Causal Drawbridges: Characterizing Gradient Blocking of Syntactic Islands in Transformer LMs
For theoretical linguists and NLP researchers, this work demonstrates how mechanistic interpretability can generate new hypotheses about linguistic representation and processing, though the findings are incremental and domain-specific.
This paper uses causal interventions in Transformer LMs to show that extraction from syntactic islands (e.g., coordinated VPs) engages the same filler-gap mechanisms as canonical wh-dependencies, but these mechanisms are selectively blocked to varying degrees, replicating human gradient judgments. The authors derive a novel hypothesis about the representation of 'and' in extractable vs. non-extractable constructions.
We show how causal interventions in Transformer models provide insights into English syntax by focusing on a long-standing challenge for syntactic theory: syntactic islands. Extraction from coordinated verb phrases is often degraded, yet acceptability varies gradiently with lexical content (e.g., "I know what he hates art and loves" vs. "I know what he looked down and saw"). We show that modern Transformer language models replicate human judgments across this gradient. Using causal interventions that isolate functionally relevant subspaces in Transformer blocks, attention modules, and MLPs, we demonstrate that extraction from coordination islands engages the same filler-gap mechanisms as canonical wh-dependencies, but that these mechanisms are selectively blocked to varying degrees. By projecting a large corpus of unrelated text onto these causally identified subspaces, we derive a novel linguistic hypothesis: the conjunction "and" is represented differently in extractable versus non-extractable constructions, corresponding to expressions encoding relational dependencies versus purely conjunctive uses. These results illustrate how mechanistic interpretability can inform syntax, generating new hypotheses about linguistic representation and processing.