A Compare Aggregate Transformer for Understanding Document-grounded Dialogue

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 1358–1367 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 1358 A Compare Aggregate Transformer for Understanding Document-grounded Dialogue Longxuan Ma, Weinan Zhang, Runxin Sun, Ting Liu Research Center for Social Computing and Information Retrieval Harbin Institute of Technology, Harbin, Heilongjiang, China {lxma,wnzhang,rxsun,tliu}@ir.hit.edu.cn Abstract Unstructured documents serving as external knowledge of the dialogues help to generate more informative responses. Previous research focused on knowledge selection (KS) in the document with dialogue. However, dialogue history that is not related to the current dia- logue may introduce noise in the KS process- ing. In this paper, we propose a Compare Aggregate Transformer (CAT) to jointly de- noise the dialogue context and aggregate the document information for response generation. We designed two different comparison mecha- nisms to reduce noise (before and during de- coding). In addition, we propose two met- rics for evaluating document utilization effi- ciency based on word overlap. Experimental results on the CMU DoG dataset show that the proposed CAT model outperforms the state-of- the-art approach and strong baselines. 1 Introduction Dialogue system (DS) attracts great attention from industry and academia because of its wide appli- cation prospects. Sequence-to-sequence models (Seq2Seq) (Sutskever et al., 2014; Serban et al., 2016) are verified to be an effective framework for the DS task. However, one problem of Seq2Seq models is that they tended to generate generic re- sponses that provids deficient information Li et al. (2016); Ghazvininejad et al. (2018). Previous re- searchers proposed different methods to alleviate this issue. One way is to focus on models’ ability to extract information from conversations. Li et al. (2016) introduced Maximum Mutual Information (MMI) as the objective function for generating di- verse response. Serban et al. (2017) proposed a la- tent variable model to capture posterior information of golden response. Zhao et al. (2017) used condi- tional variational autoencoders to learn discourse- level diversity for neural dialogue models. The Document: Movie Name: The Shape of Water. Year: 2017. Director: Guillermo del Toro. Genre: Fantasy, Drama.Cast: Sally Hawkins as Elisa Esposito, a mute cleaner who works at a secret government laboratory. ... Critical Response: one of del Toro's most stunningly successful works ... Dialogue: S1: I thought The Shape of Water was one of Del Toro's best works. What about you? S2: Yes, his style really extended the story. S1: I agree. He has a way with fantasy elements that real- ly helped this story be truly beautiful. It has a very high r- ating on rotten tomatoes, too. S2: Sally Hawkins acting was phenomenally expressiv- e. Didn't feel her character was mentally handicapped. S1: The characterization of her as such was definitely off the mark. Figure 1: One DGD example in the CMUDoG dataset. S1/S2 means Speaker-1/Speaker-2, respectively. other way is introducing external knowledge, ei- ther unstructured knowledge texts Ghazvininejad et al. (2018); Ye et al. (2019); Dinan et al. (2019) or structured knowledge triples (Liu et al., 2018; Young et al., 2018; Zhou …

Content truncated to 500 words