Abstract
The aim of this paper is to reconcile two claims that have long been thought to be incompatible: (a) that we compositionally determine the meaning of complex expressions from the meaning of their parts, and (b) that prototypes are components of the meaning of lexical terms such as fish, red, and gun. Hypotheses (a) and (b) are independently plausible, but most researchers think that reconciling them is a difficult, if not hopeless task. In particular, most linguists and philosophers agree that (a) is not negotiable; so they tend to reject (b). Recently, there have been some attempts to reconcile these claims (Prinz, Furnishing the mind: concepts and their perceptual basis 2002; The Oxford handbook of compositionality 2012; Jönsson and Hampton, Cognition 106:913–923, 2008; Hampton and Jönsson, The Oxford handbook of compositionality 2012; Schurz, The Oxford handbook of compositionality 2012), but they all adopt an implausibly weak notion of compositionality. Furthermore, parties to this debate tend to fall into a problematic way of individuating prototypes that is too externalistic. In contrast, I propose that we can reconcile (a) and (b) if we adopt, instead, an internalist and pluralist conception of prototypes and a context-sensitive but strong notion of compositionality. I argue that each of this proposals is independently plausible, and that, when taken together, provide the basis for a satisfactory account of prototype compositionality.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
A feature x is typical, relative to category A, if the probability is high that an entity has x if it belongs to A, and x is diagnostic if the probability is high that an entity belongs to A if it has x.
For example, if subjects have to decide whether something is a bird, they are faster and more accurate for sparrows than penguins regardless of whether they are shown pictures or words for sparrows and penguins (see Murphy 2002, Chap. 11). This stimulus independence challenges those who deny that a theory of lexical concepts should do double duty as a theory of linguistic meaning. Most of the data used to argue that concepts have a prototype structure can be used to argue that linguistic meaning has a prototype structure.
For readers familiar with issues about compositionality and context-sensitivity as they arise in recent discussions in Philosophy of Language, I should point out that most contextualist positions, such as Recanati’s Truth-conditional pragmatics and Carston’s Relevance theory, assume versions of compositionality that are substantially weaker than the principle I will defend. The non-compositional views defended in Philosophy of Mind by prototype theorists such as Prinz and Hampton belong to the same family as the radical contextualist views in the Philosophy of Language. These points are elaborated in Sect. 5.
This assumption is made in most empirical studies of concepts which use linguistic stimuli. As will become clear below, this assumption allows for a wide variety of views on the relation between the standing meaning of words and the meaning that they take in particular utterances.
As several authors have noted, there are performance and other sorts of limits on the actual productivity and systematicity of language. For our purposes, nothing hangs on this. For further discussion, see Del Pinal (2014, Chaps. 2, 3).
There is much debate about the basic types of [A N] modifications (Partee 1998; Morzycki 2015). Most classifications include intersective, subsective, modal, and privative modifications. In modal and privative modifications—e.g., former president and fake gun—that x is an [A N] does not entail that x is an N. The modal cases do not present any challenges to prototype theory in particular: they shift the time or world of evaluation, and do not affect the content of the head N. The privative cases are problematic for all theories, so critics of prototype compositionality justly do not appeal to them in their critiques. Still, Del Pinal (2015) argues that privative modifications (e.g., fake gun), and certain subsective modifications (e.g., bad gun) actually support the view that the meaning of head Ns includes a prototype component, on which the modifiers partly operate. So it is not inaccurate to say that just those modifications that are hard for classical theories are easy for prototype theories, and vice versa. This is why it is particularly important for prototype theories to deal with simple intersective modifications.
To be clear, most models of prototype combinatorics do not stop at the composition stage—see, e.g., Hampton’s Composite Prototype Model and Costello and Keane’s C\(^3\) Model. What is important is only that they include an initial compositional stage, even if they also try to model some post-linguistic pragmatic modifications.
The view that prototypes are conceptual components can be implemented in various ways. One can be a global or local prototype theorist: i.e., one can hold that all terms include prototypes, or one can hold that, e.g., artifact terms include prototypes but mathematical and some technical terms do not. In this paper, I remain neutral on this issue. This in no way weakens my defence of prototype compositionality. The dialectical situation is as follows. Suppose there are good reasons to hold that prototypes are conceptual components for some class, say, for natural kinds and artifact terms. As mentioned above, many theorists would still resist including prototypes as conceptual components because they believe that prototypes are not sufficiently compositional. A clear statement of this position can be found in Fodor and Pylyshyn (2015). Now, my aim is to show that prototypes are sufficiently compositional. If that is correct, we can begin investigating the potential advantages of including prototypes as conceptual components.
