Searchable Hidden Intermediates for End-to-End Models of Decomposable Sequence Tasks
This work addresses the challenge of enhancing end-to-end models for speech translation, offering a novel method that bridges the gap between cascaded and end-to-end approaches, though it is incremental in nature.
The paper tackles the problem of improving end-to-end models for complex sequence tasks like speech translation by introducing a framework that learns searchable hidden intermediates from decomposed sub-tasks, resulting in performance gains such as +6 and +3 BLEU on Fisher-CallHome test sets and +3 and +4 BLEU on MuST-C test sets.
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.