distributions are considered as cross sections of the clause.
There are differences in nominal group composition depending on the role it
fills experientially and interpersonally, as we will see in Section 9.5 below.
9.4.4 Quantity
QUANTITY is the last feature which was coded for in relation to the
determiners in the nominal group. The only factor considered here is the
presence or absence of a quantifying determiner. Although quantifying
determiners are more frequent than selection, they have a relatively low
frequency as shown in Table 9-19 and Table 9-20. Table 9-19 presents the
results across the three time periods for the GAE corpus and we can see that
regardless of counting method, there is no statistical difference in the
frequency of these determiners. This suggests that for this single speaker at
least, there may be a stable probability for this type of determiner. In other
words, for a large enough sample, the probability of any given referring
expression including selection by quantification is approximately 0.03 for this
speaker. Further investigation would be needed in order to test whether this is
in fact an accurate probability (if we accept that it is reasonable to convert a
frequency distribution to a probability measure for this data).
If we now compare these results with the results for HES and MFN, the
results in the MFN corpus show that it has the highest occurrence of
quantifying determiners (see Table 9-20). There is nothing surprising about
finding differences in the occurrence of these elements in the nominal group.
In fact, there is no reason to expect different speakers to have the same or
similar frequency of occurrence. It would be incredibly surprising if we found no
variation at all in comparing the texts. In fact, it is quite surprising that there is
no variation in the different periods for the GAE corpus. What we might hope
263
Chapter 9: A functional analysis of simple referring expressions
for is some stability in terms of register so that the results here could contribute
to the probabilities in the computer model of the theory. It would seem from
these results that the quantifier element of the nominal group is a good
candidate for a relatively steady-state distribution. Despite the fact that the
MFN corpus has a greater number of HEAD-PLUS nominal groups, the
differences do not seem as great as one might have expected them to be.
Differences in frequency of 2.3 per cent, comparing GAE and MFN using local
counting, are not very large, even if calculated as statistically significant.
Table 9-19: Distribution of the QUANTITY feature (quantifying determiners) for GAE
period 1 (GAE) period 2 (GAE) period 3 (GAE)
Local Counting
Total N 176 110 138
NO-QUANTITY 88.1% 155 87.3% 96 81.9% 113
QUANTITY 11.9% 21 12.7% 14 18.1% 25
Global Counting
Total N 653 492 528
NO-QUANTITY 23.7% 155 19.5% 96 21.4% 113
QUANTITY 3.2% 21 2.8% 14 4.7% 25
No statistically significant differences
Table 9-20: Comparative results for the QUANTITY feature (quantifying determiners) for GAE,
HES and MFN
GAEset HES MFN
Local Counting
Total N 424 395 411
NO-QUANTITY 85.8% 364 80.8% 319 88.1% 362
QUANTITY 14.2% 60 19.2% 76 11.9%• 49
Global Counting
Total N 1673 1169 552
NO-QUANTITY 21.8% 364 27.3% 319 65.6% 362
QUANTITY 3.6%• 60 6.5%• 76 8.9%• 49
Key for levels of statistical significance:
• 98% degree of confidence
264
Chapter 9: A functional analysis of simple referring expressions
The final area we will consider with respect to the constituency of the
nominal group concerns modification, both pre-head and post-head
modification. As was explained in Chapter 6, the referent of the referring
expression is sub-classified through modification. The features coded for this
part of the grammar are related to the AD_HOC_DESCRIPTION system in the
system network for thing.
Looking at modification as a percentage of all expressions analysed, we
can see as shown in
Table 9-21 that the speaker of the GAE corpus does not vary in any
significant way in terms of her frequency of use of modifiers and qualifiers.
The coded feature no-modifier was included in the counting scheme in order to
count those nominal groups that are not head-only but have no modification at
all, e.g. determiner + head as in the game. There is some degree of variation
concerning the percentage of all nominal groups having no modification in
period one, although this is only significant with respect to period two, no
statistically significant differences were found between periods one and three
or two and three. These results show remarkable consistency in the frequency
of use of both pre- and post- head modification.
265
Chapter 9: A functional analysis of simple referring expressions
Table 9-22: Comparison of frequency of modification for GAE, HES, and MFN
GAEset HES MFN
Global Counting
Total N 1673 1169 552
QUALITY 6.6%• 111 11.4%• 133 40.8%• 225
‡
QUALIFIER 5.3% 88 6.2%‡ 73 31.7%• 175
NO-MODIFIER 14.3%† 240 17.2%† 201 14.9% 82
QUALITY + QUALIFIER 0.9%* 15 1.6%* 19 13.9%* 77
Key for levels of statistical significance:
• 98% degree of confidence
† 95% degree of confidence
‡ 90% degree of confidence
* no statistical analysis available for these figures
This consistency found in Table 9-21 does not hold when comparing
different speakers and text types. This is clear in Table 9-22 where we find
that although the GAE and HES corpora are quite similar in terms of frequency
and use, the differences in use of modifier (QUALITY) is significant to the 98 per
cent degree of confidence and in the use of qualifier, the differences are
significant to the 90 per cent degree of confidence. The MFN corpus shows
more striking differences, although this is unsurprising given the type of text
and the lexical density of the referring expressions in this corpus.
The distinction made in Table 9-21 and Table 9-22 reflects the distinction
made in the system network for THING concerning the selection of an ad hoc
description (see Chapter 6). The probability associated to this feature ([ad hoc
description]) is 50 per cent, which suggests that a referent is equally likely to
be described by an ad hoc description as it to not be. However what was
found in the GAE, HES and MFN corpora respectively was a frequency
distribution of 43.4 per cent (184 of 424), 49.1 per cent (194 of 395) and 80.0
per cent (329 of 411) respectively (figures calculated from Table 9-14 on page
258).
266
Chapter 9: A functional analysis of simple referring expressions
Since in this chapter we are less concerned with explaining the differences
found and more focussed on establishing a description of referring expressions
we will now move on to the results in relation to modification in a
multidimensional view. So far we have simply considered the occurrence of
elements individually. A rather different perspective on the frequency
distribution is obtained if we consider the features coded as variables which we
can control for our own purposes. As stated above, the MODIFIER feature has
been used to code for expressions that include (or not) some kind of
descriptive modification of the referent, having already been identified as
nominal groups having more than one element. Isolating features such as
QUALITY and QUALIFIER, referring respectively to the presence of a modifier or
qualifier, we can gain a better understanding of the relationship that each has
with other elements of the nominal group and with respect to the role of the
nominal group in the clause.
Not all features and results are given in the tables as this would be far too
lengthy and detailed for presentation here. Instead the tables presented
highlight the more significant findings. This type of split analysis allows us to
see whether modifiers, for example, are ‘favoured’ in certain positions such as
Subject or whether certain features do not co-occur. In other words, this will
allow us to determine, at least in an initial investigatory way, whether all
elements in nominal groups are distributed equally or whether, for example,
the presence or absence of modification changes the distribution.
though HES and MFN appear to be more similar in this distribution, having
nearly identical results, there seems to be evidence to suggest that the cluster
frequency of referring expressions in Complement role is relatively stable. 34.0
per cent to 36.5 per cent of all referring expressions will cluster in the
Complement position of the clause, with the majority of expressions filling a
Subject role. This is somewhat surprising since in any given clause there is
the potential for only one Subject but there may in fact be one or more
Complements and/or Adjuncts or none at all.
