Evidence for Increased Discriminability in Judging the Acceptability of Machine Translations: The Case for Magnitude Estimation

Report No. ARL-MR-0720
Authors: James D. Walrath
Date/Pages: May 2009; 20 pages
Abstract: An earlier experiment, in which magnitude estimation (ME) was used as the method for judging the acceptability of machine translations, was replicated with one exception. The current study used a four-point Likert scale as the measurement methodology. In the earlier work, using ME, judges easily discriminated between two sets of machine translations, one containing 25% more correctly translated names than the other. In this study, using a Likert scale, judges were not able to make the same discrimination. These results support the theory that ME may be a superior measurement methodology for assessing the acceptability of machine translations.
Distribution: Approved for public release
  Download Report ( 0.095 MBytes )
If you are visually impaired or need a physical copy of this report, please visit and contact DTIC.
 

Last Update / Reviewed: May 1, 2009