Generating Diverse Translation from Model Distribution with Dropout

Wu, Xuanfu; Feng, Yang; Shao, Chenze

doi:10.48550/arxiv.2010.08178

Cited by 1 publication

(2 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In Freitag and Firat (2020) (Cheng et al, 2016). Despite a large quantity of corpus it can generate, its generated corpus has very low diversity (Wu et al, 2020;Sun et al, 2020;Shen et al, 2019).…”

Section: Mnmtmentioning

confidence: 99%

“…1 One possible solution, referred to as the generation-based approach, is to generate the multi-way aligned examples by distilling the knowledge of the existing NMT model, e.g., extracting German-English-French synthetic threeway aligned data by feeding the English-side sentences of German-English bilingual corpus into the English-French translation model. Although the generation-based approach can theoretically generate non-English corpus with the same size as original bilingual corpus, its generated corpus has very low diversity as the search space of the beam search used by NMT is too narrow to extract diverse translations (Wu et al, 2020;Sun et al, 2020;Shen et al, 2019), which severely limits the power of the generation-based approach.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

EAG: Extract and Generate Multi-way Aligned Corpus for Complete Multi-lingual Neural Machine Translation

Xu¹,

Zhang²,

Meng³

et al. 2022

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)

View full text Add to dashboard Cite

Complete Multi-lingual Neural Machine Translation (C-MNMT) achieves superior performance against the conventional MNMT by constructing multi-way aligned corpus, i.e., aligning bilingual training examples from different language pairs when either their source or target sides are identical. However, since exactly identical sentences from different language pairs are scarce, the power of the multi-way aligned corpus is limited by its scale. To handle this problem, this paper proposes "Extract and Generate" (EAG), a two-step approach to construct large-scale and high-quality multi-way aligned corpus from bilingual data. Specifically, we first extract candidate aligned examples by pairing the bilingual examples from different language pairs with highly similar source or target sentences; and then generate the final aligned examples from the candidates with a welltrained generation model. With this two-step pipeline, EAG can construct a large-scale and multi-way aligned corpus whose diversity is almost identical to the original bilingual corpus. Experiments on two publicly available datasets i.e., WMT-5 and OPUS-100, show that the proposed method achieves significant improvements over strong baselines, with +1.1 and +1.4 BLEU points improvements on the two datasets respectively.

show abstract

Section: Mnmtmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%