HOW MANY MEANINGS DOES A WORD HAVE ?

unitary semantic representation brings with it some serious problems.io Consider a t~pic briefly touched on already: comRositionality. According to the principle of compositionality, the meaning of the whole is a function of the meanings of the parts, and the manner of their combination. To put it another way: The meaning of a word is the contribution that the word makes, given the manner in which the word is combined with other words, to the meaning of the composite expression. For some semanticists (such as the Montague grammarians), the principle of compositionality has the status of an axiom, whose truth is self-evident, and without which no semantic investigation could be pursued at all. For in the absence of compositionality, it is argued, speakers would not be able to create an infinite number of new sentences from the finite resources of their language, and if they were to create novel sentences, their hearers would not be able to understand them. Bierwisch's model, by positing two levels of meaning, raises the question of the domain of operation of compositionality. Does compositionality operate at the "semantic" level, or at the "conceptual" level? When a word makes its contribution to the meaning of a composite expression, do'es it contribute its semantic meaning, or (one of) its conceptual meanings? It is clear that compositionality does not, and cannot, operate at the conceptual level (as understood by Bierwisch). On the two-level model, conceptual meani.ngs are context-dependent entities. While there may be general principles for the conceptual elaboration of abstract unitary meanings (Bierwisch 1983), knowing.the meaning of a word does not entail knowing the full range of conceptual interpretati.ons that may be assigned to the word. On the contrary, one of the strengths of the two-level model lies precisely in the fact that it removes the need for such rampant.-polysemy. So what is being combined can only be the "semantic" meanings of component words. However, there are a number of considerations which speak against this hypothesis: Stellenbosch Papers in Linguistics, Vol. 25, 1992, 133-168 doi: 10.5774/25-0-79

structure; when I "open the window" I manoeuvre a movable frame enclosing a sheet of glass; if I'''brick up a window" I brick up an aperture in a wall; and so on.
In brief, window, in (I), denotes eight quite different kinds of entity.2 We observe the same effect if we keep the verbal element constant and change the direct object.
If (1) lists the kinds of activities that can be performed with respect to a window, (2) lists some kinds of things that can be opened.
(2) open {the window / the door / a bottle of wine / a bottle of champagne / a can of beer / a book / a newspaper / a parcel / a pair of scissors / one's shirt / one's eyes / one's mouth / one's hand / one's arms} The 14 phrases in (2) show that open can denote as many different kinds of activity.
When I "open a bottle of wine".I insert a cork screw, rotate it, and pull; when I "open my arms", I move my arms forwards and outwards; when I "open a pair of scissors" I cause the blades of the scissors to separate; when I "open my shirt" I undo some buttons, and so on.It is not possible to "open a window" by performing the kind of activity I perform, say, when opening my arms or opening a bottle of wine.
The above examples with window and open are based on some brief remarks on these words in Lakoff (1987: 416).
A more detailed discussion of open is offered by Searle (1983: 145f.).
In the remainder of this paper I will focus specifically on the word open.
It should be borne in mind, however, that the phenomenon exemplified in (1) and (2) could have been illustrated on just about any word, picked at random from the dictionary.Thus, "eating a steak" is a quite different kind of activity from "eating an ice-cream", or "eating soup".If something is "buried under the tree", it is in a quite different location than that of the picnickers who "picnic under the tree"."Losing $1000" as a Stellenbosch Papers in Linguistics, Vol. 25, 1992, 133-168 doi: 10.5774/25-0-79 result of my wallet being stolen is a different kind of happening than "losing $1000" on the stock.exchange(only in the first case is there a chance that I may "find" my $1000 again).
Lakoff and Searle offer very different accounts of the phenomenon in question.For Lakoff, the fact that window can refer to different kinds of entity in the world appears to be sufficient reason to regard the word as polysemous.There is a long and respected tradition, in twentieth century linguistics, which is highly sceptical of polysemy.
Perhaps the most forceful statement of the "unitary meaning hypothesis" is to be found in Jakobson's essay on the Russian cases (Jakobson 1936).
Each of the Russian cases can express a.
wide range of apparently highly diverse meanings.Yet to attribute independent status to each of these particular meanings ("Sonderbedeutungen"), Jakobson argues, would lead inevitably to the disintegration of the linguistic sign, and its replacement by a multitude of form-meaning relations.
