Learning to Collaborate for Question Answering and Asking

Proceedings of NAACL-HLT 2018, pages 1564–1574 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Learning to Collaborate for Question Answering and Asking Duyu Tang‡, Nan Duan‡, Zhao Yan†, Zhirui Zhang♭, Yibo Sun§, Shujie Liu‡, Yuanhua Lv♮, Ming Zhou‡ ‡Microsoft Research Asia, Beijing, China ♮Microsoft AI and Research, Sunnyvale CA, USA †Beihang University, Beijing, China ♭University of Science and Technology of China, Anhui, China §Harbin Institute of Technology, Harbin, China {dutang,nanduan,v-zhaoya,v-zhirzhi,v-yibsu,shujliu,yuanhual,mingzhou}@microsoft.com Abstract Question answering (QA) and question gener- ation (QG) are closely related tasks that could improve each other; however, the connection of these two tasks is not well explored in lit- erature. In this paper, we give a systemat- ic study that seeks to leverage the connection to improve both QA and QG. We present a training algorithm that generalizes both Gen- erative Adversarial Network (GAN) and Gen- erative Domain-Adaptive Nets (GDAN) under the question answering scenario. The two key ideas are improving the QG model with QA through incorporating additional QA-specific signal as the loss function, and improving the QA model with QG through adding artificially generated training instances. We conduct ex- periments on both document based and knowl- edge based question answering tasks. We have two main findings. Firstly, the performance of a QG model (e.g in terms of BLEU score) could be easily improved by a QA model vi- a policy gradient. Secondly, directly applying GAN that regards all the generated questions as negative instances could not improve the ac- curacy of the QA model. Learning when to re- gard generated questions as positive instances could bring performance boost. 1 Introduction In this work, we consider the task of joint learn- ing of question answering and question genera- tion. Question answering (QA) and question gen- eration (QG) are closely related natural language processing tasks. The goal of QA is to obtain an answer given a question. The goal of QG is almost reverse which is to generate a question from the answer. In this work, we consider answer selec- tion (Yang et al., 2015; Balakrishnan et al., 2015) as the QA task, which assigns a numeric score to each candidate answer, and selects the top ranked one as the answer. We consider QG as a generation problem and exploit sequence-to-sequence learn- ing (Seq2Seq) (Du et al., 2017; Zhou et al., 2017) as the backbone of the QG model. The key idea of this work is that QA and QG are two closely tasks and we seek to leverage the con- nection between these two tasks to improve both QA and QG. Our primary motivations are twofold- s. On one hand, the Seq2Seq based QG model is trained by maximizing the literal similarity be- tween the generated sentence and the ground truth sentence with maximum-likelihood estimation ob- jective function (Du et al., 2017). However, there is no signal indicating whether or not the gener- ated sentence could be correctly answered by the input. This problem could be precisely mitigated through incorporating QA-specific signal …

Content truncated to 500 words