The effects of the position of referring expressions in the clause will
continue to be explored in the following by considering whether or not there is
a relationship between the presence or absence of modifiers and the type of
element of the clause.
Table 9-23: Frequency distribution of all referring expressions with respect to ROLE, for
GAE, HES, and MFN
9.5.2 Modifiers
268
Chapter 9: A functional analysis of simple referring expressions
Table 9-24: Results for MODIFIER based on ROLE for GAE and HES combined
Feature MODIFIER NO-MODIFIER
(N=243) (N=133)
mean σ mean σ Tstat level of significance
SUBJECT 0.099 0.299 0.195 0.398 2.659 98%
COMPLEMENT 0.601 0.491 0.639 0.482 0.728
ADJUNCT 0.300 0.459 0.165 0.373 2.905 98%
Interpretation
NO-MODIFIER exhibits significantly higher use of the feature: SUBJECT (19.50% vs. 9.90%).
MODIFIER exhibits significantly higher use of the feature: ADJUNCT (30.00% vs. 16.50%).
The use of COMPLEMENT is not significantly different.
The results from the MFN corpus shown in Table 9-25, however, indicate
an inverse trend where we find that modification is most likely to occur in
Subject role. Furthermore, nominal groups in Complement or Adjunct role are
less likely to include any type of modifier. It would seem then that with respect
to referring expressions, ad hoc descriptions (i.e. modifiers and qualifiers),
rather than being seen solely as a property of the expression itself, may in fact
be a property of the goals of the text itself. Consequently, it would seem that
the selection of ad hoc description(s) may be influenced by text variables as
well as by speaker intention.
269
Chapter 9: A functional analysis of simple referring expressions
speakers within the same text type (GAE and HES) and across different text
types (MFN) do vary in their use of modification (see Table 9-22). It is must be
noted that with respect to the differences between speakers within the same
text type, the degree of variation is considerably less than it is for different text
types.
9.5.3 Determiners
Table 9-26: Results for MODIFER based on presence of DETERMINER for GAE and HES
combined
MODIFIER NO MODIFIER
Feature (N=243) (N=133)
Mean σ Mean σ Tstat Level of Significance
DETERMINER 0.69 0.47 0.541 0.50 2.83 98%
NO-DETERMINER 0.31 0.47 0.459 0.50 2.83 98%
Interpretation
The feature MODIFIER exhibits significantly higher use of the feature: DETERMINER
(68.70% vs. 54.10%).
The feature NO MODIFIER exhibits significantly higher use of the feature: NO-
DETERMINER (45.90% vs. 31.30%).
270
Chapter 9: A functional analysis of simple referring expressions
Table 9-27: Results for MODIFER based on presence of DETERMINER for MFN
MODIFIER NO MODIFIER
Feature (N=226) (N=102)
Mean σ Mean σ Tstat Level of Significance
DETERMINER 0.553 0.498 0.686 0.466 2.285 95%
NO-DETERMINER 0.447 0.498 0.314 0.466 2.285 95%
Interpretation
The feature MODIFIER exhibits significantly higher use of the feature: NO-
DETERMINER (44.70% vs. 31.40%).
The feature NO MODIFIER exhibits significantly higher use of the feature:
DETERMINER (68.60% vs. 55.30%).
Table 9-28: Results for MODIFER based on presence of PARTICULARIZED DETERMINER for GAE
and HES combined
MODIFIER NO MODIFIER
Feature (N=243) (N=133)
Level of
Mean σ Mean σ Tstat Significance
PARTICULARIZED 0.259 0.439 0.286 0.453 0.552 -
NOT-
PARTICULARIZED 0.325 0.469 0.195 0.398 2.697 98%
Interpretation
The feature MODIFIERexhibits significantly higher use of the feature: NOT-
PARTICULARIZED (32.50% vs. 19.50%).
The use of NO MODIFIER is not significantly different.
Table 9-29: Results for modifier based on presence of PARTICULARIZED DETERMINER for MFN
MODIFIER NO MODIFIER
Feature (N=226) (N=102)
Mean σ Mean σ Tstat Level of Significance
PARTICULARIZED 0.27 0.445 0.441 0.499 3.106 98%
NOT-
PARTICULARIZED 0.221 0.416 0.186 0.391 0.718 -
Interpretation
The feature NO MODIFIER exhibits significantly higher use of the feature:
PARTICULARIZED (44.10% vs. 27.00%).
The use of MODIFIER is not significantly different.
When all determiners are included, only the GAE-HES combined corpus
shows a direct relationship between the use of modifiers and the use of
determiners. What Table 9-26 is saying is that the nominal groups with the
feature MODIFIER have a statistically significant increase in the use of
DETERMINER (all types including the indefinite article as stated above).
271
Chapter 9: A functional analysis of simple referring expressions
However, the relationship is quite the opposite when only deictic determiners
(particularized) are included in the calculations as is the case in Table 9-28.
Here the nominal groups which include a modifier exhibit a higher use of
determiners that are not coded as particularized; in other words, a nominal
group with a modifier element is less likely to also have a deictic determiner.
The MFN corpus behaves slightly differently in this respect. Table 9-27
shows the opposite set of relations as found in the GAE-HES combined corpus
shown in Table 9-26. In the MFN corpus, the nominal groups with a MODIFIER
element show a higher use of the absence of a DETERMINER and those without
a MODIFIER show a higher use of the DETERMINER feature. Whereas the
relationship between the two nominal group elements was shown to be direct
in the GAE-HES corpus, the presence of a modifier in the MFN corpus
suggests a lower probability of a determiner. Considering only deictic
determiners (coded as PARTICULARIZED DETERMINERS), a similar pattern
emerges with the MFN corpus as was found in the combined GAE-HES
corpus. The results shown in Table 9-29 indicate that nominal groups without
a modifier show an increased frequency of deictic determiners. The presence
of a modifier did not show any statistically significant relationships with deictic
determiners.
This seems to suggest that although clearly not mutually exclusive, there is
evidence that there is a tendency toward an inverse relationship in the
frequency distribution of MODIFIERS and PARTICULARIZED DETERMINERS (i.e.
modifier and deictic determiner).
The final area that will be considered here is the relationship, if any, that
quantifying determiners have with elements of the clause and/or other
elements of the nominal group. The frequency of quantifying determiners is
relatively low in all three corpora in this study. It is clear from Table 9-30 and
Table 9-31 below that the use of a quantifying determiner does not show any
increase in use of Subject, Complement, or Adjunct. In other words, nominal
groups with a quantifying determiner are not more frequent in Subject role than
272
Chapter 9: A functional analysis of simple referring expressions
Table 9-30: Results for QUANTITY based on ROLE for GAE and HES combined
NO QUANTITY
Feature (N=684) QUANTITY (N=78)
Mean σ Mean σ Tstat Level of Significance
SUBJECT 0.192 0.394 0.141 0.35 1.084 -
COMPLEMENT 0.494 0.5 0.513 0.503 0.312 -
ADJUNCT 0.313 0.464 0.346 0.479 0.598 -
Interpretation
There are no significant differences between these sets.
