1. Introduction

Which of several senses a polysemous term contributes to the meaning of an utterance depends on context, at least if there is such a thing as “polysemy”. I maintain, however, that the nature of the context-dependence of polysemy has been misunderstood.

The phenomenon of copredication, that is, sentences like ‘Lunch was delicious but lasted forever’ or ‘The book is heavy but informative’, ‘London is populous but tends to vote conservative’, intuitively involves multiple senses of polysemous terms being activated in the same context. This looks problematic. For one thing, it seems to pose problems for externalism, roughly the thesis that:

Take the sentence ‘Lunch was delicious but lasted forever’. It sounds as if ‘lunch’ means both the event and the food, but if that is right, then is the sentence made true by how things are with the event, or how things are with the food, or perhaps with the union of these things? Moreover, is there not the risk that the result will be that the sentence might turn out to be true and false, if the one or other of the predicates is, say, true of the food and false of the event?

These sorts of concerns might lead us to follow ^{Chomsky (2000)}, ^{Collins (2017}, see also his 2023 which argues that cases of copredication have only illusory truth conditions) and ^{Pietroski (2018)} to reject extensional truth-conditional semantics on account of the phenomenon of copredication. Conversely, it might lead us to follow ^{Liebesman and Magidor (2017}, ²⁰¹⁸⁾ and ^{Brody and Feiman (2024)} and reject polysemy so as to preserve extensional truth-conditional semantics. These latter deny that there is a compelling intuition in favor of the polysemy of such words as ‘lunch’ and ‘book’. In response, however, Christopher A. ^{Vogel (2025)} has argued that the real problem that copredication poses for (E) is not the intuition that copredication involves terms with more than one extension, but rather that copredicative sentences do not generate the inference patterns we would expect if (E) were true. For example, ‘London is populous, but tends to vote conservative’ would seem to entail ‘London is populous’ and ‘London tends to vote conservative’, but it would seem not to entail that ‘Something is populous, but tends to vote conservative’. If (E) were true, then the first sentence should entail all these things.

I believe that both rejecting (E) on account of polysemy and denying the existence of polysemy to rescue (E) are premature responses. Instead, like ^{Pustejovsky (1995)}, ^{Asher (2011)}, ^{Vicente (2021a}, ^2021b), ^{Ortega-Andrés (2019)}, and Ortega-Andrés and Vicente (2023), I will offer an account of the semantics of co-predication which treats it as involving polysemy, but which is compatible with (E). My account will differ however from these related views in significant details. The key to my account will be the role played by context.

By contrast with other versions of contextualism (e.g., ^{Recanati 2017}), I hold that the role of context, with respect to polysemous terms, is not merely to determine which sense or senses of the term is/are activated, but additionally, to determine something I will call a “specification” of the extension of the term.² The notion of a specification differs from the notion of sense, for one thing being more abundant than senses and, further, playing a more direct role in the truth-conditions of a sentence. In what follows, I will argue for the need to make a distinction between senses and specifications and, with that distinction in hand, I will show how the truth conditions for sentences like ‘Lunch was delicious and lasted forever’ and ‘The book is heavy but informative’ can involve genuine polysemy (such that ‘Lunch’ and ‘The book’ are picking out more than one element of the domain in their respective contexts), while nevertheless remaining determinate and consistent (i.e., no utterance is evaluated as true and false). The account I offer will also be capable of explaining our intuitions about inferences involving copredicative sentences, such as those brought out by Vogel, and it will have the resources to offer an explanation for why some cases of copredication seem acceptable, while others seem unacceptable.

1

As we have seen, there are various possible responses to the cases of copredication. There are those which, following ^{Chomsky (2000)}, take these cases to undermine extensional, truth-conditional semantics (^{Collins 2017}, ^{Pietroski 2018}). Then there are those which deny that polysemy is a genuine phenomenon (^{Liebesman and Magidor 2017}, ²⁰¹⁸; ^{Brody and Feiman 2024}).

