Saved in:
Bibliographic Details
Main Authors: Oh, Soyoung, Huang, Xinting, Pink, Mathis, Hahn, Michael, Demberg, Vera
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2506.01723
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915735014473728
author Oh, Soyoung
Huang, Xinting
Pink, Mathis
Hahn, Michael
Demberg, Vera
author_facet Oh, Soyoung
Huang, Xinting
Pink, Mathis
Hahn, Michael
Demberg, Vera
contents Idioms present a unique challenge for language models due to their non-compositional figurative interpretations, which often strongly diverge from the idiom's literal interpretation. In this paper, we employ causal tracing to systematically analyze how pretrained causal transformers deal with this ambiguity. We localize three mechanisms: (i) Early sublayers and specific attention heads retrieve an idiom's figurative interpretation, while suppressing its literal interpretation. (ii) When disambiguating context precedes the idiom, the model leverages it from the earliest layer and later layers refine the interpretation if the context conflicts with the retrieved interpretation. (iii) Then, selective, competing pathways carry both interpretations: an intermediate pathway prioritizes the figurative interpretation and a parallel direct route favors the literal interpretation, ensuring that both readings remain available. Our findings provide mechanistic evidence for idiom comprehension in autoregressive transformers.
format Preprint
id arxiv_https___arxiv_org_abs_2506_01723
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Tug-of-war between idioms' figurative and literal interpretations in LLMs
Oh, Soyoung
Huang, Xinting
Pink, Mathis
Hahn, Michael
Demberg, Vera
Computation and Language
Artificial Intelligence
Idioms present a unique challenge for language models due to their non-compositional figurative interpretations, which often strongly diverge from the idiom's literal interpretation. In this paper, we employ causal tracing to systematically analyze how pretrained causal transformers deal with this ambiguity. We localize three mechanisms: (i) Early sublayers and specific attention heads retrieve an idiom's figurative interpretation, while suppressing its literal interpretation. (ii) When disambiguating context precedes the idiom, the model leverages it from the earliest layer and later layers refine the interpretation if the context conflicts with the retrieved interpretation. (iii) Then, selective, competing pathways carry both interpretations: an intermediate pathway prioritizes the figurative interpretation and a parallel direct route favors the literal interpretation, ensuring that both readings remain available. Our findings provide mechanistic evidence for idiom comprehension in autoregressive transformers.
title Tug-of-war between idioms' figurative and literal interpretations in LLMs
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2506.01723