Incentives Of EdTech: A Systematic Review Of EduNLP Research

A systematic review of 204 educational natural language processing papers finds a persistent gap between private-sector incentives and the needs of teachers and classrooms, according to research published on arXiv. [1] The review, led by Gabrielle Gaudeau, examined papers from the Association for Computational Linguistics' Special Interest Group on Building Educational Applications in 2024 and 2025. [1] It found that teachers are systematically under-represented as beneficiaries of research, appearing as explicit beneficiaries in just 33.3 percent of the papers. [1] Learners and students were the most frequently named explicit beneficiary, appearing in 125 papers. [3] When teachers did benefit, the papers stated so explicitly 80.9 percent of the time, meaning they were almost never the unstated but evident beneficiary of research. [3] Real-world deployment of the technologies described in the literature remains rare, occurring in only 9.8 percent of the papers reviewed. [1] The analysis also found that ethical engagement tends toward acknowledgement rather than action. [1] The ethics of artificial intelligence covers topics including algorithmic biases, fairness, accountability, and transparency, particularly where systems influence human decision-making, with education identified as an application area with important ethical implications. [7] The review identifies a structural tilt in whose interests EduNLP research serves. Non-profit organisations, industry, and governmental bodies appear prominently as implicit beneficiaries — they are not named in the paper as intended beneficiaries, but the research clearly serves their interests. [3] This pattern is most visible in automated assessment research, the largest task category in the corpus, which consistently benefits learners and industry while teachers and examiners are sparsely represented. [3] The commercial relationship is direct: automated scoring tools reduce the need for human markers and hold clear value for large-scale testing organisations. [3] Grammar error correction research showed the broadest stakeholder spread, in part because such tools serve not only learners and teachers but also the general public who use writing assistance tools in everyday tasks. [3] The global teacher shortage is pushing schools toward greater reliance on artificial intelligence, the authors note, making the question of whose interests are served increasingly urgent. [2] The paper offers concrete recommendations for more responsible EduNLP research practices, drawing on exemplary papers in the corpus. [1] The findings arrive as NLP applications in education continue to expand across question answering, question construction, automated assessment, and error correction, with large language models now widely used in these tasks. [5]