By a ‘pluralist’ notion of prototypes I do not mean a ‘pluralist’ theory of concepts, in the sense defended by e.g., Weiskopf (2009). A pluralist theory of prototypes holds that the dimensions along which prototypes are organised are richer than is sometimes assumed: e.g., we not only represent how objects typically look but also how they typically come into being. In contrast, a pluralist theory of concepts holds that, depending on such things as tasks and context, different types of mental structures can serve as the surrogates of concepts. Still, nothing I say here is in tension with pluralism in this other sense.
These views share the idea that the processes which determine the meaning of complex expressions have access to general beliefs which do not come from the meaning of the parts or from their structure. If we apply these accounts to the view that prototypes are constituents of linguistic meaning, we can formulate the basic idea as follows (cf. Prinz 2012):
-
(Modal C) As a default, FL determines the meaning of complex expressions from the meaning of their syntactically immediate constituents, the way they are combined, and general beliefs. As a fallback, FL determines the meanings of complex expressions following C.
Modal C says that FL is capable of combining prototypes compositionally, but that it does so only when it lacks the relevant general beliefs. Several authors have endorsed the Modal C solution to the problem of emergent features (see e.g., Weiskopf 2009; Robbins 2002; Schurz 2012). Strictly, Modal C is less constrained than \(C^{mod}\); in practice, however, they allow very similar accounts of emergent features. The reasons I give below to constraint \(C^{mod}\) also apply to Modal C.
-
Modulation* processes fall squarely in what two system theorists of reasoning call the ‘associative system’ or ‘system 1’ (Sloman 2002; Morewedge and Kahneman 2010). On the standard view, the operations of system 1 are based on associations, not abstract rules, and operate on prototypes in an automatic and unconscious manner. These are also characteristics of pre-compositional modulations*. In the cut examples, subjects assume a particular manner of cutting, without being aware of the processing behind it. Indeed, this explains why the effects of modulations* are taken as part of the intuitive truth conditional content or meaning of utterances. In contrast, the additional relevance seeking, inferential operations of free modulation fall squarely in what two system theorists call the “rule-based system” or “system 2”. Rule-based processes are sensitive to goals and beliefs, and tend to be conscious and relatively effortful. Insofar as the general distinction between associative and rule-based systems has psychological reality, this further justifies distinguishing between modulation* and free modulation, and taking modulation* as the basic pre-compositional process.
These examples suggest that something like free modulation cannot be affecting the intuitive truth conditions or meanings assigned to expressions. This should not be confused with claims about limits on what utterances can communicate in context via traditional pragmatic means such as Gricean conversational implicatures. Naturally, if we have to determine what someone who used, say (7-a) meant to communicate in that context, we might very well say that it is something like (7-b).
In particular, the claim that free modulation over-generates meanings has been intensively debated by philosophers and linguistics. Some of the most influential arguments against free modulation can be found in the papers re-printed in Stanley (2007). For a direct response to Stanley’s main arguments and examples see Hall (2008), among others. I think some of the best and yet somewhat ignored examples against free modulation are presented in Asher (2011) and Vespoor (1996). An important (and ultimately critical) empirical investigation of whether contextualist theories with free modulation such as Recanati’s successfully predict which meaning shifts are licensed is reported in Rabagliati et al. (2011). In the end, even Recanati (2010) accepts that it is likely that, as stated, free modulation is too unconstrained, and welcomes suggestions for how to constrain it (see his Introduction).
Prototype theorists should welcome this strategy. Although most have adopted unconstrained combinatorial principles such as \(C^{mod}\), they usually did so as a response to the emergent features objection. If my proposal for reconciling compositionality with prototype theory is successful, theorists such as Prinz, Hampton and Jönsson need not adopt such radically unconstrained principles, nor depend—to defend their preferred theory of meaning—on the eventual resolution of the debate regarding over-generation and free modulation.
The results could signal a violation of DS for two reasons. First, they could signal a violation of Semantic Locality: if when computing the prototype of baby ducks or baby Peruvian ducks some emergent feature is added regarding their type of feet, this would explain why subjects are less confident that baby ducks and baby Peruvian ducks have webbed feet compared to ducks. Second, the results could also signal a violation of Uniform Modification: if the way in which weights are readjusted in the case of baby ducks is different to the case of quacking ducks, this would explain why subjects’ drop in confidence with respect to webbed feel is different in each case. In either case, DS would be violated.
