Pattern-aware Data Augmentation for Query Rewriting in Voice Assistant Systems
This work is significant for voice assistant developers and researchers, as it provides a method to improve the performance of Query Rewriting systems, especially in data-scarce scenarios, by reducing the need for extensive manual data labeling.
This paper addresses the challenge of insufficient labeled data for Query Rewriting (QR) systems in voice assistants. The authors propose a data augmentation framework that learns patterns from existing data and inversely generates rewrite candidates from labels, which effectively boosts QR performance in low-resource domains.
Query rewriting (QR) systems are widely used to reduce the friction caused by errors in a spoken language understanding pipeline. However, the underlying supervised models require a large number of labeled pairs, and these pairs are hard and costly to be collected. Therefore, We propose an augmentation framework that learns patterns from existing training pairs and generates rewrite candidates from rewrite labels inversely to compensate for insufficient QR training data. The proposed framework casts the augmentation problem as a sequence-to-sequence generation task and enforces the optimization process with a policy gradient technique for controllable rewarding. This approach goes beyond the traditional heuristics or rule-based augmentation methods and is not constrained to generate predefined patterns of swapping/replacing words. Our experimental results show its effectiveness compared with a fully trained QR baseline and demonstrate its potential application in boosting the QR performance on low-resource domains or locales.