Bilingual language models capture real cross-linguistic influence in second-language reading, suggesting that L1 knowledge shapes how non-native readers process L2 text in measurable ways.
This paper shows that bilingual language models can predict how non-native English readers process text better when trained on their native language.