273
Chapter 9: A functional analysis of simple referring expressions
Table 9-32: Results for QUANTITY based on DETERMINER for GAE and HES combined
NO QUANTITY QUANTITY
Feature (N=684) (N=78)
As stated above, there is insufficient space here to give the results for all
combinations of elements within the nominal group. There is certainly
sufficient evidence given here to suggest that further exploration would yield
interesting results.
9.6 Summary
This chapter has been concerned with the results of the study of referring
expressions in three separate corpora. The results highlight the frequency and
distribution of referring expressions within the clause and across three corpora.
The results were used to consider variation and stability of the functional
grammatical structure of these expressions with a three-way comparison:
variation and stability for one speaker in one corpus; variation and stability
between two speakers of similar texts; and variation and stability between
different speakers of very different text types. Furthermore, statistical tests
274
Chapter 9: A functional analysis of simple referring expressions
275
Chapter 10: A functional analysis of complex referring expressions
276
Chapter 10: A functional analysis of complex referring expressions
referring expressions; that is to say, from the perspective of the speaker who
wants to use a linguistic expression to refer to some object. The claim made
here is that this should lead to a better understanding of the meaning involved.
However, from the analytical perspective, the only means available to studying
the artefact of referring expressions is their formal realisation as language.
One consequence is that it may seem at times that there is confusion as to
whether the object of study is the nominal group or the referring expression.
Ideally the use of ‘referring expression’ is preferred in order to reduce the
effects of a purely structural basis for analysis. Yet the only means of
identifying a complex referring expression is by its structural representation.
This kind of double vision will often lead to a melange at times of more
functional or more structural terminology.
Despite wanting to describe complex referring expressions, a focus on
structures is necessary to some extent in order to represent the occurrence of
these expressions. The first description that will be discussed here concerns
the frequency of postmodifying structures or qualifiers. As was explained in
Chapter 8, referring expressions were identified in the texts as those
expressions representing a Participant or Circumstance role in the clause.
Complex referring expressions were identified as those referring expressions
which were realized by a nominal group that included a qualifier element.
The frequency of qualifiers in all three corpora was given in Table 9-22 in
Chapter 9 where the frequency rate for qualifiers is given as a percentage
overall based on two different calculations. This data is reproduced here in
Table 10-1 for convenient reference. The first is the frequency of qualifiers of
all referring expressions in the corpus and the second includes only those
referring expressions realised as nominal groups.
As stated above, all referring expressions were coded as either Participant
or Circumstance. Looking at Table 10-2, it is very clear that complex referring
expressions are far more likely to occur in a Participant Role as compared to a
Circumstance Roles. This distinction is explained further by considering the
distribution of complex referring expressions in terms of the interpersonal role
it fills in the clause.
The frequency of qualifiers is relatively low for the GAE and HES corpora
and considerably higher for the MFN corpus. From the results presented in
277
Chapter 10: A functional analysis of complex referring expressions
Table 10-1: Comparison of frequency of qualifiers for GAE, HES, and MFN
GAEset HES MFN
Referring Expressions
Total N 1673 1169 552
qualifier 5.3% 88 6.2% 73 31.7% 175
Nominal Groups
Total N 1510 981 466
qualifier 5.8% 88 7.4% 73 37.6% 175
278
Chapter 10: A functional analysis of complex referring expressions
speaker to be aware of the nature of the text type (although this does not
suggest that register knowledge is not significant).
279
Chapter 10: A functional analysis of complex referring expressions
Figure 10-1: Ad hoc description system for realizing qualifiers, including probabilities
(Fawcett, 2004)
The probabilities from the system network are based on the selections
already made in the network to that point entering the cultural classification
system (i.e. the probabilities are dependent). It is almost impossible to
compare with the frequencies found in this thesis as it is very difficult to
recreate these statistics without knowing precisely which options were
selected. However by converting the probabilities from the system network
into a percentage as a total of all qualifiers, the probabilities can be more
easily compared with the results from the current data.
A comparison of the results from this study with the probabilities in the
system network are presented in Table 10-4 where the number of occurrences
are presented as a percentage frequency of the total number of nominal
groups matching the features listed in the table. The probabilities given in the
system network for [by role in other situation] cover relative clauses only and
therefore do not include non-finite clauses. The system network as it stands
does not include options for these types of postmodifiers.
The probability given by the system networks for finite postmodifying
clauses seems to fit the data in the corpora. In fact when all three corpora are
combined as is the case in Table 10-4, the probability given in the system
networks matches relatively closely to the frequency of these finite
postmodifying clauses. However, as will be discussed in Section 10.3.1 below,
this does not hold when the corpora are considered separately as there is
considerable variation with respect to these frequencies (see Table 10-12).
Although this will be discussed in the relevant sections below, the remaining
280
Chapter 10: A functional analysis of complex referring expressions
probabilities do not match the results presented here so closely. One area
clearly seems to be overestimated from the probabilities in the system
networks. The probability associated to the option [ad hoc relation with other
thing] is 76.2 per cent. However in the corpora studied here, this type of ad hoc
description represents 67 per cent of all post head ad hoc descriptions.
Modifiers and qualifiers are elements of the nominal group and both
contribute to the description of the referent. However, as they are
distinguished in the grammar there must be fundamental differences
concerning their functions. The position taken here is that the qualifier is the
281
Chapter 10: A functional analysis of complex referring expressions
Table 10-5: Comparison of frequency of modifiers and qualifiers for GAE, HES, and
MFN
GAEset* HES* MFN*
Total N
(nominal groups with 487 441 431
cultural classification as
head)
MODIFIER 23.4% 114 30.2% 133 52.4% 226
QUALIFIER 18.1% 88 18.1% 80 42.0% 181
NO-MODIFIER 62.2% 303 56.2% 248 23.7% 102
1
MODIFIER + QUALIFIER 3.7% 18 4.5% 20 18.1% 78
* no statistical analysis available for these figures
1
the numbers for modifier + qualifier are included in the individual totals for
modifier and qualifier, so should be subtracted from the total
one further result that should be discussed briefly. This concerns the co-
presence of a modifier and a qualifier. Table 10-5 above also indicates the
frequency of a nominal group having both a modifier and a qualifier. Clearly
this is relatively infrequent in all three corpora, although it is especially rare in
the GAE and HES corpora.
