| 研究課題/領域番号 |
20K00807
|
|---|
| 研究機関 | 高知工科大学 |
|---|
研究代表者 |
ダニエルズ ポール 高知工科大学, 共通教育教室, 教授 (50307245)
|
|---|
| 研究分担者 |
熊井 信弘 学習院大学, 付置研究所, 教授 (00248999)
|
|---|
| 研究期間 (年度) |
2020-04-01 – 2023-03-31
|
| キーワード | speech / assessment / computer |
|---|
| 研究実績の概要 |
The computer-scored speech assessment system was fully integrated with our institution’s in-house English proficiency test. Over 400 first-year students completed the computer-scored speaking proficiency test. The computer-generated speaking scores are being compared against standardized English proficiency test scores, such as TOEIC and CASEC scores.
The initial platform was designed to assess speaking tasks that had specific ‘correct’ target phrases or sentences. This year the researchers focused on implementing open-ended speaking tasks, specifically to support extensive speaking tasks for lower-level language learners. The latest version of the system supports translation, listening, speaking and evaluation. Language learners are able to speak a phrase in Japanese, listen to the English translation, speak the English phrase and receive feedback. Learners can to listen to the pronunciation of new words or phrases used in a specific context, and can practice their production skills as they repeat the language in the L2. Practice language tasks or language tests can be generated based on the database of items generated by a learner or an entire class using this latest speech assessment system.
In addition to assessing open-ended speaking tasks, the system was improved to provide individualized feedback. Learner’s language output can now be compared with the target language to determine how similar the two are. The new feedback implementation can be used with shadowing and dictation tasks to provide learner feedback on their spoken or written production.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
The adoption of the computer-scored speaking assessment was successful with less than 5% of the 400 test-takers experiencing technical issues with the audio capture when taking the test.
In addition, a new ‘extensive speaking’ option has been completed and is functioning as intended after initial testing. The translation function makes use an API offered by Deepl.com. The translation is generally accurate, but there are instances where inaccurate translations occur. For example, the kanji ‘私の故郷’ can be read different ways- furusato, kyoko, shushin. Depending on how the kanji is read, the English translation varies in accuracy. In addition, proper nouns or katakana English also pose challenges with the translation API.
|
| 今後の研究の推進方策 |
1. The translate and speak assessment function will be integrated with the English language curriculum to assess how the developed system could support ‘extensive’ speaking activities.
2. Speaking score data collected from the in-house English proficiency test will be compared to standardized English test scores, such as TOEIC and CASEC to determine if any positive correlations emerge between test scores and between speaking and listening scores.
The feedback implementation will be tested with over 100 participants to determine how it can effectively provide accurate and useful individualized speaking feedback.
3. Computer-scored speaking data will be continually assessed to further improve the reliability of the speaking evaluation.
|
| 次年度使用額が生じた理由 |
Due to domestic and international travel restrictions in 2021, the researchers had limited opportunities to participate in academic meetings. In addition, because of the Russia/Ukrainian conflict, we were unable to employ our key coder who resides in Russia. In 2022 we will either look for alternative method to employ our Russian coder, or use another coding service.
|
