• Innovation through disruptive and scalable technology .
  • Cutting-edge AI research .
  • Accelerating innovations in research and service .
  • We strive for (and achieve) excellence! .
  • “SotA” (State-of-the-Art) .
  • Visual demo of research and service innovation
  • Human. Machine. Experience Together .

Conversational AI

Utilizing AI to talk to machines

Background

Conversational AI is a technology bridging machines and the human-being via communication. In this vein, the system processing tasks given from the conversation with users is called Conversational Agent system. For delivering natural conversation with users, Conversational Agent system requires following components: Language Understanding module, which extracts the meaning from users’ speech; Dialogue Management module, which analyzes the situation during conversation and creates answers accordingly; Task Agent implementing individual service; Chit-chat Agent conveying daily conversation; Information Agent which provides answers by searching information from the knowledge base. Furthermore, it also includes Information Extraction module which collects information for the Knowledge Base modeling of conversation system as well as extracts information.

Project

In our project, we mainly delve on these parts.

Language Understanding

For improving performance of Language Understanding based on deep learning, a method utilizing Big Data has been frequently used. In this project, we also research on Language Understanding based on Unsupervised Representation Learning with Big Data as a foundation.

Question Answering

As one type of intelligent search systems, Question Answering was experimented in numerous works. In this project, we contemplate on Question Answering by considering contexts of the conversation for making proper relation to our conversational system.

Dialogue Policy

Traditionally, conversation system used rule-based conversational strategies based on Flow Chart. However, in this method, it is difficult to control the flow of conversation if scenario becomes complicated. In this project, we delve on conversational strategies which can be applied even with the complicated tasks by implementing machine learning technology.

Conclusion

Historically, having natural conversation between human and machine has been a great challenge and demands a long period of time to propose a solution. Since communication goes beyond the mere language understanding, it should also analyze contexts as well as persons who are participating in the conversation. In line with this, we use our best endeavor to solve this problem with our machine learning technology.

References

  • [1] Devlin, Jacob, et al. "Bert: Pre-training of deep bidirectional transformers for language understanding." arXiv:1810.04805 (2018).
  • [2] Reddy, S., Chen, D., & Manning, C.D. (2018). CoQA: A Conversational Question Answering Challenge. CoRR, abs/1808.07042.
  • [3] Williams, Jason D., and Steve Young. "Partially observable Markov decision processes for spoken dialog systems." Computer Speech & Language 21.2 (2007): 393-422.

  • Matthew
  • Jun
  • Ted
  • Jay
  • Brad
  • Ryanne
  • Haven
  • Jason