These results indicate that it is unlikely that a modifier and qualifier will co-
occur in the same nominal group. This is confirmed by the results presented in
Table 10-6 and Table 10-7 where the effect of the presence or absence of a
modifier is calculated with respect to the GAE and HES combined corpus and
the MFN corpus respectively. In both cases there is an inverse relationship
between modifiers and qualifiers, where the nominal groups with a modifier are
significantly less likely to also have a qualifier. These results also suggest that
if the feature [no modifier] is selected (i.e. a nominal group does not have a
modifier), then there is a higher use of the feature [qualifier]. However this is
not quite accurate. What these results are saying is that if all nominal groups
are split between those that have a modifier and those that do not, then those
that do not have a higher frequency of a qualifier than those that do. This
should follow readily from the frequencies considered in Table 10-5 above
since we know that the co-presence of a modifier and a qualifier is rare.
Therefore it follows that qualifiers are far more likely to occur when a modifier
does not. This is not surprising for the GAE and HES corpora, although it
would not have been obvious for the MFN corpus since the frequency of
modifiers and qualifiers is very close (a difference of approximately 10 per
cent) and the frequency of both is 18.1 per cent. Therefore, statistically at
least, it could have been entirely possible for qualifiers to be more frequent in
nominal groups that have a modifier. This does suggest that in English, and
for these types of texts, there is no sort of mutual attraction between modifiers
and qualifiers. Their use is more often mutually exclusive.
283
Chapter 10: A functional analysis of complex referring expressions
Table 10-6: Results for modifier based on presence of qualifier for GAE and HES
combined
Feature modifier no modifier
(N=243) (N=133) Result
Mean σ Mean σ Tstat Level of Significance
QUALIFIER 0.144 0.352 0.97 0.171 25.456 98%
NO-QUALIFIER 0.856 0.352 0.03 0.171 25.456 98%
Interpretation
The feature [no modifier] exhibits significantly higher use of the feature: qualifier
(97.00% vs. 14.40%).
The feature [modifier] exhibits significantly higher use of the feature: no-qualifier
(85.60% vs. 3.00%).
Table 10-7: Results for modifier based on presence of qualifier for MFN
modifier no modifier
Feature (N=226) (N=102) Result
Mean σ Mean σ Tstat Level of Significance
QUALIFIER 0.341 0.475 0.99 0.099 13.665 98%
NO-QUALIFIER 0.659 0.475 0.01 0.099 13.665 98%
Interpretation
The feature [no modifier] exhibits significantly higher use of the feature: qualifier
(99.00% vs. 34.10%).
The feature [modifier] exhibits significantly higher use of the feature: no-qualifier
(65.90% vs. 1.00%).
284
Chapter 10: A functional analysis of complex referring expressions
When making the same comparison for the three text types under study
here, this notion of a reduced expression is reinforced. As Table 10-9 shows,
there is little to no variation among the three texts in terms of the frequency
distribution of type of ad hoc description. There is a very slight difference in
relative occurrence of the postmodifier type with a frequency of 2.2 per cent for
MFN and 6.7 per cent and 8.0 per cent respectively for GAE and HES. The
numbers in all cases are so low that we should not consider this as statistically
valid. It is quite striking that for the distribution of situation and minor
relationship with thing, there is no statistical difference. In other words, this
distribution seems to be constant and the ‘simplest’ type of complex referring
expression is by far and consistently the most frequent.
285
Chapter 10: A functional analysis of complex referring expressions
postmodifier 6% 5 7% 5 2% 4
10.3.1 Results for nominal groups coded with [by role in other situation]
286
Chapter 10: A functional analysis of complex referring expressions
Table 10-11: Frequency distribution of finite and non-finite clauses as qualifier for all
texts
AD-HOC-SITUATION-TYPE Frequency Overall Local
(N=336) (N=115)
finite 63 18.8% 54.8%
non-finite 52 15.5% 45.2%
287
Chapter 10: A functional analysis of complex referring expressions
288
Chapter 10: A functional analysis of complex referring expressions
There does not appear to be any strong preference for one particular type
of non-finite postmodifying clause when the corpora are combined. The
distribution of non-finite clauses for all three corpora combined is given in
Table 10-13. However, there are so few instances of each type that it is
impossible to predict whether this would remain true for a much larger study.
In de Haan’s (1989:64) study of postmodifying clauses, he found the non-finite
‘-en’ (‘-ed’) past participle clauses to form the largest category of non-finite
postmodifying clauses. However, the same study showed that “finite clauses
constitute by far the largest group within the corpus” (ibid.:62) and this was not
found to be true in all corpora studied here (see Table 10-12 above).
Table 10-14 compares the distribution of non-finite postmodifying clauses
in the combined GAE and HES corpus with the MFN corpus. Clearly the
frequencies are far too low for any conclusions to be drawn but as a
preliminary observation, it would seem that ‘-en’ clauses are considerably
more frequent in the MFN corpus. This seems to be the only noticeable
difference between the two sets. However even this observation seems
premature and a more extensive survey of these types of clauses would have
to be carried out in future research.
Table 10-13: Frequency distribution of finite and non-finite clauses as qualifier for all
texts
Frequency Overall Local
NON-FINITE TYPE (N=336) (N=52)
‘-ing’ participle 19 5.7% 36.5%
infinitive ‘to’ 18 5.4% 34.6%
‘-en’ participle 15 4.5% 28.9%
289
Chapter 10: A functional analysis of complex referring expressions
Despite the fact that non-finite postmodifying clauses are in fact more
frequent in the MFN corpus than finite postmodifying clauses (52.7 per cent
non-finite versus 47.3 per cent finite, see Table 10-14 above), the overall
number of these clauses is relatively low. Of all postmodifying clauses only
38.3 per cent are non-finite in the email corpus as compared to 52.7 per cent
in the model forest network corpus. It is difficult to make any claims about
trends or tendencies. However, it is clear that the dominant process types for
the temporary conceptual situation claimed in this thesis (μ) are material and
relational. There does seem to be one pattern that should be investigated
further in a larger scale study. This is the striking split in the distribution of the
infinitival postmodifying clauses. It is perhaps not surprising that the non-finite
-ing clause is most frequently expressing a material type of process since it is
often associated to the progressive aspect and this aspect is often linked to
material process types. However non-finite ‘-ing’ clauses do not always
express the progressive aspect (de Haan, 1989:63). One could even stretch
this to apply to the past participle postmodifying clauses as they tend to reflect
passive constructions and it does not seem surprising to see an association
between material processes and passive structures. However there is nothing
290
Chapter 10: A functional analysis of complex referring expressions
291
Chapter 10: A functional analysis of complex referring expressions
10.3.2 Results for nominal groups coded with [by relation with other
thing]
The most frequent type of qualifier in the corpora studied here is the one
which describes the referent by a minor relationship with thing (see Table 10-8
above). This category accounts for 61.0 per cent of all nominal groups with a
qualifier element. As previously explained in Chapter 8, in the coding scheme
this category included both prepositional groups and nominal groups
(traditionally noun phrase apposition). However in the system networks, an
earlier distinction is made between these two (cf. [alternative specification of
thing] and [by minor relationship with thing] in Chapter 6). The relevant
network has a relative 76.2 per cent probability associated to the feature [by
minor relationship with thing] whereas it has only a 4.8 per cent probability
associated to the feature [alternative specification of thing]. These calculations
have been determined for our purposes here in order to be able to compare
figures with the statistics given in this study. This simply means ignoring those
coded features that are not considered in the system network such as non-
finite postmodifying clauses and the feature labelled postmodifier (see Section
10.3.3).
