Professional Documents
Culture Documents
PSYCHOLOGICAL REVIEW
LANGUAGE AND THOUGHT:
ASPECTS OF A COGNITIVE THEORY OF SEMANTICS 1
DAVID R. OLSON 2
Ontario Institute for Studies in Education
The argument presented here consists of three parts. First, it is shown by
reference to the role of semantics in a transformational grammar that semantic decisions are not determined by either syntactic or semantic selection
restrictions but by the speaker's knowledge of the intended referent. Second,
it is shown that current theories of word-referent relations are inadequate
to specify the grounds for the above-mentioned semantic decisions. Finally,
by the use of a paradigm case, a theory of reference in terms of a cognitive
theory of semantics is advanced in which it is shown that a semantic decision,
such as the choice of a word, is made so as to differentiate an intended
referent from some perceived or inferred set of alternatives. The implications include an altered conception of meaning, a statement of the relative
information in words and in pictures, a hypothesis about the redundancy of
the postulation of deep structure, and revised conception of the relation
between language and thought.
257
1970 by the American Psychological Association, Inr
258
DAVID R. OLSON
THEORY OF SEMANTICS
provided by Roger Brown.3 Eve said, "I'm
eating my sweater," thereby producing an
anomaly by violating selection restrictions on
eating. Her utterance was, nonetheless, an
appropriate description of her behavior.
McCawley, however, goes on to argue, in
a way somewhat similar to that of Katz
and Fodor (1963), that the selection restrictions that limit the choices of words
that can occur in a sentence are semantic.
Selectional restrictions are definable solely in terms
of properties of semantic representations and . . .
to determine whether a constituent meets or violates a selectional restriction it is necessary to
examine its semantic representation and nothing
else [McCawley, 1968, p. 135].
259
260
DAVID R. OLSON
THEORY OF SEMANTICS
meaning of a word are not themselves given
by the dictionary entry but are generated in
the course of differentiating perceived referents. The meaning of the word horse, as
given by the Oxford dictionary, includes
such things as quadruped, mammal, and
domesticated. However, by considering a
horse in an unusual context, other properties,
not listed in the dictionary, may become distinguishing. Thus, if a person is asked to
differentiate between a whale and a horse,
an appropriate definition would be "The
horse is the terrestrial one."
The feature "terrestrial" may now be
considered to be part of the definition of
horse; note, however, that the marker
"terrestrial" is not given by the dictionary
but rather by the speaker's knowledge of
whales and horses. Anything else the
speaker knows that will permit their differentiation may be used as if it were part of
the meaning of the word. The ones that
have been frequently used have been listed
in a published dictionary but that is a poor
indication of what a speaker knows or how
he uses the language. Quine (1953) pointed
out the limitations of working from the dictionary meaning of a word: "We have so
frequently to content ourselves with a lame
partial synonym plus stage directions [p.
58]."
A fifth, and final, line of evidence for
the role of psychological knowledge in making semantic decisions comes from a consideration of ostensive definitions, the defining of a word by pointing to its referent.
Some words, such as round or black, can
be defined only in this way; in other cases,
it appears to be optional. For example, in
some way pointing to a zebra and saying,
"That is a zebra" is equivalent to the statement, "A zebra is an animal like a horse
with stripes.'' In the case of the ostensive
definition, it is clear that the information,
which in another context would be called a
semantic marker (having stripes), is in fact
perceptual information. Hence, the perception of a referent in terms of distinguishing
features provides the meaning for the word,
at least in this case. The inference that
perceptual knowledge is the basis of seman-
261
can be accounted for by the surface syntactic structure. Hence, we are led to hypothesize that the base structure of language has been postulated unnecessarily,
presumably because of an inadequate account
of cognition, the knowledge of referents.
This suggestion has been made independently by McCawley (1968) but for similar
reasons. We shall return to this point.
The suggestion that the speaker's knowledge of the intended referent provides the
basis for semantic decisions is unsatisfactory
in at least two ways. The first is illustrated
in the earlier discussion of pronominalization.
