Existence of Citation Age Bias in NLP

Determine whether the Natural Language Processing (NLP) research community exhibits a citation age bias, i.e., a systematic tendency to disproportionately cite newer papers and neglect older work, specifically within NLP as opposed to reflecting a general pattern across other artificial intelligence subfields.

Background

The paper analyzes citation age dynamics across approximately 300,000 arXiv submissions from 2013 to 2022 spanning 15 subfields, comparing AI-related computer science areas (including NLP, Computer Vision, Machine Learning, and AI) with non-AI computer science and non-CS fields. The authors find that AI subfields show marked decreases in the age of citations over the last decade, while non-CS fields often exhibit stable or increasing citation ages.

Based on this broader cross-field analysis, the authors question prior claims that NLP uniquely suffers from 'citation amnesia.' They explicitly state uncertainty about whether NLP has a specific citation age bias, suggesting that observed patterns may instead be driven by the high dynamicity of AI subfields in general.