It seems clear from Table 10-16 that when all three corpora are combined,
prepositional groups filling the qualifier element of a nominal group constitute
the majority of qualifiers and furthermore within this category, they constitute
the vast majority, with nominal group apposition having a relatively low
frequency in comparison (12.6 per cent). However, the figures obtained here
do not reflect the probabilities predicted by the system networks.
292
Chapter 10: A functional analysis of complex referring expressions
Table 10-16: Frequency distribution of ad hoc description [by relationship with other
thing]
comparative
% % from system
Feature N Local Mean Global Mean
network
The results for the combined GAE and HES corpus illustrate considerable
variation as compared to the MFN corpus with respect to the relative frequency
of postmodifying prepositional groups and nominal groups as shown in Table
10-17. The differences found in the frequency of postmodifying prepositional
groups are statistically significant to the 98 per cent degree. By calculating a
comparative frequency based on the same categories as the calculation for the
probabilities from the system network given above in Table 10-4, we can gain
a more accurate description of how the current results compare to the
probabilities in the system networks. Clearly from Table 10-18 the MFN
corpus is much closer to the frequencies predicted by the system networks.
However the results from the GAE and HES combined corpus suggest that the
frequency of both types of postmodification are overestimated in the case of
prepositional groups and underestimated for nominal group apposition.
Table 10-17: Comparison of GAE+HES and MFN for the feature [by relationship with
other thing]
293
Chapter 10: A functional analysis of complex referring expressions
Table 10-18: Comparative results for postmodifying prepositional groups and nominal
groups
System
Network
GAE+HES MFN
Comparative Comparative
Local % Local %
N % N % %
N=91 N=114
N=127 N=140
pgp 42.2% 68 53.4% 63.4% 111 79.3% 76.2%
np-
14.3% 23 18.1% 1.7% 3 2.1% 4.8%
apposition
The use of prepositions is not equal; there are clear preferences, or rather
certain prepositions are far more frequent than others (see Table 10-19). The
preposition of is by far the most frequent preposition of all, accounting for 37.2
per cent of all prepositions introducing a qualifier. Following this we find in and
for, which have frequencies 15.0 per cent and 13.3 per cent respectively.
Then with is next most frequent with a frequency of 6.7 per cent. Table 10-19
shows the frequencies for the majority of such prepositions, all others having
less than a total of ten occurrences throughout all corpora.
294
Chapter 10: A functional analysis of complex referring expressions
Table 10-20: Most Frequent Prepositions for GAE, HES and MFN
295
Chapter 10: A functional analysis of complex referring expressions
The results presented above can be compared with Biber et al. (1999). In
their study of postmodification by prepositional phrase, they found that the six
most frequent prepositions of postmodifying prepositional phrases accounted
for 90 per cent of all such prepositional phrases (ibid.:635). In this study, the
top six prepositions account for 80.5 per cent. The top three most frequent
prepositions are the same items (of, in and for), however the top six differ
slightly here in that from is among the top six here whereas it is not one of the
top six in Biber et al. (1999). Instead their results show to as one of the six
most frequent prepositions introducing a postmodifying prepositional phrase.
One rather considerable difference in the methodology of calculating these
frequencies is that Biber et al. (1999) included some phrases as postmodifiers
whereas in this study the same phrases would have be treated as quantifying
determiners (and hence no postmodification). For example, the following
296
Chapter 10: A functional analysis of complex referring expressions
10.3.2.2 Apposition
297
Chapter 10: A functional analysis of complex referring expressions
If the results are combined for GAE and HES, as shown in Table 10-18
above, the problem of low number of occurrences is resolved and a more
accurate statistical result can be obtained. With the number of occurrences
now close to 100 in each case, the difference is clear; NP-apposition is not a
frequent type of postmodifying structure. However it is especially rare in the
MFN corpus. Those found in GAE and HES would benefit from further
research as it may be that they may fall into a category that is more of an
incremental nature to referring rather than modifying (cf. ‘turn increments’,
Ford, Fox and Thompson, 2002).
Although the numbers of what has been called ‘postmodifier’ here are
small, as a percentage distribution, they are considerably less frequent in the
MFN text. This may suggest it is a feature of a more casual use of language
(e.g. something small versus something that is small). However it is clear that
these types of postmodifiers have not yet been studied in great detail. Future
research which considers the boundaries between modifiers and qualifiers
without solely structural criteria would need to examine these cases. This is
especially true for the anything wrong and something good examples as in
(130) above. The examples given in (128) and (129) seem quite similar to the
use of non-finite postmodifying clauses which were discussed in Section
10.3.1 above.
298
Chapter 10: A functional analysis of complex referring expressions
Table 10-22: Frequency of the postmodifier type of qualifier for GAE+HES and MFN
GAE + HES MFN
Feature Mean N Mean N
ad hoc situation 37.3% 60 31.4% 55
This section will briefly consider instances where the complex referring
expression includes multiple qualifiers in the nominal group. In these cases
only the first qualifier was included in the functional analysis. The numbers of
multiple qualifiers are relatively low and when only those cases of conjoined
qualifiers are considered, the frequency is even lower. Of all nominal groups
with a qualifier element, only 5.9% have more than one qualifier in the GAE
and HES combined corpus whereas in the MFN corpus only 6.6 per cent
express multiple qualifiers. These figures are presented in Table 10-23.
However in terms of grammatical complexity, multiple embedded qualifiers
were also counted. These were very infrequent in the GAE and HES
combined corpus with only 4 out of a total 168 nominal groups (2.4 per cent)
expressing this type of nominal group complexity. In contrast, this is the most
frequent type of multiple qualifiers in the MFN corpus where 17.7 per cent of all
nominal groups with a qualifier express multiple embedded qualifiers (see
Table 10-23).
Examples of the types of multiple qualifiers are given in (131) to (137)
below.
299
Chapter 10: A functional analysis of complex referring expressions
(132) the creation [of the Waswanipi Cree Model Forest [the
only Aboriginal-led model forest]]
300
Chapter 10: A functional analysis of complex referring expressions
(134) the role [of the Aboriginal community] [in the Model Forest
Network]
(136) two other moms [going just to go] [who again I know fairly
well]
(137) lots of changes [coming over the next little while] [new
people etc.]
301
Chapter 10: A functional analysis of complex referring expressions
302
Chapter 10: A functional analysis of complex referring expressions
Table 10-24: Type of role conflated with head of the nominal group
GAE+HES MFN
ROLE of head element N=161 N=175
participant 96.90% 156 0.46 97.70% 171 0.46
circumstance 3.10% 5 0.46 2.30% 4 0.46
303
Chapter 10: A functional analysis of complex referring expressions
Table 10-25: Distribution of overt situation between material and relational processes
GAE+HES MFN TOTAL
PARTICIPANT
N=161 N=175 N=336
TYPE
material 13.0% 21 16.6% 29 14.9% 50
relational 75.8% 122 79.4% 139 78.0% 261
The results in Table 10-26 show that, despite the relatively low
frequencies, there are no differences in the frequency distribution of actor and
goal-scope when comparing the GAE and HES combined corpus and the MFN
corpus. There seems to be a slight preference for the head element to be
conflated with the role of goal-scope rather than expressing a role of actor.
