Polysemy and the lexicon
How many meanings?
Some years ago I wrote a paper entitled “How many meanings does a word have?” (Taylor 1992). I began by juxtaposing some remarks by John Searle on the verb to open and observations by George Lakoff on the word window.
• Lakoff (1987: 416) pointed out that window can be used to refer to different categories of things. If someone throws a rock and breaks a window, what is broken is (most probably) a glass panel, whereas if someone paints a window, what they paint is (most probably) the wooden structure encasing the glass panels. Deliver the window, brick up the window, sit in the window again refer to different kinds of entity. For Lakoff, this was evidence that the word was polysemous, that is, it had distinct, though closely related meanings.
• Searle (1983: 145–8), discussing the verb open, noted that a person who opens a door and a surgeon who opens a wound do very different things to the entities in question. You would not, for example, open a door by making incisions in it with a surgical scalpel. The truth conditions for “open a door” and “open a wound”– the kinds of things that a person would have to do in order for the expressions to properly apply to their actions– are therefore different in the two cases. Searle, however, does not draw the conclusion that open is polysemous. Rather, he claims that the verb has a fixed and constant value. Truth conditions emerge as a consequence of the way in which we interpret the expressions, relative to what he calls “the Background”, that is, our knowledge of cultural practices, how we interact with the world, and how the world is constituted. Searle illustrates the role of interpretation relative to the Background on the example Sam opened the Sun. We can, Searle claims, readily understand this expression on the basis of the semantic content of the component words, but we cannot generate a set of truth conditions for it. Opening the Sun is not an established practice, and we have no idea what a person who “opened the Sun” is supposed to have done.
Lakoff and Searle exemplify two radically different approaches to the issue of word meaning. Each approach is problematic. Let us begin with Lakoff’s approach. Lakoff seems inclined to recognize a distinct meaning whenever a difference in the reference of a word can be detected. There are several problems with this approach:
• The number of distinct meanings which would have to be assigned to words such as window and open (and no doubt to many others), even in their literal, non-metaphorical uses, is going to be very large indeed. Given that the referents of window in paint the window, break the window, open the window, install a window, look through the window, brick up the window, sit in the window, and so on, are identifiably different, we should have to say that window is many-ways polysemous. Such “rampant polysemy” (Cuyckens & Zawada 2001: xvii) would make enormous demands on the storage capacity of the language user.
• Polysemy generates ambiguity, and ambiguity has to be resolved, presumably at a cost in cognitive processing. If a word is n-ways polysemous, any sentence containing that word is in principle n-ways ambiguous. As more and more polysemous words come together in a sentence, the ambiguity increases exponentially. Concretely: If open is n1-ways polysemous, and window is n2-ways polysemous, the expression open the window would have n1 × n2 possible readings, and could only be understood as the result of a complex process of disambiguation. This conclusion is patently counterintuitive. If anything, the more context that is added to a word, the less likely it is to be perceived as ambiguous.
• The number of distinct polysemes of a word is not only likely to be very large, it may well be indefinitely large. For example, the number of different things that can be opened, and the different sets of truth conditions associated with these various opening activities, is in principle open-ended. Tolerating large-scale polysemy is one thing, admitting indefinitely large polysemy would border on the absurd (Searle 1983: 146).
Searle’s position, which assumes a single semantic value for a word, is also not without its problems.
• If we wish to maintain that a word has a constant semantic value, we are under the obligation to spell out exactly what that meaning is. Searle’s discussion of open is instructive in this respect. Although Searle insisted that open has but a single meaning (at least, in the non-metaphorical examples which he cites), Searle nowhere states what this meaning is. Perhaps Searle considered that the meaning of open was so self-evident that it was not necessary to spell it out. But the task is anything but trivial. It is, of course, not at all difficult to state what it means to “open the door”, “open the curtains”, “open a book”, and so on. We can demonstrate these activities (given appropriate props), we can mime them, we can describe quite precisely how the patient entities (the door, curtains, etc.) are affected by the process. The difficulty is to state the meaning of open abstracting away from the particularities of these various expressions and the activities which they designate. Moreover, the proposed meaning will have to be sufficiently specific so as to differentiate the word from words of similar, though contrasting meaning. For example, it would not do to define open as involving the creation of a gap in an entity through the separation of its parts. Such a definition would also be valid for cut, in some of its uses. Neither would it do to define open in terms of gaining access to the interior of an entity. Such a definition would predict that open would be used to describe the peeling of an orange, or the cracking of nut.
• Let us assume, for the sake of argument, that a unitary meaning of a word can be identified, and has been satisfactorily characterized. We now need to state the mechanisms which give rise to the various interpretations which the word can have, in the different contexts in which it can appear. These mechanisms will need to be of a very general nature, in the sense that they are not specific to a particular lexical item as used in a particular context– otherwise, we will run the risk of introducing polysemy by the back door, as it were.
• On Searle’s account, the truth conditions of a word pertain to its contextualized interpretation, not to its unitary meaning. Presumably, other logical properties of an expression, such as its entailments and presuppositions, would also be attributes of an interpretation, it would not be possible to read them off from the (unitary) linguistic meanings of the component items and their manner of combination. This raises questions about the usefulness of unitary representations. If they are not involved in determining truth conditions or entailments, what are they for, and why should we be bothered by them?
The above remarks suggest that tolerance for extreme polysemy on the one hand, and attempts to eliminate polysemy on the other, are both seriously flawed. Both, it will be noted, raise questions of psychological plausibility. Can we reasonably suppose that speakers of a language have stored, and access, and make choices between large numbers of polysemes, and that the understanding of even very banal expressions requires a massive amount of disambiguation? Can we, on the other hand, suppose that speakers have stored schematic meanings for each lexical item (meanings which may actually be quite difficult to articulate) and that they really do apply to these all manner of elaboration strategies? Observe that the question of cognitive economy cuts both ways. Unitary meanings economize on storage but necessitate the computation of contextual interpretations. Polysemy economizes on the computation of contextual variants but increases storage demands; it also requires disambiguation procedures. We should also bear in mind that while a streamlined lexicon might appeal to some theorists on account of its “elegance”, or “parsimony”, there are no principled grounds for setting an upper limit on the amount of polysemy that the lexicon can tolerate, or for deciding, a priori, what an “optimal” amount of polysemy might be.
Given the problems associated with rampant polysemy on the one hand and unitary abstract senses on the other, a reasonable approach would be to seek a compromise. We can imagine two kinds of compromise:
• Going half-way. We try to reduce polysemy by abstracting as much as possible over individual meaning nuances of the different uses of a word. At the same time, we allow polysemy where unifying abstract senses are not feasible. In other words, we strive for as little polysemy as
possible, but permit as much as necessary, arriving at an account which
eschews both rampant polysemy and strict monosemy.
• Having it both ways. We allow both fine-grained semantic distinctions and more general, schematic meanings which capture the commonality amongst groups of specific meanings. The meanings, at various levels of schematicity, would be typically displayed in a network format, with more specific meanings standing as elaborations of more schematic meanings.
