Sequence Comparison Applied to Correction and Markup of Multi-Word Responses

Abstract

In some instructional situations, such as foreign language dictation, the degree of correctness of a student's text response can be determined without reference to grammar and semantics by comparison with a target string provided by a course author. The standard sequence comparison procedure, which assesses the distance between two strings in terms of edit costs, makes demands on machine time proportional to the product of the string lengths. This characteristic renders it impractical for real-time correction of multi-word responses on current instructional computer systems. We present a much faster but nonadmissible version of other applications, thereby bringing the technique within range of current microcomputers. The usual method for generating markup for single word responses does not generalize well to multi-word responses because it fails to recognize word boundaries, and will sometimes suggest edits that seem unnatural to users. We propose an extension which attends to word boundaries and thereby recommends corrections that appear more reasonable.