How effective is stemming and decompounding for German text retrieval?

Information retrieval systems operating on free text face difficulties when word forms used in the query and documents do not match. The usual solution is the use of a "stemming component" that reduces related word forms to a common stem. Extensive studies of such components exist for English, but considerably less is known for other languages. Previously, it has been claimed that stemming is essential for highly declensional languages. We report on our experiments on stemming for German, where an additional issue is the handling of compounds, which are formed by concatenating several words. The major contribution of our work that goes beyond its focus on German lies in the investigation of a complete spectrum of approaches, ranging from language-independent to elaborate linguistic methods. The main findings are that stemming is beneficial even when using a simple approach, and that carefully designed decompounding, the splitting of compound words, remarkably boosts performance. All findings are based on a thorough analysis using a large reliable test collection.