However, there is a path between the extremes of rejecting extensional truth-conditional semantics on one hand and rejecting polysemy on the other. For example, ^{Pustejovsky (1995)} and ^{Asher (2011)} have proposed accounts of polysemy on which the referent of ‘lunch’ is a “dot object”, that is, an object, though perhaps it is better to say a “bare particular”, in itself lacking any quality including cardinality, but which can be conceptualized in a number of different ways, for example as a delicious amount of food or as a lengthy event. Agustin Vicente and Marina ^{Ortega-Andrés have offered an alternative, “Activation Package Theory” (2019}, see also ^{Ortega-Andrés 2023}, and ^{Vicente 2021a}, ^2021b). As Vicente explains the referential component of the theory:

the meaning of a word-type is a concept, a concept being a body of knowledge about a certain category stored in long-term memory. Concepts understood in this sense are structured mental entities that support different ways of categorizing the categories they are about and of supporting inferences [...]. However, concepts also provide a different kind of possibilities of denotation, related not to different ways of categorizing certain entities or substances, but to different pieces of information stored in the concept or knowledge structure. In these cases <cases of copredication>, word tokens have denotations that [...] refer to different kinds of entities in the world. (^{Vicente 2021a, pp. 923-924})

Put another way, word-types refer to complex concepts, which are bodies of knowledge, while word tokens pick out objects in the world. This is combined with a theory of “activation” which is intended to explain how different components of a concept like lunch can be selected, such that an occurrence of ‘lunch’ can refer first to food, when selected by an occurrence of the phrase ‘...was delicious’, and then an event, when selected by an occurrence of the phrase ‘...lasted forever’. Thus, the occurrence of ‘lunch’ in ‘lunch was delicious but lasted forever’ changes its referent over the course of the sentence, and the sentence has the content lunch _food was delicious but lunch _event lasted forever.

In a recent paper, Emma ^{Borg (2024)} has classified both the “activation package theory” view and the “dot objects” view as examples of what she calls “rich common core” models of polysemy. On these accounts, the meanings of word types are complex bundles of semantic features, rather than extensions. Such accounts can, however, accommodate a purely extensional account of the meanings of individual utterances. Moreover, a rich common core model can accommodate the view that the meaning of the word type, the rich common core, may fix the extension of the utterance in a context, as is the case for activation package theory. This would seem to support (E), depending on how exactly that thesis is understood. As Borg says,

the approach is consistent with externalist semantics only, it seems, at the cost of shifting the point of connection between language and the world from standing elements of the lexicon (the rich conceptual contents associated with word types, which do not refer) to the contextually selected bundle of elements (which do refer). Semantic internalists, who are happy with the idea that we can speak of the referential properties of concrete uses of expressions may well suggest that this is really to concede the ground to them. (2024, p. 17)

I will offer another rich common core account. As such, my account also takes truth conditions to be features of utterances of sentences in context, and extensions to be features of the constituents of such utterances, and the meaning (at least in one use of ‘meaning’) of word types to be complex concepts that determine the extensions of tokens in given contexts. I will maintain, though, that my view does not “shift the point of connection between language and the world from standing elements of the lexicon...”, because on my account the standing elements of the lexicon also have extensions and, moreover, whether an utterance of a sentence is true or false is determined by how things are with these extensions. I will explain how this works in section 3. The result is that, I maintain, my position is compatible with externalism.

As a rich common core model, my proposal can also be contrasted with other sorts of accounts, like what ^{Borg calls (2024, pp. 18-21)} “thin common core models”. Such models would include radical contextualism, which François Recanati explains as follows:

Senses are what words convey in context; they are ‘intuitable’, i.e., they serve as building blocks of thoughts (thoughts being what utterances express). Lexical meanings, on the other hand, are theoretical entities: properties of word types that the linguist posits to account for the observed uses. According to Radical Contextualism, there is something intrinsically schematic and pattern-like which makes lexical meaning too abstract (and also, in a sense, too rich) to be the contributed content. (2017, p. 295)

