Professional Documents
Culture Documents
Katz, J. Recent Issues in Semantic Theory
Katz, J. Recent Issues in Semantic Theory
Katz, J. Recent Issues in Semantic Theory
KATZ
I. INTRODUCTION
The present paper has two objectives. In part, it is a reply to Uriel Wein
reich's paper 'Explorations in Semantic Theory' 1, which critically discusses
the conception of a semantic theory first published in 'The Structure of a
Semantic Theory' 2 and subsequently extended in a number of publications,
including, chiefly, my 'Analyticity and Contradiction in Natural Language' 3,
Postal's and my book An Integrated Theory of Linguistic Descriptions4, and
my 'Semantic Theory and the Meaning of "Good"'.5 Weinreich's paper
contains serious misrepresentations of the theory developed in these publi
cations, stemming from misunderstandings about its aims and basic as
sumptions. It also presents what is purported to be a preferable conception
of semantic theory. One objective of this paper is, therefore, to expose his
misrepresentation and to demonstrate that his alternative conception of
semantic theory is not preferable.
However, I have kept this paper from being exclusively critical by utilizing
the opportunity afforded by a discussion of Weinreich's paper to further
elaborate my theory of the semantic component of a linguistic description.
My replies to Weinreich's criticisms are used here as convenient vehicles
both for presenting recent thoughts of mine on semantic theory and for
further clarifying and explaining the stands I take on some major issues in
theoretical semantics. Therefore, the paper's other objective is to make
constructive contributions to the field of semantics.
* This work was supported in part by the Joint Services Electronics Program (Contract
DA36-039-AMC-02300(E)), the National Science Foundation (Grant GP-2495), the
National Institutes of Health (Grant MH-04737-05), the National Aeronautics and Space
Administration, and the U.S. Air Force (ESD Contract AF19(628)-2487); also by a grant
from the National Institutes of Health (MH-05120-04) to Harvard University, Center for
Cognitive Studies.
1 Weinreich (1966), 395-477.
2 Katz and J. A. Fodor (1963), 170-210.
3 Katz (1964b), 519-546.
4 Katz and P. Postal (1964).
5 Katz (1964a), 739-766. And, more recently, a book that Weinreich did not have available
to him when he wrote his paper: Katz (1966).
124
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY*
In this section, I will present a summary of the semantic theory set forth
in my publications cited above. The extent of Weinreich's misrepresentations
makes such a presentation necessary in order that the reader have a state
ment of the theory readily available with which to compare them. Such a
comparison makes it easier to see how Weinreich's misrepresentations, all
taken together, give a false over-all picture of my semantic theory.
My conception of semantic theory was not developed as an isolated system
of principles but as an integrated part of a broader theory about the nature
of language in general. The basic problem of this broader theory - referred
to as linguistic theory6 - is to explicate the abstract form of the knowledge
that makes a speaker competent to communicate in his language. Linguistic
communication is envisaged as a process in which syntactically structured
and acoustically realized objects serve as vehicles for transmitting meaningful
messages from one person to another. The speaker encodes his inner, private
thoughts and ideas in the form of external, publicly observable, acoustic
phenomena, and the hearer decodes such physical phenomena to obtain his
own inner experience of the same thoughts and ideas. The solution that
linguistic theory offers for this basic problem has to be a delineation of the
principles that make it possible for different speakers of the same language
to assign the same meaning to the same acoustic signal. This solution must
take the form of a model which shows how the rules of particular linguistic
descriptions pair a phonetic representation of any appropriate acoustic
signal with its proper semantic interpretation.
According to the model developed by those working on linguistic theory,
the manner in which phonetic representations are paired with semantic
interpretations is specified by the organizational structure that linguistic
theory imposes on linguistic descriptions. A linguistic description, which is
a formal reconstruction of the internalized rules that constitute a speaker's
knowledge of his language, consists of three components. One of these three
is a syntactic component of the sort described by Chomsky in a variety of
publications.7 This component generates an infinite set of abstract formal
objects, called structural descriptions, each of which specifies the full syntactic
description of a single sentence. The other two components are a phono
logical component and a semantic component, as described, respectively in
the work of Halle and Chomsky8 and in the work of Fodor, Postal, and
125
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
126
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
1 But, of course, not only new sentences. We claim that sentences which are not new are
also understood by a speaker because he obtains their meaning as a compositional function
of the meanings of their parts. The emphasis is placed on novel sentences because under
standing their meaning cannot be based on familiarity.
127
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
128
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
129
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
meaning from the words in the latter group in the same respect. Such
generalizations are expressed by including this semantic marker in the lexical
readings for the words in the former group and excluding it from the lexical
readings for those in the latter.
Derived readings also provide statements of semantic regularities. The
expressions 'one of the bachelors on my block', 'an uncle who lives in Detroit',
'my best male friend', 'the bull that left the farm last week', 'the neighborhood
priest', 'her younger brother', 'my sister's spouse', etc., are all semantically
similar in that each has the concept maleness in its meaning, while the ex
pressions 'a cheerful classmate", 'our favorite sister', 'sticks and stones', 'the
truth about city politics', 'the parent who visited the school', 'my watch', etc.,
are not similar in the relevant respect either to the expressions in the previous
list or to each other. Such semantic generalizations are obtained by projection
rules that assign derived readings containing the semantic marker (Male) to
each of the expressions in the former list. They cannot assign a derived reading
containing this semantic marker to any expression in the latter because
(Male) is not in any of the lexical readings for the components of these
expressions.
Thus, one way to think of semantic markers is as the elements in terms of
which semantic generalizations can be made. But another is to think of them
in terms of the role a set of semantic markers plays as the representation of a
sense. In this role, they provide a decomposition of a sense of a constituent,
breaking it down into its component concepts. They thus provide the formal
elements that make it possible for us to reveal the structure of complex
semantic entities: a configuration of semantic markers assigned to a con
stituent as its reading mirrors the structure of the sense of the constituent
because formal elements in the configuration stand for components of that
sense.
The semantic interpretation of a sentence is the object paired with its
phonetic representation in the linguistic reconstruction of the process by
which speakers pair meanings with speech signals. It consists of a set of
semantically interpreted underlying phrase markers together with the set of
statements that follow from them and the definitions of semantic properties
and relations in semantic theory. Thus, to complete the definition of a
semantic interpretation, semantic theory must contain an exhaustive list of
such definitions. In previous publications, we offered the following definitions
among others:
(1) A constituent of a sentence S, including S itself, is semantically anoma
lous just in case the set of readings assigned to it is empty.
(2) A constituent of S, including S itself, is semantically unique just in case
the set of readings assigned to it contains one member.
130
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
131
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
ponent and semantic theory can be submitted to empirical tests that may
either confirm or disconfirm them. If such predictions are falsified, it be
comes necessary to make suitable revisions in either the semantic component
or the relevant definition or both on order to block the false predictions.
