Show Summary Details

Page of

PRINTED FROM OXFORD HANDBOOKS ONLINE ( © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice).

date: 24 April 2019


Abstract and Keywords

Polysemy is an interesting phenomenon that concerns cases in which a word or phrase enjoys multiple, related meanings. This article distinguishes polysemy from similar phenomena and presents some tests for determining the presence of polysemy. In addition, polysemy is differentiated from other phenomena that involve potential multiplicity of meaning. Later in the article, a few potential cases of polysemy are explored. The final two sections deal with the (so-called) polysemy paradox and consider ways in which types of polysemy can be characterized and categorized. Concepts are outlined with the use of several examples, allowing polysemes and ambiguities to be examined in context.

Keywords: polysemy, ambiguity, semantics, meanings, polysemes

"Nous apellerons ce phénomène de multiplication [de sens] la polysémie"

Michel Bréal, 1897


It’s tempting to weasel out of the job of writing this article by saying that “polysemy” is roughly the same in meaning as “ambiguity” and then providing a link to an entry on “ambiguity.”1 The two words seem to converge on a meaning, roughly “multiple meaning” (though “ambiguity” derives from something close to “double meaning”). Normal usage would likely conflate the two if the term “polysemous” enjoyed anything close to the sort of popularity required to determine normal usage. Nonetheless, I resist the temptation to pass on writing this article, as there are interesting conceptual distinctions between the two, at least in technical usage, that are worth highlighting.

This article examines various aspects of the phenomenon of polysemy. Part 1 concerns issues in defining ‘polysemy’ and distinguishing it from ambiguity. Part 2 presents and evaluates some tests for polysemy. Part 3 further distinguishes polysemy from other phenomena that involve potential multiplicity of meaning. Part 4 explores a few potential cases of polysemy. Part 5 deals with the (so-called) polysemy paradox, and part 6 considers ways in which types of polysemy can be characterized and categorized.

Part 1: Premable

In the parlances of socio and cognitive linguistics, ’polysemy’ is often used in a restrictive way to characterize multiple related meanings expressed by a single word, phrase, or, more generally, a symbol. This characterization is pretty rough and contentious—it reifies meaning, uses the vague term “related,” and relies on a dubiously clear notion of word/phrase/symbol. The concerns about words/phrases/symbols are of particular importance. Consider the sentence:

1. Dusko wants to buy a big dog treat.

(1) is apparently polysemous. Even ignoring the specific and indefinite readings associated with the indefinite NP, (1) is associated with several distinct truth conditions, such as:

  1. 1a. Dusko wants to buy a dog treat that is large.

  2. 1b. Dusko wants to buy a treat appropriate for large dogs.

The truth conditions are closely related in any reasonable sense, and so, (1) looks like a good case of a polysemous sentence, with the phrase “big dog treat” being the reasonable culprit. That is, if we treat “a big dog” as a single string. However, syntactic analysis reveals two distinct linguistic structures associated with the string “a big dog”:

  1. 1a’. Dusko wants to buy [NP[DETa [APbig [NP dog treat]]]]

  2. 1b’. Dusko wants to buy [NP[DETa [[APbig dog] [NP treat]]]]

If we individuate phrases by their symbols in a language, “a big dog treat” is one string. If we think of a phrase as a pairing of symbols and (at minimum) a syntactic sturucture, then “a big dog treat” is really two distinct phrases that look like one; the apparent polysemy is generated by a syntactic ambiguity. No harm results so long as we keep careful watch on what we mean by a “phrase.” The interesting features of polysemy involve cases in which the two meanings are related by some core meaning; the ambiguity in (1) results from the structure of the sentence rather than from variation on a core meaning. We will consider polysemy as it applies to syntactically analyzed strings.

Despite the shortcomings, the characterization makes for a good starting point and an initial way to distinguish polysemy from ambiguity. The latter concerns cases in which the meanings of a term may well be highly disconnected, as in the famous sorts of cases such as:

  1. 2. John ran the marathon. (John participated in the marathon/John was the manager of the marathon)

  2. 3. Kissing dogs can be dangerous. (Dogs that are kissing can be dangerous/The act of kissing a dog can be dangerous)

  3. 4. Every dog is happier than its owner. (Every dog is happier than the owner of that dog/Every dog is happier than the owner of some salient object)

By contrast, here are some paradigmatic cases of polysemy (illustrated using related groupings):

  1. 5. John was in a bad mood.

  2. 6. Mary was in a sedan.

  3. 7. Yanis was in a lot of trouble.

  1. 8. Branka walked to the Yugo.

  2. 9. Branka walked Lana to the Yugo.

  1. 10. Ozzy took a drag from the pipe.

  2. 11. Ozzy took a million dollars from the safe.

  3. 12. Ozzy’s court case took a lot of time from her day.

  1. 13. The Danube’s mouth is full of water.

  2. 14. Mara’s mouth is full of water.

There are many more such cases, but (5)–(14) can help give us a first-pass fix on the concept. We should proceed with care, however. While “in” in (5)–(7) contributes a notion of containment, it is not clear that philosophers would be sanguine in the claim that “in” expresses multiple meanings, rather than that its meaning is underspecified and becomes specified within the context of the sentence (see part 3.2). As a matter of logic, only one of the following inferences seems analytically valid:

Bob is in a bad mood.

Part of Bob overlaps a part of a bad mood.

Bob is in Kenya.

Part of Bob overlaps a part of Kenya.

Bob is in a lot of trouble.

Part of Bob overlaps a part of a lot of trouble.

From the standpoint of logic, it doesn’t really matter whether or not the meaning of “in” in its locative sense is suitably related with “in” in its non-locative sense. It merely matters that the locative sense of “in” enjoys a different representation than the others.

