| YUNG PIKYU | 1461023: D, 中間発表 | 松本 裕治，中村 哲，新保 仁，Kevin Duh |
|
title: Implicit Discourse Connectives in Machine Translation
abstract: Usage of discourse connectives (DCs) differs across languages, thus addition and omission of connectives are common in translation. We investigate how implicit (omitted) DCs in the source text impacts various machine translation (MT) systems, and whether a discourse parser is needed as a preprocessor to explicitate implicit DCs. Based on the manual annotation and alignment of 7266 pairs of discourse relations in a Chinese-English translation corpus, we evaluate whether a preprocessing step that inserts explicit DCs at positions of implicit relations can improve MT. Results show that, without modifying the translation model, explicitating implicit relations in the input source text has limited effect on MT evaluation scores. In addition, translation spotting analysis shows that it is crucial to identify DCs that should be explicitly translated in order to improve implicit-to-explicit DC translation. On the other hand, further analysis reveals that the disambiguation as well as explicitation of implicit relations are subject to a certain level of optionality, suggesting the limitation to learn and evaluate this linguistic phenomenon using standard parallel corpora. language of the presentation: English | ||
| 松田 昇悟 | 1451098: M, 2回目発表 | 松本 裕治，中村 哲，新保 仁，進藤 裕之 |
|
title: Improving Coreference Resolution by Graph Clustering
abstract: In coreference resolution, mention-pair model is currently paid attention for using only noun phrase pairs while proposed methods are much complicated. However, mention-pair model is impossible to use information other than mention pairs. In this research, we consider a method for improvement of the accuracy by using a graph clustering in order to incorporate information other than mention pairs in simple model. language of the presentation: Japanese 発表題目: グラフクラスタリングによる共参照解析の精度改善 発表概要:共参照解析において、提案される手法が複雑化していることが問題視されており、そこで名詞句対の共参照を判定するのみであるmention-pair modelに注目が集まっている。しかし、mention-pair modelは名詞句対以外の情報を活用することができないという問題点がある。本研究では、簡易なモデルに名詞句対以外の情報を取り入れるためにグラフクラスタリングを用いて精度を向上する方法を検討する。 | ||
| 三田 雅人 | 1451103: M, 2回目発表 | 松本 裕治，中村 哲，新保 仁，進藤 裕之 |
|
title: Improving Preposition Error Detection Using Semantic Role Labeling
abstract: Preposition error occurs frequently in nonnative (L2) English writing because of its difficulty in using them properly. The preposition error detection task is also hindered by the difficulty of dealing with the various conceptual meaning expressed by prepositions. However, there is no work can deal with these various semantic role of prepositions. In this research, we present a novel approach for applying semantic role labeling which identifies shallow semantic information in a given context topreposition error detection task. Experiment result showed our approach improves theaccuracy of detection. language of the presentation: Japanese | ||
| 村松 航平 | 1451110: M, 2回目発表 | 松本 裕治，中村 哲，新保 仁，進藤 裕之 |
|
title: Improving Lexicalized Reordering Model Combination for Pivot Machine Translation
abstract:For low resource language machine translation, pivot translation, which introduces third language in addition to source and target language, is known for good solution. In several pivot translation methods, phrase table combination method, which combines source-pivot and pivot-target translation model into source-target translation model gets especially high translation performance. Phrase reordering is important problem in machine translation, and reordering problem for pivot translation is also studied. In this study, we incorporate lexicalized reordering idea into table combination method. language of the presentation: Japanese | ||
| 原 一貴 | 1451085: M, 2回目発表 | 飯田 元，藤川 和利，市川 昊平，奥田 剛，渡場 康弘 |