Views such as this are simply incompatible with (E), because the meanings of lexical meanings do not determine whether individual utterances are true or false. Two other views that ^{Borg canvases are the “literalist” view (2024, pp. 11-14)} and the “sense enumeration” view (2025, pp. 7-11). According to the literalist view, there is one or a small number of lexical meanings of a sentence, which is what an utterance of the sentence literally means, while apparent polysemy like our cases of copredication involve non-literal meaning, and such sentences are in general, strictly speaking, false. As my project is to offer an account compatible with the truth of copredicative sentences, taking them to be genuinely polysemous, I will not pursue this approach. The sense enumeration view is largely compatible with what follows, as it holds that each of the various senses of a polysemous term is among the literal meanings of the term. This view is also clearly compatible with (E). Nevertheless, what follows is intended to fall into the rich common core family of views, rather than the sense enumeration family of views, because the former has an easier time distinguishing polysemy from homophony and also is better equipped to explain why some senses of a polysemous term allow for copredication and others seem not to, a puzzle I will provide my own answer to in section 3.

My account, then, has important commonalities with the dot object and activation package views, however I maintain that it has some advantages in that it is more clearly compatible with externalism, it does not involve any commitment to a metaphysics of dot objects and it does not involve any mid-utterance change of extension, as in the Vicente and Ortega-Andrés’ account.

2

Returning now to copredication. There are many such cases, cases involving count-nouns used as singular terms (‘lunch was delicious but went on forever’), count-nouns in harness with determiners (‘the book was heavy but informative’), proper names (‘London is populous but tends to vote conservative’) and even verbs (‘She wrote movingly but illegibly’). All of these cases would seem to involve polysemous terms which are such that more than one sense of the term is at play in a given utterance. In an utterance of ‘the book was heavy but informative’, the physical book sense of ‘book’ is at play, but so is the information book sense of ‘book’.

If two different senses are at play in the same context, then it would seem that they might determine two different extensions simultaneously. If the subject term of a sentence in context, say ‘lunch’ in some utterance of ‘lunch was delicious but went on forever’ has two different extensions, then it may seem that the utterance could be both true and false if the two extensions behaved differently with respect to the predicate. There are two ways to avoid this problem. The first, is to follow Vicente and Ortega-Andrés in holding that a single term, in this case ‘lunch’, can change its denotation over the course of the utterance. Vicente, who holds that the meaning of polysemous words like ‘lunch’ and ‘book’ are knowledge structures, claims that

the knowledge structures framework has a natural implication as to the semantics of co-predicational sentences. Competent users of a certain language access all the different senses of the co-predicational term as soon as they access one of them. As they go processing the sentence, they can retrieve the sense that best matches the selectional preferences of the predicate they encounter. So, processing ‘the book is heavy and interesting’ implies first accessing the knowledge structure of ‘book’, then accessing and retrieving the sense book-physical object, and, upon encountering the predicate ‘interesting’, accessing and retrieving the sense book-informational content. (2021b, p. 18)

To give another of his own examples, the sentence ‘Brazil is a large Portuguese-speaking republic and is very high in inequality rankings but always first in the FIFA rankings’ has, as its truth conditions, ‘Brazil [place] is a large piece of land & Brazil [people] is Portuguesespeaking & Brazil [State] is a republic & Brazil [economic system] is very high in inequality & Brazil [football representative team] is always first in the FIFA rankings’ (^{Vicente 2021b, pp. 18-19}). This, I think, is wrong. I do not think that words change their denotations over the course of an utterance. The empirical evidence seems to suggest that polysemous words support activation of multiple senses to a much greater extent than homophonous words, with at least one study finding that, once activated, multiple senses of a polysemous word, at least in the case of metonyms, will remain at equivalent levels of activation for periods of at least 750 milliseconds (^{MacGregor et al. 2015}). This supports a distinction between polysemy and homophony, but the conclusion that has been drawn by some from this is that, rather than competing with one another for activation, as the different senses of homophonous words do, the different senses of polysemous words “act collaboratively to strengthen a unified representation” (^{MacGregor et al. 2015, p. 137}). Now this evidence is not decisive and I would certainly agree in some contexts that we are using a polysemous term in one sense rather than another, but I believe this suggests that we should be wary of positing too-finegrained sense-shifts.

