| 目黒 豊美 | 1361201: D, 中間発表 | 中村 哲，松本 裕治，戸田 智基，Sakriani Sakti，Graham Neubig，吉野 幸一郎 |
|
title: Non-task-oriented dialogue system based on statistical model
abstract: Our aim is to build non-task-oriented dialoguesystembased on statistical model. In this presentation, we show (1) the analysis of listening-oriented dialogue(LoD) to reveal the characteristics of LoD, (2) a dialogue control method using POMDP for LoD, (3) dialogue act tagging for microblog utterances using semantic category patterns, and (4) the fusion of rule-based and stochastic utterance generation. language of the presentation: Japanese | ||
| 杉山 享志朗 | 1451060: M, 2回目発表 | 中村 哲，松本 裕治，戸田 智基，Sakriani Sakti，Graham Neubig，吉野 幸一郎 |
|
title: An Investigation of Machine Translation Evaluation Metrics in Cross-lingual Question Answering
abstract: Through using knowledge bases, question answering (QA) system have come to be able to answer questions accurately over a variety of topics. However, knowledge bases are limited to only a few major languages, and thus it is often necessary to build QA systems that answer questions in one language based on an information source in another (cross-lingual QA: CLQA). Machine translation (MT) is one tool to achieve CLQA, and it is intuitively clear that a better MT system improves QA accuracy. However, it is not clear whether an MT system that is better for human consumption is also better for CLQA. In this study, we investigate the relationship between manual and automatic translation evaluation meterics and CLQA accuracy by creating a data set using both manual and machine translations and perform CLQA using this created data set. As a result, we find that QA accuracy is closely related with a metric that considers frequency of words, and as a result of manual analysis, we identify 3 factors of translation results that affect CLQA accuracy. language of the presentation: Japanese | ||
| 辻岡 聡 | 1451075: M, 2回目発表 | 中村 哲，松本 裕治，戸田 智基，Sakriani Sakti，Graham Neubig |
|
title: Acoustic Data-driven Pronunciation Lexicon for Non-native Speech Recognition
abstract: Non-native speech differs significantly from native speech, often resulting in a degradation of the performance of automatic speech recognition (ASR). Handcrafted pronunciation lexicons used in standard ASR systems generally fail to cover non-native pronunciations, and design of new ones by linguistic experts is time consuming and costly. This study proposes a method to automatically learn a pronunciation lexicon in an iterative fashion. We start with a seed lexicon of basic non-native pronunciations and train a grapheme-to-phoneme (G2P) converter. The resulting non-native G2P converter is used to generate non-native pronunciation variations, including pronunciations of new words. The weights of these pronunciation variations are then estimated from actual acoustic evidence of non-native speakers, and the acoustic model is updated based on these new pronunciation variations. This process is done iteratively until convergence. In experiments, we evaluate our ASR systems for speakers with three degrees of English proficiency level. The results reveal that the proposed method can cope with fluctuation and ambiguity of non-native pronunciation, and is able to achieve an improvement in recognition accuracy, particularly for low-proficiency speakers. language of the presentation: Japanese 発表題目: 非母語音声の認識のための実音声を用いた発音辞書獲得 発表概要: 国際会議などでは英語が国際共通語として用いられ，英語非母語話者の間でも英語で意思疎通を図る場面が多い． このような非母語音声を認識して会議録を作成するなどの応用技術を考えた場合，非母語音声認識を高精度に行う必要がある． しかし，非母語話者の音声は母語話者に比べ，発音の揺らぎやブレが原因となり，非母語音声の認識精度は母語音声よりも低下する問題が生じる． 非母語音声認識において，音響モデル，発音辞書，言語モデル，デコーディングの各処理系にて考慮する必要があるが，本研究では発音辞書に焦点を当てる．非母語話者の発音の揺れに対処するために，単語表記列から発音系列候補を予測するG2P（Grapheme-to-phoneme）ツールを用いて，複数の発音バリエーションを生成し，非母語音声話者の実音声から生起頻度の高い発音バリエーションを推定する手法を検討する． その発音バリエーションを発音辞書に適応した結果，発音の揺らぎやブレに対応することができ，認識精度の向上を確認できた． | ||
| 川西 誠司 | 1451040: M, 2回目発表 | 中村 哲，小笠原 司，戸田 智基，Sakriani Sakti，Graham Neubig |
|
title: Speech and Environmental Sound Recognition using Deep Neural Network
abstract: The speech recognition quality under the noisy environment is degraded, so usually we remove an environmental sound as noises. But, there are many merits provided by recognizing an environmental sound. So we try to improve speech recognition and environmental sound recognition accuracy using multi-task deep neural network and report a result of the comparison with single-task deep neural network. language of the presentation: Japanese | ||
| 出島 遼 | 1451077: M, 2回目発表 | 金谷 重彦，佐藤 嘉伸，MD.ALTAF-UL-AMIN，杉浦 忠男，佐藤 哲大 | ||||||
|
title: Automatic determination of blood flow velocity in brain microvessel of rat using a small implantable CMOS imaging device
abstract: Functional Magnetic Resonance Imaging (fMRI) and Positron Emission Tomography (PET) are often used for imaging particularly active brain locations. However, spatial resolution of these devices is at most 0.5mm and thus is not sufficient to investigate blood flow changes at micro-vascular level. In this study, we use the small implantable CMOS imaging device developed by our research group. The conventional methods for determining blood flow velocity have had limitations with inevitable variations arising due to manual measurement. In the experiment, the blood flow velocity was estimated by focusing on the movement of red blood cells in the videos. Our proposed methods applies correlation and automatically determines the blood flow velocity without physical inspection. We compare the multiple automatic determination. language of the presentation: Japanese (choose one) 発表題目: ラットの脳微小血管における血流速度自動決定法 発表概要: fMRI(functional magnetic resonance imagind)やPET(Positron Emission Tomography)を用いて脳血流を測定し活動的な脳部位の同定が行われている。しかしこれらの装置の空間分解能は、微小血管レベルでの血流変化を検討するには十分ではない。本研究では共同研究グループが独自に開発したCMOSによる脳血流イメージング装置を使用した。これまで撮影された動画より血流速度を求める手法は人の目視であり、計測者によるばらつきを無視できない問題があった。そこで画像中の赤血球の移動速度を自動検出する複数の方法を比較検討した。この結果について報告する。 | ||||||||
| 川上 陽子 | 1451036: M, 2回目発表 | 金谷 重彦，笠原 正治，MD.ALTAF-UL-AMIN，杉浦 忠男，小野 直亮，佐藤 哲大 | ||||||
| ||||||||
| 江口 遼平 | 1451019: M, 2回目発表 | 金谷 重彦，安本 慶一，MD.ALTAF-UL-AMIN，杉浦 忠男，小野 直亮 | ||||||
|
title: Unraveling the defense mechanism of Angelica acutiloba by gene expression profiling
abstract: Angelica acutiloba is a perennial herb from the family Apiaceae or Umbelliferous. Its root was used in Kampo medicine, which is a Japanese adaptation of Traditional Chinese medicine. Angelica acutiloba has the defense mechanism to insect damage, and makes BETA-CARYOPHYLLENE which has the property of inviting a bee to exterminate noxious insects. In addition, BETA-CARYOPHYLLENE is a medically effective ingredient. In a word, medically effective ingredients in Angelica acutiloba increase by insect damage. The purpose of this study is unraveling the defense mechanism of Angelica acutiloba. In this presentation, I will report the analysis results so far. language of the presentation: Japanese | ||||||||