Polysemy and Inference: Reasoning with Underspecified Representations
Authors
Griffin Pion, Elliot Schwartz, Jake Quilty-Dunn, Eric Mandelbaum and Spencer Caplan
Affiliation: CUNY
Category: Psychology
Keywords: polysemy, homonymy, lexical ambiguity, mental representation, lexical processing, inference, concepts, deduction
Schedule & Location
Date: Wednesday 3rd of September
Time: 14:30
Location: Gen. Henryk Dąbrowski Hall (006)
View the full session: Polysemy & Communication
Abstract
Lexical ambiguity has classically been categorized into two kinds (e.g., Lyons 1977). Homonyms are single word forms that map to multiple, unrelated meanings. For example, "bat" refers to both flying mammals and clubs used in baseball. Polysemes, in contrast, are single word forms that map to multiple, related senses. For instance, “breakfast" refers to both dining events and foods. While related, these senses of "breakfast" are non-identical: breakfast foods are the sorts of things that can be overcooked, and breakfast events are the sort of things that can last a short time, but not vice versa. There is mounting evidence that polysemes, unlike homonyms, follow systematic and cross-linguistic patterns (Srinivasan & Rabagliati 2015). Nevertheless, the cognitive representations supporting such patterns remain uncertain.
There is general consensus (Frisson 2015) that homonyms employ a one-to-many mapping between a single word form (“bat”) and two or more distinct senses which are stored separately (BATanimal and BATtool - see Fig. 1a). However, accounts disagree regarding the status of polysemy. One account, which we term the List View, argues that polysemes employ the same representational structure as homonyms (Klein & Murphy 2001), with the difference between “classes” of ambiguity reduced to relatedness (either semantic (Devitt 2021) or metalinguistic (Fodor & Lepore 2002)) between senses. A different account, which we term the Underspecification View, holds that polysemes, unlike homonyms, involve a one-to-one-to-many structure: a single word form (e.g., "breakfast") maps to a single underspecified representation (BREAKFASTunderspecified), and this underspecified representation in turn points to multiple determinate senses (BREAKFASTevent and BREAKFASTfood). See Frisson (2009, 2015); for a review, Falkum and Vicente (2015); Fig. 1b). If polysemes do involve an underspecified representation, there is a further question as to when this representation is activated during lexical processing.
In our paper, we develop a novel inference experiment on which the List and the Underspecification Views make divergent predictions. We thus investigate whether polysemes and homonyms have distinct mental representations. A number of other paradigms have been used to probe for processing differences between homonymy and polysemy including sensicality judgments (Frisson 2015; Klein & Murphy 2001; Klepousniotou et al. 2008), lexical decision tasks (Armstrong & Plaut, 2008, 2016; Beretta, Fiorentino, & Poeppel 2005; Klepousniotou & Baum 2007; Rodd, Gaskell, & Marslen-Wilson 2002), eye-tracked reading (Frazier & Rayner 1990; Frisson & Pickering 1999, 2007), as well as word learning and meaning extension (Rodd et al., 2012; Srinivasan & Snedeker 2011). While these studies have provided insight into the potential processing differences and usage choices between polysemy and homonymy, such tasks do not clearly identify the representational status of such differences. Importantly, these methods are largely silent on whether polysemy is limited to linguistic processing (Pietroski 2018) or whether it reflects underlying conceptual structure (Quilty-Dunn 2021; Srinivasan & Rabagliati 2015).
