AI Essay Grading Systems Systematically Penalize Non-Native English Speakers

Multiple AI-powered essay grading systems used in high-stakes standardized testing, including Educational Testing Service's e-rater and Vantage Learning's IntelliMetric, were found to exhibit systematic bias against non-native English speakers. Academic research published in 2019 demonstrated that these systems consistently assigned lower scores to essays written by non-native speakers, even when the content quality, argumentation, and ideas were equivalent to those of native speakers. The bias manifested through the algorithms' heavy weighting of linguistic features such as sentence complexity, vocabulary sophistication, and grammatical structures that favor native English writing patterns. Non-native speakers, despite demonstrating strong content knowledge and critical thinking skills, were penalized for using simpler sentence structures, more common vocabulary, or slight grammatical variations that are typical of second-language acquisition patterns. The impact was particularly significant for international students taking standardized tests like the GRE, TOEFL, and state assessment exams, where AI grading was increasingly being used to supplement or replace human scorers. Research indicated that approximately 200,000 test-takers annually could be affected by these scoring disparities, with potential consequences for college admissions, scholarship opportunities, and academic placement decisions. Educational Testing Service and other testing companies initially defended their systems, arguing that language proficiency was a legitimate component of writing assessment. However, researchers demonstrated that the bias persisted even when controlling for overall English proficiency levels, suggesting the algorithms were not accurately measuring writing quality but rather privileging specific linguistic styles associated with native speakers. The controversy intensified debates about the appropriate role of AI in educational assessment, particularly for diverse student populations. Critics argued that automated scoring systems risked perpetuating educational inequalities by systematically disadvantaging students from non-English speaking backgrounds, while proponents emphasized the need for consistent and scalable assessment methods in an era of increasing test volume.