Simultaneous speech translation methods for news and lectures in foreign languages

2016 Fiscal Year Final Research Report

• Project/Area Number: 24240032
• Research Category: Grant-in-Aid for Scientific Research (A)
• Allocation Type: Single-year Grants
• Section: 一般
• Research Field: Perception information processing/Intelligent robotics
• Research Institution: Nara Institute of Science and Technology
• Principal Investigator: Nakamura Satoshi 奈良先端科学技術大学院大学, 情報科学研究科, 教授 (30263429)
• Co-Investigator(Kenkyū-buntansha): 松本 裕治 奈良先端科学技術大学院大学, 情報科学研究科, 教授 (10211575) ; 戸田 智基 奈良先端科学技術大学院大学, 情報科学研究科, 准教授 (90403328) ; サクリアニ サクティ 奈良先端科学技術大学院大学, 情報科学研究科, 助教 (00395005) ; Neubig Graham 奈良先端科学技術大学院大学, 情報科学研究科, 助教 (70633428) ; Duh Kevin 奈良先端科学技術大学院大学, 情報科学研究科, 助教 (80637322) ; 小町 守 奈良先端科学技術大学院大学, 情報科学研究科, 助教 (60581329)
• Project Period (FY): 2012-05-31 – 2017-03-31
• Keywords: 音声情報処理 / 音声翻訳 / 音声認識 / 機械翻訳

Outline of Final Research Achievements

In this project new simultaneous speech-to-speech translation algorithms are proposed. First algorithm has a mechanism to decide to output or hold the phrases to the machine translation module until the current time based on the right probability in the phrase-based statistical machine translation. Second algorithm is able to segment the input phrase sequence based on greedy search according to POS bigram information. Third algorithm predicts next phrase or local parse tree element based on SVM with the incremental bottom-up parser. Here, the algorithm decides to output or hold the phrases again. The experiments showed that the proposed algorithms successfully realized the simultaneous speech translation. Furthermore neural machine translation algorithms with attention mechanisms are investigated. The 80 hours of J-E interpretation data, 50 hours of JP lecture transcription data, and 22 hours of J-E translation data are collected to be used for simultaneous speech translation research.

Free Research Field

知能コミュニケーション