Experiment: We developed a novel inference paradigm to test whether polysemes' posited underspecified representations cognitively exist, and if so, whether they can be used in inference. We tasked participants with judging the validity of Aristotelian syllogisms. These arguments consist of two premises sharing a middle term (M), as well as a conclusion. For example:
Premise: All M are P
Premise: Some S are M
Conclusion: Some S are P
Participants (n=28) judged a total of 56 arguments divided between four middle term categories (Polysemy, Homonymy, along with Valid and Invalid Fillers) which were fully crossed with four distinct Aristotelian syllogism forms (Dimatis, Disamis, Datisi and Darii). The Polysemy trial middle terms were drawn from Murphy (2021) with each premise written to uniquely constrain which sense of the polyseme was being used. For example:
Premise: All breakfasts are under three hours
Premise: Some smoothies are breakfasts
Conclusion: Some smoothies are under three hours
Since this argument equivocates on the senses of “breakfast", it's invalid. However, if polysemes have a single underspecified representation that can be used in inference, we predict that participants may draw the conclusion anyway. Importantly, we compare performance on polysemy trials with homonymy trials, where syllogisms equivocate on a homonym's two meanings. Crucially if homonyms, unlike polysemes, lack this underspecified element (mapping directly from word form to each distinct sense) then participants should reject the validity of these arguments; e.g.:
Premise: All bats share a common ancestor
Premise: Some sporting goods are bats
Conclusion: Some sporting goods share a common ancestor
If participants endorse the polysemy and homonymy arguments at similar rates, it would suggest both sorts of lexical items are represented in the same way. If, however, participants endorse the polysemy arguments more frequently than the homonymy arguments, it would suggest the presence of an underspecified representation for just the polysemous entries that is capable of supporting such inferences. To rule out the possibility that any differences across the middle terms stem from variable plausibility of the conclusions themselves, we first ran a norming study which matched the average truth ratings of all the conclusions across polysemy and homonymy groups.
Results: Participants judged that the conclusion followed from the premises significantly more often when the middle term is an equivocal polyseme (65%) compared to an equivocal homonym (48%) (p \< .001) (see Figure). This analysis, among others elided here for space considerations, was confirmed via mixed-effects model-comparison fitting logistic regressions to trial-level data. Our results highlight the different underlying representations involved in processing polysemous and homonymous word forms. Polysemes invoke a level of representation shared across multiple senses which can be used in inference while homonyms do not. This bolsters existing evidence that polysemy and homonymy reflect fundamentally distinct cognitive categories. Since inference is a paradigmatically conceptual, nonlinguistic process, this evidence also suggests that underspecified, polysemous representations reflect features of nonlinguistic cognition that underwrite domain-general conceptual thought.
Figures
Figure 1: Two forms of lexical ambiguity. On the List View, both homonyms and polysemes are
represented as a single word form mapping directly to its two or more constituent senses (panel a). On
the Underspecification View, homonyms are represented as on the List View (panel a), but for
polysemes, a single word form maps to an intermediate, underspecified structure which, in turn, maps to
the word form’s multiple determinate senses (panel b).
Figure 2: Bar graph of average validity response split up by word-type, where these types occurred as
middle terms (Valid Filler, Polysemy, Homonymy, and Invalid Filler). Judgments are significantly higher
(unpaired t-tests) for Valid Fillers compared with Polysemy (mean of 91% vs. 65%, p \<.001), as well as for
Polysemy compared with Homonymy (mean of 65% vs. 48%, p \< .001), but there is no difference in
judgments for Homonymy compared with Invalid Fillers (mean of 48% vs. 46%, p \= .564).
References
Armstrong, B. C., & Plaut, D. C. (2008). Settling dynamics in distributed networks explain task differences in semantic ambiguity effects: Computational and behavioral evidence. In B. C. Love, K. McRae, & V. M. Sloutsky (Eds.), Proceedings of the 30th Annual Conference of the Cognitive Science Society (pp. 273–278). Cognitive Science Society.