132
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
133
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
certain semantic relation to another. Therefore, even in the case of the two
semantic properties that Weinreich concedes to our conception of semantic
theory, semantic anomaly and semantic ambiguity, semantic theory tries to
do more than merely "detect" cases of the former and "determine" the
number of senses in cases of the latter. It also offers explanations of the
anomaly or ambiguity of a sentence on the basis of features of the repre
sentation of its meaning.
It is important to note in this connection that, although we have not as
yet given definitions for all the semantic properties and relations that play a
role in the speaker's semantic competence, the manner in which semantic
theory is formulated does commit us to the thesis that any other semantic
properties and relations about which speakers can make reliable judgments
can be defined in the same manner in which the ones considered so far have
been defined. That is, and this is a crucial point, our semantic theory is
built on the principle that all such aspects of a speaker's semantic competence
can be reconstructed and explained in terms of definitions framed in terms
of formal features of semantically interpreted underlying phrase markers.
Thus, any aspect of the speaker's semantic competence that can be expressed
in his judgments about the semantic structure of sentences can be predicted
from such definitions and the appropriate semantically interpreted under
lying phrase markers. One of the real tests of semantic theory lies in deter
mining the truth of this claim about the definability of semantic properties
and relations.
After first misrepresenting our conception of semantic theory in this way,
Weinreich proceeds to the errors of such a preoccupation with the concept
of polysemy. He writes:
But in assigning this concept so central a place, KF is guilty of two errors. In the
first place, it takes no cognizance of the obvious danger that the differentiation
of submeanings in a dictionary might continue without limit.'8
This is all what Weinreich says about the first alleged error at this point in
his paper. But he returns to it later, and we shall return to this criticism at the
appropriate point in this paper. However, we may make one observation
here. On the view that a semantic component is an explication of a speaker's
semantic competence, which Weinreich shares with us, it is quite impossible
to imagine how such infinite differentiation can be taken as a real danger.
Weinreich is here entertaining either the, possibility that the submeanitigs, or
senses, of a lexical item can in certain instances be infinite in number, or the
possibility that the senses of certain lexical items involve infinitely many
semantic distinctions, or the possibility that there are infinitely many lexical
134
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
In the second place, one would think a scientific approach which distinguishes
between competence (knowledge of a language) and performance (use of a language)
ought to regard the automatic disambiguation of potential ambiguities as a matter
of hearer performance.19
We argued that (1) has no meaningful interpretation on which 'ball' has the
sense of a social activity, even though it has this as one of its senses in the
dictionary, because of the conceptual incongruity of relating a social activity
135
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
20 Katz and J. A. Fodor (1963), the section titled 'Linguistic Description Minus Grammar
Equals Semantics'.
21 Weinreich (1966), 398.
136
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
137
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
138
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
to assure that all entries having that marker, and only those, can be introduced
into the points of a syntactic frame identified by the category symbol SxMi.25
Since this is just to say what we have often said explicitly, it is not very
hard to see why Weinreich thinks it "probably fair" to understand us in
this way. But Weinreich's manner of phrasing this matter makes it appear
as if he is according us a better interpretation than our discussion deserves.
Actually, however, our discussion of the status and role of syntactic markers
is formally much more satisfactory than his rephrasing of them.26
Weinreich next observes, quite correctly, that our formulation of the
syntactic portion of a dictionary entry is given in terms of the type of
syntactic theory presented in such early works of Chomsky's as Syntactic
Structures and 'A Transformational Approach to Syntax'.27 In this type of
theory, subclassification under a major category (i.e. Noun, Verb, etc.) and
cross-classification with respect to subclasses were handled by using re
writing rules typical of what was then called the phrase structure sub
component of the grammar. Since these early works, Chomsky and others
have discovered that rules of this sort are too weak in expressive power to
permit a satisfactory systematization of cross-classificational phenomena at
the syntactic level. Thus, it is certainly true that our previous account of
syntactic markers, like those of other linguists prior to the discovery of the
inadequacy of using phrase structure rules to handle cross-classification,
were based upon an incorrect conception of the nature of the syntactic
markers that represent the syntactic properties and relations involved in
cross-classification under major categories. But since we went along with
others in accepting the older theory in which cross-classificational properties
and relations are treated as if they were subclassificational, we cannot be
accused of being unclear with regard to whether syntactic markers that
139
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
Noun-+Proper-Noun
Noun -+Common-Noun
Proper-Noun -Count-Proper-Noun
Common-Noun- Count-Common-Noun
Proper-Noun-- Mass-Proper-Noun
Common-Noun--Mass-Common-Noun
The trouble with such rules is the following. First, there is no motivation for
the choices that we have to make to establish subclassificational distinctions
as constituting levels in the hierarchy. Second, symbols such as 'Count
Proper-Noun', 'Count-Common-Noun', 'Mass-Common-Noun', etc. are
140
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
formally speaking, as distinct from another as the symbols 'A', 'B', 'C', etc.
This means that, with these symbols, there is no way to express the relations
between subcategories, for the formalism treats the subcategories represented
by 'Count-Proper-Noun' and 'Count-Common-Noun' as bearing no more
relation to one another than the subcategories represented by 'Proper-Noun'
and 'Abstract-Noun'. As a consequence, although it will be possible, for
example, to state a transformational rule that applies only to Proper Nouns,
or one that applies only to Common Nouns, when it comes to a rule that
must apply to Count Nouns or one that must apply to Mass Nouns, it must be
stated in terms of the wholly unrelated subcategories of Proper-Noun or
Common-Noun. Accordingly, we miss a generalization thatwould increase the
simplicity and depth of the analysis. The seriousness of this failure is multiplied
ats we require further rules that apply to cases in terms of subcategories lower
down in the hierarchy. For in these cases, owing the nature of expansion by
rewriting rules, we will further increase the number of such unnecessary
distinctions that must be imported whenever a rule requires a subcategory
distinction. Therefore, if we represent cross-classificational subclassification
hierarchically, we have no motivation for the decisions about domination
that we adopt, and we are unable to formulate transformational rules so that
they make simple and revealing generalizations about the conditions under
which they connect underlying and superficial structures.
Because of this, Chomsky was led to develop a type of rule for representing
cross-classificational subclassification in syntax without imposing hier
archical ordering. This type of rule was already found in phonology29 and
Chomsky modelled his new theory of subclassification and cross-classifi
cation in syntax on phonological rules of this type. Revising his early version
of syntactic theory, he restricted the appearance of re-writing rules to what
he calls the categorical component of the grammar, which handles just the
phrase structure, and he introduced what he calls a lexicon, which is to
handle structure below the phrase level. The lexicon consists of an unordered
set of entries each of which is a set of syntactic features associated with a set
of phonological features that specify the lexical item itself. The syntactic
features are given in a complex symbol such as [+ Noun, + Common,
+Count, ...], where '+' indicates that the lexical item in question has the
feature so marked and '-' indicates that it does not. The categorical com
ponent contains, for each major category A, a rule A-- A, where A is a
dummy symbol for which complex symbols will be substituted according
to a fixed convention. The categorical component and the lexicon constitute
the base component of the syntactic component of a linguistic description.
