GL-CLeF: A Global-Local Contrastive Learning Framework for Cross-lingual Spoken Language Understanding

TL;DR

This paper introduces GL-CLeF, a contrastive learning framework that explicitly aligns multilingual representations for zero-shot cross-lingual spoken language understanding, improving transfer at sentence, token, and semantic levels.

Contribution

The paper proposes a novel contrastive learning approach with local and global components to explicitly align multilingual representations for SLU tasks, outperforming existing methods.

Findings

GL-CLeF achieves state-of-the-art performance on MultiATIS++.

The framework effectively pulls representations of similar sentences across languages closer.

Explicit alignment improves zero-shot cross-lingual transfer in SLU.

Abstract

Due to high data demands of current methods, attention to zero-shot cross-lingual spoken language understanding (SLU) has grown, as such approaches greatly reduce human annotation effort. However, existing models solely rely on shared parameters, which can only perform implicit alignment across languages. We present Global--Local Contrastive Learning Framework (GL-CLeF) to address this shortcoming. Specifically, we employ contrastive learning, leveraging bilingual dictionaries to construct multilingual views of the same utterance, then encourage their representations to be more similar than negative example pairs, which achieves to explicitly aligned representations of similar sentences across languages. In addition, a key step in GL-CLeF is a proposed Local and Global component, which achieves a fine-grained cross-lingual transfer (i.e., sentence-level Local intent transfer, token-level Local slot transfer, and semantic-level Global transfer across intent and slot). Experiments on MultiATIS++ show that GL-CLeF achieves the best performance and successfully pulls representations of similar sentences across languages closer.