alt Inc. announced "alt Dialogue Engine", capable of natural human speech, combining AIML rules, data of knowledge bases and RNN utterance generation.
Combining artificial intelligence markup language (AIML) rules, data drawn from knowledge bases, and response and question generation that uses recurrent neural networks (RNN), the alt Dialogue Engine is unique in that it can automatically classify intent of questions to return most suitable response using ranker learning technique. This is world’s first dialogue AI that incorporates all these elements for the Japanese language (see note). It is also ranked in the highest class for English dialogue AI.
Unique Characteristics of the alt Dialogue Engine
AIML rules can be used to provide the Dialogue Engine with personality and individuality as it can memorize information on the speaker.
Capable of knowledge-based question answering by extracting data from online knowledge bases like Wikipedia and Q&A web sites.
Style and tone of utterance are automatically adjusted to suit the speaker when using knowledge bases.
The use of RNN allows for completely new conversations to be created in response to questions asked.
The RNN has an internal memory state, which changes based on the questions asked and controls utterance generation. The internal state enables bringing consistency in conversation referring to entire history of the dialogue.
Answers based on the AIML rules, the knowledge bases, and the RNN can be automatically ranked into intent categories to respond with the most appropriate answer.
When two humans converse with each other, there are questions, answers, and then new topics arising in a mutual and free interplay. The Dialogue Engine also actively participates, asking its own questions to promote the flow of the conversation, as would be the case for a truly natural conversation, without simply keeping responding to questions. The RNN makes this possible, and alt Inc. continues adjust the algorithms while collecting conversation data to keep improving the quality and accuracy of the dialogue. Our goal is to realize the Dialogue Engine that will continuously evolve and grow as human grows while learning language and interaction.
NOTE: Based on internal research, referencing academic papers and announcements as of July 2018.
Kindly direct inquiries to:
Haruko Yamaguchi, +81-3-6380-7076
Marketing and PR
3-1-2 Higashi-Kanda, Chiyoda-ku, Tokyo
Vietnam Tech Lab:
11th floor, Lucky Building, 81 Tran Thai Tong Street, Hanoi, Vietnam