29 Halle (1959).
141
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
30 The treatment of the feature system given in this paragraph is one of two methods
proposed by Chomsky (1965), 75-127. At the present time, it is not known which of them
is preferable, but for the purposes of our discussion they are not sufficiently different to
warrant a comparison.
31 Weinreich (1966), 401.
32 Weinreich (1966), 401.
142
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
I will not dwell on the second of these charges. I am not sure what it means to
say that examples are inconsistent. If it is a way of saying that our treatment
of syntactic markers is inconsistent, then we have just answered this charge.
As to the first charge that our examples are "surprisingly anecdotal", I
must confess that I am puzzled that such a charge should be made at all.
Perhaps Weinreich means that the syntactic concepts we have used as ex
amples in previous publications are not representative of those available in
syntactic theory at its present state of development. However, his own
examples seem no better off than ours in this respect. Thus, interpreting
Weinreich to be referring to syntactic concepts, if correct, would mean that
he is operating on a double standard. Moreover, it would be an irrelevant
standard. Examples do not have to be representative, since they are not
intended to serve as an indication of the state of our knowledge of syntax.
They are only required to illustrate the point that they are supposed to be
examples of.
Thus, it must be that Weinreich has in mind examples of linguistic con
structions, the examples from English that we have used. In this case his
statement that our examples are "surprisingly anecdotal" criticizes us for
choosing examples that, individually, are not representative of actual speech
or that, collectively, do not comprise a corpus that is nearly extensive enough.
Thus, a set of examples that would clearly avoid this criticism would be one
whose members are literal transcriptions of natural speech and that is itself
in size somewhere on the order of the number of sentences in the books of a
well-stocked public library (or, in the case of words, somewhere on the order
of the number of words in a standard desk dictionary). Here, again, Wein
reich seems to operate on a double standard. But since Weinreich makes this
criticism repeatedly and since it is, of course, possible that both Weinreich
and we could be subject to the same valid criticism, it is important to show
that the criticism is not valid.
Let us take the two interpretations of the criticism separately. First, the
charge that our examples are not literal transcriptions of natural speech
falsely implies that they would be better as examples if they did reflect all the
linguistically irrelevant distortions characteristic of natural speech. But,
since these distortions are due to such things as memory limitations, dis
tractions, shifts of attention and interest, changes in motivation, idio
syncratic and random psychological errors, etc., this charge is based on a
failure to understand that the distinction between a theory of competence
and a theory of performance also involves a distinction between the sort of
evidence relevant to the former type of theory and the sort relevant to the
latter type. Evidence which is appropriate as a basis for judging a theory of
competence reflects only the ideal form of the speaker's linguistic knowledge,
143
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
144
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
associated with (D, C) in the dictionary, of which the lexicon is a part, then
there is no reason for not allowing this set of readings to be carried along
with the substitution so that after all lexical substitutions each lexical item
in the underlying phrase marker will be assigned its proper set of lexical
readings. Thus, the obvious and most natural change to make in the con
ception of the semantic component to bring it in line with this change of
Chomsky's in the syntactic component is to drop the convention (1)34, since
the lexical substitution rule already performs the function that (I) was
designed to perform. The postulation of (I) can be thought of as something
that was necessary to fill a gap that existed in the earlier formulation of the
syntactic component. Its removal can be thought of not only as an increase
in the simplicity of the semantic component but also as the elimination of a
certain heterogeneity in the types of rules proposed in semantic theory.
That is, it was always somewhat unclear just what the status of (I) was.
Was it to be thought of as another type of projection rule different in kind
from the others or some other type of semantic rule? To adopt the former
interpretation complicates the general definition of the notion of a projection
rule because this definition then covers a more heterogeneous collection of
projection rules. To adopt the latter interpretation complicates the general
definition of the notion of a semantic rule because this definition then covers
a more heterogeneous collection of semantic rules.
This matter brings up a related one: the elimination of type 2 projection
rules and the theoretical advantage of this. In An Integrated Theory of
Linguistic Descriptions35, Postal and I suggested the elimination of type 2
projection rules from semantic components based on the elimination of
generalized transformations from the transformational subcomponent of
the syntactic component, and Chomsky36 discovered strong reasons in favor
of eliminating generalized transformations. Postal and I showed that singu
lary transformations do not make a contribution to the meaning of a
sentence, i.e. that whatever contribution they may appear to make must be
regarded as due to some inherent feature of underlying phrase markers.
We also showed that singulary transformations characteristically produce
phrase markers that are too structurally impoverished to be the objects
of semantic interpretation, i.e. when these transformations operate, they
cause the loss of syntactic information that is essential in order for semantic
rules to operate correctly. Thus, we concluded that semantic rules must
operate exclusively on phrase markers that have not been produced by the
application of singulary transformations. The syntactic function of gener
145
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
146
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
type 2 projection rule, since its only function was to insure that the readings
of constituents in a matrix phrase marker are combined by projection rules
with the right readings of phrase markers embedded in it, and this function
is performed by the constituent structure of underlying phrase markers. The
theoretical advantage in eliminating type 2 projection rules is the same as
that obtained by the elimination of (I). Namely, it is possible to make the
definition of 'projection rule' less complicated because the collection of rules
that it has to cover is now less heterogeneous than it was when it included
type 2 projection rules. We can define projection rules as rules that operate
on a partly semantically interpreted underlying phrase marker by combining
readings assigned to grammatically related constituents when the governing
selection restriction is satisfied and assigning the combination as a derived
readingto the constituent that dominates those whose readings were combined.
Weinreich contends that the normal form for a dictionary entry on our
theory "does not discriminate between fortuitous homonymy and lexico
logically interesting polysemy".38 What he means is that we provide the
same normal form for representing "lexicologically interesting cases of
polysemy" such as 'land' meaning country and 'land' meaning real estate
(also: 'cook' meaning one who prepares food for eating and 'cook' meaning
the preparation of food for eating) and for representing cases of "fortuitous
homonomy" such as 'rock' meaning back-and-forth movement and 'rock'
meaning a concreted mass of stony material. His criticism is that we ought
not use the same normal form for both types of case because, without
different normal forms to distinguish them, we cannot exhibit the relations
between lexicologically interesting cases of polysemy.
