Neural Machine Translation with CKY-based Convolutional Attention
-
- Watanabe Taiki
- Graduate School of Science and Engineering, Ehime University
-
- Tamura Akihiro
- Graduate School of Science and Engineering, Ehime University
-
- Ninomiya Takashi
- Graduate School of Science and Engineering, Ehime University
-
- Bharata Adji Teguh
- Department of Electrical Engineering and Information Technology, Universitas Gadjah Mada
Bibliographic Information
- Other Title
-
- CKY に基づく畳み込みアテンション構造を用いたニューラル機械翻訳
- CKY ニ モトズク タタミコミ アテンション コウゾウ オ モチイタ ニューラル キカイ ホンヤク
Search this article
Abstract
<p>This paper proposes a new attention mechanism for neural machine translation (NMT) based on convolutional neural networks (CNNs), which is inspired by the CKY algorithm. The proposed attention represents every possible combination of source words (e.g., phrases and structures) through CNNs, which imitates the CKY table in the algorithm. NMT, incorporating the proposed attention, decodes a target sentence on the basis of the attention scores of the hidden states of CNNs. The proposed attention enables NMT to capture alignments from underlying structures of a source sentence without sentence parsing. The evaluations on the Asian Scientific Paper Excerpt Corpus (ASPEC) English-Japanese translation task show that the proposed attention gains 1.43 points in BLEU as compared to a conventional attention-based encoder decoder model. Furthermore, the proposed attention is at least comparable to, or better than, a conventional attention-based encoder decoder model on the FBIS Chinese-English translation task. </p>
Journal
-
- Journal of Natural Language Processing
-
Journal of Natural Language Processing 26 (1), 207-230, 2019-03-15
The Association for Natural Language Processing
- Tweet
Details 詳細情報について
-
- CRID
- 1390564238099533952
-
- NII Article ID
- 130007663695
-
- NII Book ID
- AN10472659
-
- ISSN
- 21858314
- 13407619
-
- NDL BIB ID
- 029581382
-
- Text Lang
- ja
-
- Data Source
-
- JaLC
- NDL
- Crossref
- CiNii Articles
- KAKEN
-
- Abstract License Flag
- Disallowed