机器翻译

SCORE

BLEU

WER

机器翻译:联合训练 alignment and translater

Bahdanau D, Cho K, Bengio Y. Neural machine translation by jointly learning to align and translate[J]. arXiv preprint arXiv:1409.0473, 2014.

Encoder - Decoder 当句子很长的时候,难以学习,因为需要把一个很长的序列压缩为一个固定长度的向量。

attention 向量是变长的,这个问题怎么解决?

低频词

论文:Luong M, Sutskever I, Le Q V, et al. Addressing the Rare Word Problem in Neural Machine Translation[C]. meeting of the association for computational linguistics, 2014: 11-19.

低频词:混合模型

论文:Luong M, Manning C D. Achieving Open Vocabulary Neural Machine Translation with Hybrid Word-Character Models[C]. meeting of the association for computational linguistics, 2016: 1054-1063.

解决的问题:encoder + decoder + attention 模型,对低频词效果不好,
常将低频词全部映射为一个的特殊词,然后通过后续规则解决。

方法:在 encoder 和 decoder 测加了一个对的 Character Model,这个模型和翻译模型联合优化!