MCˆ2: Multi-perspective Convolutional Cube for Conversational Machine Reading Comprehension

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 6185–6190 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 6185 MC2: Multi-perspective Convolutional Cube for Conversational Machine Reading Comprehension Xuanyu Zhang College of Information Science and Technology Beijing Normal University, Beijing, 100875, China xyz@mail.bnu.edu.cn Abstract Conversational machine reading comprehen- sion (CMRC) extends traditional single-turn machine reading comprehension (MRC) by multi-turn interactions, which requires ma- chines to consider the history of conversa- tion. Most of models simply combine pre- vious questions for conversation understand- ing and only employ recurrent neural networks (RNN) for reasoning. To comprehend con- text profoundly and efficiently from different perspectives, we propose a novel neural net- work model, Multi-perspective Convolutional Cube (MC2). We regard each conversation as a cube. 1D and 2D convolutions are integrated with RNN in our model. To avoid models pre- viewing the next turn of conversation, we also extend causal convolution partially to 2D. Ex- periments on the Conversational Question An- swering (CoQA) dataset show that our model achieves state-of-the-art results. 1 Introduction Conversation is one of the most important ap- proaches for humans to acquire information. Dif- ferent from traditional machine reading compre- hension (MRC), conversational machine reading comprehension (CMRC) requires machines to an- swer multiple follow-up questions according to a passage and dialogue history. However, these questions usually have complicated linguistic phe- nomena, such as co-reference, ellipsis and so on. Only considering conversation context profoundly can we answer the current question correctly. Recently, many CMRC datasets, such as CoQA (Reddy et al., 2019) and QuAC (Choi et al., 2018), are proposed to enable models to understand pas- sages and answer questions in dialogue. Here is an example from the CoQA dataset in Figure 1. We can observe that the second and third questions omit key information. It is impossible for both hu- Billy went to the farm to buy some beef for his brother's birthday. When he arrived there, he saw that all six of the cows were sad and had brown spots. The cows were all eating their breakfast in a big grassy meadow … … History: Who went to the farm? Billy Why? To buy some beef Q1 Q2 A1 A2 For what? His brother's birthday Q3 A3 Passage: Figure 1: An example in the CoQA dataset. mans and machines to understand such questions without dialogue history. Most of existing methods consider conversation history by prepending previous questions and an- swers to the current question, such as BiDAF++ (Yatskar, 2019), DrQA+PGNet (Reddy et al., 2019), SDNet (Zhu et al., 2018) and so on. How- ever, the latent semantic information of dialogue history is neglected. And the model may confuse some unrelated questions and answers in a sen- tence. Although FlowQA (Huang et al., 2019) utilizes intermediate representations of previous conversation, the flow mechanism can not synthe- size the information of different words in different turns of conversation simultaneously. Moreover, previous models only use recurrent neural network (RNN) as their main skeleton, …

Content truncated to 500 words