This explanation of the modifier effect is compatible with the account provided by Gagne and Spalding (2014). They explain the modifier effect as a result of subcategorisation, in conjunction with the belief that subcategories might lack some of the properties of categories. In the explanation above, we also appealed to that belief. Gagne and Spalding (2014) argue that this belief is not based on extensional knowledge, but is more like a meta-belief about the general relation between categories and subcategories. I agree with this, but would add that subjects do seem to be sensitive to the difference between typical and atypical modifiers, otherwise the decrease in confidence between the B and C cases could not be explained. Gagne and Spalding (2014) might be somewhat sceptical about this additional point. The reason is that they present an experiment which introduces a new condition, call it type E, in which the modifier of the NP is an unknown word, e.g., blika ducks have webbed feet. The result is that E cases show a decrease that is stronger than B cases. However, we can easily reconcile these results with our interpretation: subjects treat unknown modifiers as atypical modifiers (clearly a reasonable assumption); hence, E cases are treated as C cases.
In Sect. 8.1 I discuss how we can operationalise the distinction between extensional feedback or memory based and reasoning based features. For now, I will assume that there is an intuitive distinction. To see this, consider some representative examples of the phrases and emergent features used in the time-sensitive studies we will examine: peeled bananas and white, peeled apples and white, water guns and harmless, boiled celery and soft. These are all categories with which most participants are likely acquainted, unlike the examples used by Johnson and Keil, which focus on novel or surprising categories such as arctic bicycle. In addition, note that in the memory based examples it is very hard to reason causally to explain why the emergent feature is typical of the class, without appealing to direct knowledge that the class has in fact that feature.
A study which used a cross-modal lexical priming paradigm provides converging evidence for this conclusion (Swinney et al. 2007). Specifically, it showed that complex NPs prime memory based emergent features by the onset of the word that came right after the head N of the NP. The stimuli used by Swinney et al. (2007) can be accessed at: http://lcnl.ucsd.edu/publications.html.
Famously, sentences which predicate typical features of the subject such as Dutch trains are yellow generate less N400 activity at the predicate than sentences such as Dutch trains are white/sour, which do not predicate typical features.
For example, in their study of emergent features in combinations involving social concepts, Kunda et al. (1990) found that how novel or surprising subjects found the class denoted by a combination, was a good predictor of whether other subjects would give causal explanations for their emergent features.
For example, in a pilot study conducted to test this proposal, we used reasoning schemes such as (R) and (C): (R) “Since the ARCTIC ___________________ , and since BICYCLES ___________________ , then arctic bicycles have spiked tires” and (C) “Since PEELED things ___________________ , and since BANANAS ___________________ , then peeled bananas are white”. (R) uses a paradigmatic NP plus reasoning-based emergent feature pair, and (C) uses a paradigmatic NP plus memory based emergent feature pair. After each reasoning scheme, we asked subjects to rate the task on a 1 (easy/natural) to 7 (hard/weird) likert scale. NPs plus reasoning based feature pairs consistently rank as significantly easier/more natural than NPs plus memory based feature pairs. So there is some initial evidence that we can use tasks such as this to classify emergent features.
References
Asher, N. (2011). Lexical meaning in context: A web of words. Cambridge: Cambridge University Press.
Barsalou, L. W. (1987). The instability of graded structure: Implications for the nature of concepts. In U. Neisser (Ed.), Concepts and conceptual development: Ecological and intellectual factors in categorization (pp. 101–140). Cambridge: Cambridge University Press.
Chomsky, N. (2001a). Beyond explanatory adequacy. Tech. Rep. Cambridge, MA: MIT.
Chomsky, N. (2001b). Derivation by phase. In M. Kenstowitz (Ed.), Ken Hale: A life in language (pp. 1–52). Cambridge, MA: MIT Press.
Connolly, A. C., Fodor, J., Gleitman, L., & Gleitman, H. (2007). Why stereotypes don’t even make good defaults. Cognition, 103(1), 1–22.
Cook, V. J., & Newson, M. (2007). Chomsky’s universal grammar: An introduction. Oxford: Blackwell.
Del Pinal, G. (2014). The architecture of our semantic compentence: Compositional operations and complex lexical representations. Ph.D. Thesis, Columbia University, New York.
Del Pinal, G. (2015). Dual content semantics, privative adjectives and dynamic compositionality. Semantics and Pragmatics, 8(6), 1–56.
