Why Do LLMs Accept Generics? Explaining Acceptability using Psycholinguistic and Model-Derived Factors

TL;DR

This paper explores why large language models accept certain generic statements by analyzing psycholinguistic and model-derived factors. The results show that models weigh property commonality against kind distinguishability, similar to human judgment. Models prioritize factual truth in their judgments.

Why It Matters

Understanding how LLMs judge generics is crucial for improving their performance in natural language processing tasks. This research has implications for developing more accurate and informative language models. It can also inform the design of more effective language understanding systems.

Key Findings

• LLMs consider property commonality and kind distinguishability when judging generics
• Models favor characteristic properties but ignore striking ones
• Factual truth is the primary factor influencing LLM judgments