| 隆辻 秀和 | M, 2回目発表 | 知能コミュニケーション | 中村 哲, 松本 裕治, 吉野 幸一郎 |
|
title: A study of neural language generation model combining given information re-prediction model
abstract: Language generation is a task that generates sentence to appropriate form with the set of external information. In recent years, language generation with neural network has achieved the capability of generates sentence natural and fluent than the classical model. but, this method does not promise generated response reflects given information fully because only trained with word predict error. We propose a neural language generation model with information re-predicting model and objective function handles both word and information prediction error. We present experiments with DSTC2 corpus and evaluate generation quality in terms of informativeness and naturalness. language of the presentation: Japanese | |||
| 田中 翔平 | M, 2回目発表 | 知能コミュニケーション | 中村 哲, 松本 裕治, 吉野 幸一郎, 須藤 克仁 |
|
title: Conversational Response Re-ranking Based on Event Causality and Role Factored Tensor Event Embedding
abstract: We propose a novel method for selecting coherent and diverse responses for a given dialogue context. The proposed method re-ranks response candidates generated from conversational models by using event causality relations between events in a dialogue history and response candidates (e.g., ``be stressed out'' precedes ``relieve stress''). We use distributed event representation based on the Role Factored Tensor Model for a robust matching of event causality relations due to limited event causality knowledge of the system. Experimental results showed that the proposed method improved coherency and dialogue continuity of system responses. However, naturality of some responses were lowered due to the overgeneralization of events. We describe the examples and discuss the solutions. language of the presentation: Japanese 発表題目: 因果関係と事態分散表現を用いた雑談対話応答のリランキング 発表概要: 本研究では，対話履歴に対し一貫した多様な応答を選択する手法を提案する．提案手法では対話履歴に対する一貫性を保つため，対話モデルより生成された応答候補を，対話履歴と応答候補の間に存在する因果関係（ストレスが溜まる → 発散する，など）を用いてリランキングする．この際，因果関係の認定には統計的に獲得された因果関係ペアを用いるが，対話中に存在する全ての因果関係を被覆するような辞書を用意することは難しい．そこで，Role Factored Tensor Model を用いて事態を分散表現に変換することで，因果関係知識のカバレージを向上させ，因果関係知識と対話中の因果関係の頑強なマッチングを実現した．自動評価，人手評価の結果，提案手法は応答の一貫性や対話継続性を向上させることが確認できた．一方で，事態の過汎化に由来する応答の自然性低下が見られる場合もあった．これらの問題についても例示し，解決の方向性について論じる． | |||
| 帖佐 克己 | M, 2回目発表 | 知能コミュニケーション | 中村 哲, 松本 裕治, 須藤 克仁 |
|
title: Japanese-English Real-time Neural Machine Translation using Connectionist Temporal Classification
abstract: Real-time Translation is a task that outputs translated words before inputting an entire source sentence. In this task, there is a trade-off between translation accuracy and delay time, hence the system needs to determine the output timing appropriately. In the proposed method, we add meta token '<wait>' which is output instead of delaying translations to target-side vocabulary. We also introduce Connectionist Temporal Classification to the loss function. This function can consider all sequences which can be transformed into a correct sequence by removing the delay tokens. Therefore, this function can optimize the translation model and the output timing-control model simultaneously and the output timing is decided adaptively. The results of the experiments with two kinds of corpora showed that the proposed method can decide output timing adaptively and achieved comparable translation accuracy to the previous method. language of the presentation: Japanese | |||
| 木下 泰輝 | M, 2回目発表 | 知能コミュニケーション | 中村 哲, 松本 裕治, 田中 宏季 |
|
title: Measuring affective sharing between two people from simultaneously recorded EEG signals
abstract: Empathy plays an important role in human social interaction. For example, It promotes stronger relationships and collaboration. As empathy relies on inter-brain neural synchronization, psychophysiological measurement of empathy should be possible. In this study, we measured affective sharing, one of the main components of empathy, from EEG signals. To elicit affective sharing, we conducted an experiment in which participants communicated using facial expressions of joy, sadness, and neutrality. EEG signals were simultaneously recorded from both participants during the experiment. The result showed the correlations of the EEG powers were significantly higher under the joy and sadness conditions in the alpha-mu band. This result demonstrates that it is possible to measure affective sharing in response to emotional faces from the correlation of EEG powers. To our knowledge, this is the first EEG hyperscanning study that investigates affective sharing in response to emotional faces. language of the presentation: Japanese 発表題目: 同時計測された脳波信号からの2名における感情共有の測定 発表概要: 共感は人間の社会的インタラクションにおいて重要な役割を持っている． 例えば，共感により，より強固な人間関係や協力体制が築かれる．共感は2名以上の脳における同期現象によって引き起こされることから，生体信号による計測が可能であると考えらえる． 本研究では，共感の主要な要素の一つである感情共有を，脳波信号により計測した． 実験環境において感情共有を誘発するため，被験者は2名1組となり，喜び，悲しみ，無感情の3種類の表情を用いたコミュニケーションを行った． 実験中，脳波信号が2名から同時に計測された． 分析の結果，脳波信号のパワーの相関が，喜びと悲しみの条件において有意に高くなることが確認された． この結果から，脳波信号のパワーの相関によって感情共有が測定できる可能性が示された． | |||