In (8)–(9), the thematic structure of “walk” is altered to include a theme in (9), but the meaning of “walk” in both cases involves a common pediatric core. “Took” in (10)–(12) is an interesting case. In (10), “taking” involves an action in which the pipe is acted upon but not by way of transference. In (11), there is a clear notion of transference—the million dollars is transferred from the safe. In (12), the subject is at best metaphorically the recipient of time. Nonetheless, there seems to be a fairly common core meaning to “take” in each—the subject affects a loss from the referent of the indirect object of the sentence. Finally, in (13) and (14), both the Danube and Mara have mouths, but the river mouth seems to get its sense by transformation from the notion of a mouth as it applies to animated objects that consume.2

Notice that some terms are both ambiguous and polysemous:

  1. 15. Donatien climbed onto the bank after swimming.

  2. 16. Donatien went to the bank.

  3. 17. Donatien banks with Justine.

  4. 18. You can bank on Donatien to see Justine.

The interpretation of “bank” as a riverside comes to English from the Norse term “banki,” meaning ridge. This meaning is etymologically unrelated to its meaning as a type of financial institution, which comes from the Latin”bancus” via the Normans, who used the term “baunk” to mean bench (which served as a reliable location of money handlers in the marketplace). The verb “bank,” meaning reliable in some respect, is related by extension to the verb “bank” meaning to provide the services of a bank, which derives from the noun. Thus, it seems that “bank” is both polysemous and ambiguous, depending on how one clusters the meanings associated with “bank.”

There are many projects in philosophical semantics to which the difference between ambiguity (in the sense of converging unrelated meanings) and the more restrictive notion of polysemy is simply irrelevant. Multiplicity of meaning is multiplicity of meaning, and any project committed to clear and precise representations of meaning probably shouldn’t be troubled by the fine points of just how related the multiplicity of meanings are. And in general, taxonomy based on the relatedness of distinct meanings is a pretty dull affair for anyone but the committed lexicographer. However, philosophers, linguists, psycho-linguists, and advertisers, interested in the linguistic processes by which words acquire new and interesting meanings and extensions of meanings, should be very interested in the distinction. Commonalities of meaning may sometimes be adventitious, but often polysemy is reflective of transformations of that produce a new meaning for a term from a former meaning. These transformations, if they can be characterized in a theoretically productive way, may offer some insight into the production of new representations from older ones (see part 6). Alternatively, in some cases, it may offer us insight into how to think about terms that are underdetermined with respect to meaning. It would be interesting to know what someone knows, for example, when they know the meaning(s) of the word “in” such that they can interpret (5)–(7) appropriately. This may help us get a better grasp on the appropriate notion of “related” meaning in the characterization of “polysemy” that we offered initially.

A few notes of caution before proceeding. First, etymology is not a perfectly reliable arbiter of polysemy. A meaning that is etymologically related to another need not be appropriately similar to the initial meaning, as the two may drift apart over time to the point that they are no longer suitably related. An example of this is “cardinal”—its ornithological and ecclesiastical interpretations turn out to be historically related, as the color of the bird was reminiscent of the robes of the ecclesiastical prince.3 Furthermore, a word may acquire a new sense because, historically, one group misunderstood another’s use of the term.

Second, it is arguable whether or not polysemy should be seen as coming in degrees.4 A bad argument for this conclusion goes as follows: polysemous terms enjoy related, similar meanings. Similarity comes in degrees. Therefore, polysemy comes in degrees. But the gradeability of a phrase does not follow from the gradeability of a condition for the correct application of the phrase. “Giant” doesn’t seem gradeable even though it is a necessary condition for being a giant that one is tall.5 But whether or not polysemy comes in degrees, there are definitely cases in which it is vague whether or not two meanings for a term or phrase constitute polysemy. It will be often hard to tell whether or not a term t is merely ambiguous or polysemous.

Third, we will have to make some simplifying assumptions as we proceed. We will not worry about reification of meaning nor about how fine or coarsed grained meanings. We will not worry (too much) about the semantics-pragmatics distinction except where necessary. In particular, I won’t fuss too much over whether in any given case we can determine between a putative meaning being semantically determined or pragmatically calculable. We will worry mostly about what we can use to test for polysemy and how we might try to account for it.

Relatedly, we will not be concerned with questions regarding whether or not metaphors and metonymy generally are to be treated as legitimate meanings or derived ones, except where necessary. Thus, the fact that “leaf” can be used to refer to (roughly) memory in the following passage:

  • Kind gentlemen, your pains,
  • Are register’d, where every day I turn
  • The leaf to read them (Macbeth)

isn’t sufficient to show that “leaf” is polysemous. The fact that “the ham sandwich” can be used to refer to a customer who ordered a ham sandwich will not be sufficient to show that “the ham sandwich” is polysemous (Nunberg (1995)). Hopefully a successful theory of polysemy will shed light on these and related processes. The success of a theory of polysemy won’t depend on it.

Part 2: Testing for polysemy

Linguists and philosophers have developed various tests for polysemy and ambiguity.6 Most of these tests involve attempts to “freeze” a single meaning of the putatively polysemous phrase and then see if that frozen meaning can be used to express the multiple meanings. Unfortunately, this makes most tests unsuitable for distinguishing polysemous phrases from merely ambiguous phrases, since both enjoy multiple meanings.

A note on the tests: these are tests, not knockdown arguments. Most of them require a somewhat slippery notion of interpretation, namely: Can the sentence in question express multiple meanings? Of course, with some creativity and perhaps some practice, competent language users can find themselves detecting meanings in unexpected places. What the tests presuppose is that the reader can access something like interpretations of a sentence that aren’t deviant. Or, to use a non-technical term, the test presupposes that the test participants don’t engage in shenanigans.

