Researchers from Osaka University have developed a new method for dialogue systems. The method is called lexical acquisition through implicit confirmation. It is a method for a computer to acquire the category of a word it does not know over multiple dialogues by confirming if the predictions are correct during the flow of a conversation.
Conversational robots, chatbots and voice assistant apps have been popping up everywhere in the last few years. But these computer systems basically answer questions based on what has been preprogrammed into the computer. There is another method where a computer learns from humans by asking simple and repetitive questions. But if the computer only asks questions like, “What is XYZ?” in order to grow, researchers found that users lose interest in talking to the computer.
The group was led by Professor Komatani. They developed an implicit confirmation method where the computer acquires the category of an unknown word during a conversation with humans. The method aims for the system to predict the category of an unknown word from user input during the conversation, in order to make implicit confirmation requests to the user and the user responds to these requests. This way, the system acquires knowledge about words while still holding a conversation.
With this method, the system decides if the prediction is correct or not by using the response following each request and the context by using machine learning techniques. Along with this, the system’s decision performance improved by taking the classification results gained from dialogues with other users into consideration.
Chatbots in the market speak to anyone in the same manner. But, as systems like chatbots are becoming more popular, computers will be required to speak by learning from a new conversational partner according to the situation. The group’s research results are a novel approach toward the realization of dialogue systems where a computer can become smarter through a conversation with humans and lead to the development of dialogue systems with the ability to customize responses to the user’s situation.
The paper on this research was presented at the SIGDIAL 2017 in August 2017.