Armstrong, B. C., & Plaut, D. C. (2016). Disparate semantic ambiguity effects from semantic processing dynamics rather than qualitative task differences. Language, Cognition and Neuroscience, 32(3), 321-344. https://doi.org/10.1080/23273798.2016.1171366
Beretta, A., Fiorentino, R., & Poeppel, D. (2005). The effects of homonymy and polysemy on lexical access: An MEG study. Cognitive Brain Research, 24(1), 57–65. https://doi.org/10.1016/j.cogbrainres.2004.11.006
Devitt, M. (2021). Semantic polysemy and psycholinguistics. Mind & Language, 36(1), 134-157. https://doi.org/10.1111/mila.12282
Falkum, I. L., & Vicente, A. (2015). Polysemy: Current perspectives and approaches. Lingua, 157, 1-16. https://doi.org/10.1016/j.lingua.2015.02.002
Fodor, J., & Lepore, E. (2002). The compositionality papers. Oxford University Press.
Frazier, L., & Rayner, K. (1990). Taking on semantic commitments: Processing multiple meanings vs. multiple senses. Journal of Memory and Language, 29(2), 181-200.
Frisson, S. (2009). Semantic underspecification in language processing. Language and Linguistics Compass, 3(1), 111-127. https://doi.org/10.1111/j.1749-818x.2008.00104.x
Frisson, S. (2015). About bound and scary books: The processing of book polysemies. Lingua, 157, 17-35. https://doi.org/10.1016/j.lingua.2014.07.017
Frisson, S., & Pickering, M. J. (1999). The processing of metonymy: Evidence from eye movements. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25(6), 1366–1383.
Frisson, S., & Pickering, M. J. (2007). The processing of familiar and novel senses of a word: Why reading Dickens is easy but reading Needham can be hard. Language and Cognitive Processes, 22(4), 595-613.
Klein, D. E., & Murphy, G. L. (2001). The representation of polysemous words. Journal of Memory and Language, 45(2), 259-282.
Klepousniotou, E., & Baum, S. R. (2007). Disambiguating the ambiguity advantage effect in word recognition: An advantage for polysemous but not homonymous words. Journal of Neurolinguistics, 20(1), 1-24.
Klepousniotou, E., Titone, D., & Romero, C. (2008). Making sense of word senses: The comprehension of polysemy depends on sense overlap. Journal of Experimental Psychology: Learning, Memory, and Cognition, 34(6), 1534–1543.
Lyons, J. (1977). Semantics. Cambridge University Press.
Murphy, E. (2021). Linguistic representation and processing of copredication [Doctoral dissertation, University College London].
Pietroski, P. (2018). Conjoining meanings: Semantics without truth values. Oxford University Press.
Pietroski, P. M. (2025). One word, many concepts: Endorsing polysemous meanings. In E. Lepore & U. Stojnić (Eds.), Oxford handbook to contemporary philosophy of language. Oxford University Press.
Quilty-Dunn, J. (2021). Polysemy and thought: Toward a generative theory of concepts. Mind & Language, 36(1), 158-185. https://doi.org/10.1111/mila.12328
Rodd, J., Gaskell, G., & Marslen-Wilson, W. (2002). Making sense of semantic ambiguity: Semantic competition in lexical access. Journal of Memory and Language, 46(2), 245-266. https://doi.org/10.1006/jmla.2001.2810
Rodd, J. M., Berriman, R., Landau, M., Lee, T., Ho, C., Gaskell, M. G., & Davis, M. H. (2012). Learning new meanings for old words: Effects of semantic relatedness. Memory & Cognition, 40, 1095-1108. https://doi.org/10.3758/s13421-012-0209-1
Srinivasan, M., & Rabagliati, H. (2015). How concepts and conventions structure the lexicon: Cross-linguistic evidence from polysemy. Lingua, 157, 124-152. https://doi.org/10.1016/j.lingua.2014.12.004
Srinivasan, M., & Snedeker, J. (2011). Judging a book by its cover and its contents: The representation of polysemous and homophonous meanings in four-year-old children. Cognitive Psychology, 62(4), 245-272.
Srinivasan, M., & Snedeker, J. (2013). Polysemy and the taxonomic constraint: Children's representation of words that label multiple kinds. Language Learning and Development, 10(2), 97-128.
