Reproduce “Modeling Coverage for Neural Machine Translation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Zhaopeng Tu, Zhengdong Lu, Yang Liu, Xiaohua Liu, Hang Li

The authors add a running record of attention to each source word in neural machine translation, feeding that coverage into later attention choices to guide the model toward less-attended material without guaranteeing complete translation. In Chinese–English experiments, gated neural coverage with dimension 10 reached average BLEU 30.14, versus 28.32 for GroundHog, across three NIST test sets.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.