OWL: Probing Cross-Lingual Recall of Memorized Texts via World Literature

TL;DR

This study introduces OWL, a multilingual dataset to evaluate how well large language models memorize and recall texts across different languages, revealing significant cross-lingual memorization capabilities.

Contribution

The paper presents OWL, a novel dataset for probing cross-lingual memorization in LLMs, and evaluates models' ability to recall aligned texts across multiple languages.

Findings

LLMs can recall memorized content across languages.

Perturbations slightly decrease recall accuracy.

GPT-4o identifies titles/authors 69% of the time.

Abstract

Large language models (LLMs) are known to memorize and recall English text from their pretraining data. However, the extent to which this ability generalizes to non-English languages or transfers across languages remains unclear. This paper investigates multilingual and cross-lingual memorization in LLMs, probing if memorized content in one language (e.g., English) can be recalled when presented in translation. To do so, we introduce OWL, a dataset of 31.5K aligned excerpts from 20 books in ten languages, including English originals, official translations (Vietnamese, Spanish, Turkish), and new translations in six low-resource languages (Sesotho, Yoruba, Maithili, Malagasy, Setswana, Tahitian). We evaluate memorization across model families and sizes through three tasks: (1) direct probing, which asks the model to identify a book's title and author; (2) name cloze, which requires predicting masked character names; and (3) prefix probing, which involves generating continuations. We find that LLMs consistently recall content across languages, even for texts without direct translation in pretraining data. GPT-4o, for example, identifies authors and titles 69% of the time and masked entities 6% of the time in newly translated excerpts. Perturbations (e.g., masking characters, shuffling words) modestly reduce direct probing accuracy (7% drop for shuffled official translations). Our results highlight the extent of cross-lingual memorization and provide insights on the differences between the models.