Bilingual models better predict L2 reading when L1 matches
Original title: Bilingual Language Models Capture Cross-Linguistic Influence in Second-Language Reading
The advantage appears in late eye-tracking measures, depends on script overlap, and peaks for intermediate-proficiency readers.
The paper asks whether bilingual language models reflect first-language biases during second-language reading. Models are compared on eye-tracking data from readers with 12 different L1 backgrounds reading English. Matching the reader's L1 beats both a monolingual English baseline and bilingual models trained on other L1s. The gain is concentrated in late measures, varies with L1-L2 script overlap, and is strongest among intermediate-proficiency readers.
- Evaluation uses English eye-tracking data from readers with 12 L1 backgrounds.
- Matched-L1 bilingual models outperform a monolingual English baseline and other-L1 bilingual models.
- The predictive gain is strongest in late-stage eye-tracking measures.
- The effect is modulated by L1-L2 script overlap and peaks at intermediate proficiency.
From the original article
1 min read at arxiv.orgWhile monolingual language model (LM) surprisal is widely established as a predictor of reading times, it cannot capture the non-native (L2) reading dynamics that are shaped by the reader's first language (L1). In this work, we investigate whether the cross-lingual transfer within bilingual models aligns with human L1-driven biases during L2 reading.
By Abderrahmane Issam · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.