The computational principles that underlie human concept learning have been debated in the literature for decades. Here, we formalize and test a new perspective that is grounded in rate-distortion theory (RDT), the mathematical theory of optimal (lossy) data compression, which has recently been gaining increasing popularity in cognitive science. More specifically, we characterize optimal conceptual systems as solutions to a special type of RDT problem, show how these optimal systems can generalize to unseen examples, and test their predictions for human behavior in three foundational concept-learning experiments. We find converging evidence that optimal compression may account for human concept learning. Our work also lends new insight into the relation between learnability and compressibility; integrates prototype, exemplar, and Bayesian approaches to human concepts within the RDT framework; and offers a potential theoretical link between concept learning and other cognitive functions that have been successfully characterized by efficient compression.

Scalar implicatures, one of the signatures of pragmatic reasoning, are believed to arise from competing alternative utterances, which the listener knows that the speaker could have used to express a strengthened meaning. But do scalar implicatures also arise in the presence of nonce objects, for which no alternative name is known? We conduct a series of experiments assessing the degree of scalar strengthening driven by familiar and nonce objects. We find that nonce objects can derive scalar implicatures as strongly as familiar objects in simple reference games. Our experiments also reveal an asymmetry in the relative strengths of familiar- and nonce-driven inferences: relative to the prior, participants preferentially interpret the name of a shared feature as referring to an object with an additional nonce feature over an object with an additional familiar feature, suggesting that familiar alternatives exert greater scalar pressure than nonce alternatives. We also present exploratory model simulations suggesting that our results may be explained by rationally reasoning about a high-cost description of the novel object. Our findings support the idea that novel lexical entries may be generated from one-shot encounters and spontaneously used in pragmatic inference.

Systems of personal pronouns (e.g.,`you' and `I') vary widely across languages, but at the same time not all possible systems are attested. Linguistic theories have generally accounted for this in terms of strong grammatical constraints, but recent experimental work challenges this view. Here, we take a novel approach to understanding personal pronoun systems by invoking a recent information-theoretic framework for semantic systems that predicts that languages efficiently compress meanings into forms. We find that a test set of cross-linguistically attested personal pronoun systems achieves near-optimal compression, supporting the hypothesis that efficient compression shapes semantic systems. Further, our best-fitting model includes an egocentric bias that favors a salient speaker representation, accounting for a well-known typological generalization of person systems (`Zwicky's Generalization') without the need for a hard grammatical constraint.

Across many semantic domains, cross-linguistic regularities in categorization systems (e.g., color or kinship terms) have been taken to reflect constraints on how humans perceive and conceptualize the world. Such conceptual representations are often assumed to be universal, and independent of an individual's experience with a particular language. However, in most cases, representational constraints have not been observed empirically on language-independent grounds. This study comes to fill in this gap. We use a card sorting task to provide the first empirical evidence for a common, language-independent representation of pronominal referents, shared by speakers of different languages.

Compositionality is considered a key hallmark of human language. However, most research focuses on item-level compositionality, e.g., to what extent the meanings of phrases are composed of the meanings of their sub-parts, rather than on language-level compositionality, which is the degree to which possible combinations are utilized in practice during language use. Here, we propose a novel way to quantify the degree of language-level compositionality and apply it in the case of English adjective-noun combinations. Using corpus analyses, large language models, and human acceptability ratings, we find that (1) English only sparsely utilizes the compositional potential of adjective–noun combinations; and (2) LLMs struggle to predict human acceptability judgments of rare combinations. Taken together, our findings shed new light on the role of compositionality in language and highlight a challenging area for further improving LLMs.

Iterated models of pragmatic reasoning, such as the Rational Speech Act model (RSA; Frank & Goodman, 2012), aim to explain how meaning is understood in context. We propose an optimal experiment design approach for teasing apart such models, in which contexts are optimized for differentiating model predictions in reference games. We use this approach to compare RSA with RD-RSA (Zaslavsky et al., 2020), a recent variant of RSA grounded in Rate-Distortion theory. First, we show that our optimal experiment design approach finds cases in which the two models yield qualitatively different predictions, in contrast to previous experimental settings for which these models generate similar predictions. Next, we test the models on newly collected experimental data using our optimal design. Our results show that in this experimental setting RD-RSA robustly outperforms the standard RSA model. This finding supports the idea that Rate-Distortion theory may characterize human pragmatic reasoning.

Abstract: It has been proposed that semantic systems evolve under pressure for efficiency. This hypothesis has so far been supported largely indirectly, by synchronic cross-language comparison, rather than directly by diachronic data. Here, we directly test this hypothesis in the domain of color naming, by analyzing recent diachronic data from Nafaanra, a language of Ghana and Côte d’Ivoire, and comparing it with quantitative predictions derived from the mathematical theory of efficient data compression. We show that color naming in Nafaanra has changed over the past four decades while remaining near-optimally efficient, and that this outcome would be unlikely under a random drift process that maintains structured color categories without pressure for efficiency. To our knowledge, this finding provides the first direct evidence that color naming evolves under pressure for efficiency, supporting the hypothesis that efficiency shapes the evolution of the lexicon.

How do humans produce and comprehend language in pragmatic ways? A variety of models of pragmatic inferences have been proposed, and these models are typically evaluated on their ability to account for human inferences in reference game experiments. However, these experiments are not tailored to target theoretical differences between models or clearly tease apart model predictions. We propose an optimal experiment design approach to systematically construct reference games that can optimally differentiate between models of human pragmatic reasoning. We demonstrate this approach and apply it to four models that have been debated in the literature: Grammar-based, Iterated Best Response (IBR), Rational Speech Act (RSA), and a recent variant of RSA grounded in rate-distortion theory (RD-RSA). Using these optimal reference game experiments, we find empirical evidence favoring iterated rationality models over the grammar-based model, as well as support for the relevance of rate-distortion theory to human pragmatic inferences. These results suggest that our optimal reference game design framework may help adjudicate between computational theories of pragmatic reasoning.

Colour naming across languages has traditionally been held to reflect the structure of colour perception. At the same time, it has often, and increasingly, been suggested that colour naming may be shaped by patterns of communicative need. However, much remains unknown about the factors involved in communicative need, how need interacts with perception, and how this interaction may shape colour naming. Here, we engage these open questions by building on general information-theoretic principles. We present a systematic evaluation of several factors that may reflect need, and that have been proposed in the literature: capacity constraints, linguistic usage, and the visual environment. Our analysis suggests that communicative need in colour naming is reflected more directly by capacity constraints and linguistic usage than it is by the statistics of the visual environment.

Gibson et al. () argued that color naming is shaped by patterns of communicative need. In support of this claim, they showed that color naming systems across languages support more precise communication about warm colors than cool colors, and that the objects we talk about tend to be warm-colored rather than cool-colored. Here, we present new analyses that alter this picture. We show that greater communicative precision for warm than for cool colors, and greater communicative need, may both be explained by perceptual structure. However, using an information-theoretic analysis, we also show that color naming across languages bears signs of communicative need beyond what would be predicted by perceptual structure alone. We conclude that color naming is shaped both by perceptual structure, as has traditionally been argued, and by patterns of communicative need, as argued by Gibson et al. -although for reasons other than those they advanced.