ゼミナール発表

日時: 9月27日(火)2限(11:00-12:30)


会場: L1

司会: 中島 悠太
魚谷 果那 1551017: M, 2回目発表 環境知能学 萩田 紀博,加藤 博一,神原 誠之
title: Social common sense Modeling of a queue in public space for a service robot
abstract:In all services provided, inside or outside of the field queue is generated by people, each of standing position is formed from a consensus of the crowd. The consensus of the crowd, from the parameters of the order or the like into the standing position and the queue of people in the queue, which is a set of thinking that person to determine what number. In humans, but can provide services obtained in order from the crowd of consensus, in the case of service provision by a robot, it is not possible to measure the consensus of the crowd, which may produce customer dissatisfaction. We propose a social common sense modeling that determines the next person to provide services, to achieve the equality of services provided by the robot.
language of the presentation:Japanese

 
大石 陽波 1551020: M, 2回目発表 環境知能学 萩田 紀博,加藤 博一,神原 誠之
title: Learning clerk behaviors by watching others for a communication robot
abstract: In recent years, expectations to robots are increasing in daily life environment, such as shopping malls, offices, rail stations. For example, a robot introduced to a retail shop and restaurant needs to be able to conduct barker, advertasement, propagation, service, guidance. In this study, I propose a technique for teaching and being executed the tasks to conduct as a shop clerk to a communcation robot. I aim to make it possible that a shopkeeper who don't have expertise such as programming ability teaches clerk behaviors and the robot learns them by watching others.
language of the presentation: Japanese
 
光 将人 1551043: M, 2回目発表 環境知能学 萩田 紀博,加藤 博一,神原 誠之
title: TV Chat Robot with a Natural Conversation by Time Shifting
abstract: This paper proposes a communicative robot that facilitates talk with a user who lacks a chance of verbal communication with others for any reasons (e.g. the elderly who live alone). As our first trial, we have developed a system which is designed to support a user to talk while watching a TV program. To achieve natural response timing, the proposed system includes three response functions: backchannel, repetition and machine answering. In this paper, in order to realize a natural communication with a robot, the proposed system overcomes the problems which are delay of robot utterance generated by SNS comments and overlap of utterances by time sifting function.
language of the presentation: Japanese
発表題目: *** この部分を発表題目に ***
発表概要: *** この部分を発表概要に ***
 

会場: L2

司会: 丁 明
MARTINEZ ANDER 1561028: D, 中間発表 自然言語処理学 松本 裕治,中村 哲,新保 仁,進藤 裕之,能地 宏
title: Improving Neural Machine Translation on resource-limited pairs using auxiliary data of a third language
abstract: In the recent years interest in Deep Neural Networks (DNN) has grown in the field of Natural Language Processing, as new training methods have been proposed. The usage of DNN has achieved state-of-the-art performance in various areas. Neural Machine Translation (NMT) described by Bahdanau et al. (2014) and its successive variations have shown promising results. DNN, however, tend to over-fit on small data-sets, which makes this method impracticable for resource-limited language pairs. This article combines three different ideas (splitting words into smaller units, using an extra dataset of a related language pair and using monolingual data) for improving the performance of NMT models on language pairs with limited data. Our experiments show that, in some cases, our proposed approach to subword-units performs better than BPE (Byte pair encoding) and that auxiliary language-pairs and monolingual data can help improve the performance of languages with limited resources.
language of the presentation: English
 
池田 大志 1551006: M, 2回目発表 自然言語処理学 松本 裕治,中村 哲,新保 仁,進藤 裕之
title: Japanese Text Normalization with Encoder-Decoder Model
abstract: Text normalization is the task of transforming lexical variants to their canonical forms. We model the problem of text normalization as a character sequence to sequence leaning problem and present the neural encoder-decoder model for solving it. However, lots of corpora are required to train the encoder-decoder model and such large annotated noisy-clean pairs almost do not exist. To address this issue, we propose an method to augment data, which converts existing resources into synthesized non-standard forms by hand-crafted rules. We conducted an experiment to demonstrate that the synthesized corpus contributes to stably train the encoder-decoder model and improve Japanese text normalization.
language of the presentation: Japanese
 
岩元 文 1551016: M, 2回目発表 自然言語処理学 松本 裕治,中村 哲,新保 仁,進藤 裕之,能地 宏
title: Coreference resolution on Scientific Domain
abstract: Coreference resolution on scientific domain is necessary for further NLP tasks such as information extraction from scientific texts. To get coreference system for this purpose, we used ACL anthology annotated data insted of general domain data such as newspaper. However, there are some difficulties in this domain, e.g. mistake of part-of-speech tags because of the unseen words, parsing mistake from too long sentence. We changed annotated mention span from entire noun phrase to minimum noun phrase that includes dependency head(head NP).
language of the presentation: Japanese
 
榎本 陽一 1551019: M, 2回目発表 自然言語処理学 松本 裕治,中村 哲,新保 仁,進藤 裕之,能地 宏
title: Concept identification of Abstract Meaning Representation parsing
abstract:Abstract meaning Representation (AMR) parsing is the problem of mapping natural language strings into meaning representations and Concept identification is part of AMR parsing process.In this process,sentence will be mapped to graph fragments. In test data of AMR annotation data,there is a lot of the words that appear only in the test data. I propose efficient method of Concept identification taking measures to cope with such a word.
language of the presentation: Japanese