Dancing Between Success and Failure: Edit-level Simplification Evaluation using SALSA

TL;DR

This paper introduces SALSA, a comprehensive human annotation framework for evaluating text simplification, along with a new automatic metric LENS-SALSA, revealing detailed insights into model and human performance.

Contribution

The paper presents SALSA, a novel fine-grained annotation framework for text simplification evaluation, and develops LENS-SALSA, an automatic metric trained on these annotations.

Findings

GPT-3.5 performs more quality edits than humans

Fine-grained annotations reveal differences in simplification strategies

Word-level quality estimation shows promising results

Abstract

Large language models (e.g., GPT-4) are uniquely capable of producing highly rated text simplification, yet current human evaluation methods fail to provide a clear understanding of systems' specific strengths and weaknesses. To address this limitation, we introduce SALSA, an edit-based human annotation framework that enables holistic and fine-grained text simplification evaluation. We develop twenty one linguistically grounded edit types, covering the full spectrum of success and failure across dimensions of conceptual, syntactic and lexical simplicity. Using SALSA, we collect 19K edit annotations on 840 simplifications, revealing discrepancies in the distribution of simplification strategies performed by fine-tuned models, prompted LLMs and humans, and find GPT-3.5 performs more quality edits than humans, but still exhibits frequent errors. Using our fine-grained annotations, we develop LENS-SALSA, a reference-free automatic simplification metric, trained to predict sentence- and word-level quality simultaneously. Additionally, we introduce word-level quality estimation for simplification and report promising baseline results. Our data, new metric, and annotation toolkit are available at https://salsa-eval.com.