Evaluating Automatic Detection of Misspellings in German.

Anne Rimrott and Trude Heift Simon Fraser University This study investigates the performance of a spell checker designed for native writers on misspellings made by second language (L2) learners. It addresses two research questions: 1) What is the correction rate of a generic spell checker for L2 misspellings? 2) What factors influence the correction rate of a generic spell checker for L2 misspellings? To explore these questions, the study considers a corpus of 1,027 unique misspellings from 48 Anglophone learners of German and classifies these along three error taxonomies: linguistic competence (competence versus performance misspellings), linguistic subsystem (lexical, morphological or phonological misspellings), and target modification (single-edit misspellings (edit distance = one) versus multiple-edit misspellings (edit distance > 1)). The study then evaluates the performance of the Microsoft Word® spell checker on these misspellings. Results indicate that only 62% of the L2 misspellings are corrected and that the spell checker, independent of other factors, generally cannot correct multiple-edit misspellings although it is quite successful in correcting single-edit errors. In contrast to most misspellings by native writers, many L2 misspellings are multiple-edit errors and are thus not corrected by a spell checker designed for native writers. The study concludes with computational and pedagogical suggestions to enhance spell checking in CALL.