Non-deterministic and emotional chatting machine: learning emotional conversation generation using conditional variation
- PDF / 1,246,188 Bytes
- 9 Pages / 595.276 x 790.866 pts Page_size
- 54 Downloads / 202 Views
(0123456789().,-volV)(0123456789().,-volV)
ORIGINAL ARTICLE
Non-deterministic and emotional chatting machine: learning emotional conversation generation using conditional variational autoencoders Kaichun Yao1 • Libo Zhang2
•
Tiejian Luo1 • Dawei Du1 • Yanjun Wu2
Received: 21 January 2020 / Accepted: 2 September 2020 Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract Conversational responses are non-trivial for artificial conversational agents. Artificial responses should not only be meaningful and plausible, but should also (1) have an emotional context and (2) should be non-deterministic (i.e., vary given the same input). The two factors enumerated, respectively, above are involved and this is demonstrated such that previous studies have tackled them individually. This paper is the first to tackle them together. Specifically, we present two models both based upon conditional variational autoencoders. The first model learns disentangled latent representations to generate conversational responses given a specific emotion. The other model explicitly learns different emotions using a mixture of multivariate Gaussian distributions. Experiments show that our proposed models can generate more plausible and diverse conversation responses in accordance with designated emotions compared to baseline approaches. Keywords Chatting machine Conditional variational autoencoders Non-deterministic Neural dialog
1 Introduction Understanding the emotional content of conversations and empathizing accordingly is a challenge for artificial conversational agents. Having emotional intelligence, i.e., to enable machine to understand affect and emotion [21], has been a long-term goal for artificial intelligence. Moreover, to express the diverse emotional contents of conversation is another important factor to generate successful artificial conversational agents. To build an interactive human like chatbot, it is absolutely essential to equip the machine with the ability of expressing and understanding emotions and learning diversity of natural languages.
Kaichun Yao and Tiejian Luo were contributed equally and should be considered as co-first authors. & Libo Zhang [email protected] 1
School of Computer Science and Technology, University of Chinese Academy of Sciences, Beijing, China
2
State Key Laboratory of Computer Science, Institute of Software Chinese Academy of Sciences, No. 4 North Road, Beijing, China
The success of deep neural networks in natural language processing tasks [2, 3] promotes the exploration of the paradigm of neural dialogue generation greatly. In existing conversation-generating systems based on the neural network techniques, an encoder-decoder framework [27] has shown great potential in modeling open-domain conversations [4, 5], However, a vanilla encoder-decoder model is prone to generate dull and generic responses. To improve the quality of responses in the conversation generation, latest efforts include diversity promoting objective functions [13], diverse decoding [15], topic-intr
Data Loading...
