CEval: A Benchmark for Evaluating Counterfactual Text Generation

TL;DR

This paper introduces CEval, a comprehensive benchmark for evaluating counterfactual text generation methods, unifying metrics, datasets, and baselines to facilitate consistent and fair comparison.

Contribution

CEval provides a standardized benchmark with datasets, metrics, baselines, and open-source tools for evaluating counterfactual text generation methods.

Findings

No method perfectly balances counterfactual accuracy and text quality.

High-performing counterfactual methods often produce lower-quality text.

Large language models with simple prompts generate high-quality but less accurate counterfactuals.

Abstract

Counterfactual text generation aims to minimally change a text, such that it is classified differently. Judging advancements in method development for counterfactual text generation is hindered by a non-uniform usage of data sets and metrics in related work. We propose CEval, a benchmark for comparing counterfactual text generation methods. CEval unifies counterfactual and text quality metrics, includes common counterfactual datasets with human annotations, standard baselines (MICE, GDBA, CREST) and the open-source language model LLAMA-2. Our experiments found no perfect method for generating counterfactual text. Methods that excel at counterfactual metrics often produce lower-quality text while LLMs with simple prompts generate high-quality text but struggle with counterfactual criteria. By making CEval available as an open-source Python library, we encourage the community to contribute more methods and maintain consistent evaluation in future work.