2020/05/28 by Ali Ahmadvand, Ahmadvand, Ali, Jason Ingyu Choi +3 · 1 citation
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Speech and dialogue systems #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2005.13804
openalex publication_date 2020/05/28 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Classifying the general intent of the user utterance in a conversation, also\nknown as Dialogue Act (DA), e.g., open-ended question, statement of opinion, or\nrequest for an opinion, is a key step in Natural Language Understanding (NLU)\nfor conversational agents. While DA classification has been extensively studied\nin human-human conversations, it has not been sufficiently explored for the\nemerging open-domain automated conversational agents. Moreover, despite\nsignificant advances in utterance-level DA classification, full understanding\nof dialogue utterances requires conversational context. Another challenge is\nthe lack of available labeled data for open-domain human-machine conversations.\nTo address these problems, we propose a novel method, CDAC (Contextual Dialogue\nAct Classifier), a simple yet effective deep learning approach for contextual\ndialogue act classification. Specifically, we use transfer learning to adapt\nmodels trained on human-human conversations to predict dialogue acts in\nhuman-machine dialogues. To investigate the effectiveness of our method, we\ntrain our model on the well-known Switchboard human-human dialogue dataset, and\nfine-tune it for predicting dialogue acts in human-machine conversation data,\ncollected as part of the Amazon Alexa Prize 2018 competition. The results show\nthat the CDAC model outperforms an utterance-level state of the art baseline by\n8.0% on the Switchboard dataset, and is comparable to the latest reported\nstate-of-the-art contextual DA classification results. Furthermore, our results\nshow that fine-tuning the CDAC model on a small sample of manually labeled\nhuman-machine conversations allows CDAC to more accurately predict dialogue\nacts in real users' conversations, suggesting a promising direction for future\nimprovements.\n