A new method developed at Osaka University for dialogue systems exploits lexical acquisition through implicit confirmation, allowing computers to understand the category of new words over multiple dialogues, by confirming the meaning in the flow of conversation.

In the past years, many chatbots, robots and voice assistant apps have been developed; such tools are capable of interacting with users, by answering to questions with preprogrammed information.

To overcome this limited availability of answers, a method that allows computer to learn from human users has been developed in the past. However, such method works with computers asking questions such as “What is abc?” in order to acquire knowledge, causing users to lose interest in talking with the artificial interface.

To overcome this issue and allow computers to learn ‘naturally’, while having a conversation, a group led by Professor Komatani at Osaka University developed an implicit confirmation method that allows computers to understand the category of unknown words during conversation with human users.

The innovative method attempts to enable computers to make implicit confirmation requests to the user, and to have the user respond to these requests, in a way that allows the acquisition of knowledge naturally during dialogues.

During the interactions, the system seeks confirmation about the guessed meaning, deciding whether the prediction is correct or not through the user responses to request and their context, by using machine learning techniques (machine Learning is a method using algorithms to analyse data, learn from it, and then make a determination or prediction). Moreover, this system’s decision performance improved by taking the classification results obtained from dialogues with other humans into consideration.

Chatbots currently available in the market follow standard speech patterns, according to the preprogrammed words, speaking to anyone in the same way. However, as dialogue systems are predicted to grow in popularity in the future, computers will be required to speak by learning from a conversational partner according to the situation.

This group’s research results represent an innovative approach towards the realisation of dialogue systems in which a computer can become smarter through conversation with humans and will lead to the development of dialogue systems with the capacity of customising answers according to the user and the situation.

This research project was carried out as a part of joint research with Honda Research Institute Japan Co., Ltd.

Written by: Pietro Paolo Frigenti

Journal Reference: Ono, K., Takeda, R., Nichols, E., Nakano, M., & Komatani, K. (2017). Lexical Acquisition through Implicit Confirmations over Multiple Dialogues. In Proceedings of the 18th Annual SIGdial Meeting on Discourse and Dialogue (pp. 50-59).