Weinreich does suggest that we might try to exhibit the similarity among
the senses in cases of lexicologically interesting polysemy, such as the simi
larity between the two senses of 'land' and the similarity between the two senses
of 'cook', by employing branching at the appropriate syntactic marker to
indicate overlapping and divergent portions of their senses.39 But, then,
147
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
Weinreich goes on, the two senses of 'rock', which are not similar, would
have to be handled in the same manner. Thus,
148
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
comes from the lexical reading for the verb 'refuse' in the underlying sentence
structure for
Alternatively, I see no reason not to handle cases that are not syntactically
productive by having a separate dictionary entry for each form when, as in
the case of 'rock', the syntactic markers in their entries in the lexicon already
specify a feature distinction. Thus it will be possible to interpret the diction
ary as providing an explicit statement of the morphemes of the language by
representing each in terms of a unique (D, C) in the lexicon, an explicit state
ment of the senses of each morpheme by representing them in terms of the
set of lexical readings associated with its entry in the lexicon, and an explicit
statement of the similarities and differences between senses, both within a
dictionary entry and across dictionary entries, by representing such relations
in terms of identity and distinctness of semantic markers in lexical readings.
41 Katz and P. Postal (1964), Chapter 4, section 3. Cf. also footnote 39 of this paper.
149
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
Weinreich next claims that our attempt to draw a distinction between syntac
tic and semantic markers does not succeed. But, again, his failure to represent
our semantic theory correctly makes his criticism irrelevant to the real issue
and causes him to overlook the essential distinction between these two sorts
of markers in the theory of language.
Weinreich's first step in formulating this criticism makes it appear as if
semantic and syntactic classification of lexical items proceed according to the
same principles. About semantic classification, he writes,
A general criterion of economy would presumably require that there be as few
markers (primes) as possible; hence, the analyst should aim to add markers only
when failure to do so would result in a failure to mark ambiguities or anomalies of
sentences. The general principle would seem to be that no semantic marker should
appear in the path of any dictionary entry unless it also appears in the selection
restriction of at least one other entry.42
150
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
reading can be motivated by the need to predict any one of this large variety
of semantic properties and relations.
About syntactic classification, Weinreich writes,
The reasons [for increasing the delicacy of subcategorization in syntax] turn out to
be precisely the same as those for semantics: a subcategorization step is taken if
failure to do so would make the grammar generate (a) ill-formed expressions or (b)
ambiguous sentences.43
This is, of course, a serious misrepresentation, too. That the goals of the
syntactic component which depend on subcategorization go far beyond just
the marking of ill-formed expressions and ambiguous sentences is clear from
a consideration of the concept of a transformation. Transformations are
rules for mapping underlying phrase markers describing the deep structure
of sentences into derived phrase markers and ultimately into superficial
phrase markers describing the surface structure of sentences. Each is
stated in terms of a structural condition that determines the set of phrase
markers that fall in the domain of the transformation and a sequence or
formal operations that convert phrase markers to which the transformation
has been applied into a derived phrase marker. The syntactic markers
that express subcategorization features, such as +Count, +Common, etc.,
provide part of the information which determines whether or not a given
phrase marker satisfies the structural condition of a given transformation.
From this it follows that subcategorization features, in part, control the
assignment of derived constituent structure and thus determine the superficial
phrase markers. Consequently, such features must be chosen and justified
on the basis of the full range of predictions about sentences that are made by
their superficial phrase markers. Weinreich would have a case if such pre
dictions were restricted to predictions about ill-formedness and syntactic
ambiguity, but, since, in fact, such predictions go well beyond these two
types, covering every aspect of the segmentation and classification of forma
tives, he has no case whatsoever. We can put the point in another, perhaps
more revealing, way. If the prediction of ill-formedness and syntactic
ambiguity were the only things that had to be determined from superficial
phrase markers, as they are on Weinreich's account, then we could dispense
with such phrase markers entirely. The output of the transformations could
be just integers, 0 in case there is no superficial phrase marker (the sentence
is ill-formed), 1 in case the sentence is syntactically unambiguous, and some
integer n (n> 1) in case the sentence is syntactically unambiguous n-ways.
But, obviously, from the fact that a string of formatives is assigned an integer
151
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
152
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
153
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
case they have the same semantic representation requires that the vocabulary
of markers from which the elements of semantic representations are drawn
not include any markers that serve to syntactically differentiate synonymous
constructions.
Weinreich took our claim that semantic markers provide a means of
representing the conceptual content of constituents to be a statement of a
criterion for deciding whether a marker is syntactic or semantic.44 Our claim
was not intended as such, however. On our theory, the criterion governing
these decisions has to do with predictive success: whether a semantic com
ponent that treats a marker as syntactic by failing to employ it or one that
treats it as semantic by employing it is more successful in predicting semantic
properties and relations of sentences. Our claim that semantic markers
provide a means of representing the conceptual content of constituents was
actually intended to indicate that they represent components of the thoughts
and ideas conveyed by acoustic realizations of grammatical concatenations
of phonetic segments, not formal properties that determine the concatenation
of phonetic segments and the underlying organization of these segments into
grammatically related constituents. Perhaps this point can be appreciated
better if it is explicitly emphasized that subclassification markers in the
lexicon of the syntactic component express properties of one kind of object
whereas semantic markers represent the elements out of which another kind
of object - the meaning of the first - is composed. In this respect, semantic
markers are analogous to features in phonology, for phonological features
too represent elements out of which an interpretation of syntactic objects (in
their case, however, an interpretation in terms of pronunciation, not con
ceptualization) is constructed. Phonological features and semantic markers
provide theoretical vocabularies from which the interpretations, in terms of
sound and meaning respectively, are built. But, in a crucial respect, semantic
markers differ from both phonological and syntactic features: they, unlike
the latter ones, are not features at all. This is another reason why the theo
retical vocabularies of semantic theory and syntactic theory do not overlap.
A feature is a binary, or n-ary, distinction expressed in the form of a single,
unanalyzed symbol. A semantic marker, on the other hand, is a configuration
of symbols with internal structure. A semantic marker represents a concept
that forms part of the senses of linguistic constructions, and so it must have
an internal structure that represents, or is isomorphic to, the structure of the
concept it represents. Thus, semantic markers are, in this respect, more like
the labelled bracketings found in phrase markers, which are configurations of
symbols with an internal structure that represents the internal structure of
154
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
the strings of morphemes they bracket. But, of course, the difference here is
that such bracketings represent the formal arrangement of formal objects,
rather than, as in the case of semantic markers, the structure of concepts.
Weinreich next argues as follows:
155
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
upon the being to which they may refer. In order to clearly recognize that
this is a logical matter, not simply a grammatical one, consider the case of
'ship' which must receive the syntactic feature + Feminine so that sentences
such as
(10) The ship met her doom on the rocks
will be marked as syntactically well-formed. It is entirely obvious that from
156
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
157
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
(15) John (the desk) is in the room but she will be gone soon
(16) Sally (the ship) is in the room but she will be gone soon.
