Do Captioning Metrics Reflect Music Semantic Alignment?

TL;DR

This paper critically examines whether traditional language generation metrics like BLEU and ROUGE are suitable for evaluating music captioning, highlighting their poor correlation with human judgments and vulnerabilities to syntactic variations.

Contribution

It reveals the inadequacy of existing metrics for music captioning evaluation and advocates for developing more appropriate, semantically aligned assessment methods.

Findings

Traditional metrics do not correlate well with human judgments.

Existing metrics are vulnerable to syntactic changes.

A need for reevaluating evaluation standards in music captioning.

Abstract

Music captioning has emerged as a promising task, fueled by the advent of advanced language generation models. However, the evaluation of music captioning relies heavily on traditional metrics such as BLEU, METEOR, and ROUGE which were developed for other domains, without proper justification for their use in this new field. We present cases where traditional metrics are vulnerable to syntactic changes, and show they do not correlate well with human judgments. By addressing these issues, we aim to emphasize the need for a critical reevaluation of how music captions are assessed.