The Dual-Route Model of Induction

Sheridan Feucht, Eric Todd, Byron C Wallace, David Bau

Keywords: interpretability, induction heads, in-context learning, ICL, detokenization

TL;DR: We find that LLMs can do in-context copying in two different ways: either by copying individual tokens verbatim, or by copying entire word meanings (which may span multiple tokens).

Abstract: Prior work on in-context copying has shown the existence of *induction heads*, which attend to and promote individual tokens during copying. In this work we discover a new type of induction head: *concept-level* induction heads, which copy entire lexical units instead of individual tokens. Concept induction heads learn to attend to the ends of multi-token words throughout training, working in parallel with token-level induction heads to copy meaningful text. We show that these heads are responsible for semantic tasks like word-level translation, whereas token induction heads are vital for tasks that can only be done verbatim (like copying nonsense tokens). These two "routes" operate independently: we show that ablation of token induction heads causes models to paraphrase where they would otherwise copy verbatim. By patching concept induction head outputs, we find that they contain language-independent word representations that mediate natural language translation, suggesting that LLMs represent abstract word meanings independent of language or form.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the COLM Code of Ethics on https://colmweb.org/CoE.html

Author Guide: I certify that this submission complies with the submission instructions as described on https://colmweb.org/AuthorGuide.html

Submission Number: 36