This would all be good and well if we could reliably spot shenanigans. Since we often can’t, sometimes we will have to agree to disagree (sometimes with our former selves) about what the test shows and how reliable we take the test to be. The tests concerning lexical meaning are only so accurate.

2.1 Conjunction reduction

One way to test for polysemy is to conjoin phrases under the scope of the ambiguous expression. Here’s an example of conjunction reduction concerning the non-polysemous case of “skateboarded”:

  1. 19. John skateboarded to work.

  2. 20. John skateboarded to the bar.

  3. 21. John skateboarded to the store and the bar.

(21) is the conjunction reduction of (19) and (20). We will illustrate by example with a non-polysemous case and a polysemous one. “Uncle” can apply to brothers of one’s mother(s) or father(s). If we want to test whether “uncle” is thereby ambiguous, we can use conjunction and a single use of “uncle” and see if both senses are expressible, as in:

  1. 22. John and Bill each have an uncle.

(22) seems to be true relative to a case in which John‘s mother (but, let’s say, not his father) has a brother and Bill‘s father (but not his mother) does too. This provides evidence that the meaning of “uncle” is unspecific with respect to which parent’s brother is in question, as opposed to polysemous. By contrast, consider a paradigmatically ambiguous case such as “credit.” (23) seems incapable of expressing the distinct senses:

  1. 23. John and Steve each got credit from the bank for their work.

(23) can mean that both John received recognition for his work and Steve did, too, or that each one received financial compensation for his work but it cannot (sans shenanigans) mean that one received compensation and the other received recognition. This is easily explained—there is only one occurrence of “credit” and it is incapable of contributing two semantic values to a single semantic composition.

A version of conjunction reduction can be used to test for some cases of polysemy. One can use certain predicates to help privilege the different meanings under conjunction, and non-shenanigan interpretation should be hampered. The relevant property of hampered interpretation is sometimes known as “zeugma.” For example:

  1. 24. Sarah was in a bad mood, a sedan, and a lot of trouble.

(adapted from Ryle (1949))

(24) sounds like a bad joke—a paradigm of zeugma. You’d probably know what someone meant if they asserted (24), but it would take some shenanigans. It’s not nonsense, but it is difficult to give it a coherent “straight” interpretation. Similarly for the other cases:

  1. 25. Ozzy took a hit from the pipe and a million from the safe.

  2. 26. The Danube’s and Branka’s mouths are full of water.

So far, so good.

A problem for the conjunction reduction test involves the sensitivity of zeugma judgments to other factors in the sentence. As noted by Lewandowska-Tomaszczyk (2007) (following Cruse 1986), the following two are different with respect to zeugma:

  1. 27. Judy’s dissertation is thought provoking and yellowed with age.

  2. 28. Judy’s dissertation is still thought provoking although yellowed with age.

The difference between “and” and “although” seems to concern the implied discourse relations that hold between the two predicates. “And” suggests parallelism, while “although” suggests contrast. In other words, conjunction reduction may well give misleading results depending on the conjunction used. Similar considerations appear in the literature concerning generics and habituals. It’s tempting to think of “bees” as polysemous between an individual-level interpretation (particular bees) and a kind-level interpretation. Clearly these two interpretations are related but different. Nonetheless, one apparently fails to get obviously zeugmatic interpretations from conjoining predicates that apply to different levels, such as “thrive in warm environments” (type-level predication) and “are swarming my porch” (individual-level predication):

  1. 29. Bees thrive in warm environments and are swarming my porch.

These cases look like a problem for the conjunction reduction test, depending on how one thinks we should treat the putative polysemy in generics. One might think that this provides evidence against ambiguity in bare plurals and for attributing the apparent polysemy to the predicates. For contrast, consider:

  1. 30. Bees are suffering colony collapse and stinging my dog.

But as an initial diagnostic, it works pretty well. One has to be careful in applying the test to ensure that no other sources of zeugma are present. The best remedy is to use several tests and see how the results come out. Don’t settle for one!

2.2 Ellipsis and gapping

One way to test for ambiguity is to employ ellipsis or gapping where appropriate. Ellipsis is best explained by example:

  1. 31. Sulla was a dictator and Caesar was too.

  2. 32. The Huns invaded the Roman Empire and the Goths did so too.

“Too” and “did so too” express “a dictator” and “invaded the Roman Empire,” respectively.

Gapping provides a means of eliding over certain verbs by leaving them phonologically null. For example:

  1. 33. Sam took the train and Barbara the ferry.

  2. 34. Barry gave Diana an olive and Sarah a peach.

Notice that (34) is ambiguous between an interpretation according to which Sarah received a peach from Barry and one according to which Sarah gave Diana a peach.

Ellipsis can and is used to test for ambiguity. Elliptical phrases seem to retain the semantic value of the phrase they elide, so use of ellipsis seems to prevent the range of interpretations that a non-elided version of the sentence would allow. Compare the two interpretations of ‘run’:

  1. 35. William tried to run the Boston marathon and Brooke did too.

(35) is interpretable (without shenanigans) as

  1. 36. William tried to participate in the Boston marathon as a runner and Brooke tried to participate in the Boston marathon as a runner.

or as:

  1. 37. William tried to be the head organizer of the Boston marathon and Brooke tried to be the head organizer of the Boston marathon.

It is hard to get the following interpretations without shenanigans:

  1. 38. William tried to be the head organizer of the Boston marathon and Brooke tried to participate in the Boston marathon as a runner.


  1. 39. William tried to participate in the Boston marathon as a runner and Brooke tried to be the boss of the Boston marathon.

Finally, while it would be a bit misleading to put it this way, (40) enjoys all the meanings expressed by (35)–(39):

  1. 40. William tried to run the Boston marathon and Brooke tried to run the Boston marathon.

