Searchable Hidden Intermediates for End-to-End Models of Decomposable Sequence Tasks

TL;DR

This paper introduces a novel end-to-end framework that learns searchable hidden intermediates in sequence models, leveraging decomposed sub-tasks and beam search to improve performance in complex tasks like speech translation.

Contribution

It proposes a framework that exploits compositionality to learn and improve hidden intermediates, enabling better search and adaptation in end-to-end sequence models.

Findings

Outperforms previous state-of-the-art by +6 and +3 BLEU on Fisher-CallHome test sets.

Achieves +3 and +4 BLEU improvements on MuST-C English-German and English-French datasets.

Demonstrates the effectiveness of searchable hidden intermediates in complex sequence tasks.

Abstract

End-to-end approaches for sequence tasks are becoming increasingly popular. Yet for complex sequence tasks, like speech translation, systems that cascade several models trained on sub-tasks have shown to be superior, suggesting that the compositionality of cascaded systems simplifies learning and enables sophisticated search capabilities. In this work, we present an end-to-end framework that exploits compositionality to learn searchable hidden representations at intermediate stages of a sequence model using decomposed sub-tasks. These hidden intermediates can be improved using beam search to enhance the overall performance and can also incorporate external models at intermediate stages of the network to re-score or adapt towards out-of-domain data. One instance of the proposed framework is a Multi-Decoder model for speech translation that extracts the searchable hidden intermediates from a speech recognition sub-task. The model demonstrates the aforementioned benefits and outperforms the previous state-of-the-art by around +6 and +3 BLEU on the two test sets of Fisher-CallHome and by around +3 and +4 BLEU on the English-German and English-French test sets of MuST-C.