However the split is reasonably close to 50-50. There were no instances of a
beneficiary role although there easily could have been as in for example the
man who was given £50 (invented example).
305
Chapter 10: A functional analysis of complex referring expressions
Looking at the different roles of the head element for relational types of
situation in Table 10-28, we find that for both corpora, the role of carrier is
vastly preferred over the role of attribute. This makes intuitive sense.
However the use of possessed as an additional role for the head element
seems notably more frequent in the MFN corpus.
Table 10-28: Participant roles related to attributive and possessive types of relational
situations
GAE+HES MFN
Although the frequency for mental and verbal process type situations is
very low, we will just briefly consider the relevant participants that appeared in
the data. The distribution of participants among mental and verbal type
situations is given in Table 10-29. The only comment that can be made with
so few instances is that the results suggest that ‘phenomenon’ is far more
likely than ‘senser’ in mental type situations and there is some indication that
the head element is more likely to be represented as ‘verbiage’ than any other
participant in verbal type situations.
Table 10-29: Frequency distribution of type of role for mental and verbal situations
System Feature Count Global Mean
MENTAL TYPE phenomenon 13 3.9%
To this point, we have only considered cases whether the head element of
the nominal group was deemed to have an additional role as a participant.
However some instances were identified where the head element is
represented as having a circumstance role. These cases are relatively rare as
we saw in Table 10-24 above (3.1 per cent in the email corpus and 2.3 per
cent in the MFN corpus). The difference between these two types of role for
the head element in complex referring expressions is illustrated in (138) and
(139) respectively. Both examples are taken from the HES corpus.
In the cases where the head element has the additional role of
circumstance, in all cases but one the corresponding situation was best
307
Chapter 10: A functional analysis of complex referring expressions
Although circumstance roles conflated with the head element are very
infrequent, further specification was nevertheless conducted in an attempt to
see if any patterns emerged. The results are displayed in Table 10-30 where
location is the most frequent type of role that was found. In order to gain a
better understanding of these types of expressions a much larger sample must
be studied in order to increase the frequency of instances. It remains unknown
whether this would lead to a higher frequency in terms of a percentage of all
complex referring expressions. What this type of role suggests is that the
speaker, in referring to a referent, has decided to refer to it by its location in
space. Given that the corpora analysed in this study constitute an exchange
between interactants who are not sharing the same physical space (i.e. an
asynchronous exchange), this strategy would likely be ineffective or at least
unnecessary. It would be interesting to compare similar results for face to face
casual conversation.
308
Chapter 10: A functional analysis of complex referring expressions
Table 10-30: Frequency distribution of type of circumstance role for the head element
of the nominal group
Feature Count Local Mean Global Mean
57.1%
location 4 1.2%
space 2 28.55% 0.6%
time 2 28.55% 0.6%
manner 1 14.3% 0.3%
cause 0 0.0%
accompaniment 2 28.6% 0.6%
extent 0 0.0%
matter 0 0.0%
We will now consider the description of what has been called covert
relational situations.
such cases, the head element of the nominal group is conflated with a role
similar to carrier. Thus the referent is seen as a carrier for some description.
In structural terms, the qualifiers in these cases tend to be filled by
prepositional phrases.
The covert relational situations constitute 42.6 per cent of all relational type
additional situations in the email corpus and 45.3 per cent in the MFN corpus.
The comparison between the two corpora shown in Table 10-31 represents a
frequency of the referent as carrier in a covert relational situation as a
percentage of all complex referring expressions. There are no statistically
significant differences between the frequencies of head/carrier in each corpus.
The role assigned to the head element of the nominal group is constant in
these cases; in other words the referent is always represented in the referring
expression as a covert carrier. Therefore there is little more to say about its
transitivity relations except for the role contributed by the qualifier which we will
now consider.
310
Chapter 10: A functional analysis of complex referring expressions
the referent) by cause and matter in the MFN corpus. However, as stated
above, the limited number of instances in each corpus makes it difficult to
obtain any firm patterns. Nevertheless there are certainly pointers for future
research in this area.
Table 10-33: Comparison of GAE+HES combined corpus and MFN corpus for roles
contributed by the qualifier in covert relational situations
CV ROLE GAE + HES MFN
description by time 5.6% 9 1.1% 2
description by space 13.0% 21 9.1% 16
description by manner 0.0% 0 1.1% 2
description by cause 5.6% 9 9.1% 16
description by
accompaniment 1.2% 2 3.4% 6
description by matter 6.2% 10 12.0% 21
In Chapter 6 the system network for THING was presented in as much detail
as possible for the purposes of understanding its role in the production of
referring expressions. It is a highly detailed and complex network. The
311
Chapter 10: A functional analysis of complex referring expressions
312
Chapter 10: A functional analysis of complex referring expressions
these expressions has been very challenging. There is one problem however
that seems to be inherent to developing a system network. The system is
meant to represent the semantic options available to speakers in producing
language. However, the system networks are usually developed by applying
an analytical framework that is in fact based on the same system network.
There is here great potential for introducing bias into the research.
The analyst, unlike the functional theoretician, is forced first and foremost
to identify structures; language must be parsed into meaningful units before
any analysis can begin. The criterion used in this thesis to identify a referring
expression was based on the function of the expression in the clause as a
Participant or a Circumstance. However once these expressions or units were
identified, structural criteria were used to determine whether or not the
referring expression referred to a ‘thing’, ‘event’ or ‘description’ (see the
discussion of this methodology in Chapter 8). This type of flip-flop between
function and structure seems unavoidable. However such an approach reflects
the integrated nature of function and structure in language. The main
distinction that must be maintained theoretically at least is the difference
between claims that are relevant to the theorist building the system networks
which model semantic options in language production and those that are
necessary, useful and/or relevant for the analyst.
Based on the analysis presented in this chapter, one might conclude, for
example, that the system networks should be revised so that they can account
for the covert relational situations identified in the corpora. Such a proposal
might lead to a system network like the one illustrated in Figure 10-3 below.
This figure represents a combination of the existing network for ad hoc
description (Fawcett, 2004) and the analytical categories developed here (see
Chapters 7 and 8 and Section 10.4.2 above). Consequently it is a hybrid
representation which mixes the semantic options available to the speaker and
the analytical categories defined by the analyst. The problem with this is that it
is based on the framework of the analyst. It is the analyst who found the
relations to be covert (not clearly identifiable). However, from the speaker’s
perspective, the additional role that he or she assigned to the referent (and
represented by the head element of the nominal group) is always known since
in terms of referring, it is the role that the referent has in the ad hoc situation. It
313
Chapter 10: A functional analysis of complex referring expressions
is not ‘covert’ to the speaker; it is only covert to the analyst. This is a rather
large problem when trying to reconcile the differences in approaches to the
same phenomenon. It would seem then that the option available to the
speaker in these cases is between expressing this relation explicitly or not.
