表題番号：2023C-085
日付：2024/03/30
研究課題計算機の文章読解能力向上に関する研究
| 研究者所属（当時） | 資格 | 氏名 | |
|---|---|---|---|
| （代表者） | 理工学術院 基幹理工学部 | 教授 | 河原 大輔 |
- 研究成果概要
- To improve the text understanding abilities of computers, we conducted studies on training foundation models using large text corpora and developing and evaluating application systems through fine-tuning these models. For the training of the foundation models, we investigated the impact of filtering methods for large text corpora on downstream tasks, constructed models that learned the syllable count for literature generation, and examined knowledge-integrated models using Mixture of Experts (MoE). For application systems, we developed systems for generating interesting senryu (a type of haiku) and playing word chain games, etc. For evaluation, we automatically constructed a Japanese Winoground dataset for evaluating Japanese multimodal models. Through these research and development efforts, we believe that we have taken a step forward in the study of text understanding by computers.