Lexical Repetitions Lead to Rote Learning: Unveiling the Impact of Lexical Overlap in Train and Test Reference Summaries
This addresses the issue of rote learning in summarization models for NLP researchers, offering an incremental evaluation and training improvement.
The paper tackles the problem of summarization models memorizing training data, showing that lexical overlap between training and test summaries leads to up to a 5x difference in ROUGE-2 scores and reproduces factual errors. It proposes limiting lexical repetitions in training to improve generalization on novel cases while maintaining average performance.
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.