Admittedly, (40) is a weird way to express things, because the redundancy is bound to pique the reader’s curiousity as to why the speaker put things that way rather than using ellipsis (suggesting a violation of Grice’s maxim of manner). But the basic idea is that the two phrases are not semantically linked.

This is easily explained on one assumption:

Preserve: The semantic value of the ellipsis copies is identical to the meaning of the elided phrase.

There are issues concerning the best way to make Preserve precise.7 One particular issue is that the eliding term can mask a fair bit of syntactic structure. But the intuitive idea is clear—a phrase contributes a semantic value and Preserve ensures that it is the same one that is contributed by the elliptical phrase.

Syntactic considerations prevent the use of ellipsis to test some cases of polysemy. There are constraints on ellipsis—in particular, that one cannot elide a predicate without its arguments. For example, the following is ungrammatical:

  1. 41. *William tried to run the Boston marathon and Brooke did so the New York Marathon.

Nonetheless, we can use gapping to test for polysemy:

  1. 42. William ran the Boston Marathon and Brooke the New York Marathon.

(42) does not allow for cross-interpretations. If we use predicates that try to force cross-interpretation, we should expect mild zeugma. We may have to get a little fancy for some of the cases, especially where a preposition is the target of analysis.

  1. 43. Yanis is in Paris and Alexis Athens.

  2. 44. Yanis is in a bad mood and Alexis an angry mood.

  3. 45. ?? Alexis is in a sedan, Jerry a hatchback and Yannis a bad mood.

(45) pretty clearly shows that one cannot co-ordinate the psychological and locative notions of “in.” Similarly for “took”:

  1. 46. Ozzy took a perilous path and Kelly an easy trail.

  2. 47. Ozzy took a million dollars and Kelly a silver spoon.

  3. 48. ??Ozzy took a perilous path and Kelly a silver spoon.

We have to get even fancier and use gapping on possessives for cases like “mouth”:

  1. 49. The Danube’s mouth holds more water than the Mississippi’s.

  2. 50. ?The Danube’s mouth holds more water than Mara’s.

The applicability of these tests is obviously limited by the fragility of our judgments. And unlike standard ambiguity cases, the interpretations in polysemic cases are, by definition, more closely related, which in some cases may make distinguishing them more difficult.

2.3 Contradiction tests

One striking way to tell whether or not a phrase has multiple meanings is to try to force a coherent interpretation by priming a contradictory interpretation. For example, consider whether or not (51) has a non-shenaningan, non-contradictory interpretation:

  1. 51. Brooke ran the Boston marathon, but she didn’t run the Boston marathon.

(51) is clearly interpretable as non-contradictory in English, albeit a misleading way to put things. Compare:

  1. 52. Kenny is my uncle but he’s not my uncle.

Try it: we predict that only shenanigans will avail you of a non-contradictory interpretation of (52) 8

It’s a little harder to run this test in some of our paradigm cases of polysemy. But let’s try. Consider (and please avoid shenanigans by treating “trouble” as a relative term):

  1. 53. Yanis was in trouble but he wasn’t in trouble.

We can hear (53) as non-contradictory, but one has to pay attention. The hard part is to make sense of what “in trouble” could mean if we force the “in” to mean the same thing it does when it precedes “a sedan.” That reading is absurd, of course, but (I think) not incoherent. Similarly:

  1. 54. Ozzy took a million dollars but he didn’t take a million dollars.

  2. 55. Ozzy took a hit from the pipe but he didn’t take a hit from the pipe.

These are harder, but we can get the non-contradictory meaning. However, it is more difficult to detect the presence of shenanigans in this case. Finally:

  1. 56. Branka’s mouth is full of water but her mouth isn’t full of water.

It is not easy to tell in (56) whether or not we can recover a non-contradictory interpretation. It’s easiest to do so if one imagines a strange case in which Branka somehow manages to serve as the wide entry point of a river.

The contradiction test is not without its problems. But it’s noteworthy that the contradiction test attempts to do the opposite of the other two tests—instead of trying to isolate one meaning of a phrase, it attempts to force an interpreter to take two occurrences of the word and provide the distinct meanings.

Part 3: What polysemy might not be

It is useful to distinguish polysemy from related cases of multiple meanings, and to look at a few cases in which polysemy has been invoked to explained cases that are apparently non-polysemous. It will be a highly selective exploration, sadly, as polysemy has been called in to save the day in many cases, especially by philosophers eager to notice polysemy everywhere. We will take heed of Kripke’s (1977) dictum about ambiguity:

It is very much the lazy man’s approach in philosophy to posit ambiguities when in trouble.

It’s equally lazy to posit polysemes when in trouble. We will look at some phenomena that can be easily confused for polysemy and consider some reasons against assimilating them.

3.1 Context sensitivity

It’s a well-established fact that some terms require contextual input in order to fix their reference. The classics (“I,” “here,” “now”) aside, terms like “coming” and “going” require a reference point supplied by context that can shift around depending on how the speaker’s location or a relevant reference point of the discourse proceeds. That’s why “come over” makes sense as an invitation, but “go over” doesn’t, unless the speaker plans to be at the location when the addressee gets there.

These terms shift their reference fairly easily, but they aren’t polysemous, at least not on this count—though both have non-locative senses that make them ripe candidates for polysemy and ambiguity! They are context sensitive. Other cases, however, are harder to tell. Ram Neta (2007) points out that there is contextual variation in what it takes to see someone. Consider (57):

  1. 57. I see Mt. Rushmore

In some contexts (57) seems true if one sees the very top of Mt. Rushmore and recognizes it. In other cases, (57) is false unless, say, Lincoln’s face is clearly visible. This seems to pattern with context sensitivity, but it isn’t obvious that “see” isn’t ambiguous or polysemous between “a sign that identified the object was in the visual field” and “a relatively large proportion of the object’s surface was in the visual field.” The problem for us is that our tests from section 2 will render nearly all context-sensitive cases as polysemous, but for reasons of contextual reference shift rather than multiple meanings. For example,

  1. 58. I saw Mt. Rushmore but I didn’t see Mt. Rushmore.