This brings me to the second alternative for avoiding contradiction. This involves adopting a view defended by some of the opponents of polysemy, namely ^{Liebesman and Magidor (2017}, ²⁰¹⁸⁾. Liebesman and Magidor think that physical books can be informative and information books can have paradigmatically physical properties like being on the top shelf or being heavy. On their view, ‘the book was heavy but informative’ is about either an information book or physical book, but not both. I agree with the first of these claims, but disagree with the second. On my view, ‘the book’ in the utterance ‘the book was heavy but informative’ is picking out both an information book and a physical book. The reason that this does not lead to the sentence coming out both true and false is that, in the relevant utterance, ‘heavy’ picks out a property that can be true of both information and physical books, as too does ‘informative’. Why the words behave this way in such utterances has everything to do with context.

On my view, context’s role with respect to polysemy is not solely to determine which sense is activated. It also plays a role in determining how a term’s extension(s) contribute to a sentence’s truth conditions. To get an intuitive handle on my proposal, let us briefly recall one use to which context has been put.

Contextualist accounts of the concept of knowledge are wellknown. One such account is David ^{Lewis’s (1996)}). Lewis holds that knowledge is always characterized by the same intuitive gloss: P is known by X if and only if X’s evidence for P eliminates every possibility that not P. However, the truth-conditions for the right-hand side of this biconditional are underdetermined (1996, p. 551). That is because the truth-conditions for this clause are context-dependent. Contexts, according to Lewis, are more or less precise, and correspondingly the range of the quantified expression ‘every possibility that not P’ is more or less wide-ranging. In a maximally precise context, ‘every possibility that not P’ will mean every possibility that not P.³ In such a context, little if anything will count as knowledge: only those things that are known with a certainty resistant to any skeptical doubts whatever (in the strongest possible sense of ‘any skeptical doubts’!). In a less precise context, ‘every possibility that not P’ will range over a more restricted set of possibilities with respect to P, and will exclude more skeptical scenarios from this range, so they will not be relevant for determining whether P is known. The result is that, in a less precise context, more things are known. The Lewis proposal for the concept of knowledge is helpful for understanding my proposal for recovering single and determinate truth conditions for utterances of sentences that involve words used in more than one sense simul taneously. The key notion that I borrow from Lewis is the notion that contexts can be more and less precise, with consequences for the domain and the ranges of predicates.

The basic proposal is this: in a maximally precise context, every object in the domain is treated as an individual on its own, but in a less precise context, two or more objects in the domain of discourse are grouped together to be treated as one. Which objects are liable to be treated together as one in a less precise context is determined by lexical meaning, that is the “rich common core” introduced above.⁴ A polysemous lexical item, or at least lexical item which supports copredication, is one whose meaning allows its various extensions to be treated as one in some contexts. Thus, in a context in which ‘the book was heavy but informative’ involves polysemy, the physical book and the information book are being treated as one. Before I explain what it means for objects to be “treated as one”, let us note the role of context here.

As stated above, contexts can be more or less precise. This is partly a matter of the presence of certain words in the sentence (that is to say, context is in part composed of the coercive effects of certain terms), but it is more than this. Consider the following case. In a metaphysics lecture on the distinction between types and tokens, the lecturer may illustrate the distinction by talking about information books and their various copies. To demonstrate that information books cannot be numerically identical with their copies, she might point out that copies have physical properties like weight, while information books don’t. She may conclude her example by holding aloft a heavy copy of some interesting book and saying ‘it is not the case that this book is heavy but informative’. Would she be speaking truthfully? I think she would, because in this case, her word ‘book’ picks out the copy individually and the information book not at all. That is so because the context is precise, and the context is made precise by the fact of its being a metaphysics lecture. This same speaker, speaking of this same copy might in a more casual context say ‘this book is heavy but informative’, and again, I suggest, she would be speaking truthfully, because her use of ‘book’, in this less precise context would pick out the copy and the information book simultaneously and they would be treated as one. Thus, utterances of the same sentence type can differ in truth value because of the nature of the context. So one of the things context does, I propose, is to determine the level of precision. As the example just given illustrates, this means that context determines if just one or more than one sense of a polysemous term is selected, and which senses these are. But in cases where more than one sense is activated, context will need to do more than just determine which senses these are.

