Zero-Shot Cross-lingual Semantic Parsing

TL;DR

This paper introduces a zero-shot cross-lingual semantic parser that leverages a multi-task encoder-decoder model to transfer knowledge across languages without relying on parallel data or high-quality translation tools.

Contribution

It presents a novel approach that removes the need for parallel data and translation systems, enabling semantic parsing in new languages through language-agnostic encodings.

Findings

Outperforms translation-based baselines in cross-lingual parsing tasks.

Achieves performance close to supervised upper-bound in some languages.

Demonstrates effective knowledge transfer without parallel data.

Abstract

Recent work in cross-lingual semantic parsing has successfully applied machine translation to localize parsers to new languages. However, these advances assume access to high-quality machine translation systems and word alignment tools. We remove these assumptions and study cross-lingual semantic parsing as a zero-shot problem, without parallel data (i.e., utterance-logical form pairs) for new languages. We propose a multi-task encoder-decoder model to transfer parsing knowledge to additional languages using only English-logical form paired data and in-domain natural language corpora in each new language. Our model encourages language-agnostic encodings by jointly optimizing for logical-form generation with auxiliary objectives designed for cross-lingual latent representation alignment. Our parser performs significantly above translation-based baselines and, in some cases, competes with the supervised upper-bound.