An Efficient Approach based on BERT and Recurrent Neural Network for Multi-turn Spoken Dialogue Understanding

Weixing Xiong, Li Ma, Hongtao Liao

Abstract

The main challenge of the Spoken Language Understanding (SLU) is how to parse efficiently natural language into effective meanings, such as its topic intents, acts and pairs of slot-values that can be processed by computers. In multi-turn dialogues, the combination of context information is necessary to understand the user's objectives, which can be used to avoid ambiguity. An approach processing multi-turn dialogues, based on the combination of BERT encoding and hierarchical RNN, is proposed in this paper. More specifically, it combines the current user's utterance with each historical sequence to formulate an input to the BERT module to extract the semantic relationship, then it uses a model derived from the hierarchical-RNN for the understanding of intents, actions and slots. According to our experiments by testing with multi-turn dialogue dataset Sim-R and Sim-M, this approach achieved about 5% improvement in FrameAcc compared with models such as MemNet and SDEN.

Download


Paper Citation


in Harvard Style

Xiong W., Ma L. and Liao H. (2020). An Efficient Approach based on BERT and Recurrent Neural Network for Multi-turn Spoken Dialogue Understanding.In Proceedings of the 12th International Conference on Agents and Artificial Intelligence - Volume 2: ICAART, ISBN 978-989-758-395-7, pages 793-800. DOI: 10.5220/0009101207930800


in Bibtex Style

@conference{icaart20,
author={Weixing Xiong and Li Ma and Hongtao Liao},
title={An Efficient Approach based on BERT and Recurrent Neural Network for Multi-turn Spoken Dialogue Understanding},
booktitle={Proceedings of the 12th International Conference on Agents and Artificial Intelligence - Volume 2: ICAART,},
year={2020},
pages={793-800},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0009101207930800},
isbn={978-989-758-395-7},
}


in EndNote Style

TY - CONF

JO - Proceedings of the 12th International Conference on Agents and Artificial Intelligence - Volume 2: ICAART,
TI - An Efficient Approach based on BERT and Recurrent Neural Network for Multi-turn Spoken Dialogue Understanding
SN - 978-989-758-395-7
AU - Xiong W.
AU - Ma L.
AU - Liao H.
PY - 2020
SP - 793
EP - 800
DO - 10.5220/0009101207930800