Commonsense and Named Entity Aware Knowledge Grounded Dialogue Generation

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1322 - 1335 July 10-15, 2022 ©2022 Association for Computational Linguistics Commonsense and Named Entity Aware Knowledge Grounded Dialogue Generation Deeksha Varshney†, Akshara Prabhakar‡∗ Asif Ekbal† †Department of Computer Science and Engineering, Indian Institute of Technology Patna, India ‡Department of Information Technology, National Institute of Technology Karnataka, Surathkal {1821cs13,asif}@iitp.ac.in akshblr555@gmail.com Abstract Grounding dialogue on external knowledge and interpreting linguistic patterns in dialogue history context, such as ellipsis, anaphora, and co-references is critical for dialogue comprehension and generation. In this paper, we present a novel open-domain dialogue generation model which effectively utilizes the large-scale commonsense and named entity based knowledge in addition to the unstructured topic-specific knowledge associated with each utterance. We enhance the commonsense knowledge with named entity-aware structures using co-references. Our proposed model utilizes a multi-hop attention layer to preserve the most accurate and critical parts of the dialogue history and the associated knowledge. In addition, we employ a Commonsense and Named Entity Enhanced Attention Module, which starts with the extracted triples from various sources and gradually finds the relevant supporting set of triples using multi-hop attention with the query vector obtained from the interactive dialogue-knowledge module. Empirical results on two benchmark dataset demonstrate that our model significantly outperforms the state-of-the-art methods in terms of both automatic evaluation metrics and human judgment. Our code is publicly available at https://github.com/deekshaVarshney/CNTF; https://www.iitp.ac.in/-ai-nlp-ml/resources/ codes/CNTF.zip. 1 Introduction Neural language models usually focus on fewer language components such as sentences, phrases, or words for text analysis. However, language acts on a much broader scale - there is frequently a central theme to a conversation, and the speakers share common information in order to comprehend one another. Information is frequently reused, how- ever to avoid overuse, same things and persons are ∗Work done during an internship at IIT Patna referred in the dialogue multiple times by using relevant expressions. A dialogue becomes coherent and speakers can understand each other when all of this information is delivered in a structured, logical, and consistent manner. Semantic understanding of dialogues can be aided by commonsense knowledge or world facts. Additionally, as a key human language phenomena, co-reference simplifies human languages while be- ing a significant barrier for machines to under- stand, particularly for pronouns, which are diffi- cult to parse due to their weak semantic mean- ings (Ehrlich, 1981). Grounded response gener- ation approaches (Ghazvininejad et al., 2018; Di- nan et al., 2018) can provide replication of facts in open-domain settings, whereas commonsense knowledge is critical for creating successful inter- actions since socially constructed commonsense knowledge is the collection of contextual details that humans are expected to understand and use during a conversation. Despite demonstrating efficacy in empirical eval- uation, past work has a few significant drawbacks. There is no explicit representation of entities, se- mantic relations, or conversation structures, in par- ticular. To solve such restrictions, asking a conver- sation model to identify relevant structures in dia- …

Content truncated to 500 words