An Effective Natural Language Understanding Model Using Deep Learning and PyDial Toolkit

Abstract

In the current world, where we want spoken dialogue systems, to understand natural language utterances by human beings in a very much human way, an effective natural language understanding model plays a crucial role. In the model, we have proposed and used bi-directional LSTM (Long short-term memory network) based Recurrent neural network for slot identification, and we have developed an ontology for a subset of ATIS airlines dataset, describing the slots under different categories (request able, user-request able, in formable) to guide the Spoken dialogue system to help the user finish the task. The novelty in this research comes from the fact that a complete methodology has been proposed to build the NLU model for any domain based on deep learning which can understand utterances in simpler context and understand multi-word slots. With the following methodology, a natural language understanding model can be built for a spoken dialogue system for any domain. In the proposed model, the bi directional lstm model obtained had an accuracy of over 86%. The proposed technique is easy to use and can be helpful in building basic voice based search agents for any domain with good performance.

Related Papers

• → QuesBELM: A BERT based Ensemble Language Model for Natural Questions(2020)3 cited
• → Spoken Language Characterization(2007)11 cited
• → Automatic Data Expansion for Customer-care Spoken Language Understanding(2018)5 cited
• End-to-End Spoken Language Understanding using RNN-Transducer ASR.(2021)
• → On joint training with interfaces for spoken language understanding(2021)