Casual Conversation Technology Achieving Natural Dialog with Computers

In recent years, voice recognition agents such as NTT DOCOMO’s “Shabette Concier” have become popular. ShabetteConcier is a voice agent capable of responding to task-related utterances such as “send mail” or “call,” and can answer questions such as “how high is Mount Fuji?” or “what is the highest mountain in the world?” It can also respond to casual conversation with utterances such as “I love you” or “hello.” It is highly convenient for users to be able to simply make utterances to perform a task or request particular information. Nevertheless, Shabette-Concier is not only used to enjoy these conveniences – users also talk to it using a wide range of day-to-day chat, suggesting that user desire for casual conversation is very high. However, Shabette-Concier is only able to give precise replies to utterances within the bounds of assumptions, and does not have sufficient variation in its responses. Therefore, we believe we can offer casual conversation as popular content to users and expand communication module installations in new devices such as robots, games and vehicles, and apply this technology to a range of businesses to satisfy user demand for casual conversation technologies. To respond to the user demand for casual conversation, we have developed a casual conversation system based on the technical achievements of NTT Media Intelligence Laboratories. With this system, we have aimed to enable natural conversation between computers and human beings, using utterance data created from large-scale data to generate a rich range of responses – the system does not repeat one-off utterances with users, but enables multiple and varied exchanges. This article describes an overview of the system and the dialog technology it uses.