recurrent base と違い、 positional encoding を加えているだけ • 一方で、順序情報が重要と考えられる commonsense reasoning タスクで BERT 特徴が使えたという話も ◦ Attention Is (not) All You Need for Commonsense Reasoning [7] (ACL 2019) • 大量のデータで unsupervised に学習できれば獲得す ることも可能? or 順序なくてもある程度解ける?
Effectively? An Empirical Study, C. Sankar et al., ACL 2019, https://arxiv.org/abs/1906.01603 [2] Learning End-to-End Goal-Oriented Dialog, A. Bordes et al., ICLR 2017, https://arxiv.org/abs/1605.07683 [3] Personalizing Dialogue Agents: I have a dog, do you have pets too?, S. Zhang et al., ACL 2018, https://arxiv.org/abs/1801.07243 [4] DailyDialog: A Manually Labelled Multi-turn Dialogue Dataset, Y. Li et al., IJCNLP 2017, https://arxiv.org/abs/1710.03957 [5] Learning Symmetric Collaborative Dialogue Agents with Dynamic Knowledge Graph Embeddings, H. He et al., ACL 2017, https://arxiv.org/abs/1704.07130 [6] Self-Supervised Dialogue Learning, J. Wu et al, ACL 2019, https://arxiv.org/abs/1907.00448 [7]Attention Is (not) All You Need for Commonsense Reasoning, T. Klein and M. Nabi, ACL 2019, https://arxiv.org/abs/1905.13497