Ethan Perez: New! "Unsupervised Question Decomposition for Question Answering": https://arxiv.org/pdf/2002.09758.pdf
We decompose a hard Q into several, easier Qs with *unsupervised learning*, improving multi-hop QA on HotpotQA without extra supervision.
w/@PSH_Lewis @scottyih @kchonyc @douwekiela (1/n) https://t.co/Z8ADXlLHoR
4 replies, 269 likes
Ethan Perez: Now with code from @facebookai, based on XLM and @huggingface transformers! https://github.com/facebookresearch/UnsupervisedDecomposition
And with blog post: https://medium.com/@ethanperez18/unsupervised-question-decomposition-for-question-answering-9b81c5f7a71d
Have fun training your own models to decompose questions into easier sub-questions... fully unsupervised!
2 replies, 205 likes
Kyunghyun Cho: a super fun project with many interesting observations and lessons on multi-hop question-answering (though, I'm really not a fan of this term "multi-hope")
0 replies, 38 likes
Patrick Lewis: 🚀You can break down complex questions into a list of simple questions without any labelled data.
🚀Strong, robust results on HotPotQA, and don't need any supporting fact annotation.
A pleasure working with @EthanJPerez @scottyih @kchonyc & @douwekiela! https://arxiv.org/abs/2002.09758
1 replies, 25 likes
Douwe Kiela: Can we learn to decompose difficult questions into easier ones, without supervision? Turns out we can! Great job by @EthanJPerez
0 replies, 11 likes
Found on Feb 25 2020 at https://arxiv.org/pdf/2002.09758.pdf