Decomposing Swedish Compounds Using Memory-Based Learning

Swedish morphology differs significantly from English in several ways. This is something which makes natural language processing based on the English language not always applicable for Swedish material. One area where there is a difference is compounding. The word-forming process of compounding is very productive in Swedish. The compounds are mostly written as one word, without the segmentation point marked in any way. Thus segmentation has to be done in order to interpret the compounds. In this study I have implemented a decomposer which finds the segmentation point in Swedish compounds, making it easier to handle compounds in natural language processing. Brodda’s algorithm for heuristic compound segmentation guided the work. The decomposer is implemented in TiMBL, a memory-based learner.