arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1906.01282cs.CL
收藏

基于格的神经机器翻译Transformer编码器

Lattice-Based Transformer Encoder for Neural Machine Translation

Shanghai Jiao Tong University(上海交通大学);National Institute of Information and Communications Technology (NICT)(日本信息通信研究机构(NICT))

Fengshun Xiao, Jiangtong Li, Hai Zhao, Rui Wang, Kehai Chen

摘要:

神经机器翻译(NMT)采用确定性序列作为源表示。然而,无论是词级还是子词级分割都有多种选择,可以使用不同的词分割器或不同的子词词汇量来分割源序列。我们假设分割的多样性可能会影响NMT的性能。为了将不同的分割与最先进的NMT模型Transformer相结合,我们提出了基于格的编码器,以在训练过程中自动探索有效的单词或子单词表示。我们提出了两种方法:1)格位置编码和2)格感知的自我注意。这两种方法可以结合使用,相互补充,进一步提高翻译性能。实验结果表明,基于格的编码器在字级和子字级表示上优于传统的Transformer编码器。

英文摘要:

Neural machine translation (NMT) takes deterministic sequences for source representations. However, either word-level or subword-level segmentations have multiple choices to split a source sequence with different word segmentors or different subword vocabulary sizes. We hypothesize that the diversity in segmentations may affect the NMT performance. To integrate different segmentations with the state-of-the-art NMT model, Transformer, we propose lattice-based encoders to explore effective word or subword representation in an automatic way during training. We propose two methods: 1) lattice positional encoding and 2) lattice-aware self-attention. These two methods can be used together and show complementary to each other to further improve translation performance. Experiment results show superiorities of lattice-based encoders in word-level and subword-level representations over conventional Transformer encoder.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