Minimal meaningful propositions alignment in student response comparisons

Abstract

In an intelligent educational system, automatic sentence alignment has a pivotal role in determining a foundation for clustering, comparing, summarizing and classifying responses. In this paper, we go beyond sentence alignment by splitting the reference and the student responses into single clauses, which are then aligned using finegrained semantic components (facets). This detailed analysis will enable automated educational systems to become highly scalable, domain-independent and to enrich the classroom experience. The results are very promising, showing a significant increase in terms of F1-score, compared to the best performing baseline.