In task-oriented conversation systems, natural language generation systemsthat generate sentences with specific information related to conversation floware useful . We propose a simple one-stage framework to generate utterances directly from MR (Meaning Representation) Our model is based onGPT2 and generates utterances with flat conditions on slot and value pairs, which does not need to determine the structure of the sentence . We evaluate various systems in the E2E dataset with 6 automatic metrics . Our system is asimple method, but it demonstrates comparable performance to previous systemsin automated metrics . In addition, using only 10\% of the data set without anyother techniques, our model achieves comparable performance, and shows the possibility of performing zero-shot generation

Author(s) : Joosung Lee

Links : PDF - Abstract

Code :
Coursera

Keywords : generation - utterances - performance - representation - conversation -

Leave a Reply

Your email address will not be published. Required fields are marked *