Collaboration or Corporate Capture? Quantifying NLP's Reliance on Industry Artifacts and Contributions

• DOI: 10.48550/arxiv.2312.03912
• Publication Date: 2023-01-01
• Authors: Will Aitken; Mohamed Abdalla; Karen Rudie; Catherine Stinson
• Keywords: FOS: Computer and information sciences; Computer Science - Computation and Language; Computation and Language (cs.CL)

The advent of transformers, higher computational budgets, and big data has engendered remarkable progress in Natural Language Processing (NLP). Impressive performance industry pre-trained models garnered public attention recent years made news headlines. That these are is noteworthy. Rarely, if ever, academic institutes producing exciting new NLP models. Using critical for competing on benchmarks correspondingly to stay relevant research. We surveyed 100 papers published at EMNLP 2022 determine whether this phenomenon constitutes a reliance publications. find that there indeed substantial reliance. Citations artifacts contributions across categories least three times greater than publication rates per year. Quantifying does not settle how we ought interpret the results. discuss two possible perspectives our discussion: 1) Is collaboration with still absence an alternative? Or 2) free inquiry been captured by motivations research direction private corporations?