Translate and Label! An Encoder-Decoder Approach for Cross-lingual Semantic Role Labeling

TL;DR

This paper introduces a cross-lingual encoder-decoder model that translates and annotates sentences with semantic roles without needing parallel data, improving SRL in resource-poor languages across various settings.

Contribution

The paper presents a novel encoder-decoder approach for cross-lingual SRL that does not require parallel data and can generate high-quality SRL annotations in multiple languages.

Findings

Model performs well in monolingual and multilingual settings.

Generated SRL data enhances resource-poor language performance.

Manual evaluation confirms high-quality sentence translation and annotation.

Abstract

We propose a Cross-lingual Encoder-Decoder model that simultaneously translates and generates sentences with Semantic Role Labeling annotations in a resource-poor target language. Unlike annotation projection techniques, our model does not need parallel data during inference time. Our approach can be applied in monolingual, multilingual and cross-lingual settings and is able to produce dependency-based and span-based SRL annotations. We benchmark the labeling performance of our model in different monolingual and multilingual settings using well-known SRL datasets. We then train our model in a cross-lingual setting to generate new SRL labeled data. Finally, we measure the effectiveness of our method by using the generated data to augment the training basis for resource-poor languages and perform manual evaluation to show that it produces high-quality sentences and assigns accurate semantic role annotations. Our proposed architecture offers a flexible method for leveraging SRL data in multiple languages.