Skip to main navigation Skip to search Skip to main content

Sentence-State LSTMs For Sequence-to-Sequence Learning

  • Xuefeng Bai
  • , Yafu Li
  • , Zhirui Zhang
  • , Mingzhou Xu
  • , Boxing Chen
  • , Weihua Luo
  • , Derek Wong
  • , Yue Zhang*
  • *Corresponding author for this work
  • Westlake University
  • Alibaba Group Holding Ltd.
  • University of Macau

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Transformer is currently the dominant method for sequence to sequence problems. In contrast, RNNs have become less popular due to the lack of parallelization capabilities and the relatively lower performance. In this paper, we propose to use a parallelizable variant of bi-directional LSTMs (BiLSTMs), namely sentence-state LSTMs (S-LSTM), as an encoder for sequence-to-sequence tasks. The complexity of S-LSTM is only O(n) as compared to O(n2) of Transformer. On four neural machine translation benchmarks, we empirically find that S-SLTM can achieve significantly better performances than BiLSTM and convolutional neural networks (CNNs). When compared to Transformer, our model gives competitive performance while being 1.6 times faster during inference.

Original languageEnglish
Title of host publicationNatural Language Processing and Chinese Computing - 10th CCF International Conference, NLPCC 2021, Proceedings
EditorsLu Wang, Yansong Feng, Yu Hong, Ruifang He
PublisherSpringer Science and Business Media Deutschland GmbH
Pages104-115
Number of pages12
ISBN (Print)9783030884796
DOIs
StatePublished - 2021
Externally publishedYes
Event10th CCF Conference on Natural Language Processing and Chinese Computing, NLPCC 2021 - Qingdao, China
Duration: 13 Oct 202117 Oct 2021

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13028 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference10th CCF Conference on Natural Language Processing and Chinese Computing, NLPCC 2021
Country/TerritoryChina
CityQingdao
Period13/10/2117/10/21

Keywords

  • Bi-directional LSTMs
  • CNN
  • Neural machine translation
  • Sentence-State LSTMs
  • Transformers

Fingerprint

Dive into the research topics of 'Sentence-State LSTMs For Sequence-to-Sequence Learning'. Together they form a unique fingerprint.

Cite this