The occurrence of 'she' in (16) but not (15) can refer to the subject of the
first clause as well as someone else.
The two conventional criteria to which Weinreich is referring here are, on the
one hand, what is thought of as meaning (or sense) and, on the other, what is
thought of as reference (or denotation). He goes on to argue that we are
caught in a dilemma. According to Weinreich, distinguishers are not moti
vated by considerations having to do with meaning because they do not
express conceptual components of senses as do semantic markers, and
distinguishers are not motivated by considerations having to do with refer
ence because our semantic theory does not try to handle the problem of how
linguistic constructions can refer to objects, actions, events, etc. in the
world. The mistake in this argument is the falsity of its assumption that a
theory of meaning which does not say everything about the solution to the
problem of reference cannot say anything about it. There is no a priori reason
to suppose that a semantic component is precluded from treating some aspects
158
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
of reference and every empirical reason to suppose that it will treat some, for
it is quite clear that meaning and reference are closely related.
One of the more significant aspects of the distinction between competence
and performance is that the explanation of how linguistic constructions relate
to the world is a matter of formulating a psychological model which accounts
for the manner in which speakers employ their competence to refer to the
furniture of the universe. Theories of competence cannot describe such uses
of language, but they certainly must specify all the purely linguistic in
formation upon which referential uses of language depend. The syntactic
component must represent some of the information presupposed in refer
ential distinctions. For example, the information about pronoun-noun
relations that underlies the ambiguity of
159
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
160
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
The theory of distinguishers is further weakened when we are told (KF, fn. 16) that
"certain semantic relations among lexical items may be expressed in terms of
interrelations between their distinguishers". Although this contradicts the definition
just quoted [ cf. above], one may still suppose that an extension of the system
would specify some special relations which may be defined on distinguishers. But
the conception topples down completely in Katz's own paper (1964b), where
contradictoriness, a relation developed in terms of markers, is found in the sentence
Red is green as a result of the distinguishers! Here the inconsistency has reached
fatal proportions.53
161
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
in deciding that the elements are distinguishers. For example, if the elements
in this turn out to be required elsewhere in some selection restriction, we
have not made the right choice. Determining whether they are required in the
formulation of a selection restriction is a matter of constructing the semantic
component to predict the semantic properties and relations of sentences, as
discussed above.
Hence, it is certainly wrong for Weinreich to say that there is no motivated
way to make decisions about the occurrence of distinguishers in lexical
readings, unless, of course, he accepts nothing as motivating such decisions
that falls short of providing a discovery procedure for them. If he does not
insist on something that strong, I do not see how he can fail to accept
considerations of the sort just sketched as providing appropriate motivation.
If he does insist on a discovery procedure, he will be in the unfortunate
position of rejecting reasonable empirical constraints because they do not
meet impossible demands.
162
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
56Weinreich considers the adoption of "... a more powerful conception of the semantic
interpretation process, in which features of a selection restriction of a word Z would be
transferred into the path of another word W, when it is constructed with Z." He then goes
on to say, "This is the solution adopted by Katz and Postal..., and it is the general solution
which we will elaborate in ?3.3" (Weinreich, 1966, 407).
163
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
tinction between lexical readings and selection restrictions. But, not only is
there nothing here to undermine this distinction, the very concept of the
semantic marker (Selector) actually presupposes it. For (Selector) is defined on
the basis of an operation performed on the semantic marker part of one
reading, on the basis of a selection restriction in another: the first reading is
converted into a new reading for the same constituent by the addition of
semantic markers that are found in the selection restriction of the second.
Without the distinction which Weinreich regards as made untenable by the
concept of the semantic marker (Selector) it would not be possible to know
which of the semantic markers in the second reading are to be included in the
set that replaces (Selector) in the first or where these semantic markers
appear in the first.
164
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
165
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
With such a rule, we economize every lexical reading that reconstructs a sense
166
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
Given that the semantic markers in lexical readings are unordered, Wein
reich levels another criticism. He considers the two sentences:
The paths of (1) cats, (2) chase, and (3) mice - although amalgamated in the order
1 + (2 + 3) in (18i) and as (2 + 1) + 3 in (18ii), would yield the same unordered set
of features {I 2 3}, as the amalgamated paths; for, as we have seen, there is neither
ordering nor bracketing of elements in a KF path. For similar reasons, the theory
is unable to mark the distinction between three cats chased a mouse and a cat
chased three mice, between (bloody + red) + sunset and bloody + red + sunset, and
so on for an infinite number of crucial cases.
For KF, the meaning of a complex expression (such as a phrase or a sentence) is
an unstructured heap of features - just like the meaning of a single word. The
projection rules as formulated in KF destroy the semantic structure and reduce the
words of a sentence to a heap.6'
61 Weinreich (1966),410.
62 Katz and J. A. Fodor (1963), 198-200.
63 Katz (1964a).
167
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
64 The symbol 'V' in this semantic marker represents exclusive disjunction, either (Male)
or (Female) but not both.
65 Katz (1964a).
168
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
has also overlooked, the semantic differences between sentences such as (25)
and (26).
In order to see how a semantic component avoids the difficulty that Wein
reich mistakenly attributes to us, we will consider the way that a semantic
component for English would distinguish the meanings of (25) and (26).
Neglecting lexical readings for 'cat', 'mouse', the plural morpheme, and
other readings for other senses of 'chase', we may start with the reading for
the most familiar sense of 'chase': (((Activity of X) (Nature: (Physical))
((Motion) (Rate: (Fast)) (Character: (Following Y)) (Intention: (Trying to
catch ((Y) ((Motion))), <SR>. We consider this lexical reading in some
detail in order to convey an idea of the considerations involved in formulating
semantic representations of senses.
The semantic marker (Activity) distinguishes 'chase' in the intended sense
from state verbs, such as 'sleep', 'wait', 'suffer', 'believe', etc., and from
process verbs, such as 'grow', 'freeze', 'dress', 'dry', etc., and classifies it
together with other activity verbs, such as 'eat', 'speak', 'walk', 'remember',
etc. (Activity) is qualified as to nature by the semantic marker (Physical).
This indicates that chasing is a physical activity and distinguishes 'chase'
from verbs like 'think' and 'remember' which are appropriately qualified in
their lexical readings to indicate that thinking and remembering are mental
activities. But (Activity) is not further qualified, so that, inter alia, 'chase' can
apply to either a group or individual activity. In this respect, 'chase' contrasts
with 'mob' which is marked (Type: (Group)) - hence, the contradictoriness
of
(27) Mary mobbed the movie star (all by herself)
169
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
does not imply that the motorist is fleeing at all. Finally, note that 'chase' is
not an achievement verb in the sense of applying just to cases where a definite
goal is obtained, since it is not necessary for the person to actually catch the
one he is chasing for him to have chased him, as is indicated by the non
anomalousness of sentences such as
170
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
where 'R1' is the reading for 'cats' and 'R2' is the reading for 'mice'. It is
obvious from this example alone that Weinreich's criticism that a semantic
component designed in terms of our conception of semantic theory would
assign the same semantic representation to sentences which differ in meaning
is totally false. His criticism appears applicable only when a realistic portrayal
of projection rules and semantic markers is obscured by one that omits just
those aspects of projection rules and semantic markers that makes the
criticism inapplicable.67
Before concluding this section, I would like to indicate the theoretical
power of dummy markers such as 'X' and 'Y' which act as categorized
variables for readings of subjects and objects. My account will be brief and is
intended more in the way of a sketch than a finished presentation.
