|Abstract:||Noun phrase internal agreement errors and erroneously split compounds are two error types in Swedish texts that have become increasingly more frequent during the last decade. In this paper we present empirical results from studying occurrences of such errors in running text. The results show interesting frequency patterns in the realization of errors. For example, occurrences of agreement errors generally seem to deviate from a possible correction in only one morphological feature. Thus, we find error examples such as ett bil (the article ett deviates in gender from bil, Eng. car), but no examples such as ett bilarna (the article ett deviates in gender, number and species from the noun bilarna, Eng. the cars). Possible explanations for the rise of such error patterns are discussed.|
We also show how agreement errors and split compound errors can be detected and corrected by Granska, a Swedish grammar checker being developed at the Royal Institute of Technology in Stockholm. From a Swedish perspective, detection of these errors is an important task since they are typically not handled in their morphological richness by grammar checkers originally made for English. However, the task is far from trivial, so problems in detecting errors and possible solutions are discussed.