Causal interventions expose implicit situation models for commonsense language understanding
Takateru Yamakoshi, James L. McClelland, Adele Eva Goldberg, Robert D. Hawkins · 2023
Accounts of human language processing have long appealed to implicit "situation models" that enrich comprehension with relevant but unstated world knowledge.Here, we apply causal intervention techniques to recent transformer models to analyze performance on the Winograd Schema Challenge (WSC), where a single context cue shifts interpretation of an ambiguous pronoun.We identify a relatively small circuit of attention heads that are responsible for propagating information from the context word that guides which of the candidate noun phrases the pronoun ultimately attends to.We then compare how this circuit behaves in a closely matched "syntactic" control where the situation model is not strictly necessary.These analyses suggest distinct pathways through which implicit situation models are constructed to guide pronoun resolution 1 . AJoe paid the detective because has received the final report on the case.Joe paid the detective because has delivered the final report on the case. BJoe paid the detectives because