Fodor, J. (1999). Concepts: Where cognitive science went wrong. New York: Oxford University Press.
Fodor, J., & Lepore, E. (1996). The pet fish and the red herring: Why concepts still can’t be prototypes. Cognition, 58(2), 329–343.
Fodor, J., & Pylyshyn, Z. (1988). Connectionism and cognitive architecture: A critical analysis. Tech. Rep. Rutgers Center for Cognitive Science.
Fodor, J. A., & Pylyshyn, Z. W. (2015). Minds without meanings. Cambridge, MA: The MIT Press.
Gagne, C. L., & Spalding, T. L. (2014). Subcategorization, not uncertainty, drives the modification effect. Language, Cognition and Neuroscience, 29(10), 1283–1294.
Gleitman, L. R., Connolly, A. C., & Armstrong, S. L. (2012). Can prototype representations support composition and decompositions? In M. Werning, W. Hinzen, & E. Machery (Eds.), The Oxford handbook of compositionality (Chap. 20). Oxford: Oxford University Press.
Hall, A. (2008). Free enrichment or hidden indexicals? Memory & Language, 23, 426–456.
Hamill, R., Nisbett, R. E., & Wilson, T. D. (1980). Insensitivity to sample bias: Generalizing from atypical cases. Journal of Personality and Social Psychology, 39(4), 578–589.
Hampton, J. (1997). Conceptual combination. In K. Lamberts & D. Shanks (Eds.), Knowledge, concepts, and categories (pp. 133–160). Cambridge, MA: The MIT Press.
Hampton, J. (2006). Concepts as prototypes. In B. H. Ross (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. 46, pp. 79–113). Amsterdam: Elsevier.
Hampton, J. A. (1987). Inheritance of attributes in natural concept conjunctions. Memory & Language, 16, 579–591.
Hampton, J. A., & Jönsson, M. L. (2012). Typicality and compositionality: the logic of combining vague concepts. In M. Werning, W. Hinzen, & E. Machery (Eds.), The Oxford handbook of compositionality (Chap. 18). Oxford: Oxford University Press.
Horwich, P. (1998). Concept constitution. Philosophical Issues, 9, 15–19.
Jacobson, P. (2012). Direct compositionality. In W. H. Marcus Werning & E. Machery (Eds.), The Oxford handbook of compositionality (Chap. 5, pp. 109–128). Oxford: Oxford University Press.
Johnson, C., & Keil, F. (2000). Explanatory understanding and conceptual combination. In F. Keil & R. Wilson (Eds.), Explanation and cognition (pp. 327–359). Cambridge: The MIT Press.
Jönsson, M. L., & Hampton, J. A. (2008). On prototypes as defaults (comments on Connolly, Fodor, Gleitman and Gleitman 2007). Cognition, 106, 913–923.
Kahneman, D. (1973). Attention and effort. Englewood Cliffs, NJ: Prentice-Hall.
Kahneman, D. (2011). Thinking fast, thinking slow. New York: Farrar, Straus, and Giroux.
Kunda, Z., Miller, D. T., & Claire, T. (1990). Combining social concepts: The role of causal reasoning. Cognitive Science, 14, 551–577.
Kutas, M., & Federmeier, K. D. (2011). Thirty years and counting: Finding meaning in the n400 component of the event-related brain potential. Annual Review of Psychology, 62, 621–647.
Leslie, S. J. (2015). Hillary Clinton is the only man in the Obama administration: Dual character concepts, generics, and gender. Analytic Philosophy, 56(2), 111–141.
Machery, E. (2011). Concepts: A tutorial. In R. Belohlavek & G. J. Klir (Eds.), Concepts and fuzzy logic (Chap. 2). Cambridge, MA: The MIT Press.
Margolis, E., & Laurence, S. (1999). Concepts and cognitive science. In E. Margolis & S. Laurence (Eds.), Concepts: Core readings (Chap. 1, pp. 3–82). Cambridge, MA: The MIT Press.
McElree, B., Murphy, G. L., & Ochoa, T. (2006). Time course of retrieving conceptual information: A speed-accuracy trade-off study. Psychonomic Bulletin & Review, 13(5), 848–853.
Morewedge, C. K., & Kahneman, D. (2010). Associative processes in intuitive judgment. Trends in Cognitive Sciences, 14(10), 435–440.
Morzycki, M. (2015). Modification. Cambridge: Cambridge University Press (in press).
