Cross-Lingual Transfer of Cognitive Processing Complexity

TL;DR

This study demonstrates that multilingual models like XLM-RoBERTa can predict cross-lingual sentence complexity patterns using eye-tracking data, revealing their sensitivity to structural features across diverse languages.

Contribution

The paper shows that XLM-RoBERTa, trained only on English, can generalize to predict structural complexity in 13 different languages using eye-tracking data.

Findings

XLM-RoBERTa predicts eye-tracking patterns across 13 languages.

The model is biased towards sentence length but also captures structural differences.

It detects more complex structures beyond simple word order.

Abstract

When humans read a text, their eye movements are influenced by the structural complexity of the input sentences. This cognitive phenomenon holds across languages and recent studies indicate that multilingual language models utilize structural similarities between languages to facilitate cross-lingual transfer. We use sentence-level eye-tracking patterns as a cognitive indicator for structural complexity and show that the multilingual model XLM-RoBERTa can successfully predict varied patterns for 13 typologically diverse languages, despite being fine-tuned only on English data. We quantify the sensitivity of the model to structural complexity and distinguish a range of complexity characteristics. Our results indicate that the model develops a meaningful bias towards sentence length but also integrates cross-lingual differences. We conduct a control experiment with randomized word order and find that the model seems to additionally capture more complex structural information.