Reference-free Evaluation Metrics for Text Generation: A Survey

TL;DR

This survey reviews reference-free evaluation metrics for natural language generation, addressing the challenges of reference-based methods and exploring diverse approaches applicable across various NLG tasks.

Contribution

It provides a comprehensive overview of reference-free metrics, their applications, and potential future research directions in text generation evaluation.

Findings

Highlights the limitations of reference-based metrics

Summarizes various reference-free evaluation approaches

Identifies promising future research directions

Abstract

A number of automatic evaluation metrics have been proposed for natural language generation systems. The most common approach to automatic evaluation is the use of a reference-based metric that compares the model's output with gold-standard references written by humans. However, it is expensive to create such references, and for some tasks, such as response generation in dialogue, creating references is not a simple matter. Therefore, various reference-free metrics have been developed in recent years. In this survey, which intends to cover the full breadth of all NLG tasks, we investigate the most commonly used approaches, their application, and their other uses beyond evaluating models. The survey concludes by highlighting some promising directions for future research.