Lexical Repetitions Lead to Rote Learning: Unveiling the Impact of Lexical Overlap in Train and Test Reference Summaries

TL;DR

This paper investigates how lexical overlap between training and test summaries causes models to memorize data, leading to poor generalization, and proposes limiting lexical repetitions during training to enhance model robustness and novelty in summarization.

Contribution

It introduces a fine-grained evaluation protocol based on lexical similarity and demonstrates that limiting lexical repetitions during training reduces rote learning and improves generalization.

Findings

Significant performance gap based on lexical similarity levels.

Limiting lexical repetitions reduces factual errors and rote memorization.

Enhanced generalization on novel and recent news summaries.

Abstract

Ideal summarization models should generalize to novel summary-worthy content without remembering reference training summaries by rote. However, a single average performance score on the entire test set is inadequate in determining such model competencies. We propose a fine-grained evaluation protocol by partitioning a test set based on the lexical similarity of reference test summaries with training summaries. We observe up to a 5x (1.2x) difference in ROUGE-2 (entity recall) scores between the subsets with the lowest and highest similarity. Next, we show that such training repetitions also make a model vulnerable to rote learning, reproducing data artifacts such as factual errors, especially when reference test summaries are lexically close to training summaries. Consequently, we propose to limit lexical repetitions in training summaries during both supervised fine-tuning and likelihood calibration stages to improve the performance on novel test cases while retaining average performance. Our automatic and human evaluations on novel test subsets and recent news articles show that limiting lexical repetitions in training summaries can prevent rote learning and improve generalization.