DEXTER: Deep Encoding of External Knowledge for Named Entity Recognition in Virtual Assistants

TL;DR

This paper introduces DEXTER, a deep encoding method for external knowledge to improve named entity recognition in noisy voice assistant inputs, achieving significant error reduction and aiding related semantic parsing tasks.

Contribution

The paper presents a novel NER system that incorporates external knowledge and indirect labeling techniques, specifically designed for noisy voice assistant data, outperforming baseline models.

Findings

6% reduction in NER error rate

Up to 5% improvement in semantic parsing error rate

Effective handling of noisy, user-derived data

Abstract

Named entity recognition (NER) is usually developed and tested on text from well-written sources. However, in intelligent voice assistants, where NER is an important component, input to NER may be noisy because of user or speech recognition error. In applications, entity labels may change frequently, and non-textual properties like topicality or popularity may be needed to choose among alternatives. We describe a NER system intended to address these problems. We test and train this system on a proprietary user-derived dataset. We compare with a baseline text-only NER system; the baseline enhanced with external gazetteers; and the baseline enhanced with the search and indirect labelling techniques we describe below. The final configuration gives around 6% reduction in NER error rate. We also show that this technique improves related tasks, such as semantic parsing, with an improvement of up to 5% in error rate.