| 研究課題/領域番号 |
19K00850
|
|---|
| 研究機関 | 会津大学 |
|---|
研究代表者 |
BLAKE John 会津大学, コンピュータ理工学部, 上級准教授 (80635954)
|
|---|
| 研究分担者 |
Mozgovoy Maxim 会津大学, コンピュータ理工学部, 准教授 (60571776)
|
|---|
| 研究期間 (年度) |
2019-04-01 – 2023-03-31
|
| キーワード | NLP / visualization / grammatical explanations / contextualized grammar / academic writing / scientific writing |
|---|
| 研究実績の概要 |
During the 2021 academic year, we developed and refined the algorithms to identify various language features in the language visualizer. This included improvements to the automatic detection of linking words and classifying them as prepositions, (e.g. despite), conjunctions (e.g. but) and adverbial transitions (e.g. however) based on their grammatical form in context. This is a non-trivial task for linking words that only take one part of speech. However, many words occur in two different forms, (e.g. however [conj] hard, we). We also improved the algorithms identifying tense and voice. We have developed more mulitimodal explanatory materials in both English and Japanese. For each langage item we created a slideshow, bilingual audio explanations, and are currently creating video versions.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
3: やや遅れている
理由
The restictions due to the pandemic caused difficulties in both the development and evaluation of the software program, and in the development and evaulation of the multimodal explanations. Where research assistants, who are undegraduate or graduate students, were working alone at their workstations, there was no undue impact caused by covid-19. However, given the need to train the assistants, test the usability of the software and jointly create video materials; social distancing slowed down or halted progress at times. In addition, the fear of covid-19 impacted the willingness of part-time assistants to collaborate on site, meaning that the project was understaffed during this year. Despite this, we managed to develop the scripts for video to be created in the following academic year.
|
| 今後の研究の推進方策 |
In this academic year, the main thrust of the software development phase is on refining two functionalisites in the software, namely automatic visualisation of coherence and cohesion. This process is non-trivial, given the length of the natural language pipeline needed to achieve this. At each step in the pipeline, the potential for errors (e.g. false positive and false negative) increases. It is likely to be necessary to postprocess the initial results of the pipeline with tailormade algorithms to increase the accuracy of the system. The main focus of the multimodal explanation development phase is on recording, editing and subtitling videos. Once completed these videos can be linked to the language feature detector and visualizer to enable users to get explanations on demand.
|
| 次年度使用額が生じた理由 |
Primarily to cover the costs associated with video production. The lion's share of the funding is likely to cover the honoriums paid to teaching and research assistants.
|
