2024-03-29T09:29:43Zhttps://ipsj.ixsq.nii.ac.jp/ej/?action=repository_oaipmhoai:ipsj.ixsq.nii.ac.jp:001837922023-04-27T10:00:04Z01164:04179:09105:09268
Improving Neural Machine Translation with Linearized Dependency Tree DecoderImproving Neural Machine Translation with Linearized Dependency Tree Decodereng機械翻訳(2)http://id.nii.ac.jp/1001/00183704/Technical Reporthttps://ipsj.ixsq.nii.ac.jp/ej/?action=repository_action_common_download&item_id=183792&item_no=1&attribute_id=1&file_no=1Copyright (c) 2017 by the Information Processing Society of JapanNara Institute of Science and TechnologyNara Institute of Science and TechnologyNara Institute of Science and TechnologyAn, Nguyen LeAnder, MartinezYuji, MatsumotoIn spite of achieving significant performance in recent years, there are some existing issues that Sequence-to-Sequence Neural Machine Translation still does not solve completely. Two of them are translation for long sentences and the over-translation. To address these two problems, we propose an approach that utilize more syntactic information, syntactic dependency information, so that the output is generated based on more abundant information. In addition, the outputs are presented not as a simple sequence of tokens but as a linear grammatical tree structure. In addition, the output of the model is presented not as a simple sequence of tokens but as a linearized tree construction. Experiments on the Europarl-v7 dataset of French-to-English translation demonstrate that our proposed method can produce dependency relations between words in the target language and improve BLEU scores by 1.57 and 2.40 on datasets consisting of sentences with up to 50 and 80 tokens, respectively. Furthermore, the proposed method also solved the ineffective translation for long sentences and repetition problems in Neural Machine Translation.In spite of achieving significant performance in recent years, there are some existing issues that Sequence-to-Sequence Neural Machine Translation still does not solve completely. Two of them are translation for long sentences and the over-translation. To address these two problems, we propose an approach that utilize more syntactic information, syntactic dependency information, so that the output is generated based on more abundant information. In addition, the outputs are presented not as a simple sequence of tokens but as a linear grammatical tree structure. In addition, the output of the model is presented not as a simple sequence of tokens but as a linearized tree construction. Experiments on the Europarl-v7 dataset of French-to-English translation demonstrate that our proposed method can produce dependency relations between words in the target language and improve BLEU scores by 1.57 and 2.40 on datasets consisting of sentences with up to 50 and 80 tokens, respectively. Furthermore, the proposed method also solved the ineffective translation for long sentences and repetition problems in Neural Machine Translation.AN10115061研究報告自然言語処理(NL)2017-NL-23310172017-10-172188-87792017-10-12