Beyond the Rosetta Stone: Unification Forces in Generalization Dynamics

Carter Blum, Katja Filippova, Ann Yuan, Asma Ghandeharioun, Julian Zimmert, Fred Zhang, Jessica Hoffmann, Tal Linzen, Martin Wattenberg, Lucas Dixon, Mor Geva

Keywords: multilinguality, factual recall, training dynamics

Abstract:

Large language models (LLMs) struggle with cross-lingual knowledge transfer: they hallucinate when asked in one language about facts expressed in a different language during training.
This work introduces a controlled setting to study the causes and training dynamics of this phenomenon by training small Transformer models from scratch on synthetic multilingual datasets.
We identify a learning phase wherein a model develops either separate or unified representations of the same facts across languages, and show that unification is essential for cross-lingual transfer.
We also show that the degree of unification depends on how strongly a fact is associated with a particular language, and on how easy it is to identify the language.
Based on these insights, we develop methods to modulate the level of cross-lingual transfer by manipulating data distribution and tokenization, and we introduce metrics and visualizations to characterize their effects on unification.
Finally, we show that our measures of representational unification correlate with cross-lingual factual accuracy in LLMs, such as Gemma.
Our work shows how controlled settings can shed light on pre-training dynamics and suggests new directions for improving cross-lingual transfer in LLMs.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Submission Number: 1105