2021 Fiscal Year Final Research Report
A study of Information Access by making use of Latent Spaces shared across Modalities and Languages
| Project/Area Number |
19K11980
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 60080:Database-related
|
|---|
| Research Institution | Toyohashi University of Technology |
|---|
Principal Investigator |
Akiba Tomoyosi 豊橋技術科学大学, 工学(系)研究科(研究院), 准教授 (00356346)
|
|---|
| Project Period (FY) |
2019-04-01 – 2022-03-31
|
| Keywords | 情報アクセス / 情報検索 / 系列変換モデル / 潜在空間 / 音声認識 / 機械翻訳 / 音声翻訳 / 音声ドキュメント検索 |
|---|
| Outline of Final Research Achievements |
Thanks to the recent development on information technology, spoken and written modalities have become to be used exchangeably for human communication, and then to be recorded in repository without distinction. On the other hand, a variety of languages are spoken and written in the world. In order to access information recorded in such various modalities and languages, some IR methods for crossing them are indispensable. This study aims to develop an information access method to connect various information form such as modalities and languages and to make use of each of them complementarily. To this end, we developed several information transformation methods using sequence-to-sequence models with the help of latent spaces shared with the various information form.
|
| Free Research Field |
自然言語処理, 音声言語処理
|
| Academic Significance and Societal Importance of the Research Achievements |
言葉の情報は音声およびテキストの両モダリティで分け隔てなく表現されつつあり、かつ世界規模では多様な言語で表現されている。モダリティや言語の違いから多様な表現形態を取りうる言語情報を、系列変換モデルという統一的な枠組みをベースに、相互に変換する手法を実現するとともに、系列変換モデルを効果的に学習する手法を開発した。本研究の成果により、現実世界の不均質な情報の利用が促進されるとともに、個々の表現形態の利点を活用する技術の開発が進むと考えられる。
|