In some cases where two noun phrases have
the same intended referent such as, "Batman
was told that Bruce Wayne would be killed,"
"Mr. Trudeau was told that the Prime
Minister would be invited," the substitution
of a pronoun for the second noun phrase may
not be appropriate. Hence, it is not simply
a matter of identity of intended referents;
one referent may have many names. Second,
262
DAVID R. OLSON
REFERENCE (Meaning)
SIGNS
REFERENTS
(Words, Perceptual
Codes, Etc.)
THEORY OF SEMANTICS
cat ... " rather than "The white cat ... "
or "The animal . . . ." Similarly, they argue
that a statement is a symbolization of a
reference (p. 82) and reference may more
or less approximate the structure of the
referents; that is, ideas may correspond more
or less to the world. However, any one
reference may be symbolized in more than
one way and Ogden and Richards give little
help in specifying how one chooses one way
rather than another. They do, however, indicate that the objective in communication
is to arrive at the same referent. This criterion would be used to assess the adequacy
of a statement and the equality or synonymy
of a paraphrase.
The relativity of language was indicated
by Ogden and Richards' Fourth Canon"a
symbol refers to what it is actually used to
refer to [p. 193]"and in their theory of
definition"all definitions are essentially ad
hoc [p. 111]." Words are relevant to some
purpose or situation, and consequently are
applicable only over a restricted field or
"universe of discourse [p. 111]." This
point was greatly expanded by Wittgenstein's (1958) construal of language as a
language game, a game played within certain
rules permitting and forbidding certain
moves, which is played in some context for
some purpose.
The implication of this assumption is that
one can read things into language unnecessarily such as "meaning" or "thinking" or
"asserting." Rather, Wittgenstein insists
that language is a tool that may be used for
certain purposes and these uses vary for
each language game. Thus, it becomes impossible to give "the meaning" of a word
that holds in all contexts because these contexts are continuously changing; the meaning is inconsistent outside the framework of
the particular game involved. Thus, "the
meaning of a word is its use in the language
[p. 20]." "And the meaning of a name
is sometimes explained by pointing to its
bearer [p. 21]." Again, "nothing has so
far been done when a thing has been named.
It has not even got a name except in the
language game [p. 24]." We draw boundaries between events and name them only
263
264
DAVID R, OLSON
Event
Case I
Case Z
Case 3
Case 4
o
o
o
o
Alternative
Utterance
D0B
.. the round ,
white one
...(look under )
the round, white,
wooden block
that is about one
inch across ,..
Word-Referent Relationships
1. Words do not name things. In the
paradigm case the thing is a block, yet few
if any of the speakers so designate it.
2. Words do not name intended referents.
In the paradigm case, the gold star is under
the same block, the same intended referent,
yet the utterances differ in each case. The
conclusion is that words designate, signal,
or specify an intended referent relative to
the set of alternatives from which it must
be differentiated. In the language of information theory, we would say that statements reduce alternatives or uncertainty.
The term "information" shall be used for
any perceptual or linguistic cue that reduces
the number of alternatives to the intended
referent. The meaning of an utterance is,
therefore, the information provided by the
utterance to a listener. Thus, the meaning
of an utterance is dependent on the context
of alternatives. In some cases, a single
phoneme could have meaning if it permits
the choice between alternatives (Lyons,
1968). It is now possible to revise the conception of an ostensive definition. An ostensive definition is given, not simply by pointing to a referent, but by indicating the
referent relative to a set of alternatives.
3. An utterance does not exhaust the potential features of the perceived referent.
Thus, in the paradigm case, none of the
speakers say wooden, or physical, object;
they specify the object to the level required
by the listener to differentiate the intended
THEORY OF SEMANTICS
referent from the alternatives. If the set of
alternatives were changed, the utterance
would be changed, perhaps to include those
adjectives. In general, it would seem to be
the case that an utterance never exhausts all
the possibilities of the perceived referent.
One would find new properties of any event
as long as the event was put among different
sets of alternatives. Recall the earlier illustration of the horse contrasted with the
whale in which case a defining character of
the horse becomes "terrestrial," a feature
that has not up to this time been included
in the dictionary. Since not all the potential
information of the referent as perceived is
signaled by a word, we conclude that words
(or utterances) neither symbolize, stand for,
nor represent referents, objects, or events.
