Probing Semantic Alignment, Lexical Invariance, and Syntactic Influence in LLM Metaphor Processing
A new research paper explores how large language models (LLMs) process metaphors, examining semantic alignment, lexical invariance, and syntactic sensitivity. The study found that while LLMs perform well on metaphor tasks, their interpretations can show semantic drift and are sensitive to syntactic changes. The research suggests that strong performance on metaphor benchmarks may not always indicate a deep, integrated semantic understanding. AI
IMPACT Highlights limitations in interpreting LLM benchmark performance for metaphor understanding.