Investigating the use of Paraphrase Generation for Question Reformulation in the FRANK QA system

TL;DR

This study evaluates paraphrase generation methods to enhance question variety in the FRANK QA system, revealing limitations due to data errors and parser constraints that hinder improvements.

Contribution

It provides an analysis of paraphrase generation effectiveness and highlights the impact of data quality and system limitations on question reformulation.

Findings

Paraphrase generation effectiveness is affected by data errors.

Cleaning LC-QuAD 2.0 improves evaluation accuracy.

Paraphrase methods have limited impact due to FRANK's parser limitations.

Abstract

We present a study into the ability of paraphrase generation methods to increase the variety of natural language questions that the FRANK Question Answering system can answer. We first evaluate paraphrase generation methods on the LC-QuAD 2.0 dataset using both automatic metrics and human judgement, and discuss their correlation. Error analysis on the dataset is also performed using both automatic and manual approaches, and we discuss how paraphrase generation and evaluation is affected by data points which contain error. We then simulate an implementation of the best performing paraphrase generation method (an English-French backtranslation) into FRANK in order to test our original hypothesis, using a small challenge dataset. Our two main conclusions are that cleaning of LC-QuAD 2.0 is required as the errors present can affect evaluation; and that, due to limitations of FRANK's parser, paraphrase generation is not a method which we can rely on to improve the variety of natural language questions that FRANK can answer.