Figure 10-3: Proposed system network for AD HOC DESCRIPTION based on analytical
criteria
An initial proposal for revising the system network in more functional terms
is given in below in Figure 10-4. The main focus developed here is on the
additional role assigned to the referent (and hence to the head of the nominal
group in the resultant nominal group). It would seem important for the system
network to reflect the options available to the speaker in terms of selecting the
way in which this role is realized in relation to the situation created for the
purposes of referring (i.e. μ). Therefore it is reasonable to propose that the
presence of all true qualifiers results in an additional role being assigned to the
referent and that this role is represented by the head element of the nominal
group. In the production of the complex referring expression, the speaker must
feel the need to construct a secondary (ad hoc) situation, μ, for the purposes of
referring to the referent in question. As a result, the main feature needed in an
314
Chapter 10: A functional analysis of complex referring expressions
DESCRIPTION. This is shown in Figure 10-4. As all qualifiers are seen here as
creating an ad hoc situation with the referent, the initial option is whether to
describe the referent by an ad hoc situation or not. This is represented in the
network in Figure 10-4 by the simple system [by ad hoc description by
situation] which illustrates the following two options: [by role in other situation]
OR [not by role in other situation].
The revised system network for AD HOC DESCRIPTION in Figure 10-4 states
that given the selection of [ad hoc description] and [by role in other situation],
two systems are entered in parallel. The first, labelled here as ROLE, identifies
the type of role assigned to the referent in the ‘other situation’ (called μ in this
thesis). In this system two options are available and a selection must be
made between [participant] and [circumstance]. This will determine the role
that is assigned to the referent in the additional situation, μ. As a consequence
of placing this option at this point in the network, it will lead to a preselection of
the required roles and processes when the full system network is re-entered to
produce the qualifier (e.g. clause, prepositional group) (see Chapter 5 and 6).
This may seem overly complicated but it is needed here as it is part of the
options available to the speaker in referring to the referent.
315
Chapter 10: A functional analysis of complex referring expressions
Table 10-34: Examples of terminal features for [expressed situation] and [implied
situation]
expressed situation
[by reduced situation] the person going abroad
a company named Acme
someone to travel with
our supply of clean water
a web-based database of indicators
a little house with a yard
[by fully defined situation] a nice long message i typed to you
the building she wants
other parents that I know and trust
implied situation
[by non-finite situation] more of a desire to work hard and excel
a hard time getting out of the driveway
a chance to have your summer holiday
a valuable tool to monitor stand level and landscape level
change
[by relation with other thing] a key element of the Network’s success
a hotel room downtown
plans for this summer
plans with Bill
increasing concerns about climate change
It may be clear from the examples in Table 10-34 that some structures
appear in more than one category. This is seen as having both positive and
316
Chapter 10: A functional analysis of complex referring expressions
negative implications. On the positive side, it reflects the fact that the
categorisation was made on functional criteria rather than purely structural
ones. On the negative side, it suggests that further specification is needed in
order to complete the system networks and in order to be able to account for
the differences in form. One terminal feature has not changed from the
original network presented in Figure 10-2 and this concerns qualifiers realized
as relative clauses. The only difference now is that the relevant feature is now
labelled as [by fully defined situation]. These types of qualifier are the only
ones expressing fully defined situations, realized as fully expressed clauses.
The feature [by reduced situation] refers to those complex referring
expressions where the additional situation, μ, is expressed but in a reduced
form. In structural terms, this feature will lead to inserting into the nominal
group a qualifier that is realized by either a non-finite clause or a prepositional
group. It should be noted that the selection of [by role in other situation] will
have an insertion rule associated to it that will insert a qualifier element in the
nominal group.
Further research will need to explore whether the so-called ‘missing
elements’ in the reduced situations are due to some kind of ellipsis or whether
an alternative representation would account for these forms more
appropriately. The option of describing the referent by an implied situation is
needed in the network in order to represent the option the speaker has of not
explicitly making reference to the type of situation in which the referent is
involved for the purposes of referring. There seem to be two ways of doing
this. The first is [by non-finite situation]. This differs from [by reduced situation]
because the type of process in which the referent is involved in the additional
situation, μ, is not explicitly stated. The same is true for [by relation with other
thing] where the qualifier is realized by either a nominal group or a
prepositional group. There must be a relation between the referent and the
other thing but the speaker has chosen not to state it explicitly.
There is insufficient space for a discussion of the realization rules that
would produce the formal representation of the systemic options. It is
assumed that many of the existing ones can be recycled since the formulation
here is more of a restructuring of the functional representation rather than a
completely new account. The only new structure added here is the non-finite
317
Chapter 10: A functional analysis of complex referring expressions
10.6 Summary
318
Chapter 11: Conclusion
11. Conclusions
As first stated in Chapter 1, this thesis was not a study of the nominal
group and yet it was. This statement summarises the main paradox
experienced throughout this research. The nominal group is an area of the
grammar which is particularly complex and nominal postmodification is one of
the main unresolved areas. Since almost all studies of postmodification rely
on primarily structural criteria, it was felt in this research that by looking at the
problem in a different way, a new understanding may be gained for some of
the complexities and unresolved issues. Ironically even the most functional
linguists remain focussed on the structural identity of the nominal group (or
noun phrase). Consequently, one of the main goals of this thesis was to
develop a functional approach the nominal group.
There is no single functional term in Systemic Functional Linguistics for
these expressions (cf. Halliday’s (1969) ‘participating entity’). This is very
surprising since in SFL meaning is meant to hold a central place in the theory.
Meaning is represented in the system networks which attempt to model
relations among meaning. As Neale (2002:305) concludes, “a ‘part-of-speech’
analysis of a corpus, which is based on traditional grammatical classifications
in terms of classes at the level of form, is a less useful method for this kind of
model building”. In this thesis, an attempt has been made to shift the focus
from traditional grammatical classifications to more functional classifications.
In a sense then this thesis could be summed up generally as broadening
Systemic Functional Linguistic theory to develop a systemic and functional
approach to the nominal group by foregrounding its meaning as a referring
expression.
The remainder of this chapter is organised as follows. The next section
examines the main contributions of the thesis. Then, in the third section the
limitations of the current research will be discussed. Finally, the last section
will consider some directions for future research.
319
Chapter 11: Conclusions
This thesis has made contributions in three main areas. The first of these
is in the area of descriptive linguistics. It offers a full descriptive account of the
occurrences of referring expressions in three corpora. This description
provides a good account of the frequency distribution of the componence of
the nominal group, offering specifically not only an account of the relationship
among the elements of the nominal group but also an account of how the
frequency of the elements varies according to the position of the referring
expression (nominal group) in the clause (e.g. Subject, Complement, or
Adjunct).
The second area in which this thesis contributes is theoretical linguistics.
