Word meaning in minds and machines.

Area of Science:

• Cognitive Science
• Computational Linguistics
• Psychology

Background:

• Recent advancements in Natural Language Processing (NLP) have endowed machines with significant linguistic capabilities.
• Psychologists are increasingly examining NLP models, comparing their outputs to human psychological judgments like similarity, association, and comprehension.
• This has prompted inquiry into whether NLP models can function as viable psychological theories of human cognition.

Purpose of the Study:

• To compare human and machine word meaning representation.
• To evaluate the extent to which current NLP systems model human psychological judgments.
• To identify limitations of current NLP models and propose directions for future development.

Main Methods:

• Comparative analysis of word meaning representations between humans and contemporary NLP systems.
• Evaluation of NLP model outputs against psychological judgments including similarity, association, priming, and comprehension.
• Assessment of NLP models' reliance on text-based patterns versus grounding in human desires, goals, and beliefs.

Main Results:

• Contemporary NLP systems demonstrate considerable success in modeling human word similarity.
• However, these systems fall short in capturing other crucial aspects of human word meaning.
• Current models are overly dependent on statistical patterns in large text corpora, lacking grounding in perception, action, and human intentionality.

Conclusions:

• NLP models show promise as tools for understanding human word similarity but are not yet comprehensive psychological theories.
• Future NLP systems need to integrate grounding in perception and action to develop more human-like conceptual understanding.
• Developing NLP with a stronger conceptual basis is essential for more accurate modeling of human word meaning and cognition.