They serve rather to differentiate some perceived event from some set of alternatives.
How this partitioning is accomplished would
require extensive analysis, but it may be
suggested that it is accomplished on the
basis of some distinguishing feature or features. As a general point, it may be argued
that the view that an utterance partitions a
set of alternatives is equivalent to the view
that utterances specify different aspects or
distinguishing features of the perceived
referent. The advantage of this translation
in this case is that these features are presumably what in other contexts would be
called semantic markers. The limitation of
the translation is that if perceptual distinguishing features may be taken as a model,
then that set of features is not exhaustive,
with new features becoming distinguished
as the perceived event is set among a new
context of alternatives.
4. Having defined meaning as information that permits a choice or differentiation
among alternatives, it becomes possible to
state the advantages of words over the direct
perception of events. A word or an utterance, since it not only specifies the perceived
referent but also the set of excluded alternatives, contains more information than the
simple perception of the event itself. The
exception to this is the case in which the perception of the event occurs in the context
of all of the alternatives from which it is
265
266
DAVID R. OLSON
THEORY OF SEMANTICS
also determine whether some pronouns may
be deleted or must be retained. In the
present paradigm case, ". . . the black one,"
one was retained as a place holder in the
surface structure even if it contained no information. In other languages with a different surface structure, such as French, for
example, the utterance would simply be
"C'est sous le blanc."
Another example comes from Whorf
(1956). In describing a flash of light
We are obliged to say "it flashed" or "a light
flashed," setting up an actor it, or a light, to perform what we call an action, flash. But the flashing and the light are the same; there is no thing
which does something, and no doing, Hopi says
only rephi [p. 262-263],
267
268
DAVID R. OLSON
269
THEORY OF SEMANTICS
sense data gathered from the referent. Second, even the "simple" perceptual process is
not immediate but rather involves a large
component of expectancies and inferred alternatives (Bruner, 1957; Garner, 1966;
Sokolov, 1969). That is, even perception
involves the nonpresent. Third, already
considered have been some cases in which
the intended referent was neither present
nor obvious, such as "Prime Minister";
other cases, such as walk/run, walk/drive,
mind/body, shall be considered presently.
It also remains to be seen whether or not
'the model presented here would actually
help in the solution of such a problem as the
meaning of the word "good," a problem
that has intrigued philosophers for so long
(Katz, 1964). It is an appropriate inference from the model that perceptual information, cognition, underlies semantic decisions, and that perceptual information is
gathered by experience in the referential
world.
Syntactic Considerations
First, some of the analyses have proceeded
on the basis of a speaker's choice of a single
word. What then is the syntactic status of a
single word? It is hypothesized that the
appropriate level of analysis is a message
a message being construed as any utterance
that specifies the event relative to the set
of alternatives. A single word is then an
utterance in which all the other sentence
constituents are assumed or agreed on and
hence have been replaced by pronouns or
simply deleted. An example of this is
Wittgenstein's language game in which the
builder simply calls out "Block," instead of
"Please give me a block, not a pillar, slab,
or beam." In other words, the present
analysis of a single word should be construed as the analysis of message. In other
cases, a single phoneme, if it partitions
alternatives to the intended referent, may be
considered as a message (Lyons, 1968).
Second, it is important to show that the
alternatives from which the intended referent
is to be differentiated determines semantic
decisions not only for the adjectives relied
on in the paradigm case, but also for other
NP + VP
/
NP + VP
NP + VP.
270
DAVID R. OLSON
271
THEORY OF SEMANTICS
a word to differentiate an event from a set
of alternatives; the meaning of the word is
its use in partitioning a set of alternatives.