So far, I have described the multiple extensions of polysemous terms in imprecise contexts as being “treated as one”. In the next section, we will have to provide a technical unpacking of what it is to be treated as one.

3

The lexical meaning of a term fixes its extension(s) in a context. A polysemous term in a copredicative sentence, in an imprecise context, has more than one extension. Yet, in such a context, the truth conditions for such a sentence are determinate and single (there are not different sets of truth conditions for the different extensions). That is because, in some way, the different extensions are “treated as one”. But in what way?

Peter Geach offers a useful notion for our purposes, the notion of a “specification”:

The word “word”, I pointed out, is ambiguous. It may mean “token word” or “type word”, or “dictionary-entry word”, or various other things. Despite this, I may specify as the universe of discourse the words in a given volume in my room at Leeds; for I could give each word in the volume a proper name and get a finite list of them. The ambiguity I have just mentioned is an ambiguity over what shall count as the same word; but since a list specifying a domain anyhow need not be nonrepetitive, this need not worry us. (1973, p. 294)

Geach’s point is that the cardinality of the set of words depends on how we have “specified” the domain, and consequently the extension of the term ‘word’. We can specify the extension of the term ‘word’ in such a way that each type word and each token word are treated separately, but we needn’t do this. Although Geach mentions the possibility of repetitions in this passage, his actual concern is with the possibility of “compressions” remaining in the specification (see ^{Dummett 1996, p. 314}), and indeed the claim that we can always precisify the referents of our words more finely plays a key role in his argument for his well-known theory of relative identity (Geach 1967, 1980, p. 216, see also ^{Molto 2024}). Thus, if I list all the words in all the books in my office, I do not necessarily decide, for each entry, if it is a name for a word type or for a word token, it might be both at the same time. The point is that we may express domains or subsets of domains precisely, by listing each element separately, or less precisely, by grouping some of them together.

Geach talks about using a list of names to specify a domain, thus we can list the extension of ‘city’ by rhyming off all the names of cities ‘London’, ‘Paris’, ‘Rome’, and so on. If any of these names are being used in the senses as in our example copredicative sentence ‘London is populous but tends to vote conservative’, then that name will introduce into the domain several distinct entities at once: the population, the landmass, the political entity, and so on. But the cardinality of cities is determined not by its extension. London, the population, and London, the landmass, are both in the extensions of ‘city’ in this case, but they are not counted as different things, they are both counted as London.

It is not necessary, though, that we use names to specify the domain, as Geach does. Our interest in Geach’s notion of a specification is primarily to allow us to distinguish between two things: the domain itself and the different partitions of the domain which group elements of the domain together in a more or less fine-grained way. Using the formal language of set theory will be more helpful for our purpose than using proper names, as it will allow us to show exactly what entities are being treated as one, and thus in the account we offer, a specification of the domain will be represented by a partition of the domain, and a specification of the extension of a singular term will be represented by a set of the extensions of the term that are being treated as one. Thus, for a polysemous name a, in a context where it picks out both x and y, the expression ‘(x,y)’ will serve as a specification of the extension of a.

The purpose of introducing this notion of a specification is to allow us to reduce multiple extensions, in the case of polysemous terms, to single entities with which to evaluate the truth or falsity of sentences, and this will be key for avoiding the worry raised earlier that allowing multiple extensions will lead to unwanted contradictions.

The terms of our language will include the names a, b, the monadic predicates F ₁, G ₁, the dyadic predicates F ₂, G ₂,⁵ and the connectives ¬^, &, v , → ^, ↔ . The formation rules for sentences will be those of the language of first-order logic without quantification.⁶ We will also need the following four components: D, C, S, and I, where D is a domain, C is a set of contexts which index terms of L ₀ such that all terms of the same sentence token have the same index. S is a set of functions from C onto partitions of D.

