Joint Speech Translation and Named Entity Recognition

TL;DR

This paper introduces multitask models that jointly perform speech translation and named entity recognition, outperforming cascade baselines without sacrificing translation quality or efficiency.

Contribution

It presents a novel joint modeling approach for speech translation and NER, improving NER performance while maintaining translation quality and efficiency.

Findings

Joint models outperform cascade baselines in NER accuracy

No degradation in translation quality with joint models

Maintains computational efficiency of direct speech translation

Abstract

Modern automatic translation systems aim at place the human at the center by providing contextual support and knowledge. In this context, a critical task is enriching the output with information regarding the mentioned entities, which is currently achieved processing the generated translation with named entity recognition (NER) and entity linking systems. In light of the recent promising results shown by direct speech translation (ST) models and the known weaknesses of cascades (error propagation and additional latency), in this paper we propose multitask models that jointly perform ST and NER, and compare them with a cascade baseline. The experimental results show that our models significantly outperform the cascade on the NER task (by 0.4-1.0 F1), without degradation in terms of translation quality, and with the same computational efficiency of a plain direct ST model.