Consider the examples:
67Weinreich (1966, 410) says, quite correctly, that the issue here is "how is the difference
in grammer concretely related to the difference in total meaning?" But, then, goes on to
say, in the very next sentence, "On this KF is silent. What is particularly ironic is that an
enterprise in semantics inspired by the most sophisticated syntactic research ever under
taken should end up with a fundamentally asyntactic theory of meaning." But, even if our
theory does give the wrong predictions by treating cases like (25) and (26) as paraphrases,
as we have been at great pains in the text to show it does not, still, Weinreich's conclusion
that the theory is asyntactic, that it is silent on the issue of the relation of syntactic and
semantic structure in sentences, would be false. For the whole conception of projection
rules, as has been said over and over again, is based on the idea that the combination of
readings, both with respect to which readings are combined and with respect to how they
are combined, is determined, except for selection, by the grammatical relations in the
sentence, and nothing else. On this, Weinreich is silent.
171
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
'Z' is a variable for a reading of the indirect object ('Mary' in (31) and 'John'
in (32)) and 'W' is a variable for a reading that represents the sum of money
which the buyer exchanges for the object bought. In sentences such as
172
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
where 'R[ ]' stands for the reading of the constituent enclosed in the
brackets, 't_n' represents a time point earlier than the speech point (i.e. the
point in time when the utterance of the sentence occurs) and 't-n+ m' rep
resents a time point later than the one represented by the time-designation
t -n
173
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
When one considers the phrase eat bread and eat soup, one realizes that eat has a
slightly different meaning in each phrase: in the latter expression, but not in the
former, it covers the manipulation of a spoon. Continuing the procedure applied
in KF to polysemous items such as ball and colorful, we would have to represent the
dictionary entry for eat by a branching path, perhaps as in (20):
In Section III of the present paper, I pointed out why the possibility of such
infinite polysemy must be dismissed in the construction of a dictionary: the
dictionary is a reconstruction of one aspect of the speaker's semantic compe
tence, and since speakers are only equipped with finite brains, the mechanism
reconstructed by a dictionary can, in principle, store only finitely many bits
of information about the lexical characterization of a particular item. Since
there is obviously no rule that would enumerate each and every one of the
respects in which the "activity symbolized by eat" can be "recognizable
different" from case to case, it follows that Weinreich's problem about infinite
polysemy is spurious.
But why should Weinreich have manufactured such a pseudo-problem? The
only answer that seems to explain his doing so is that he fails to understand
the distinction between meanings of words and the actual things, situations,
events, etc. to which they can refer. For it is obviously true that the various
actions which can correctly be called eating may differ in just the ways that
Weinreich suggests, i.e. they may be performed with spoons, fingers, chop
sticks, knives, shovels, or whatever else strikes our fancy inside or outside
the boundaries of etiquette. But, and this is the fundamental point, they are,
nonetheless, eating in the same sense of the term. That is, they are all equiva
lent actions in so far as 'eating' applies to each with equal appropriateness and
the same meaning. Weinreich has not understood that meaning is an
abstraction away from all the variable features of what can be referred to by
a term, that in the meaning of a word we find represented only those features
of a thing, situation, event, action, etc. by virtue of which it is a thing, situ
ation, event, action etc, of a given type. The reductio ad absurdum of Wein
reich's position is the consequence that, on his position, no word or ex
pression can ever be used with the same meaning in which it is used at any
174
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
other time, since there is always some difference in what is referred to from
one case to the next.
Presumably, Weinreich's failure to distinguish semantically irrelevant
referential features from genuine semantic considerations stems from his
feeling that the distinction between them is hard to draw. Excluding se
mantically irrelevant referential features from having a representation in
lexical readings will, according to Weinreich,
175
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
70 Weinreich (1966), 412. Here Weinreich makes a very serious error. The position taken
by Postal and myself, and by Chomsky (1964), does eliminate certain optional trans
formations, viz. the generalized transformations, and it does change certain singulary
transformations from optional status to obligatory status, but it does not do away with
optional transformations. Certain singularies, those that deal with stylistic variants, such
as the cases 'He looked up the book' and 'He looked the book up' or 'I like only one of the
books you have' and 'Of the books you have, I like only one', still remain optional
transformations.
176
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
177
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
73Whether or not projection rules will be needed that deal with the readings of more than
two constituents depends, of course, on whether there are n-ary (for n > 2) grammatical
relations defined in the theory of syntax. Weinreich (1966, 413) makes it appear as if
everyone knew of such cases, and we simply failed to provide a treatment for them.
Actually, however, there is, up to this time, little basis for thinking that there are
such grammatical relations. If subsequent work in grammar finds some, then it will be
necessary to construct a type 1 projection rule for them, but finding a new case does not
necessarily affect the theory of projection rules, just as finding a new transformation does
not necessarily affect the theory of transformational grammar.
74 Weinreich (1966, 143) says, "Such notions as 'modifier' and 'head' ... have no status
in the theory at all; they beg a question in disguise and are probably undefinable without
reference to semantics." To the first point: if the theory he is referring to is ours, he again
exhibits his failure to understand the notion of a projection rule; if the theory is syntactic
theory, it is just false to say these notions have no status, for they specify the character of
the attribution relation. To the second point: Weinreich does not tell us what question is
supposed to be begged. To the last point: no argument is given as to why Chomsky's view
that grammatical relations such as this one are defined solely in terms of configurations of
syntactic markers in phrase markers, and so without reference to semantics, is incorrect.
75 Cf. Katz and P. Postal (1964), and also Ross (unpublished).
76 Weinreich (1966), 413.
77 Weinreich (1966), 413.
178
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
179
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
The "radical newness" of this approach escapes me, since this compositional
approach to semantics is exactly what I have been urging for the last few
years. The explanation for this seems to me to be the following: Wein
reich first misrepresented our theory almost to the point of unrecognizability
and then, misled by his own misrepresentation, inadvertantly adopted the
main idea of our semantic theory as the basis of his own "radically new
approach".
In what follows below, I will examine the various theses that comprise
Weinreich's proposal for a semantic theory. I will try to show that where he
offers a theoretically tenable thesis, it is taken from our semantic theory in
unmisrepresented form and that where he departs from our theory, his thesis
is not theoretically tenable, or else constitutes merely a terminological variant
of one of the notions in our theory.
