Automated Essay Grading

Additional Info

• Reliability and Validity
• Reading Comprehension
• Essay Grading

The Essay Component of WriteToLearn uses the KAT engine to score student essays. The KAT engine includes as a component Latent Semantic Analysis (LSA), a technique that understands the meaning of text automatically in much the same way as humans. LSA derives its knowledge of language by machine analyzing large bodies of text--over a billion words of running text can be analyzed by LSA. The Essay Component uses LSA as well as additional computational measures of the KAT engine to arrive at an overall score for a student essay as well as trait scores, such as the widely used 6 traits. The Essay Component also flags anomalous essays that require a human decision, such as off topic essays.

The Essay Component of WriteToLearn has been trained to score prompts based on human scored essays. The Essay Component statistically mimics the human scores with reliabilities equal to or exceeding the human-to-human agreement. This is true of both holistic and trait scores. Empirically, the holistic score is more reliable than trait scores, and usually the trait scores are highly correlated with the overall score. But educators often want to give students focused feedback on how to improve the next version of an essay, and trait scores are one way of doing that. The Essay Component also flags grammatical and spelling errors and provides students with suggested corrections.

Please contact us if you are interested in receiving an electronic copy of our research paper Automatic Essay Assessment.