Cohesion Grading Decisions in a Summary Evaluation Environment: A Machine Learning Approach

Abstract

The work presented in this paper has been carried out in the context of a summary writing environment provided with automatic grading. Regarding summarisation discourse, some of the most relevant variables identified in previous work are comprehension, adequacy, use of language, coherence, and cohesion. This work is focused on cohesion. The described exploratory study starts from basic automatic measures of cohesion to further analyse which of them best reflects human expert overall cohesion grades for learner summaries written in the Basque language. For this purpose, 45 basic cohesion measures are compared to overall human cohesion grades. Machine Learning techniques are used to select the best combination for cohesion grading.


Back to Table of Contents