We certainly can interpret a non-contradictory reading of (58) but it’s troubling that one may need to monkey around with “see.” Focal stress applied to “see”, for example, is not obviously semantically innocent. Such is the nature of these tests: they are helpful, but they are subject to all the contextual and interpretive effects that infects the study of language generally.

3.2 Underdetermination

Ludlow (2015) offers a nice case to motivate the idea that sometimes a word’s meaning is underdetermined. ESPN rated Secertariat the 35th best athlete of the 20th century. One sports radio station raised the (philosophical) question as to whether or not it was right to think of horses as athletes, or if being human was a necessary condition of being an athlete. I don’t really know how to settle this, and when I consult my concept “athlete,” no clear answer pops out. Ludlow and others have been inclined to claim that the word “athlete” is simply underdetermined—there is no good answer to the question “Is Secretariat an athlete?” that can be settled outside of discourse participants making localized semantic decisions (rather than discoveries) regarding the term “athlete.” The word isn’t sufficiently fixed in meaning for Secretariat to be either in or out of its extension without further contextual constraints. Different conversations may modulate the meaning of “athlete” to include Secretariat, while others will not; both are consistent with any core meaning of “athlete.”

Whether or not Ludlow is right about underdetermination, it’s important to separate the type of phenomenon he attempts to characterize from polysemy. The idea isn’t that “athlete” has several meanings, nor that “athlete” is vague. The idea is that “athlete” and related terms have (at most) a meaning that is underdetermined and is modulated by discourse participants. This is presumably a different phenomenon than that exhibited by the clearly distinct senses of “in.”

3.3 Purpose-driven narrowings of extensions

Consider the noun “knife.” “Knife” may not strike one as particularly polysemous, even if it is vague and perhaps underdetermined. (I’m not sure whether or not a sword counts as a knife, but apparently bayonettes do.)9 If at dinner someone asks me for a knife and I pass them a bayonette, they will likely consider their request unsatisfied. On the other hand, responding to the same verbal request with a butter knife when clearing thick brush would meet with a similar level of annoyance. This may well appear to be evidence for the claim that “knife” is ambiguous—fans of Crocodile Dundee will remember Paul Hogan pulling out a bowie knife after being threatened with a switch blade and (in line with the contradiction test!) claiming “That’s not a knife, THIS is a knife!”

It’s tempting to claim that “knife” is polysemous, but at a cost—just about every non-functional word will end up being polysemous. Nearly any substantive term has satisfiers that are only contextually appropriate satisfiers in a range of contexts. That’s why the genie-who-grants-wishes-in-inappropriate-manners trope is a trope. But whatever is transpiring in the knife case seems to be different than the case of “in” or “took” or “mouth.” In the knife case, it seems that we should understand requests for knives as coming along with certain expectations about how the request will be honoured given the goals at hand. It’s not that “knife” ceases to refer to bayonettes at a dinner table, it’s that satisfaction of requests and commands are goal relative. In other words, you don’t change the meaning of a term by miscomprehension or by being a jerk. (See Fara (2000) for an application of this line of thought to vagueness.)

The case does reveal something interesting, however—there is often no easy way to reconstruct what a speaker meant to communicate from merely knowing the meaning of the sentence they expressed. I may know that Alice asked for a knife, but with only that in mind, I may well be unable to know what range of things she would have acknowledged as adequate satisfiers of her request.

3.4 Granularity

An interesting and systematic (seeming) polysemy corresponds roughly to the type-token distinction that philosophers cherish, though it is more general. Philosophers have noticed that (59) is ambiguous between a type and a token reading:

  1. 59. We own the same car.

(59) can express the fact that each of us owns a distinct car that is similar with respect to some kind, or that we are both co-owners of one single car. How closely they have to correspond in similarity is an open question. But interestingly, the two senses cannot always be accessed felicitously:

  1. 60. ?My car skidded on ice and hit the same car.

One cannot read (60) as saying, say, that my Honda hit another Honda. It’s tempting to think that “same” is the culprit, allowing for sameness across different levels of grain, from the very fine to the very coarse.10

Part 4: Putative cases of polysemy

4.1 Deferred reference

Deferred reference occurs when one uses a phrase that prima facie refers to one thing to refer to another. The example of ‘the ham sandwich’ to refer to a customer is one case of this phenomenon. Nunberg (1993) provides the example of a condemned prisoner who utters:

  1. 61. I am traditionally allowed to order whatever I like for my last meal. (p. 20)

“I” here seems to refer to someone in the prisoner’s position, not the prisoner herself, as there are no traditions regarding her. Similarly, if there is always a party after the big game in a college town, one might utter the day before:

  1. 62. Tomorrow is always the biggest party day of the year.

(62) is pretty clearly incoherent on the view that “tomorrow” is directly referring to a particular day after the day of the context.

The semantics-pragmatics distinction matters here. Kaplan (1989) and Recanati (1993) deny that examples such as (61) and (62) display anything of much semantic interest, preferring instead to call on pragmatic resources to explain the relevant readings. This isn’t the place to settle the issue, and it is a complex one; but for our purposes it seems clear that interpreting “I” and “Tomorrow” in (61) and (62) require supplying a meaning that is highly related to the traditional indexical meaning assigned to “I” and “tomorrow.”

4.2 Pros hen

Aristotle notes an interesting systematic type of multiplicity of sense that he labels “pros hen” in the Metaphysics:

There are many senses in which a thing may be said to ‘be,’ but all that ‘is’ is related to one central point, one definite kind of thing, and is not said to ‘be’ by a mere ambiguity. Everything which is healthy is related to health, one thing in the sense that it preserves health, another in the sense that it produces it, another in the sense that it is a symptom of health, another because it is capable of it.