Jakobson therefore set himself the task of identifying, for each of the cases, a unitary, and highly abstract "general meaning" ("Gesamtbedeutung")', which gets fleshed out in the range of particular meanings according to the context of its use.5 In recent years, the "one form -one meaning" position has again been forcefully stated by the German linguist Manfred  Bierwisch 1988;Herweg 1988Herweg , 1989;;Wunderlich 1991).6Often in open polemics with prototype approaches to word meaning, with their postulation of extensive PolysElmy relations extending from a assumed "central sense"7, these studies have proposed unitary meanings for a range of German and English prepositions.For a flavour of the nature of the issues debated, compare the uses of in in the following expressions.In the absence of conventional expectations in this regard, there may be no standard interpretation of an expression.On hearing of ~mouse in the armchairs, we have no clear expectation as to whether the mouse is located in the hollow region defined by the seat, arms and back of the chair, or whether the mouse is located inside the upholstery.
The two-level model is not to be dismissed lightly.), until eventually the child's repertoire comes to approximate to that of the standard adult speaker (Taylor 1989: Ch. 13).
These various considerations all point, it seems to me, in the same direction, namely that in understanding a sentence a person need not, and typically does not, access for each component lexical item in the sentence a highly abstract unitary meaning, which then must be fleshed out according to the context in which the word is used.Rather, words are typically polysemous, to greater or lesser degrees; secondly, each established sense is "concrete" rather than abstract, that is to say, the semantic Once polysemy is admitted, do we not run the risk of "polysemy inflation" (Herweg 1989: 106), of which Bierwisch, Searle, Jakobson, and many others, have been so critical?
In response to this argument, :it needs to be pointed out that the postulation of a range_of specific meanings does not per g preclude the possibility that the different meanings may be perceived to be related, in some respects.Indeed, the very essence of polysemy (as opposed to chance homonymy) is traditionally said to reside in the relatedness of the separate meanings.Furthermore, the perceived commonality between distinct meanings may permit the emergence of a more abstract meaning, of which the particular meanings are instances.These considerations lie behind Langacker's "network model" of category structure./õ -----------0 In the last analysis, the language user has to learn which instantiations, and which extensions, are conventionally sanctioned in the language.
Finally, the network model assigns no special status to higher order schemas.There are certainly no grounds for supposing that a schema is a more "linguistic" kind of entity, while the instantiations of a sche~a are of a more "conceptual" or "encyclopaedic" nature.
Each node in the network reflects acts of categorisation by the language user; more abstract and more concrete meanings alike reduce ultimately to the language user's ability to perceive similarities.
So far, I have assumed that "similarity" has to do with the sharing of some common aspects; with respect to Fig. 1 (d) Fig. 3 incorporates some senses that Searle (1983) had singled out as "non-literal".These senses may be seen as extensions of the "make accessible" schema.Thus, to "open a discussion" is to make it possible for people to participate in the discussion.
The example shows that "non-literal" senses do not require a fundamentally different kind of treatment, in terms of the network model, from "literal" senses.
(e) Fig. 3 was constructed by the writer on the basis of his intuitions.
Other linguists, working on their intuitions, might possibly come up with different proposals.l8 It is important to emphasise, therefore, that the network depicted in Fig. 3 is meant more as a hypothesis than as a definitive account.Given the subjective procedure by which Fig. 3 was constructed, it is legitimate to ask whether any kinds of evidence could be adduced that might justify the proposed content and configuration.
Here we touch on a fundamental aspect of the network model, a full discussion of which would go beyond the scope of this paper.
Various kinds of potential evidence for the structure of a network may be mentioned, however.Firstly, the relative "distance" of nodes from each other, and the grouping of nodes in "clusters", may be investigated experimentally by eliciting from speakers subjective judgements of meaning similarity, and then subjecting these judgements to hierarchical cluster analysis.
This statistical technique has been employed by Schulze (1989) in Secondly, longitudinal data from first language acquisition ~ay enable the researcher to track the gradual growth of a network for an individual speaker over time (Taylor 1989: 253f.).Diachronic data may serve a similar function with respect to the development of a network in the speech community (Geeraerts 1983).Cross-language data may also be of interest.
Afrikaans oopstel lexicalises one of the higher order schemas proposed for English ~en ("make accessible to the public"), while Italian extends the network proposed for English by the incorporation of an additional schematic sense, viz."make (an electrical appliance) functional".
A crucial empirical issue (the crucial issue, even) raised by the network model concerns the depth in the network at which meanings of lexical items are accessed in the production and comprehension of utterances.
On hearing that a person opened the window, at what depth in the network is the meaning of open accessed?I have presented some arguments against the view that it is some highly abstract sense, common to all uses to open, that is accessed.But equally, it seems unlikely that very low nodes are accessed, e.g.those appropriate for, say, opening a hinged window in contrast to opening a sash, or a sliding window.
