Make Every Example Count: On the Stability and Utility of Self-Influence for Learning from Noisy NLP Datasets
This addresses data quality issues for NLP practitioners, offering a task-agnostic filtering method, but it is incremental as it builds on existing self-influence and curriculum learning approaches.
The paper tackles the problem of data quality as a bottleneck in NLP by proposing self-influence scores for data cleaning, showing that this method improves downstream performance in machine translation, question answering, and text classification with concrete gains, such as up to 2.1 BLEU points in translation.
Increasingly larger datasets have become a standard ingredient to advancing the state-of-the-art in NLP. However, data quality might have already become the bottleneck to unlock further gains. Given the diversity and the sizes of modern datasets, standard data filtering is not straight-forward to apply, because of the multifacetedness of the harmful data and elusiveness of filtering rules that would generalize across multiple tasks. We study the fitness of task-agnostic self-influence scores of training examples for data cleaning, analyze their efficacy in capturing naturally occurring outliers, and investigate to what extent self-influence based data cleaning can improve downstream performance in machine translation, question answering and text classification, building up on recent approaches to self-influence calculation and automated curriculum learning.