Question Answering-2

Question Answering - 2 (QA2): We plan to conduct the second question answering (QA2) subtask, which is the same as QA at PoliInfo-3. Although the scores in terms of “correspondence” and “well-formed” were relatively high (more than 70%) in QA, the “content” score was low (less than 50%). Because there is room for improvement, we decided to continue the same task. We also plan to connect QA2 with AV at PoliInfo-4. While using answers that participants output in QA2 as training data for AV, classifiers participants build in AV will be used for automatic evaluation in QA2. Therefore, we expect these subtasks encourage each other.

Input:

• Questions of Q&A sessions in the newsletters (JSON format)

• The minutes of the Tokyo Metropolitan Assembly (JSON format)

Output:

• Short answer of the question in a few lines (JSON format)

Evaluation:

• ROUGE (for automatic evaluation)

• Quality questions (for human evaluation)