Cross-Lingual Knowledge Distillation for Answer Sentence Selection in Low-Resource Languages

TL;DR

This paper introduces Cross-Lingual Knowledge Distillation (CLKD), a method for training Answer Sentence Selection models in low-resource languages using English teachers, achieving competitive results without target language labeled data.

Contribution

The paper proposes CLKD, a novel cross-lingual distillation approach, and introduces new multilingual datasets for evaluating AS2 in low-resource languages.

Findings

CLKD outperforms or rivals supervised fine-tuning methods.

Introduces Xtr-WikiQA and TyDi-AS2 datasets for multilingual AS2 evaluation.

Demonstrates effectiveness across diverse languages and models.

Abstract

While impressive performance has been achieved on the task of Answer Sentence Selection (AS2) for English, the same does not hold for languages that lack large labeled datasets. In this work, we propose Cross-Lingual Knowledge Distillation (CLKD) from a strong English AS2 teacher as a method to train AS2 models for low-resource languages in the tasks without the need of labeled data for the target language. To evaluate our method, we introduce 1) Xtr-WikiQA, a translation-based WikiQA dataset for 9 additional languages, and 2) TyDi-AS2, a multilingual AS2 dataset with over 70K questions spanning 8 typologically diverse languages. We conduct extensive experiments on Xtr-WikiQA and TyDi-AS2 with multiple teachers, diverse monolingual and multilingual pretrained language models (PLMs) as students, and both monolingual and multilingual training. The results demonstrate that CLKD either outperforms or rivals even supervised fine-tuning with the same amount of labeled data and a combination of machine translation and the teacher model. Our method can potentially enable stronger AS2 models for low-resource languages, while TyDi-AS2 can serve as the largest multilingual AS2 dataset for further studies in the research community.