Automatic Amharic Essay Scoring System Using Latent Semantic Analysis
No Thumbnail Available
Date
2010-11
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Essay assessment is considered to play a central role in educational process as essays are the
most useful tool to assess learning outcomes. Consequently, essays have been incorporated into
many of the standard testing programs like SAT (Systematic Aptitude Test), GRE (Graduate
Record Examination), TOFEL (Test of English as a Foreign Language) and GMAT (General
Management Aptitude Test).
Though the importance of essay assessment is elevated, the process of assessment using human
evaluators is extremely labor intensive and time consuming. Hence, automatic essay scoring
systems are developed to overcome time, cost, and generalizability issues in manual essay
assessment.
Currently, a number of automatic essay scoring systems using different techniques are available
commercially or as a result of research in the field. PEG (Project Essay Grader), E-rater
(Electronic Essay rater) and IEA (Intelligent Essay Assessor) are among the most common
commercially available automatic essay scoring systems for English language but efforts are also
made for other languages as well JESS (Japan Essay Scoring System) and AEA (Automatic
Essay Assessor for Finnish) to mention some.
This study is an attempt to develop similar system for Amharic language, the working language
of Federal Democratic Republic of Ethiopia, to factual types of essays. The study used Latent
Semantic Analysis which is an information retrieval technique to develop the model. LSA is a
novel application used to evaluate essay based on the extent to which an essay can be matched
against other essays scored by human raters. To achieve this large number of pre-graded essay
corpus in three domains are prepared from different educational institutions and used for
developing the model and conducting the experiment.
The research conducted a detail set of experiments to measure the performance of the proposed
system using the percentage of adjacent agreements between the system score and human score.
The result of the experiment varies with the domains involved and found to be 62%, 59% and
52% agreement in three domains which is considered very promising being the first attempt and
paves a way for other researchers to participate in automatic essay scoring system.
Keywords: Essay Assessment, Automatic Amharic Essay Scoring, Latent Semantic Analysis
Description
Keywords
Essay Assessment; Automatic Amharic Essay Scoring; Latent Semantic Analysis