Consider first his statement of the general relation between the semantic
and syntactic components in a linguistic description:
... we follow Chomsky on the important principle that the transformational
processing contributes nothing meaningful to a sentence, and that the operations of
the semantic component, leading to the semantic interpretation of a sentence,
should be defined entirely on the deep structure of the sentence.80
180
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
181
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
the absence of any possible justification for his method of writing dictionary
entries, it cannot be considered anything more than a less economical,
notational alternative.
At this point, Weinreich himself introduces ordering of semantic markers
in lexical readings, the very thing that he was formerly so eager to do away
with on grounds of simplicity. However, he does not want every set of sem
antic markers to be ordered and offers a rationale for ordering those that he
thinks must be treated in this way. Thus, he proposes to further characterize
his notion of a dictionary entry by the condition that a dictionary entry can
contain either clusters of semantic features, i.e. just sets of them, or con
figurations of semantic features, i.e. ordered sets of them, or (?) both. By
way of giving a rationale for introducing such ordering, he writes:
Suppose the meaning of daughter is analyzed into the components 'female' and
'offspring'. Anyone who is is a daughter is both female and an offspring; we
represent the features 'female' and 'offspring' as a cluster. But suppose the meaning
of chair is represented in terms of the features 'furniture' and 'sit'. Whatever is a
chair is 'furniture', but it is not sitting: it is 'to be sat on'. We would represent this
fact by saying that the features 'furniture' and 'sitting' form a configuration.84
A lexical reading of this sort, together with the rest of the system of evalu
ation semantic markers enables us to explicate the fact that the semantic
information that a chair is furniture is somehow different from the semantic
information that it is used for sitting. Thus, Weinreich's claim that ordering
is needed among semantic features in order to represent this difference
appropriately is false. Moreover, notice that ordering of the sort Weinreich
proposes is not even a suitable means of representing such differences in type
of semantic information. An ordering of symbols is an arrangement of them
into a sequence in accord with some principle that provides an interpretation
for differences in the positions of the sequence. For example, if we order
transformational rules in the customary fashion, we do so on the principle
that a transformation that is positioned after another can only be applied
182
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
after the other has been applied. But in Weinreich's use of ordering here there
is no principle for positioning symbols that represent semantic features. That
is, there is no linguistic interpretation for the formal arrangement in which a
symbol a precedes another b, which may, in turn, precede another c, and so
forth, beyond the stipulation that a semantic feature so ordered is not to be
taken in the straightforward way that is suggested by the symbol that labels
it. Thus, in the case of 'chair' Weinreich's arrow is simply used as a con
ventional symbol to indicate that the feature (sitting) should really have been
the feature (to be sat on). Why he did not simply use the feature (to be sat on)
without the circumlocution of (sitting) and ordering I cannot say. There is
no systematic import for this circumlocution in which ordering is used
speciously, whereas for the notion of an evaluation semantic marker con
siderable evidence for its systematic import has been presented.86
With respect to these notions of clusters and configuration, Weinreich
defines the notion of linking. The idea underlying this notion of linking is
something Weinreich calls "a basic tenet" of his approach, viz.
that the semantic structures of complex expressions and of simplex expressions are
in principle representable in the same form. ... This principle explains the possibility
of freely introducing new words into a language L by stipulating their meanings
through expressions couched in words of language L.87
86 Katz (1964a).
87 Weinreich (1966), 419.
88 Weinreich (1966), 419.
183
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
184
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
and observes, quite correctly, that (36) is not contradictory even though the
words 'small' and 'big' are antonymous.92 But he goes on, after this obser
vation, to give a wholly unsatisfactory account of the matter by saying that
these words
... are not simply predicated of a single entity, elephant. It seems, therefore, that the
features in a linking cluster (at least for some features and some constructions) are
ordered (with the features listed first being more "emphatic" or foregrounded);
and the associative rule would be suspended, so that ((a, b), c) : (a, (b, c)). Sub
clusters of features can then be shown to display tendencies toward contextual
specialization: the a of (a, b) would not be identical with the a of some (a, c):
littleness in elephants would be different from littleness otherwise.93
185
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERRQLD J. KATZ
which each case is different. What does it mean for an elephant to be little,
for a dog to be little, for a flea, etc.? How are little elephants, dogs, fleas, etc.,
qua little elephants, dogs, fleas, etc., different in size? Moreover, what is
common to littleness in one thing as compared with littleness in another?
It is thus quite apparent that there are overwhelming difficulties in Wein
reich's makeshift treatment of these problems in terms of the notion impure
linking.
It is, however, unnecessary to face these difficulties because, on our con
ception of semantic theory, there is a better way to handle cases like (36), (37),
and others that come up in connection with a semantic analysis of adjectives
such as 'little'. We divide adjectives semantically into two kinds, what I call
relative adjectives, such as 'little', 'big', 'small', 'heavy', 'tall', 'expensive', etc.,
and what I call absolute adjectives, such as 'red', 'carnivorous','spotted',
'sickly', 'scenic', etc. Relative adjectives further divide into size adjectives,
weight adjectives, cost adjectives, height adjectives, and so on. We concentrate
our attention on relative size adjectives, although the general treatment to be
given for them applies mutatis mutandis to the other types of relative adjec
tives. To begin with, we notice that what is said in (36) in its generic sense is
essentially the same as what is said by
skyscraper
man
(39) The flea is big
United States
-tarantula
can be correspondingly rendered as,
skyscraper building
man human
(40) The flea is big for a insect
United States country
tarantula _ -spider
We notice that elephants are compared with animals, skyscrapers with
buildings, a man with humans, a flea with insects, The United States with
countries, a tarantula with spiders, and so on, and that the class of entities
with which something is compared is a category to which that thing belongs.
Further, it is clear that the semantic marker which represents this category
186
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
will appear in the reading for the subject noun. Thus, we can say that relative
size adjectives fix the object to which they are attributed at some position on
a scale of possible sizes whose calibration is determined by the meaning of
the noun they modify. That is, a sentence of the form
[greater] [size]
( less in weight than (an average 1))
Thus, the lexical reading for 'big' and other relative size adjectives will
contain a relative semantic marker, which in the case of 'big' will be (Greater
in size than (an average s)). Y will be that semantic marker which represents
the category of entities with which the subject of the sentence is compared in
size, weight, cost, height, etc. The question is now how to determine Y in the
semantic interpretation of a particular sentence in a natural way.
We already know that I is one of the semantic markers in the reading for
the subject noun phrase of the sentence and I represents one of the semantic
categories to which this subject belongs. However, this is not enough infor
mation to uniquely determine E because the reading of the subject will contain
more than one semantic marker representing a semantic category. But
reflecting on the facts presented in (38) and (40) it seems reasonable to say
that I is in some sense the lowest semantic category in the subject's reading.