(Metaphysics Γ.2)

The idea is something like the following: a healthy person produces healthy urine, but the urine is not healthy in the same sense that the person is. We can imagine urine that is healthy—if we found out that some urine was alive and could be in distinct stages of health. But in the typical case, healthy urine is urine characteristic of the urine produced by a healthy person.

“Healthy” seems to offer a paradigm case of polysemy—“health” has a core meaning (in the sense of “healthy person”) and related meanings that seem pretty clearly derived from the core meaning. Notice that the tests confirm the multiple meanings:

  1. 63. ?His diet and his urine are healthy.

  2. 64. ?He’s healthy and his urine is too.

  3. 65. His urine is healthy but his urine isn’t healthy.

(63) displays zeugma—it is quite odd on a univocal interpretation. (64) is odd because of the comparison of a person and urine, but putting that aside, it is also odd on a univocal interpretation. (65) is coherent if one imagines urine being considered as the product of a person and then as, say, the diet of a person. It’s interesting in this regard to consider that there are constraints on pros hen ambiguity that are surprising. “Unhealthy” seems to enjoy the same range of readings as “healthy,” but “sick” and “ill” do not—compare “unhealthy diet” with “sick diet” and “unhealthy urine” with “ill urine.” Similarly, there is no easily accessible notion of “tall diet” or “tall urine” in the sense of diets and urine indicative of or conducive to becoming tall. Whatever processes allow the extension of “health” into derivative notions is not one that is easily applicable to all adjectives.

Part 5: The polysemy “paradox”

Theorists worried about polysemy have (mis)labeled several questions regarding the nature of polysemy “the polysemy paradox.” Taylor (2003) writes:

The paradox is that, whereas polysemy raises all kinds of theoretical and methodological issues for semanticists, and practical issues for lexicographers and for workers in natural language processing and automatic translation, speakers of a language rarely experience polysemy to be a problem at all.

(Taylor 2003b, p. 647).

This may not strike one as much of a paradox; but it does raise an interesting question. If the model of linguistic understanding runs on the basis of assigning interpretations to eachother’s utterances, then a theory that posits polysemy too freely will have a hard time explaining why we don’t find the polysemies hard to resolves. The “paradox” only has bite if polysemy is ubiquitous. So, one reasonable question to ask is just how ubiquitous polysemy really is. Very similar questions arise with context-sensitive language, ambiguity, metaphor, simile, metonymy, etc. But it’s at least reasonable to think that there will be different answers for the question each linguistic phenomenon presents for communication.

We can separate the relevant questions into two distinct questions. First, what is the right theoretical framework within which to theorize about polysemy? Second, how do linguistic agents represent polysemous meanings—as three separate lexical entries? As one entry with an underdetermined meaning? As one entry with a core meaning that is subject to meaning-altering operations?

The questions may have very different answers, but we can try to see how the answer to the first sheds some potential light on the second. Traditional semantics treats meanings as assignments to representations and thus, different meanings require either different representations or treating representations as having their semantic contributions determined by some further factor (i.e., by the context of utterance, intentions of the speaker, syntactic context, etc.). On this view, one is fairly constrained to thinking of polysemy in terms of distinct lexical entries that are perhaps derived from one representation designated as “core” but that are distinct, or as a single representation that in context received a fixed value. Thus, we might think of “in” as having the physical containment sense as a core from which the other meanings are somehow derived. Alternatively one may try to represent the meaning of “in” as a given by a representation that locates something within a structured set of points where the interpretation of the strucutrue is given by context (often suggested by the complement of the preposition).

On the other hand, theorists in the Cognitive Semantics literature have found various ways to reject key ideas of the traditional theory.11 One type of approach treats meanings as a matter of prototypes or exemplars or underdetermined theories, where the traditional notion of something falling under a meaning involves a matter of similarity between the prototype and the object under consideration. Thus, if we think of the meaning of “dog” as a set of dog-like properties (typically four legged, barking creatures that make good pets), we can explain (partly) how people judge things to be dogs or non-dogs and why there are difficult cases where the animal in question has a number of prototypical dog features but not others.12 Add to this that the network of relations and their importance can vary in systematic or contextually adventitious ways, and one may try to explain polysemy not as a derivation of one sense from another but as a matter of the availability of different similarity metrics that one can exploit. If “mouth” is treated as expressing a prototype, then perhaps the notion of the “mouth” of a river can fit under the relevant concept by highlighting the similarity of entry of water into the river to an entry of a body designed to take in substances. On this view, the notion of polysemy as multiple related senses becomes somewhat blurred, as it is not clear how one ought to individuate senses. A more fine-grained approach risks collapsing into the traditional view.

In any case, neither of these approaches about the nature of linguistic representation and polysemy seem to offer an answer, on their own, to the question of how in the face of large amounts of polysemy we manage to communicate using polysemous words and phrases. The answer, so far as I can tell, is still elusive.

Part 6: Categorizing polysemy

We have looked at a variety of cases of polysemy without much regard to how to categorize them. It’s worth noticing that polysemes seem to come in different varieties, even in the examples we have looked at so far. One key difference we may use is the difference between systematic and idiosyncratic polysemes. The former seem to involve cases like the ones we’ve seen and then some. Here are some examples of things that fall in the category of systematic polysemy:

  • Individual and kind interpretations

  • Count and mass noun

  • Locative prepositions

  • Institution and institution location (UC Davis raised its fees, UC Davis is in Davis)

  • Person and position (the Prime Minister is in Ottawa, the Prime Minister is appointed by the Governor General)

It is an interesting question why these robust pairings suffer exceptions that seem idiosyncratic: not all count nouns can be used as mass nouns and vice versa, and the language seems to put restrictions on this process. Nunberg (1995) points out that certain terms for fruit in French can be used to refer to brandies made out of the fruit, but in English this isn’t the case (i.e., “?I drank a prune”).