Murphy, G. L. (1990). Noun phrase interpretation and conceptual combination. Journal of Memory and Language, 29, 259–288.
Murphy, G. L. (2002). The big book of concepts. Cambridge, MA: The MIT Press.
Pagin, P., & Pelletier, J. (2007). Content, context, and composition. In Context-sensitivity and semantic minimalism: New essays on semantics and pragmatics (p. 25).
Partee, B. (1998). Lexical semantics and compositionality. In L. Gleitman & M. Liberman (Eds.), An invitation to cognitive science (Vol. 1). Cambridge: The MIT Press.
Prinz, J. (2012). Regaining composure: A defense of prototype compositionality. In M. Werning, W. Hinzen, & E. Machery (Eds.), The Oxford handbook of compositionality (Chap. 21). Oxford: Oxford University Press.
Prinz, J. J. (2002). Furnishing the mind: Concepts and their perceptual basis. Cambridge, MA: MIT Press.
Putnam, H. (1997). Is semantics possible? In Mind, language and reality: Philosophical papers (Vol. 2). Cambridge: Cambridge University Press.
Rabagliati, H., Marcus, G. F., & Pylkkanen, L. (2011). Rules, radical pragmatics and restrictions on regular polysemy. Journal of Semantics, 28(4), 485–512.
Radford, A. (2004). Minimalist syntax: Exploring the structure of English. Cambridge: Cambridge University Press.
Recanati, F. (2006). Indexical concepts. In M. Garcia-Carpintero & J. Macia (Eds.), Two-dimensional semantics. Oxford: Claredon Press.
Recanati, F. (2010). Truth-conditional pragmatics. Oxford: Oxford University Press.
Robbins, P. (2002). How to blunt the sword of compositionality. Nous, 36, 313–334.
Rosch, E. (1973). Natural categories. Cognitive Psychology, 4, 532–547.
Rosch, E. H. (2011). Slow lettuce: Categories, concepts, fuzzy sets, and logical deduction. In R. Belohlavek & G. J. Klir (Eds.), Concepts and fuzzy logic (Chap. 4). Cambridge, MA: The MIT Press.
Roth, E. M., & Shoben, E. J. (1983). The effect of context on the structure of categories. Cognitive Psychology, 15, 346–378.
Sassoon, G. W. (2011). Adjectival vs. nominal categorization processes: The rule vs. similarity hypothesis. Belgium Journal of Linguistics, 24, 104–147.
Schurz, G. (2012). Prototypes and their composition from an evolutionary point of view. In M. Werning, W. Hinzen, & E. Machery (Eds.), The Oxford handbook of compositionality (Chap. 26). Oxford: Oxford University Press.
Sloman, S. (2002). Two systems of reasoning. In T. Gilovich, D. Griffin, & D. Kahneman (Eds.), Heuristics and biases: The psychology of intuitive judgment (Chap. 22). Cambridge: Cambridge University Press.
Smith, E. E., Medin, D. L., Rips, L. J., & Keane, M. (1988). Combining prototypes: A selective modification model. Cognitive Science, 12, 485–527.
Stanley, J. (2007). Language in context. Oxford: Oxford University Press.
Swinney, D., Love, T., Walenski, M., & Smith, E. E. (2007). Conceptual combination during sentence processing. Psychological Science, 18(5), 397–400.
Tabossi, P. (1988). Effects of context on the immediate interpretaion of unambiguous nouns. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14(1), 153–162.
Taylor, J. R. (2009). Linguistic categorization. Oxford: Oxford University Press.
Vespoor, C. (1996). Lexical limits on the influence of context. In Proceedings of the eighteenth annual conference of the Cognitive Science Society. Mahwah, NJ: Lawrence Erlbaum Associates Publishers.
Weiskopf, D. (2009). Atomism, pluralism and conceptual content. Philosophy and Phenomenological Research, LXXIX(1), 131–164.
Acknowledgments
For helpful comments and discussions of earlier drafts of this paper, I am grateful to Luca Barlassina, Akeel Bilgrami, Brian H. Kim, Karen S. Lewis, Eleonore Neufeld, Daniel Rothschild, and Achille Varzi. I am also extremely grateful to two anonymous referees whose excellent comments led to substantial improvements of the paper. This work was supported by the Alexander von Humboldt Foundation and by BMBG Grant No. 01UG1411.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Del Pinal, G. Prototypes as compositional components of concepts. Synthese 193, 2899–2927 (2016). https://doi.org/10.1007/s11229-015-0892-0
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11229-015-0892-0