研究実績の概要 |
In August of FY1, a joint meeting with the group of Prof. Andre Freitas at Manchester University, who are also working on development of intelligent dialog system, was organized, to discuss research directions and possible collaborations for the research on dialog systems. The research plan was refined based on the discussion. The research achievements in FY1 (the first fiscal year) was two fold: (1) a UI (User Interface) for corpus development, (2) a baseline dialog system. First, a UI for corpus development was developed based on two pre-existing open source resources: the PubAnnotation annotation storage system and the TextAE text annotation editor. Their interfaces were adapted to enable development of dialog corpora. Second, a baseline system of search-oriented dialog was developed, based on an open source pre-existing system, LODQA (Linked Open Data Question Answering). LODQA originally did not support dialog processing, but it was newly implemented. Thanks to its modular design, the intent detection module (and also other modules) is easily swappable. It is expected that the baseline system can serve as a testbed for further development of a dialog system. Using the UI and the baseline system, the development of a corpus of search-oriented dialog began in FY1, which will be completed in early FY2. Overall, it is determined that fundamental resources for development of a dialog system were prepared in FY1.
|
今後の研究の推進方策 |
We have made two changes to the research plan. First, due to the reduction of the research budget, the plan to buy a GPU workstation is cancelled. Instead, we are planning to use a rental workstation service in FY2 and FY3. Second, there was a change to the development schedule: the development of the baseline system, which was planned to be developed in FY2 was developed in FY1, and instead, the development of corpus, annotation guidelines and corpus annotation was postponed to FY2. As the result, the development schedule we are now working on is as follows: (R1) UI for corpus development, Baseline system, (R2) Corpus development, Annotation Guidelines, Annotation, UI for experiment, Intent detection (R3) Intent detection (continued), UI for user study, User study, Publication. The change was made because we have determined that by developing the baseline system first, the development of the corpus and corpus annotation could be conducted more efficiently. Following the changed schedule, the corpus and annotation will be developed in FY2, and the nature of intent detection will be explored over FY2 and FY3.
|