LLMs Learn Verb Meanings via Syntax, Not Co-occurrence, Study Finds
neuranna · x · 2026-09-29
A Princeton team (Xiaomeng Zhu, RTom McCoy, Bob Frank et al.) published "The Structural Sources of Verb Meanings Revisited: Large Language Models Display Syntactic Bootstrapping" in Open Mind, testing the classic syntactic bootstrapping hypothesis using LLMs.
- Question: when LLMs learn verb meanings, does syntactic structure or co-occurrence information matter more?
- Finding: verb learning depends more on syntactic information than co-occurrence, consistent with classic hypotheses in the child language acquisition literature.
- Implication: LLMs can serve as computational models for testing human language acquisition theories.
More from Research
- Sampling the SK Model at β<1: New Polynomial-Time Algorithm Closes a 2022 Open Problem — canondetortugas · 2026-09-29
- Testing Jev on Reasoning-Intensive Regression: Wicked Fast but Classification-Focused — dbreunig · 2026-09-29
- MIT Scale ML seminar returns: talks from Jane Street, Anthropic, TML and more — a1zhang · 2026-09-29
- VLS: training-free VLM steering of frozen robot policies wins CoRL and CVPR workshop awards — micoolcho · 2026-09-29
- HCOMP 2026 Talk: Adapting Socio-Technical Congruence to Human-Agent Collectives — windx0303 · 2026-09-29
- Google Trends' #1 US region for every query is tiny Cheyenne, Wyoming — likely bot traffic — lilyraynyc · 2026-09-29