Spencer Caplan

A persistent puzzle throughout the cognitive and social sciences is how people manage to learn social conventions from the sparse and noisy behavioral data of diverse actors, without explicit instruction. Here, we show that the dominant theories of social learning perform poorly at capturing how individuals learn conventions in coordination experiments that task them with matching their behaviors while interacting in social networks. Across experiments, participants’ choices systematically deviate from both imitation and optimization. Instead, they follow a categorical, two-stage learning process: they behave probabilistically until they acquire enough information about each other to trigger a mental threshold and then their choices stabilize. We effectively estimate this threshold using the tolerance principle (TP), a parameter-free equation developed to model how children learn rules in language. We show that threshold-based agents produce social learning that is more accurate than imitating and optimizing agents, while also providing a better model of how a critical mass of dissenters can overturn conventions. The superior performance of our model holds when comparing against a variety of optimization approaches, including Bayesian inference. Furthermore, in a preregistered dyadic experiment requiring people to infer nonlinguistic behavioral patterns amid controlled levels of noise in observed signals, TP outperforms all other models at reproducing learning rates among human participants. These findings offer compelling evidence that a simple, mathematical threshold underlies individual and social learning, from grammatical rules to behavioral conventions.

Logical reasoning is one of humanity's most powerful abilities. A widespread assumption across psychology, linguistics, and philosophy holds that reasoning operates over concepts that refer to objects and properties in the world, yet this has rarely been tested empirically. We introduce a novel paradigm that exploits lexical ambiguity to differentiate candidate representations for human inference: word-forms, reference-fixing concepts, or more abstract "underspecified representations" that constrain meaning without fully determining it. Across three experiments (N=158), participants evaluated deductive arguments equivocating over polysemous or homonymous terms. Although these arguments are logically equivalent under standard analyses, participants reliably accepted non-truth-preserving polysemous arguments while rejecting homonymous ones. This asymmetry—the Uncommon Sense Effect—points to underspecified representations as the default basis of logical inference. This challenges a foundational assumption about meaning, and opens new empirical avenues for studying the link between what we think and why we think it.

Is language designed for communicative and functional efficiency? G. K. Zipf famously argued that shorter words are more frequent because they are easier to use, thereby resulting in the statistical law that bears his name. Yet, G. A. Miller showed that even a monkey randomly typing at a keyboard, and intermittently striking the space bar, would generate “words” with similar statistical properties. Recent quantitative analyses of human language lexicons (Piantadosi et al., 2012) have revived Zipf's functionalist hypothesis. Ambiguous words tend to be short, frequent, and easy to articulate in language production. Such statistical findings are commonly interpreted as evidence for pressure for efficiency, as the context of language use often provides cues to overcome lexical ambiguity. In this study, we update Miller's monkey thought experiment to incorporate empirically motivated phonological and semantic constraints on the creation of words. We claim that the appearance of communicative efficiency is a spandrel (Gould & Lewontin, 1979), as lexicons formed without the context of language use or reference to communication or efficiency exhibit comparable statistical properties. Furthermore, the updated monkey model provides a good fit for the growth trajectory of English as recorded in the Oxford English Dictionary. Focusing on the history of English words since 1900, we show that lexicons resulting from the monkey model provide a better embodiment of communicative efficiency than the actual lexicon of English. We conclude by arguing for the need to go beyond correlational statistics and to seek direct evidence for the mechanisms that underlie principles of language design.

Lexical ambiguity has classically been categorized into two kinds. Homonyms are single word forms that map to multiple, unrelated meanings (e.g., “bat” meaning baseball equipment or a flying mammal). Polysemes are single word forms that map to multiple, related senses (e.g., “breakfast” meaning a plate of food or an event). Yet there is a longstanding debate as to whether polysemy and homonymy reflect distinct cognitive representations. Some (e.g., Fodor & Lepore, 2002; Klein & Murphy, 2001) posit that they do not — merely describing differing patterns of usage — while others (e.g., Falkum & Vicente, 2015; Pietroski, 2018) argue that polysemes, but not homonyms, involve an underspecified representation that is neutral with respect to the form’s multiple senses. While some extant experimental evidence supports the latter view (Klepousniotou, Titone, & Romero, 2008; Srinivasan, Berner, & Rabagliati, 2019), there has not yet been clear evidence of the representation of lexical ambiguity affecting domain-general reasoning. Using a novel inference paradigm, we compare participants’ dispositions to endorse deductive, Aristotelian arguments with equivocating polysemes versus comparable arguments with equivocating homonyms. We find that participants endorse the former substantially more than the latter, a phenomenon that we dub the "Uncommon Sense Effect". Our results provide direct evidence that polysemes and homonyms have underlyingly distinct mental representations — in particular that polysemes uniquely invoke an underspecified representation that allows for rule-based inferences across distinct senses.