Orthogonal Language and Task Adapters in Zero-Shot Cross-Lingual Transfer

TL;DR

This paper introduces orthogonal language and task adapters for zero-shot cross-lingual transfer in multilingual transformers, showing their effectiveness varies by task and language, especially for complex tasks like NLI.

Contribution

It proposes orthogonal adapters that encode task- and language-specific info orthogonal to pretrained knowledge, enhancing cross-lingual transfer performance.

Findings

Orthogonal adapters improve zero-shot transfer, especially for NLI.

Optimal adapter configurations depend on task and language.

Orthogonality constraints show promise in fine-tuning multilingual models.

Abstract

Adapter modules, additional trainable parameters that enable efficient fine-tuning of pretrained transformers, have recently been used for language specialization of multilingual transformers, improving downstream zero-shot cross-lingual transfer. In this work, we propose orthogonal language and task adapters (dubbed orthoadapters) for cross-lingual transfer. They are trained to encode language- and task-specific information that is complementary (i.e., orthogonal) to the knowledge already stored in the pretrained transformer's parameters. Our zero-shot cross-lingual transfer experiments, involving three tasks (POS-tagging, NER, NLI) and a set of 10 diverse languages, 1) point to the usefulness of orthoadapters in cross-lingual transfer, especially for the most complex NLI task, but also 2) indicate that the optimal adapter configuration highly depends on the task and the target language. We hope that our work will motivate a wider investigation of usefulness of orthogonality constraints in language- and task-specific fine-tuning of pretrained transformers.