I is a mapping of terms of L₀ such that:

Finally, we have the truth conditions for sentences of L₀:

These truth conditions should look familiar, they are essentially the same as the truth conditions for the language of first-order logic, the only difference being that the interpretation function has as its value a specification of extension rather than an extension. But the basic idea is easy to grasp. In the semantics for first-order logic, ‘F(a)’ is true if and only if the extension of ‘a’ is part of the extension of ‘F’. If we allow that ‘a’ might have more than one extension, we cannot say that, at least without taking the risk that ‘F(a)’ will come out true and false simultaneously. Instead we say that ‘F(a)’ is true if and only if the specification of the extensions of ‘a’ is a part of the specification of the extension of ‘F’. As the specification of the extensions of ‘a’ is a single item, it is either a part of the specification of ‘F’ or it is not, at least given our stipulation that the context indexing of all terms in a sentence must be the same; it cannot be both and so there is no risk of contradiction.

Let us apply this to a case of copredication. We will use the example ‘London is populous, but tends to vote conservative’. Take the name ‘London’, the type word. It’s meaning is a body of knowledge, just as Vicente and Ortega-Andrés have suggested, but it also has extensions. The extensions of ‘London’ include the group of people, the political entity, the geographical location, and whatever else the body of knowledge is knowledge of. There is no requirement that competent speakers be able to elucidate all the extensions of ‘London’, to know the meaning of ‘London’ is just to have a sufficient grasp of the body of knowledge. When the extensions of ‘London’, the lexical item, are taken together as a set, I call it the ‘specification of the extension’ of ‘London’, or alternatively, the ‘interpretation’ of ‘London’. An interpretation of a word is its contribution to its truth conditions.

Now enters context. Context, as we have said, is more or less precise. This precision is determined by, among other things, background information or assumptions about the nature of the conversation and the interests of interlocutors (for example whether an utterance occurs in the course of small talk or in the course of a graduate seminar), but also by lexical choice. The effect of using a phrase like ‘London, you know, the geographical location’ is to increase the precision of the context. The meaning of London, the body of knowledge, is in some sense receptive to context. Changes in context, or more specifically, increased precision in context, requires more of the body of knowledge to be attended to. Most conversations about London, in fact most conversations generally, I suggest, are fairly imprecise and only very basic information in the structured body of knowledge which is the meaning of ‘London’ needs to be accessed, with the result that the different senses are not distinguished from one another. The extensions of an utterance of ‘London’ in such a context are all of the different Londons (the group of people, the political entity, the geographical location, etc). The interpretation of that utterance of ‘London’ is the set of all the Londons, and as per the first of the truth-conditions offered above, ‘London is populous’ is true if and only if this set is an element in the set by which ‘populous’ is interpreted.

But what happens if the interlocutor says ‘but by ‘London’ I didn’t mean the people who lived in London’? Typically, this would lead to a precisifying of the context. The result would be a greater attention to the minutiae of our body of knowledge, the meaning of ‘London’ and, in result, the need to distinguish London from the people of London. The population of London is now listed separately in the specification of the domain of discourse and our next utterance of ‘London’, in this new context now has a slightly reduced extension. The interpretation of ‘London’, which determines the truth or falsity of sentence utterances in this context is the set of all the Londons, except for the group of people. This process can go on of course. We may find ourselves in a maximally fine-grained context where all the different Londons are separate elements in the specification of the domain and our utterance of ‘London’ is picking out only one. The movement between contexts also does not follow a pre-ordained pattern. Different items may be kicked out of the interpretation of ‘London’ in different sequences, either one at a time or in groups. The key idea is that the body of knowledge which is the concept London determines how the extension of the lexical item ‘London’ is restricted in a given context to give us the extension of an utterance of ‘London’ in this context.

I believe this account has a number of salient advantages. I will finish by listing three. First, I think it is compatible with (E), the externalist thesis that the meaning of a sentence determines its truthconditions, in any given context. Moreover, although my account is an example of a “rich common core” account, it avoids Borg’s concern that such accounts conflict with the spirit of externalism because they “shift the point of connection between language and the world from standing elements of the lexicon...”. On the account just offered, there is a direct link between every entry in the lexicon and the world. Polysemous type-words are mapped onto a set of all the objects to which they refer. The role of the rich common core is to determine whether context reduces that set, in a context, by identifying some of the polysemously-picked-out items to stand as separate elements of the specification of the domain, requiring a different term to refer to them. In some contexts, the less precise ones, the whole extension of the lexical item will contribute to making the sentence true. As precision grows, the extension of the utterance of the lexical item will become a smaller and smaller subset of the lexical item’s extension. I hold that this is still an externalist account.

