Span-ConveRT: Few-shot Span Extraction for Dialog with Pretrained Conversational Representations

TL;DR

Span-ConveRT is a lightweight dialog slot-filling model that uses span extraction and pretrained conversational models, excelling in few-shot learning scenarios, and is supported by a new restaurant booking dataset.

Contribution

The paper introduces Span-ConveRT, a novel span extraction approach for dialog slot-filling that effectively leverages pretrained conversational models for few-shot learning.

Findings

Span-ConveRT outperforms models trained from scratch and BERT-based span extractors in few-shot settings.

Leveraging pretrained conversational knowledge improves slot-filling performance.

RESTAURANTS-8K dataset provides a new benchmark for dialog span extraction in restaurant booking.

Abstract

We introduce Span-ConveRT, a light-weight model for dialog slot-filling which frames the task as a turn-based span extraction task. This formulation allows for a simple integration of conversational knowledge coded in large pretrained conversational models such as ConveRT (Henderson et al., 2019). We show that leveraging such knowledge in Span-ConveRT is especially useful for few-shot learning scenarios: we report consistent gains over 1) a span extractor that trains representations from scratch in the target domain, and 2) a BERT-based span extractor. In order to inspire more work on span extraction for the slot-filling task, we also release RESTAURANTS-8K, a new challenging data set of 8,198 utterances, compiled from actual conversations in the restaurant booking domain.