Is Cross-Lingual Transfer in Bilingual Models Human-Like? A Study with Overlapping Word Forms in Dutch and English

TL;DR

This study investigates whether bilingual language models exhibit human-like cross-lingual activation patterns, finding that shared embeddings influence effects but may limit their explanatory power for bilingual reading.

Contribution

It demonstrates how embedding sharing conditions affect cross-lingual effects in bilingual models, highlighting the importance of lexical encoding for human-like processing.

Findings

Models show cross-lingual effects mainly when embeddings are shared.

Facilitation occurs for both cognates and false friends with shared embeddings.

Frequency influences cross-lingual effects more than form-meaning consistency.

Abstract

Bilingual speakers show cross-lingual activation during reading, especially for words with shared surface form. Cognates (friends) typically lead to facilitation, whereas interlingual homographs (false friends) cause interference or no effect. We examine whether cross-lingual activation in bilingual language models mirrors these patterns. We train Dutch-English causal Transformers under four vocabulary-sharing conditions that manipulate whether (false) friends receive shared or language-specific embeddings. Using psycholinguistic stimuli from bilingual reading studies, we evaluate the models through surprisal and embedding similarity analyses. The models largely maintain language separation, and cross-lingual effects arise primarily when embeddings are shared. In these cases, both friends and false friends show facilitation relative to controls. Regression analyses reveal that these effects are mainly driven by frequency rather than consistency in form-meaning mapping. Only when just friends share embeddings are the qualitative patterns of bilinguals reproduced. Overall, bilingual language models capture some cross-linguistic activation effects. However, their alignment with human processing seems to critically depend on how lexical overlap is encoded, possibly limiting their explanatory adequacy as models of bilingual reading.