Second, I think this account offers a natural approach to explaining why some copredicative sentences seem acceptable and others seem unacceptable. We have looked at a number of acceptable examples already. A popular apparently unacceptable example is the sentence ‘the newspaper fired its editor and fell off the table’. This sentence has a strong zeugmatic effect. What explains this effect, when sentences like ‘the book is heavy but informative’ seem unexceptional? The short answer, on my account, is that something in the first sentence causes a precisifying of context, such that we attend to a wider range of the information in the body of knowledge connected with the word ‘newspaper’ and the result is that newspaper businesses and newspaper copies are forced to be distinguished from one another. What is it about the first sentence? I doubt that the presence of either ‘...fired its editor’ or ‘...fell off the table’ on their own guarantee a precisification of context. The sentence ‘I edited the book that just fell’ seems unexceptional to me. However, the meaning of the word ‘newspaper’ seems more receptive to precisification than ‘book’, as there is something perhaps very slightly odd about ‘I edited the newspaper that just fell’ and more so ‘I was fired as editor of the newspaper that just fell’. However, I suggest that neither of these are as zeugmatic as ‘The newspaper fired its editor and fell off the table’. The reason, I think, is that sentence structure as well as lexical choice contributes to context, and that the effect of starting the sentence with ‘The newspaper’ is to contribute to a precisification of the context with respect to newspapers. In short, I think it is a combination of features of the sentence, both lexical and structural which mean that any utterance of it will force a precisification of the context. Full explanations of zeugma would need to be provided on a case-by-case basis, but it should be clear why they happen on my account.

A similar resolution immediately presents itself with respect to the issue Vogel has raised about the unique inferential profile of acceptable copredicative sentences. Once again, ‘London is populous, but tends to vote conservative’ would seem to entail ‘London is populous’ and ‘London tends to vote conservative’, but it would seem not to entail that ‘Something is populous, but tends to vote conservative’. Vogel thinks this is a problem for (E) in particular, but my account would explain our intuitions about ‘Something is populous, but tends to vote conservative’ in the same way it explains our intuitions about ‘The newspaper fired its editor and fell off the table’. Some feature or features of these sentences contribute to precisifying the context. In the former case, it would seem to be the use of the quantifier ‘something’, which would seem to force a precisification of context. In an imprecise context, ‘Something is populous, but tends to vote conservative’ is entailed by ‘London is populous, but tends to vote conservative’, given the nature of interpretation laid out earlier in this section and the natural truth conditions for existentially quantified sentences. However, this is perfectly compatible with the fact that one could never truly express ‘Something is populous, but tends to vote conservative’, even if it is true in a maximally imprecise context, because an utterance of the sentence itself precisifies the context, and the sentence is always false in more precise contexts. I suggest that this response is available to explain all the unique features of the inferential profile of copredicative sentences. I take this to be another advantage of the account offered here.

4

Polysemy is a contextual phenomenon in a number of different ways, given the phenomenon of copredication. First, whether a polysemous term activates one or more than one sense at the same time is a matter of context. Second, which sense or senses are activated is a matter of context. The meaning of a lexical item determines its extension(s) in a given context. In imprecise contexts, polysemous terms may pick out a wide range of different objects in the world at the same time, but these objects represent a single element in the “specification” of the domain of discourse for this context. More precise contexts involve more of the entities that have been grouped together as single elements in the less precise specifications of the domain being separated out as independent elements in more precise specifications. These more precise specifications allow us to interpret terms more precisely in more precise contexts. Uses of specific words, phrases and sentence constructions can all contribute to changing the context, often serving to precisify it. This explains some of the curious features of copredicative sentences that have been discussed in the literature.