NLPIR/ICTCLA2018 ACADEMIC SEMINAR 4th ISSUE COMPLETED
Last week, Wang Gang gave us a presentation on the topic of Know What You Don’t Know: Unanswerable Questions for SQuAD.
After the presentation, several questions were asked. The Q&As are listed as following:
Q: Why ‘rank sentences that answer a question higher than sentences that do not’ ?
A: This is a strategy used while constructing sentence selection datasets. Detailed information can be learned from the following paper.
Wang, N. A. Smith, and T. Mitamura. 2007. What is the jeopardy model? a quasi-synchronous grammar for QA. In Empirical Methods in Natural Language Processing (EMNLP).
Q: How do you think about this dataset ?
A: I think it is very good. Compared with SQuAD 1.1, this new version challenge all the current models. The authors hope their work could encourage the development of new reading comprehension models that know what they don’t know, and therefore understand language at a deeper level.
Q: Why the score may be lower at SQuAD 2.0 than 1.1?
A: There are more unanswerable questions in SQuAD 2.0. So the gap between human and machine is getting bigger.
The powerpoint of the presentation can be download.
4_powerpoint.pdf(882 KB)
IMG_1256
