Ping ZhongFollow


Conversational agents such as chatbots or virtual assistants currently have a single speaking/utterance style when responding to user prompts or questions. Tasks where the user is looking for specific information or answer to a question benefit from a direct, clear answer. However, such answers are not suitable when users are looking for a conversation with a chatbot, as opposed to finishing a certain task quickly and efficiently. This disclosure describes a chatbot that tailors conversational style based on user prompts and/or preferences. In some implementations, the utterance style of answers (e.g., generated by an LLM or using other techniques) is customized by adding an extra module for style embedding. The style embedding module can be pre-trained separately or can be trained jointly with a natural conversational model.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.