Fast Interleaved Bidirectional Sequence Generation

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Independence assumptions during sequence generation can speed up inference, but parallel generation of highly inter-dependent tokens comes at a cost in quality. Instead of assuming independence between neighbouring tokens (semi-autoregressive decoding, SA), we take inspiration from bidirectional sequence generation and introduce a decoder that generates target words from the left-to-right and right-to-left directions simultaneously. We show that we can easily convert a standard architecture for unidirectional decoding into a bidirectional decoder by simply interleaving the two directions and adapting the word positions and self-attention masks. Our interleaved bidirectional decoder (IBDecoder) retains the model simplicity and training efficiency of the standard Transformer, and on five machine translationtasks and two document summarization tasks, achieves a decoding speedup of ∼2× compared to autoregressive decoding with comparable quality. Notably, it outperforms left-to-right SA because the independence assumptions in IBDecoder are more felicitous. To achieve even higher speedups, we explore hybrid models where we either simultaneously predict multiple neighbouring tokens per direction, or perform multi-directional decoding by partitioning the target sequence. These methods achieve speedups to 4×–11× across different tasks at the cost of <1 BLEU or <0.5 ROUGE (on average).
Original languageEnglish
Title of host publicationProceedings of the Fifth Conference on Machine Translation
Place of PublicationOnline
PublisherAssociation for Computational Linguistics
Pages501-513
Number of pages13
ISBN (Print)978-1-948087-81-0
Publication statusPublished - 19 Nov 2020
EventFifth Conference on Machine Translation - Online Conference
Duration: 19 Nov 202020 Nov 2020
http://www.statmt.org/wmt20/

Conference

ConferenceFifth Conference on Machine Translation
Abbreviated titleWMT 2020
CityOnline Conference
Period19/11/2020/11/20
Internet address

Fingerprint

Dive into the research topics of 'Fast Interleaved Bidirectional Sequence Generation'. Together they form a unique fingerprint.

Cite this