一作ACL 2026Findings· Findings of ACL
Localized Cultural Knowledge is Conserved and Controllable in Large Language Models
Veniamin Veselovsky · Berke Argın · Benedikt Stroebl · Chris Wendler · Robert West · James Evans · Thomas L. Griffiths · Arvind Narayanan
摘要
Large language models (LLMs), like human language learners, show patterns influenced by their dominant training language. Just as humans display language patterns influenced by their native tongue (semantic accents) when learning new languages, LLMs often default to English-centric responses even...