Multilingual AMR-to-Text Generation

TL;DR

This paper introduces multilingual AMR-to-text generation models that produce natural language in 21 languages from structured AMR data, leveraging cross-lingual embeddings and pretraining, and outperforming single-language baselines in most cases.

Contribution

It presents a novel approach to multilingual AMR-to-text generation that extends previous English-focused work to 21 languages using advanced multilingual models.

Findings

Multilingual models outperform single-language baselines in 18 languages.

Human evaluation shows generated text is fluent and captures morphology and word order.

Models successfully generate diverse languages with varied linguistic properties.

Abstract

Generating text from structured data is challenging because it requires bridging the gap between (i) structure and natural language (NL) and (ii) semantically underspecified input and fully specified NL output. Multilingual generation brings in an additional challenge: that of generating into languages with varied word order and morphological properties. In this work, we focus on Abstract Meaning Representations (AMRs) as structured input, where previous research has overwhelmingly focused on generating only into English. We leverage advances in cross-lingual embeddings, pretraining, and multilingual models to create multilingual AMR-to-text models that generate in twenty one different languages. For eighteen languages, based on automatic metrics, our multilingual models surpass baselines that generate into a single language. We analyse the ability of our multilingual models to accurately capture morphology and word order using human evaluation, and find that native speakers judge our generations to be fluent.