Lexical and Algorithmic Stemming Compared for 9 European Languages with Hummingbird SearchServerTM at CLEF 2003

Hummingbird participated in the monolingual information retrieval tasks of the Cross-Language Evaluation Forum (CLEF) 2003: for natural language queries in 9 European languages (German, French, Italian, Spanish, Dutch, Finnish, Swedish, Russian and English) find all the relevant documents (with high precision) in the CLEF 2003 document sets. For each language, SearchServer scored higher than the median average precision on more topics than it scored lower. In a comparison of experimental SearchServer lexical stemmers with Porter’s algorithmic stemmers, the biggest differences were for the languages in which compound words are frequent (German, Dutch, Finnish and Swedish). SearchServer scored significantly higher in average precision for German and Finnish, apparently from its ability to split compound words and find terms when they are parts of compounds in these languages. Most of the differences for the other languages appeared to be from SearchServer’s lexical stemmers performing inflectional stemming while the algorithmic stemmers often additionally performed derivational stemming; these differences did not pass a significance test.