Domain-general categorisation explains constrained cross-linguistic variation in noun classification

Area of Science:

• Linguistic cognition
• Cognitive psychology
• Computational linguistics

Background:

• Grammatical systems frequently encode distinctions like person, number, and animacy.
• However, concepts like color, despite visual salience, are typically absent from grammatical agreement systems.
• This absence has led some to hypothesize domain-specific constraints on grammar.

Purpose of the Study:

• To investigate whether domain-general cognitive capacities, rather than domain-specific constraints, explain the prevalence of animacy over color in grammatical systems.
• To test the hypothesis that a bias for animacy over color in learning and categorization underlies its common grammatical encoding.

Main Methods:

• Learning of a miniature artificial agreement system with animacy- and color-based agreement rules.
• A non-linguistic sorting task to assess categorization biases.
• Analysis of natural language corpus data to evaluate the learnability of animacy-based categories.
• Experimental manipulation of feature predictiveness in a learning task.

Main Results:

• A significant bias favoring animacy over color was observed during the learning of the artificial agreement system.
• A parallel animacy-over-color bias was found in the non-linguistic sorting task.
• Animacy-based classification in natural language corpora yields more distinct and compact categories, enhancing learnability.
• Preliminary evidence suggests that when animacy is less predictive, learning animacy-based noun classes is impaired.

Conclusions:

• Domain-general cognitive principles, particularly categorization biases favoring highly predictive features, likely explain the common grammatical encoding of animacy over color.
• These findings challenge the notion of domain-specific constraints and support a cognitive explanation for cross-linguistic regularities in grammar.