Error-Analysis Morphology
Abstract:
Error-recognition in morphology has even seen less attention than phonology. However the identification of orthographic errors is an important research area.
Spelling correction is a standard feature nowadays in wordprocessors. Usually a word with the smallest so called editing distance is suggested for correction.
Possibly for this reason there has not been much research in this area with regard to CALL.
Note however that a simple spelling correction module is not useful for a learner of especially inflecting languages.
Usually there are some default inflection-paradigms, which are learned by the student and are then applied according to his knowledge.
The problematic cases are the nondefault ones, which an intelligent system should correct accordingly.
A program should ideally not only state that "er gehte" is wrong and may be corrected to "er geht" (dropping the final "e" as the smallest change to the original), but it might present a message saying that "gehte" is wrong because the verb "gehen" is irregular and the correct past tense form is therefore "ging".
What is required in these cases is a module using the morphology of a language to determine the cause of errors from inflection paradigms.
There has been some research on the relation between the phonology and the orthography of a language as it is usually assumed e.g. for German, that there is a certain close relationship between these two areas.
A common approach is the Soundex approach (
Mitton, Roger, 1996). The idea is that a misspelled word not in the lexicon is reduced to a highly simplyfied phonological form via a small set of rules. A comparison with the reduced representation of correct words should reveal a set of candidates.
This of course only works if the string in question is definetly not in the lexicon and the spelling error is actually based on a phonologically similar form.
Finally there has been a concept developed for a mapping of the German syllabel system and orthography (
Maas, Utz, 1992).
Based on this a program was developed not for foreign language learners but for young children learning to write. The ruled based system produces exercises training the recognition of the internal structure of words and the corresponding spelling especially for lenghend and shortend syllabels.