(Consequently, open the window is not felt to be ambiguous between these kinds of activity.)I suggest, rather, that the network is accessed -in default cases, at least -at an intermediate level in the network, namely at the level of "basic level concepts".
Again, this proposal is to be taken as a hypothesis -one, furthermore, that should be accessible to techniques of psycholinguistic investigation.
The notion of the basic level concept has been empirically investigated by Eleanor Rosch and her colleagues, primarily in connection with nominal categories (Rosch et al. 1978; see also Lakoff 1987).The basic level is that level in a taxonomic hierarchy at which we normally think and talk about objects and Stellenbosch Papers in Linguistics, Vol. 25, 1992, 133-168 doi: 10.5774/25-0-79 situations.
The object on which I am sitting as I compose this paragraph would normally be cate~orised .as a "chair", and not, say, as a "piece of furniture", even though, quite obviously, the object is a piece of furniture."Chair" is a basic level category, whereas "furniture" is not.There is evidence that in the processing of an ambiguous word, a listener accesses, if only briefly, each of the different senses of the word (Foss and Jenkins .1973, Conrad 1974, Holmes et al. 1977, Swinney 1979) For examples of the prototype approach, see Hawkins (1988), Brugman (1989), and Schulze (1989); also Taylor (1991). 8 The example is from Pribbenow (1989). 9 During a class discussion of this example, one student related that her family had been bothered by the afternoon sun coming through a west-facing window, and had decided to "paint the window black", i.e. to paint over the glass portion of the window.
10 For a ~ore extensive critique of the two-level model, see Taylor (to appear).food served in the course of a meal.
(Note also that one would probably say of a baby that is not yet consuming solid food that the baby is "eating well", not that the baby is "drinking well". Likewise of an invalid who is allowed only liquid food.)

13
A further task, perhaps, for practitioners of the two-level approach, might be to derive the unitary abstract meaning of aufschlagen from the composition of the unitary abstract meanings of its component morphemes, auf and schlagen.For some experimental data bearing on this process, see Anderson and Ortony (1975), and Garnham (1979).

20
The difficulty of coming up with general word definitions has been noted many times; see e.g.Fodor's (1981) discussion of the verb paint, and Shanon's (1988) comments on Fodor's analysis.
Fodor observes that a simple definiti.onlike "cause (a surface) to be convered with paint" is inadequate" since not every example of a person causing something to be covere!dwith paint counts as an instance of painting.A painter, in dipping his brush into a can of paint, covers the surface of his brush with paint, but has not thereby "painted the paint brush".
that in denotes differ,~nt kinds of relations in the (a), (b) and (c) expressions.In the (a) examples the one object is wholly located within a hollow region defined by the exterior sides of the reference object; in the (b) examples the one object is located within the material substance of the reference object; while in the (c) examples the one object is supported, or held in position, through partial containment in the reference objec~.In spite of these differences, Herweg (1989) proposes a unitary meaning for in, namely that one object is located within a space defined by the outer exterior of another object.The different interpretations of in follow from (conceptual) knowledge of the Stellenbosch Papers in Linguistics, Vol. 25, 1992, 133-168 doi: 10.5774/25-0-79 dimensional and topological properties of objects, and the manner in which they are typically aligned with respect to each other.
Aufschlagen has the sense "open" only with respect to the opening of one's eyes, the opening of printed materials, or the forceful opening of a door, a nut, etc.The problem, on the two-level approach, would be to come up with a unitary definition of aufschlagen which could be conceptually elaborated by just these kinds of opening activities -as well as by the other kinds of activities denoted by aufschlagen, such as the raising of prices, the pitching of tents, and a person's establishment of a place of residence. 13Stellenbosch Papers in Linguistics, Vol. 25, 1992, 133-168 doi: 10.5774/25-0-79 (f) Finally, there is the question of learnability.On Bierwisch's model, each particular use of a word is interpreted through an interaction of conceptual knowledge with an abstract linguistic meaning.But how does the child acquiring his mother tongue come to know the abstract linguistic meaning of a word?Does the child, on hearing a new word for the first time, instantaneously construct a mental representation of the word's linguistic meaning, from which the full range of conceptual variants may be generated?This seems highly implausible.(In view of what was said above on the differences between English open and Italian aprire, we would have to suppose that the English child, on first encountering open in, say, the context open your mouth, instantaneously constructs a different semantic representation than the Italian child, who first encounters aprire in the context apri la bocca.)More likely, what the child first acquires is the specific meaning that the word has in the context in which the word is first heard.On first encountering open in the context open your mouth, the child infers that open denotes the kind of activity that one performs with respect to one's mouth.At this stage in the acquisition process, open designates only this specific kind of activity.Subsequently, the use of the word is extended, to cover many other kinds of activity (open the window, open one'~eyes, etc.