On the other hand, Garner (1966) has
shown that an event is perceived in terms of
the set of alternatives from which it is
differentiated. Moreover, as was suggested
above, it is plausible that these alternatives
have their effect by determining which set
of features will be distinguishing (Gibson,
1965; Kinsbourne & Hartley, 1969; Smith,
in press). Both of these cognitive systems,
language and perception, then appear to involve search for features which will distinguish an event from the perceived or inferred alternatives. However, can it be
assumed that the features which distinguish
perceptual alternatives are those that are
used in the language? Certainly not all
features are named. Moreover, we have
considered how the same intended referent
would be encoded in language differently
depending on the contrasting set of alternatives, producing, for example, in one case
"cow" and in another "animal." But does
this imply that the speaker perceived the intended referent differently in the two contexts? Further, both Huttenlocher (1968)
and Osgood (1969) have provided interesting evidence that various grammatical and
semantic effects including transformations
are influenced by the structure of the perceptual situations. I believe it would be
wrong, however, to conclude that the structure of perception and the structure of langUtterance
Cose l
This is a square
Case 2
This is a square
Case 3
This is a square
Case 4
This is a square
Event
a
a
a
a
uage are identical. For one thing, the structure of perception underlies performance in
various media, including drawing and mathematics, not just ordinary language. I have
suggested elsewhere (Olson, 1970) that it
is performatory attempts in various media
which provide the occasion for selecting
further information from the perceptual or
referential world. Important conceptual and
empirical work remains to be done in this
area.
A second important problem is the development of word meaning. An indication of
how this problem could be attacked can be
given by an example. Consider the meaning
of the word square, in the contexts shown
in Figure 3.
For Case 1, the utterance is ambiguous in
that the perception of the square is ambiguous. What coding or what distinguishing features of the object are important in
its being called or recognized as a square?
The closed figure? The four-sided figure?
The ink-drawn figure?
The perception of the square and the
meaning of the word square become more
differentiated, or more informative, when
the square is perceived in the context of
some contrasting alternative, as in Case 2.
The contrasting alternative indicates the distinguishing features in terms of which the
object is to be perceived. The meaning of
the word is now the percept of the square
in terms of that alternative, in this case,
perhaps, "four-sided." A different alternaAlternative
A
O
OAO
Meaning
Ambiquous
4 -sided
Straight -edged
Straight -edged
4 -sided
Symmetric
272
DAVID R. OLSON
tive would have produced a different perception and a different meaning, such as
"straight-edged" in Case 3. The perceptual
knowledge of squares and the meaning of
the word "square" develops as the set of
features increases, thus permitting the differentiating of squares from increasingly
large sets of alternatives, as in Case 4.
Learning a concept through the verbal or
semantic system is simply the converse of
this process. From the sentence, "A square
is a four-sided figure," the child obtains
information both about the square and about
the figures from which it is to be differentiated. Learning concepts ostensively and
learning them verbally are perfectly symmetrical processes. Language has the practical advantage that one need not have immediately available all of the physical alternatives that would be necessary to permit
the child to arrive at the critical distinguishing features.
In general, the problem of the differentiation of word meaning is the same problem
that is sometimes called conceptual development and it warrants much more extensive
and critical study from this point of view.
The third issue, and the one of most concern to the writer, is the implication of this
view of language for the question of the
relationship between language and thought.
A tentative outline of part of the argument
may be given.
First, notice that language is merely the
specification of an intended referent relative
to a set of alternatives. All of that information is perceptual and it was available to
the speaker before he generated the utterance. Therefore, to the speaker there is no
information in an utterance. Speaking is
redundant. Hence, language does not restructure thought.6
Second, however, notice that to the listener
none of these points hold. The listener does
not know what to look at (or to think
about) ; listening to an utterance provides
information both about the intended referent
and the alternatives from which it must be
6
differentiated. There is, therefore, considerable information to the listener in an utterance; the utterance restructures his perception. That is, language influences his
thought. The means by which language influences his thought may be called instruction.
The failure to notice this asymmetry between the speaker and listener in the influence of language on thought has led to
considerable confusion. The advantageous
effects of language were attributed, inappropriately, to the speaker. To account for
this advantageous effect, new levels of cognition were postulated such as symbolic
representation (Ausubel, 1968; Bruner,
1966; Vygotsky, 1962) and verbal mediation (Kendler & Kendler, 1962). From the
point of view developed in this article, such
postulations were unnecessary; rather a
theory of instruction would be required.
