Learning by Semantic Similarity Makes Abstractive Summarization Better

TL;DR

This paper evaluates the quality of summaries generated by pre-trained language models like BART using human assessments, revealing that model-generated summaries often outperform reference summaries on certain metrics.

Contribution

It introduces a human evaluation approach to assess summarization quality and discusses the implications of dataset characteristics and model capabilities on summarization performance.

Findings

Model-generated summaries score higher than reference summaries in human evaluations.

Pre-trained language models demonstrate strong generalization abilities.

Dataset characteristics influence the perceived quality of summaries.

Abstract

By harnessing pre-trained language models, summarization models had rapid progress recently. However, the models are mainly assessed by automatic evaluation metrics such as ROUGE. Although ROUGE is known for having a positive correlation with human evaluation scores, it has been criticized for its vulnerability and the gap between actual qualities. In this paper, we compare the generated summaries from recent LM, BART, and the reference summaries from a benchmark dataset, CNN/DM, using a crowd-sourced human evaluation metric. Interestingly, model-generated summaries receive higher scores relative to reference summaries. Stemming from our experimental results, we first argue the intrinsic characteristics of the CNN/DM dataset, the progress of pre-trained language models, and their ability to generalize on the training data. Finally, we share our insights into the model-generated summaries and presents our thought on learning methods for abstractive summarization.