|
title: Netspec: A Testing Platform Design and Prototype for OpenFlow Network
OpenFlow network is a network controlled by a controller program. In order to ensure the behavior of such a network, it is important to test the controller properly. A method of unit-testing the controller has been established in the same manner as a popular software development process. However, in the development of a controller program, it is not enough to only unit-test the controller program. Validation of the behavior of the entire OpenFlow network, which is not possible currently, is also required. In my research, I propose Netspec which is a novel testing platform for OpenFlow network. Netspec validates the behavior of the OpenFlow network against the requirements. language of the presentation: Japanese | ||
| 松田 裕貴 | 1551103: M, 2回目発表 | 安本 慶一，藤川 和利，荒川 豊，諏訪 博彦，藤本 まなと |
|
title:A Development of Participatory Mobile Sensing Platform for Variety Use Cases
abstract:Because of popularization of smartphones in which many sensors are embedded, "participatory sensing", the urban sensing that utilizes users' smartphone, becomes a reality. However, there are some remaining problems such as "how to motivate a user's continuous participation", "how to correct data uploaded from various smartphones", and "how to deal with various sensing tasks". In this presentation, we we introduce the past research results, and show the development plan of the participatory mobile sensing platform as the foundation for evaluating our solutions against these problems. language of the presentation: Japanese 発表題目:多様なユースケースに適用可能なユーザ参加型モバイルセンシングプラットフォームの開発 発表概要:多種多様なセンサを搭載したスマートフォンの普及により，一般ユーザがスマートフォンを用いた都市センシングに寄与する「ユーザ参加型センシング」が現実のものになってきた．しかしながら，ユーザ参加型センシングを成功させるためには，「ユーザの持続的な参加を促す手法」「ユーザ離脱防止手法」「センシングデータの個体差の校正手法」「多様なセンシングタスクへの対応」などいくつかの課題が残されている．本発表では，これらの問題に対するこれまでの取り組みを取り上げるととともに，課題解決手法を評価するための基盤となるユーザ参加型センシングプラットフォームの開発計画を示す． | ||
| 政木 勇人 | 1451096: M, 2回目発表 | 佐藤 嘉伸，金谷 重彦，大竹 義人，横田 太 |
|
title: Automated liver segmentation from 3D MRI for quantitative evaluation of liver fibrosis
abstract: Techniques currently used for definitive diagnosis of liver fibrosis are invasive and burden to the patient is high, thus non-invasive techniques are preferable. The purpose of this study is automated liver segmentation from MR image for quantitative evaluation of liver fibrosis. Problem of automated liver segmentation from MR image is that non-uniformity of the intensity is unavoidable even in the same tissue due to its imaging principle in MR image. As far as I know there has been no work that focused on automated liver segmentation from MR image with intensity correction. I propose automated method that combines intensity correction using probabilistic atlas and automated liver segmentation method from CT image using statistic atlas. Proposed method was performed to 45 cases of contrast-enhanced MR image of the liver. As a result, Dice coefficient of proposed method was 0.94±0.03. Significant accuracy improvement with a significant level of 0.01 was observed to the paired T-test between results of corrected and non-corrected MR image. In the future work, quantitative evaluation of liver fibrosis will be performed using segmentation result of proposed method and non-image information such as blood test. language of the presentation: Japanese 発表題目: 肝線維化症の定量評価に向けた3次元MRIにおける肝臓領域の自動抽出 発表概要: 現在，臨床で肝線維化と呼ばれる肝臓疾患の確定診断に用いられる方法は侵襲的であり患者への負担が大きく，非侵襲な手法が求められている．本研究の目的は，MR画像から肝線維化症の定量評価を行うため，MR画像からの肝臓領域自動抽出を目的とする．MR画像からの肝臓領域自動抽出の問題点として，MR画像はCT画像と比べ撮像の原理上濃淡分布にむらがあることが挙げられる．この濃淡むらを考慮した肝臓領域自動抽出を行っている研究は私の知る限りない．本研究では，肝臓の確率アトラスを用いた濃淡むらの補正手法と統計アトラスを利用したCT画像から肝臓領域の自動抽出を組み合わせた自動手法を提案した．肝臓造影MR画像45症例に適用し結果，Dice係数0.94±0.03の精度で抽出できた．補正なしと補正ありの抽出結果に対しt検定を行った結果，p値0.01以下で有意に精度が向上した．今後，自動抽出結果に加え血液検査といった生化学的情報を用いて肝線維化症の定量評価を行う予定である． | ||