But a similar point may be made about
the postulation of deep structure in the language. If the functions usually attributed to
the deep structure, such as the semantic
component and the effects of syntactic selection restrictions, can be accounted for
without implicating the deep linguistic structure, there remains little virtue in postulating such a level. McCawley7 (1968) has
in fact made such a proposal on the basis
of semantic considerations. He suggests
that a grammar may consist of a "formationrule component" which specifies the semantic representations or intended referents and
a "transformational component" which relates these semantic representations to the
surface structure.
The effects of deleting the higher levels
of cognition and the deeper levels of language would simplify considerably the problem of the relation between language and
thought.
7
McCawley gives credit for the idea of abandoning the concept of deep structure to G. Lakoff and
J. R. Ross (rnimeo entitled, "Is deep structure
necessary?" Cambridge, Mass., 1967).
REFERENCES
AUSTIN, J. L. How to do things with words.
New York: Oxford University Press, 1962.
AUSUBEL, D. P. Educational psychology: A cog-
THEORY OF SEMANTICS
nitive view. New York: Holt, Rinehart &
Winston, 1968.
BERNSTEIN, B. Elaborated and restricted codes:
Their social origins and some consequences.
American Anthropologist, 1964, 66 (Part II),
55-69.
BROWN, R. Words and things. Glencoe, 111.:
Free Press, 1958.
BRUNER, J. S. On perceptual readiness. Psychological Review, 1957, 64, 123-152.
BRUNER, J. S. On cognitive growth: I and II.
In J. S. Bruner, R. R. Oliver, P. M. Greenfield,
et al. (Eds.), Studies In cognitive growth. New
York: Wiley, 1966.
CHOMSKY, N. Syntactic structures. The Hague:
Mouton, 1957.
CHOMSKY, N. Aspects of the theory of syntax.
Cambridge, Mass.: M. I. T. Press, 1965.
CHOMSKY, N. Language and mind. New York:
Harcourt, Brace & World, 1968.
COLLINS, A. M., & QUILLIAN, M. R. Retrieval
time from semantic memory. Journal of Verbal
Learning and Verbal Behavior, 1969, 8, 240-247.
FIRBAS, J. On defining the theme in functional
sentence analysis. Travaux Llngulstiques de
Prague, 1964, 1, 267-280.
FRAISSE, P. Why is naming longer than reading?
Ada Psychologies 1969, 30, 96-103.
FRAZER, J. G. Golden Bough: Study in magic and
religion. 13 vols. New York: St. Martin,
1911-1936.
GARNER, W. R. To perceive is to know. American Psychologist, 1966, 21, 11-19.
GIBSON, E. J. Learning to read. Science, 1965,
148, 1066-1072.
HALLIDAY, M. A. K. Notes on transitivity and
theme in English. Part 2. Journal of Linguistics, 1967, 3, 199-244.
HUTTENLOCHER, J. Constructing spatial images:
A strategy in reasoning. Psychological Review,
1968, 75, 550-560.
JOHNSON, N. M. E. An experimental investigation into the role of language in discrimination,
memory and transfer in pre-school children. Unpublished doctoral dissertation, Ontario Institute
for Studies in Education, 1969.
Joos, M. The five clocks. New York: Harcourt,
Brace & World, 1961.
KATZ, J. J. Semantic theory and the meaning of
"good." Journal of Philosophy, 1964, 61, 739766.
KATZ, J, J., & FODOR, J. A. The structure of a
semantic theory. Language, 1963, 39, 170-210.
KENDLER, H. H., & KENDLER, T. S. Vertical and
horizontal processes in problem solving. Psychological Review, 1962, 69, 1-16.
KINSBOURNE, M., & HARTLEY, D. Distinctive feature analysis in children's perception of simple
shapes. Paper presented at the meeting of the
Society for Research in Child Development,
Santa Monica, March 1969.
LEVY-BRUHL, L. La mentalite primitive. Paris:
Alcan, 1922.
273