Not All Errors are Equal: Learning Text Generation Metrics using Stratified Error Synthesis

TL;DR

SESCORE is an unsupervised, model-based NLG evaluation metric that uses iterative error synthesis and severity scoring to achieve high correlation with human judgments across diverse tasks without requiring human-annotated training data.

Contribution

The paper introduces SESCORE, a novel unsupervised metric that leverages error synthesis and severity scoring to evaluate NLG outputs effectively without human annotations.

Findings

SESCORE outperforms existing unsupervised metrics on multiple NLG tasks.

SESCORE achieves comparable performance to supervised metrics like COMET.

SESCORE improves correlation with human judgments on WMT translation datasets.

Abstract

Is it possible to build a general and automatic natural language generation (NLG) evaluation metric? Existing learned metrics either perform unsatisfactorily or are restricted to tasks where large human rating data is already available. We introduce SESCORE, a model-based metric that is highly correlated with human judgements without requiring human annotation, by utilizing a novel, iterative error synthesis and severity scoring pipeline. This pipeline applies a series of plausible errors to raw text and assigns severity labels by simulating human judgements with entailment. We evaluate SESCORE against existing metrics by comparing how their scores correlate with human ratings. SESCORE outperforms all prior unsupervised metrics on multiple diverse NLG tasks including machine translation, image captioning, and WebNLG text generation. For WMT 20/21 En-De and Zh-En, SESCORE improve the average Kendall correlation with human judgement from 0.154 to 0.195. SESCORE even achieves comparable performance to the best supervised metric COMET, despite receiving no human-annotated training data.