That is, it seems clear that no more abstract or general categories than those
indicated in (38) and (40) can be chosen in an equivalent rendering of (36)
and (39). For example, we cannot render 'The skyscraper is big' as 'The sky
scraper is big for a physical object' or in a more extreme case 'The flea is big'
as 'The flea is big for an animal'. Also, if you want to deny the claim that
fleas are big, you cite big insects, not horses or buildings. Hence, we can
contextually define 'V' to be that semantic marker in the reading of the
187
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
subject that represents the lowest semantic category as compared with any
of the other semantic categories represented by semantic markers in this
reading, where the lowest category is determined by the redundancy rules of
the dictionary and definition (ii) of page 166 in Section IX of this paper.
According to this definition of 'X', 'E' becomes that semantic marker in the
reading of the subject that ranks lowest in this hierarchy when the reading
of the subject and relative adjective are combined by a projection rule. Thus,
the relevant differences between the cases of (39) are indicated by the semantic
markers (Building), (Human), (Insect), (Country), and (Spider) and the
common feature of the meaning of these sentences is indicated by the fact
that in their readings these semantic markers occupy the same position, i.e.
each is a value of 'X' in semantic markers of the form (Greater in size than
(an average s)).
As we observed earlier, comparative sentences with a relative adjective
such as (37) involve a common concept for judging the things compared, i.e.
either a common concept of size, or of weight, or cost, or height, or whatever
it is. Moreover, the judgment in these cases is not made in terms of some
broader class of entities an average member of which serves as the standard of
greater, less, or equality. For example, (37) means that the elephant is smaller
on an absolute scale of size than the dog, not that the elephant is smaller
than an average animal and that the dog is bigger than that average animal.
Thus, such judgments are absolute judgments of size, weight, cost, height,
etc. between the objects compared. These facts are easily accomodated in the
above treatment of relative adjectives by the addition of a further condition
to the definition of 's'. The condition is this: in cases of as A as and Aer
comparatives, the component semantic marker (an average 1) in the relative
semantic marker of the reading for A becomes the reading for the noun
phrase subject of the second sentence, i.e. the reading for the subject of the
constituent sentence. Hence, in (37), the reading that replaces (an average Y)
would be that for the noun phrase 'the dog', and so we have as one of the
semantic markers in the reading for the predicate of (37), viz. 'is littler than
the dog', the semantic marker (Smaller in size than (D)), where '(D)' is the
reading for 'the dog'. Superlative sentences of the form 'NP1 is the Aest of
NP2' require a replacement of (an average X) by (Any R), where 'R' is the
reading for NP2. But these are matters of detail.
Returning to the main point, it must now be clear that this way of handling
relative adjectives makes Weinreich's solution both unnecessary and un
desirable. Our solution explicates the semantic structure involved in con
structions with relative adjectives whereas Weinreich's merely indicates
differences in meaning in certain of the constructions in question. Moreover,
ours deals with cases, such as (37), which Weinreich's cannot handle, and it
188
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
189
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
In view of the highly tentative nature of these suggestions, we refrain from offering
a notation for this type of construction.98
Hence, again we can conclude that Weinreich's claim for the necessity of
ordering is not based upon any argument or evidence whatsoever. Without
any substantive analysis, it is certainly not possible to regard cases of
modalization as cases that require ordering of the sort that Weinreich intends
to use to handle other types of non-linking constructions. But here in the case
96 Weinreich (1966), 424. I say on this dubious hypothesis because the hypothesis is
obviously false. The sentence 'John fixes teeth but is not a dentist' is not contradictory, as it
would have to be were the hypothesis true, since it might not be John's profession to
treat, repair and care for people's teeth.
97 Weinreich (1966), 426-428.
98 Weinreich (1966), 428.
190
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
191
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
This conflicts with our claim that a syntactic category such as noun does not
have an inherent semantic interpretation but that particular constituents
that are categorized as nouns in phrase markers have semantic structure
which is derived from the meanings of their component constituents. But
choice in this conflict is painfully clear: Weinreich's principle must be rejected
as empirically false because it implies that the meaning of every noun involves
the concept of thingness or substantiality. The list ot counter-examples is
indefinitely large, including such cases as 'virtue', 'socialism', 'shadow'
'truth', 'zero', 'relations', 'principles', 'itches', 'lap', 'pain', and so on.
It is so obvious that this principle of Weinreich's cannot be supported that
we have given an explicit reply to it only because Weinreich bases a major
feature of his theory on it. Given this principle that major morpheme classes
have inherent semantic features, these semantic features can appear in the
derivation of a sentence prior to the insertion of lexical entries. This Wein
reich allows, and to avoid redundancy, introduces theoretical machinery to
distribute the semantic features associated with major morpheme classes
down to the constituents they dominate in phrase markers. The semantic
calculatorl03, the basic mechanism of Weinreich's theory of the semantic
component, is nothing but a device for distributing semantic features down
a phrase marker. That is, semantic features associated with nodes labeled
with the symbol 'Sentence' or symbols for phrases of one type or another
or symbols for major categories (as Weinreich puts it "non-terminal nodes
of the categorical component") - where the association has been effected by
''semantically" revised phrase structure rules - are distributed down to the
morphemes that are dominated by those nodes in phrase markers.
There are two sorts of arguments given by Weinreich to motivate the
semantic calculator. One, which is based on the principle of inherent semantic
features for major morpheme classes, requires no further comment. The
other is that certain semantic features cannot otherwise be gotten into the
representation of a sentence's meaning at the level of terminal elements in
underlying phrase markers so they must be introduced higher up as features
of the whole sentence, one of its phrases, or one of its major categories.
Weinreich writes:
The introduction of semantic features into non-terminal nodes of the categorical
component also suggests a more elegant treatment of pro-forms, questions, and
imperatives than the one sketched by Katz and Postal (1964).... Katz and Postal
have shown the usefulness of postulating an I(mperative) and a Q(uestion) mor
pheme which function as "sentence adverbials". Since, however, these are elements
which have no segmental representation on any level, it is natural to introduce
192
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
RECENT ISSUES IN SEMANTIC THEORY
193
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
JERROLD J. KATZ
REFEREN CES
194
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms
Recent Issues in Semantic Theory
Author(s): Jerrold J. Katz
Source: Foundations of Language, Vol. 3, No. 2 (May, 1967), pp. 124-194
Published by: Springer
Stable URL: http://www.jstor.org/stable/25000269
Accessed: 06-07-2016 18:02 UTC
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.
Springer is collaborating with JSTOR to digitize, preserve and extend access to Foundations of Language
This content downloaded from 142.103.160.110 on Wed, 06 Jul 2016 18:02:55 UTC
All use subject to http://about.jstor.org/terms