The e-rater engine, built by Educational Testing Service (ETS), is an automated essay-scoring system that uses natural language processing to evaluate written responses. According to ETS, it analyzes features tied to writing proficiency - grammar, mechanics, word use and complexity, style, organization, and development - and was trained by learning how those features relate to the scores human raters assign.
In high-stakes use, e-rater does not replace human graders. ETS runs it in conjunction with human ratings on the Analytical Writing sections of the GRE General Test and on the TOEFL iBT writing prompts, where it serves as a check on human scores and flags discrepancies. The same engine also powers lower-stakes practice and feedback products that give learners real-time comments on their writing.
Automated essay scoring is one of the oldest applied NLP tasks in education - the idea dates to Ellis Page’s work in the 1960s - and e-rater is among the most widely deployed examples. It is also a useful contrast to the AI-writing detectors that arrived after ChatGPT: e-rater grades the quality of writing, whereas tools like Turnitin’s detector and GPTZero try to guess who, or what, produced it.