Computers scorinc GMAT essays? Impossible! Or is it?

Author

Hedberg, Sara Reese

Volume

14

Issue

3

fYear

1999

Firstpage

5

Lastpage

7

Abstract

The author was quite skeptical when she first heard that computers were scoring the essay section of the Graduate Management Admissions Test (GMAT-the SAT of graduate business schools). How, even with the decade and a half that she has written about the science and practice of artificial intelligence, could a computer possibly evaluate something as subjective as an essay? The system, called e-rater is able to score at the same rate as a human reader, and can capture the rubric humans use. There are two types of GMAT essay questions-issue analysis or argument. For an issue essay, the writer must respond to a general question; providing reasons and examples to support a position. The argument essay in this context means a persuasive essay built on a rational presentation of points on a given subject. The system runs through a sample set of 270 essays for each topic. It extracts features that are analyzed using linear regression and natural language processing. The results are used to score any essay on the topic. The system is used side by side with a human reader for GMAT essay scoring

Keywords

educational administrative data processing; linguistics; management education; natural languages; statistical analysis; GMAT essay questions; GMAT essay scoring; Graduate Management Admissions Test; argument essay; artificial intelligence; computer scoring; e-rater; essay section; graduate business schools; human reader; issue analysis; linear regression; natural language processing; persuasive essay; rational presentation; Artificial intelligence; Computational linguistics; Councils; Educational institutions; Humans; Natural languages; Production systems; Teleprinting; Testing; Yarn;

fLanguage

English

Journal_Title

Intelligent Systems and their Applications, IEEE

Publisher

ieee

ISSN

1094-7167

Type

jour

DOI

10.1109/5254.769874

Filename

769874