| 金丸 智史 | 1351031: M, 2回目発表 | 松本 裕治, | 中村 哲, | 新保 仁, | 進藤 裕之 |
|
title:Machine Translation considering topics.
abstract:In statistical machine translation, there is a problem that for polysemous words the translation that appears frequently in the training data tends to be selected without considering the context. Here we solve this problem with topic models. This presentation shows the experimental result of applying the previous studies to English-Japanese translation and future direction. language of the presentation: Japanese | |||||
| 増田 優 | 1351098: M, 2回目発表 | 松本 裕治, | 中村 哲, | 新保 仁, | 進藤 裕之 |
|
title: Improving Japanese shift-reduce dependency parsing between clauses
abstract: This presentation proposes a method that improves Japanese shift-reduce dependency parsing between clauses. It is known that long distant dependencies such as those between clauses are difficult to analyze correctly. Especially, shift-reduce algorithm tends to select closer modifiee candidate by its nature. We experimented to improve the accuracy with using features specialized in dependency between clauses and features from more distant modifiee candidate. The result was better than the baseline (shift-reduce algorithm with normal features). We will also talk on the next direction. language of the presentation: Japanese | |||||
| 山嵜 麿与 | 1351109: M, 2回目発表 | 松本 裕治, | 中村 哲, | 新保 仁, | 進藤 裕之 |
|
title: Multi document summarization for the purpose of the readability improvement
abstract: The purpose of this presentation is to improve readability in a text summarization model with sentence extraction and ordering. In the model we propose, it is formulated as a traveling salesman problem the multiple document summarization problem, and performs decoding according to the weight of the readability and informativeness. We show a method of using a neural tensor network, to improve the readability weight. language of the presentation: Japanese | |||||
| 呉 暁一 | 1361024: D, 中間発表 | 松本 裕治, | 中村 哲, | 新保 仁, | 進藤 裕之 |
|
title: Hierarchical Semi-Hidden Markov Model and its Applications
abstract: This work proposed a Hierachical Semi-Hidden Markov Model which extended from normal HMM. We also give new solutions of 3 generalized classical problems from normal HMM. Also, we discussed the applications of proposed model. Japanese 発表題目: 階層的半隠れマルコフモデル及び自然言語処理への応用 発表概要: 本研究では、隠れマルコフモデルの拡張モデルとして階層的半隠れマルコフモデルを提案すし、一般的な隠れマルコフモデルから継承された三つの問題の一般的な解き方を考案する。更に、提案モデルの自然言語処理への応用を検討する。 | |||||
| 宮原 一喜 | 1361012: D, 中間発表 | 伊藤 実, | 松本 健一, | 楫 勇一, | 関 浩之 |
|
title: Node Query Preservation for Deterministic Linear Top-Down Tree Transducers
abstract: What kind of information we have to preserve? This is a fundamental question that comes up when we transform (or translate) some data structure. The solution of the problem depends on what the information we have to recover from the transformed data. Query preservation is one of the formulation of such information preservation. Intuitively, this means that information retrieved from source data by a query can be also retrieved from the transformed data, by some query that are determined by the transformation and the source query. In this study, we examine node query preservation problems of data tree transformations. The decidability results and complexities of the problems for linear top-down data tree transducers are presented. Furthermore, we discuss an application of the query preservation to machine translation. language of the presentation: Japanese | |||||
| 小出 真子 | 1361003: D, 中間発表 | 池田 和司, | 松本 健一, | 久保 孝富 | |
|
title: Constructive Understanding of Difference in Visual Perception between Art Professionals and Novices Using Models of Neural Circuit abstract: Art professionals have superior ability for evaluation of paintings. This can be revealed by constructive approach using models of visual system. First, we hypothesized that art professionals have knowledge to evaluate abstract paintings and therefore their fixations were less attracted to merely salient features. To examine that, eye fixations of each of art professionals and novices were recorded when viewing abstract paintings and the distribution was compared with a predicted map of salient locations in early visual system. As results, fixations of the art professionals showed lower consistency with saliency map than those of the novices. Results imply that art professionals may have profound knowledge to evaluate paintings. Next we examine in which stage of visual system art professionals have superiority to identify painter styles. For that, we calculated discriminability of painter styles by using each restricted model of visual system to be compared with identification by art professional and novices. In this presentation, as the first step, we examined performance of models of the early visual system to classify styles of four painters. Results showed that features of the models were effective to classify painter styles except for between two painters having similar style. The results imply that identification of distinctly separated styles is already represented in the early visual system, although that of subtle difference of styles may be related to the higher visual system. Thus we use biologically inspired models of neural system to reveal differences of visual processing between art professionals and novices. language of the presentation: Japanese | |||||
| CHINDAMAIKUL KHANA | 1351118: M, 2回目発表 | 飯田 元, | 安本 慶一, | 市川 昊平 | |
|
title: Automatic Approach to Prepare Information for Constructing an Assurance Case
abstract: We address the problem of constructing an assurance case and propose an approach to extract information from an issue tracking system that can be used to construct an assurance case. This approach takes advantage of information searching to identify relevant information, which can also be used as materials for constructing an assurance case. This paper gives an overview of an approach and reports the result of the experiment and gives an evaluation. Experimental results suggest that this approach can be effective in reducing the time and the cost for constructing an assurance case with acceptable quality. language of the presentation: English | |||||