Mechanistic Understanding and Mitigation of Language Confusion in English-Centric Large Language Models
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909793927561216 |
|---|---|
| author | Nie, Ercong Schmid, Helmut Schütze, Hinrich |
| author_facet | Nie, Ercong Schmid, Helmut Schütze, Hinrich |
| contents | Language confusion -- where large language models (LLMs) generate unintended languages against the user's need -- remains a critical challenge, especially for English-centric models. We present the first mechanistic interpretability (MI) study of language confusion, combining behavioral benchmarking with neuron-level analysis. Using the Language Confusion Benchmark (LCB), we show that confusion points (CPs) -- specific positions where language switches occur -- are central to this phenomenon. Through layer-wise analysis with TunedLens and targeted neuron attribution, we reveal that transition failures in the final layers drive confusion. We further demonstrate that editing a small set of critical neurons, identified via comparative analysis with a multilingual-tuned counterpart, substantially mitigates confusion while largely preserving general competence and fluency. Our approach matches multilingual alignment in confusion reduction for many languages and yields cleaner, higher-quality outputs. These findings provide new insights into the internal dynamics of LLMs and highlight neuron-level interventions as a promising direction for robust, interpretable multilingual language modeling. Code and data are available at: https://github.com/ercong21/lang_confusion. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_16538 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Mechanistic Understanding and Mitigation of Language Confusion in English-Centric Large Language Models Nie, Ercong Schmid, Helmut Schütze, Hinrich Computation and Language Language confusion -- where large language models (LLMs) generate unintended languages against the user's need -- remains a critical challenge, especially for English-centric models. We present the first mechanistic interpretability (MI) study of language confusion, combining behavioral benchmarking with neuron-level analysis. Using the Language Confusion Benchmark (LCB), we show that confusion points (CPs) -- specific positions where language switches occur -- are central to this phenomenon. Through layer-wise analysis with TunedLens and targeted neuron attribution, we reveal that transition failures in the final layers drive confusion. We further demonstrate that editing a small set of critical neurons, identified via comparative analysis with a multilingual-tuned counterpart, substantially mitigates confusion while largely preserving general competence and fluency. Our approach matches multilingual alignment in confusion reduction for many languages and yields cleaner, higher-quality outputs. These findings provide new insights into the internal dynamics of LLMs and highlight neuron-level interventions as a promising direction for robust, interpretable multilingual language modeling. Code and data are available at: https://github.com/ercong21/lang_confusion. |
| title | Mechanistic Understanding and Mitigation of Language Confusion in English-Centric Large Language Models |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2505.16538 |