Within systematic polysemy, there is a potential sub-categorization of rules that produce the polysemes: constructional polysemy and sense extensions. Constructional polysemes are characterized by local syntactic or semantic contexts. The case of “walked” in (8) and (9) is a case of a constructional polysemy where the arguments determine the particular sense of the verb with respect to goal. It’s interesting in this context to compare cases such as “Drink the night away,” “sang his heart out” with “drank beer” and “sang a song” (see Jackendoff 2002; pp. 167–172), where it looks like perhaps the construction is responsible for the polysemy rather than any term involved. Sense extensions, by contrast, are operations on the lexical meaning. The conversion of count nouns into mass nouns can be characterized by Pellettier’s (1975) notion (attributed to David Lewis) of “grinding”—take any count noun and imagine a case in which something of that kind is put through a grinder and one gets a mass interpretation (“there was automobile all over the wall!”). Perhaps a universal packager can do something similar for mass nouns, though there may be resistance if the implied packaging is not made obvious (“give me a brackish water” is obviously less easy to interpret than “give me a beer”). These rules are defined over classes of lexical items and are subject to potentially arbitrary lexical constraints that seem linguistically based. Nunberg and Zaenen (1992) contrast this type of sense extension, which seems rule based, with lexical license, which allows systematic construction of new senses on the basis of specifically indexed relationships. For example, producing edible meat terms from the animal they are produced from is subject to arbitrary constraints.

One hope for explaining polysemy and its categorization comes from exploring the idea that polysemes can be constructed by specifying semantic chains and hoping that these chains form regular production mechanisms from a core meaing to a derived one. Jackendoff (2002; 341–342) considers the case of “smoke.” “Smoke” is verbally polysemous between causing something to give off smoke and forcing smoke onto something. Witness the zeugma in:

  1. 66. ?Rafik smoked the cigarette and the turkey.

Given that both are related to the noun ‘smoke’, one might consider the polysemy here to be a matter of extension from the noun naming the wispy substance to both notions. It is difficult to see, however, just what the relevant process is that goes from the noun to each sense. Moreover, it is a challenge to prototype approaches to see just how both notions of smoke form one interestingly unified prototype as opposed to two separately lexicalized meanings that are produced somewhat adventitiously.

Idiosyncratic polysemy occurs in cases in which there is a surprising application of a term in a novel manner. As Wechsler (2015: 8) points out, the adjective “topless” began to be applied to “meeting” to produce “topless meeting.” Perhaps surprisingly this was used to denote meetings at which participants are banned from bringing laptops. The processes by which idiosyncratic polysemy arises are harder to discern, and these sorts of polysemes are quite common, if indeed they are polysemes. One might be inclined to think of them as metaphoric extensions.

It wouldn’t be surprising if this sort of polysemy was produced by at least similar processes as those implicated in metaphor and metonymy.

Part 7: Conclusion

Obviously we have just scratched the surface of the relevant considerations regarding polysemy. It forms yet another part of the tapestry of meaning modulation that must be accounted for in the final story of language acquisition, interpretation, and understanding. But for all the vast amount of work done on the topic, it still remains elusive just how to give polysemy a reasonable unified characterization that makes it an interestingly unified phenomenon. Modern corpus searches have incorporated some ideas from the literature concerning polysemy and seek to test theses regarding semantic relatedness (see Glynn and Robinson (2014)). Coupled with neuroscientific work on how people process ambiguity, polysemy, and metaphor, we can anticipate a promising and exciting theoretical future.


Aristotle, Metaphysics, W. D. Ross (trans.) (1984). in The Complete Works of Aristotle: The Revised Oxford Translation, J. Barnes (ed.) Princeton: Princeton University Press.Find this resource:

Blank, A. and Koch, P. (eds.) (1999). Historical Semantics and Cognition. Berlin: Mouton de Gruyter.Find this resource:

Bréal, M. (1897). Essai de Sémantique. Paris.Find this resource:

Croft, W. and Cruse, D. A. (2004). Cognitive Linguistics. Cambridge: Cambridge University Press.Find this resource:

Cruse, D. A. (1986). Lexical Semantics. Cambridge: Cambridge University Press.Find this resource:

Fara, D. G. (2000). “Shifting Sands: An Interest-Relative Theory of Vagueness.” Philosophical Topics, 28(1), 45–81.Find this resource:

Fiengo, R. and May, R. (1994). Indices and Identity, Cambridge, MA: MIT Press.Find this resource:

Fodor, J. (1980). “The Present Status of the Innateness Controversy.” In Representations: Philosophical essays on the foundations of cognitive science. Cambridge, MA: MIT Press, pp. 257–316.Find this resource:

Fodor, J. (1998). Concepts: Where Cognitive Science Went Wrong. Oxford: Oxford University Press.Find this resource:

Fodor, J., & Lepore, E. (1996). “The Red Herring and the Pet Fsh: Why Concepts Still Can’t Be Prototypes” Cognition 58, 253–270.Find this resource:

Geeraerts, D. (1989). “Prospects and Problems of Prototype Theory.” Linguistics 27, 587–612.Find this resource:

Geeraerts, D. (1997). Diachronic Prototype Semantics: A Contribution to Historical Lexicology. Oxford: Oxford University Press.Find this resource:

Glynn, D. and Robinson, J. A. (eds.). (2014). Corpus Methods for Semantics: Quantitative Studies in Polysemy and Synonymy (Human Cognitive Processing). Philadelphia: John Benjamins Publishing Company.Find this resource:

Hobbs, J. (1985.) “Granularity,” Proceedings of the 9th ICAI, Los Angeles, CA, 432–435.Find this resource:

Jackendoff, R. (2002). Foundations of Language: Brain, Meaning, Grammar, Evolution. Oxford: Oxford University Press.Find this resource:

Kamp, H. and B. Partee. (1995). “Prototype Theory and Compositionality,” Cognition 57, 129–191.Find this resource:

Kaplan, D. (1989), “Demonstratives,” in Joseph Almog, John Perry, and Howard Wettstein (eds.), Themes from Kaplan. Oxford: Oxford University Press.Find this resource:

Kripke, S, 1977. “Speaker’s Reference and Semantic Reference,” Midwest Studies in Philosophy 2, 255–276.Find this resource:

Lakoff, G. (1982). “Categories: An Essay in Cognitive Linguistics.” In The Linguistic Society of Korea (ed.), Linguistics in the Morning Calm: Selected Papers from SICOL-1981. Seoul: Hanshin, pp. 139–193.Find this resource:

Lakoff, G. (1987). Women, Fre, and Dangerous Things: What Categories Reveal about the Mnd. Chicago: University of Chicago Press.Find this resource:

Lewandowska-Tomaszczyk, B, (2007). “Polysemy, Prototypes and Radial Categories.” The Oxford Handbook of Cognitive Linguistics, Dirk Geeraerts and Hubert Cuyckens (eds.), Oxford: Oxford University Press, pp. 139–169.Find this resource:

Ludlow, P. (2015). Living words. New York: Oxford University Press.Find this resource:

Pelletier, F. J. (1975). Non-singular reference: some preliminaries. In Philosophia Vol. 5(4), 451–465.Find this resource:

Neta, Ram. “(2007). Contextualism, and a Puzzle about Seeing.” Philosophical Studies 134(1), 53–63.Find this resource:

Nunberg, G. (1993). “Indexicality and Deixis,” in Linguistics and Philosophy 16, 1–43.Find this resource:

Nunberg G. (1995). “Transfers of Meaning.” Journal of Semantics 12, 109–133.Find this resource:

Nunberg, G, and Zaenen, A. 1992. “Systematic Polysemy in Lexicology and Lexicography.” Paper presented at Euralex II, Tampere, Finland.Find this resource:

Osherson, D., & Smith, E. (1981). “On the Adequacy of Prototype Theory as a Theory of Concepts.” Cognition 9, 35–58.Find this resource:

Recanati, F. (1993). Direct Reference: From Language to Thought. Oxford: Blackwell.Find this resource:

Ryle, G. (1949). The Concept of Mind. University of Chicago Press. London: Hutchinson.Find this resource:

Taylor, J. R. (2003a). Linguistic Categorization. Oxford: Oxford University Press. First edition, 1989.Find this resource:

Taylor, J. R. (2003b). “Polysemy’s Paradoxes.” Language Sciences 25(6), 637–655.Find this resource:

Tsohatzidis, S. (Ed.) (1990). Meanings and Prototypes: Studies on Linguistic Categorization. London: Routledge.Find this resource:

Tuggy, D. (1993). “Ambiguity, Polysemy, and Vagueness.” Cognitive Linguistics 4(3), 273–290.Find this resource:

Ungerer, F., & Schmid, H.-J. (1996). An Introduction to Cognitive Linguistics. London: Longman.Find this resource:

Violi, P. (1997). Significato ed esperienza. Milan: Bompiani. [Meaning and Experience, Trans. Jeremy Carden, 2001, Bloomington: Indiana University Press]Find this resource:

Wechsler, S. (2015). Word Meaning and Syntax: Approaches to the Interface. Oxford: Oxford University Press.Find this resource:

Zwicky, A. and Sadock, J. (1975). “Ambiguity Tests and How to Fail Them,” in Syntax and Semantics, 4 Kimball (ed.), New York: Academic Press, pp. 1–36.Find this resource:


(2) It’s not clear to me that the reading of “mouth” as applied to rivers isn’t a case of metaphor rather than polysemy. But it is pretty clearly a dead, or conventionalized metaphor (akin to “lip of the glass”).

(4) See Tuggy (1993), who argues that polysemy comes in degrees and sits somewhere in a continuum between ambiguity and unitary meaning assignment.

(5) One might think that the phrase “more of a giant” reveals gradability in “giant.” But this would be mistaken, as the phrase “more of a Nixon than a Kennedy” reveals. Presumably “more of a” is used to make comparisons regarding how closely the subject of the comparison resembles the object in salient manners.

(6) See Zwicky and Sadock (1975) and Cruse (1986) for extensive discussion of the tests.

(7) Ellipsis gives rise to its own ambiguities, as in:

  • () The governor loves her sister and Mary does too.

“Does too” can express the property of loving the governor’s sister or of loving Mary’s sister (which seems distinct even if Mary’s sister is the governor). This ambiguity is known as strict/sloppy identity. This is consistent with Preserve, since it concerns the interpretation of the sentence targeted by ellipsis. See Fiengo and May (1994).

(8) One example of shenanigans involves focal stress on the second occurrence of uncle, resulting in a reading on which “uncle” means something like paradigmatic uncle who buys presents and acts in generally accepted avuncular manners. We are advised to take note of this (and similar) readings and try to avoid letting them mislead us. Similarly for (53) in which focal stress on the second occurrence of “trouble” can result in meaning something like serious trouble. It would be nice to have an exhaustive list of possible shenaningans.

(10) See Hobbs (1985) for relevant considerations regarding the phenomenon of granularity in linguistic interpretation.

(12) Fodor (1980, 1998) argues against treating prototypes as either rmeanings or concepts, citing worries about compositionality. See Osherson and Smith (1981) and Kamp and Partee (1995) for related considerations and Fodor and Lepore (1996) for a response.