This thesis proposes a new theoretical model for analysing complex referring
expressions in Systemic Functional Linguistics. By extending the concept of
transitivity, and by considering the multifunctional nature of referring
expressions, this thesis shows that the meanings expressed in complex
referring expressions can be best accounted for in terms of the situation
created by the presence of a qualifier in the nominal group. The relationship
between the head and qualifier in the nominal group can then be expressed in
a way that is analogous to the transitivity system for the clause.
The third area of contribution is with respect to methodology. This thesis
offers a useful description of the methodology developed for adapting the
features from the system networks to an analytical coding scheme. One of the
main criticisms of other large scale studies of the nominal group (or noun
phrase) is the lack of transparency (cf. Section 2.5). It is often very difficult to
know exactly how the categories were determined and how the statistics were
calculated. The methodology offered here is highly transparent making the
research presented here available for comparison with other studies. More
important than this though is the contribution concerning modelling choice in
the analytical framework. The methodological approach developed here was
largely inductive. It set out to build a model of analysis that maintains, to the
extent possible, the features from the system networks. This will be discussed
again in Section 11.3 below.
320
Chapter 11: Conclusions
All research projects must set boundaries in order for the main body of
work to be completed. Consequently this imposes limitations in terms of the
range of phenomena that can be addressed and in terms of the amount of
information that can be presented as a result.
Concerning the first of these, this thesis was limited to an in depth study of
referring expressions as realized by nominal groups. Other types of referring
expressions were excluded. Furthermore the study was restricted to
considering the nominal group within the clause and consequently text and
discourse level functions were excluded. Future work would need to extend
beyond structural criteria and consider linguistic reference in a much broader
sense.
A second limitation concerning the range of phenomenon addressed
relates to the actual amount of data analysed. Depending on one’s
perspective analysing a total of nearly 14 000 words and 3 000 nominal groups
may either seem like a substantial amount or a limited amount. Since, due to
the nature of the exploration of the expressions under study, all units had to be
coded by hand rather than by automated software, the amount of data was
suitably large when the research was defined. However after completing the
analysis, it is clear that due to the relatively infrequent use of certain functions,
a much larger corpus would be needed to further investigate the phenomena
discovered in this thesis. If the frequency rate of complex expressions can be
assumed to be relatively stable, then the size of the corpus analysed here
would have to be multiplied by 100 at least. This would mean over a million
words and over 300 000 nominal groups. However it must be emphasised that
these figures would constitute a bare minimum.
Another limitation faced in this research concerns the challenges in
developing the methodology. As discussed in Chapters 8 and 10, there is no
obvious or direct association between the semantic options in the system
networks and the analytical framework. The reasons for this are related to the
nature of each aspect of the theory. The system networks approach the
phenomenon in question in a non-linear way. The process of modelling the
production of a referring expression is recursive in nature and therefore the
321
Chapter 11: Conclusions
This thesis has opened up several directions for future research. This
section will highlight some of the main ones.
The first follows from one of the limitations described in the previous
section. Further research should develop in more detail the system network
for ad hoc description. This should be done in two main areas.
The first is to reconsider the distinction between prehead and posthead
modification. This is a purely structural position to take and the system
features should be developed further in the direction of modelling the relevant
meanings. In other words, there may be cases of what is structurally classed
as a postmodifier but is actually functioning like a prehead modifier. If so, the
similarities in meaning should be reflected in the system networks.
322
Chapter 11: Conclusions
323
Chapter 11: Conclusions
This thesis has accomplished what it set out to accomplish within the
confinements of the constraints imposed by the nature of the thesis itself. The
most important future research goal is that this thesis will be used as the basis
for further developing and refining the ideas that it represents.
324
References
References
Aarts, F. (1971) ‘On the distribution of noun phrase types in English clause
structure.’ Lingua 26, 281-293
Abney, S. (1987) The English noun phrase and its sentential aspect.
Unpublished PhD thesis, Massachusetts Institute of Technology (MIT)
Biber, D., Conrad, S., and Reppen, R. (1998) Corpus linguistics: investigating
language structure and use. Cambridge: Cambridge University Press
Biber, D., Conrad, S., Finegan, E., Johansson, S., and Leech, G. (1999)
Longman Grammar of Spoken and Written English. Harlow: Longman
Bloor, T, and Bloor, M. (2004) 2nd edn. The Functional Analysis of English: A
Hallidayan Approach. London: Arnold
325
References
Brennan, S. and Clark, H.H. (1996) ‘Conceptual pacts and lexical choice in
conversation.’ Journal of Experimental Psychology: Learning, Memory
and Cognition 22, 1482-1493
Cleland, A.A., & Pickering, M.J. (2003) ‘The use of lexical and syntactic
information in language production: Evidence from the priming of noun-
phrase structure.’ Journal of Memory and Language 49, 214-230
Dale, R., Geldof, S., and Prost, J-P. (2005) ‘Using natural language generation
in automatic route description.’ Journal of Research and Practice in
Information Technology 37, (1) 89-105
326
References
327
References
Fawcett, R., Tucker, G., and Lin, Y. (1993) ‘How a systemic functional
grammar works; the role of realization in realization.’ In From Planning
to Realization in Natural Language Generation ed. by Horacek, H. and
Zock, M. Pinter Publishers: London: 114-186
Fontaine, L. (2004a) ‘Your money or your life: Safety and Corporate Crime in
the workplace.’ Conference presentation at the British Conference on
Forensic Linguistics/Language and Law, Gregynog, Wales July 2004
Fontaine, L. and Kodratoff, Y. (2002) ‘La notion de ‘concept’ dans les textes
spécialisés: une étude comparative entre la progression thématique et
la texture des concepts.’ In ASp: la revue du G E R A S (Groupe
d’Etude et de Recherche en Anglais de Specialité) 37/38, 59-83
Ford, C., Fox, B. and Thompson, S. (2002) ‘Constituency and the grammar of
turn increments.’ in The Language of Turn and Sequence. ed. by Ford,
C., Fox, B. and Thompson, S. Oxford: Oxford University Press: 14-38
328
References
329
References
330
References
Lin, Y., Fawcett, R., and Davies, B. (1993) ‘GENEDIS: the discourse generator
in COMMUNAL.’ ed. by Sloman, A., Hogg, D., Humphreys, G., Ramsay,
A. and Partridge, D. In Prospects for Artificial Intelligence: Proceedings
of AISB 93 the 9th biennial conference of the Society for the Study of
Artificial Intelligence and the Simulation of Behaviour. Amsterdam: IOS
Press: 148-157
Martin, J. (1997). ‘Du bon usage des corpus dans la recherche sur le discours
spécifique.’ ASp (15/18) 75-84
Moore, A.W. (1993) (ed.) Meaning and Reference. Oxford: Oxford University
Press
331
References
Quirk, R., Greenbaum, S., Leech, S., and Svartvik, J. (1985) A comprehensive
grammar of the English language. London: Longman
Reiter, E., Sripada, S., Hunter, J., Yu, J., and Davy, I. (2005) ‘Choosing words
in computer-generated weather forecasts.’Artificial Intelligence 67, 137-
169
332
References
333
Appendix
334