Whilst I’m not particularly convinced about the merits of chatbots, it’s hard to dispute their role in getting AI tools onto the market in the past year.  Such tools are only as useful as their ability to understand and respond successfully to dialogue.

A recent paper from Osaka University aims to make AI-based systems better at lexical acquisition.  Their method utilizes implicit confirmation to allow a system to understand the category of a word across multiple exchanges.  It does this by confirming whether or not its predictions are true in real-time as the conversation flows.

Live chat

This is a marked improvement on most chatbots in existence today, which rely upon pre-programmed responses to questions posed by users.  A more advanced method will see the system attempt to learn from each exchange by asking simple and repetitive questions, but this runs the risk of alienating users.

The method produced by the Osaka team revolves around an implicit confirmation method that allows the AI system to understand the category of a previously unknown word introduced to the conversation by the human.  The system allows the AI to better predict the kind of word it is from user input during the conversation.  It’s a way for the system to become smarter without having to make explicit requests of the user.

The system gauges the response of the user to help it determine whether it’s predictions were correct or not.  It then takes it’s learnings and feeds them into subsequent conversations, thus hopefully become smarter with each engagement.

Chatbots seem likely to be here to stay, so work such as this that can make their interactions smarter and more effective has to be a good thing.  It will be interesting to see how effective the team are at getting the method integrated into future iterations of chatbot.