| 11 |
Talking with AIs about views from a vehicleCasual-dialog system based on scenery and nearby information 
|
|---|
This is a study of a partner dialogue system for mobile vehicles that uses the ever-changing, real-time view from the car as a topic of conversation. This system uses a deep-learning based dialogue model using the largest scale of Japanese dialogue data developed by NTT to realize natural dialogue. This system integrates scenery images from vehicle and spots around the car's location to talk about scenery around the vehicle. By sequentially incorporating information about the area around the vehicle's location, we are realizing a new experience of driving while enjoying the pleasure of sharing the "now" with a knowledgeable dialogue system.
[1] H. Sugiyama, M. Mizukami, T. Arimoto, H. Narimatsu, Y. Chiba, H. Nakajima, T. Meguro, “Empirical analysis of training strategies of Transformer-based Japanese chit-chat systems,” arxiv:2109.05217, 2021.
[2] K. Mitsuda, R. Higashinaka, H. Sugiyama, M. Mizukami, T. Kinebuchi, R. Nakamura, N. Adachi, H. Kawabata, “Fine-tuning a pre-trained Transformer-based encoder-decoder model with user-generated question-answer pairs to realize character-like chatbots,” IWSDS, 2021.
Hiroaki Sugiyama / Interaction Research Group, Innovative Communication Laboratory
Email: cs-openhouse-ml@hco.ntt.co.jp