Dictionary-independent translation in CLIR between closely related languages

This paper presents results from a study, where fuzzy string matching techniques were used as the sole query translation technique in Cross Language Information Retrieval (CLIR) between the closely related languages Swedish and Norwegian. It is a novel research idea to apply only fuzzy string matching techniques in query translation. Closely related languages share a number of words that are cross-lingual spelling variants of each other. These spelling variants can be translated by means of fuzzy matching. When cross-lingual spelling variants form a high enough share of the vocabulary of related languages, the fuzzy matching techniques can perform well enough to replace the conventional dictionary-based query translation. Different fuzzy matching techniques were tested in CLIR between Norwegian and Swedish and it was found that queries translated using skipgram matching and a combined technique of transformation rule based translation (TRT) and n-grams perform well. For the best fuzzy matching query types performance difference with respect to dictionary translation queries was not statistically significant.