representation already makes intrinsic reference to what Bierwisch Stellenbosch Papers in Linguistics, Vol. 25, 1992, 133-168 doi: 10.5774/25-0-79 would call "conceptual" information, and to what Searle would designate as aspects of "the Background".The speaker of English, that is, knows what is meant by open the window, not in virtue of his knowledge of a unitary, abstract meaning of open, but in virtue of his familiarity with the practice of opening windows, and his knowledge that this practice is conventionally designated by the expression open the window.14Having rejecting a unitary meaning of open, the question arises whether we are not now in danger of coming back full circle to the extreme polysemy position that Searle found to be "absurd", If, as I have argued, opening a window and opening one's arms are two different kinds of activity -a fact which requires us to recognise two distinct senses of open -then by the same token opening a hinged window and opening a sliding window are also two different kinds of activity, as are opening a window hinged on the left and opening a window hinged on the right, and so on and so forth. fig.1 and [B].A study of open, however, points to the need to recognise another kind of similarity, viz.similarity by association.Compare the uses of QPen in the following pairs of expressions: , open the office open the lid, open the ja.r open the cork, open the bottle The members of ~ach pair could refer to exactly the same kind of activity.If one opens the door (to the office), one has thereby opened the office; likewise, one ope,ns a jar by opening the lid, and one opens a bottle by opening the cork.In open the Stellenbosch Papers in Linguistics, Vol. 25, 1992, 133-168 doi: 10.5774/25-0-79 (c) The very generality of the higher order schemas confirms a point made earlier, concerning tbe conventionality of a network.There are many ways in which one could conceivably " Papers in Linguistics, Vol. 25, 1992, 133-168 doi: 10.5774/25-0-79 his study of prepositional polysemy.

14
In view of the primacy of the specific meaning encoded by open the window, it is of little concern that the expression may not be strictly compositional, in terms of hypothesised general meapings of the component words, nor that its logical properties may differ from those of open one'~arms, or open the discussion.15 It is in such terms that the "indeterminacy" of an utterance is to be explained •.Suppose -cf.footnote 9 -that I wish to convey that I painted over the glass poption of a window.This is not a standard practice, nor is there in English an Stellenbosch Papers in Linguistics, Vol. 25, 1992, 133-168 doi: 10.5774/25-0the completeness of Fig. 3 is made.Lower nodes in the network are not shown, neither are intransitive, and a range of more peripheral, metaphorical uses of open considered.18 It is worth mentioning that Fig. 3 went through several versions, s6me of which bore little resemblance to the final product!19

21
The objection raised by Searle against extensive polysemy -that extensive polysemy leads to an exponential increase in sentence ambiguity -thus turns out to be invalid.As Searle rightly pointed out, our interpretation of open the window is possi?le in virtue of our familiarity with the relevant practice.But interpretation does not need to proceed on a prior activation of more general meanings of open and window.What a person mentally accesses is the particularised sense of ~, as this word is used with respect of windows, i.e. in the collocation open the window.Stellenbosch Papers in Linguistics, Vol. 25, 1992, 133-168 doi: 10.5774/25-0-79 If you open a letter or parcel, you cut or tear the envelope, or remove its wrapping.EG.I'll open the mail .
open g container -lack this ready imagibility.If I try to form Stellenbosch Papers in Linguistics, Vol. 25, 1992, 133-168 doi: 10.5774/25-0-79 Stellenbosch Papers in Linguistics, Vol. 25, 1992, 133-168 doi: 10.5774/25-0-79 Stellenbosch Papers in Linguistics, Vol. 25, 1992, 133-168 doi: 10.5774/25-0-79 3. Stellenbosch Papers in Linguistics, Vol. 25, 1992, 133-168 doi: 10.5774/25-0-79 would see the moral of Fodor's example as follows: It is indeed difficult, if not i~possible, to come up with a definition of (to) paint that is schematic for all and only the specific uses of the word.Yet a person can readily form a very clear mental image of what it means to "paint the ceiling", "paint the window", "paint one's nails", etc.It is precisely in terms of such basic level activities (or "practices") that a person knows what paint means, and in terms of which paint is to be defined.