Amharic-English information retrieval with pseudo relevance feedback

• We describe cross language retrieval experiments using Amharic queries and English language document collection from our participation in the bilingual ad hoc track at the CLEF 2007. Two monolingual and eight bilingual runs were submitted. The bilingual experiments designed varied in terms of usage of long and short queries, presence of pseudo relevance feedback (PRF), and three approaches (maximal expansion, first-translation-given, manual) for word sense disambiguation. We used an Amharic-English machine readable dictionary (MRD) and an online Amharic-English dictionary in order to do the lookup translation of query terms. In utilizing both resources, matching query term bigrams were always given precedence over unigrams. Out of dictionary Amharic query terms were taken to be possible named entities in the language, and further filtering was attained through restricted fuzzy matching based on edit distance. The fuzzy matching was performed for each of these terms against automatically extracted English proper names. The Lemur toolkit for language modeling and information retrieval was used for indexing and retrieval. Although the experiments are too limited to draw conclusions from, the obtained results indicate that longer queries tend to perform similar to short ones, PRF improves performance considerably, and that queries tend to fare better when we use the first translation given in the MRD rather than using maximal expansion of terms by taking all the translations given in the MRD.

Subject headings

NATURVETENSKAP  -- Data- och informationsvetenskap (hsv//swe)

NATURAL SCIENCES  -- Computer and Information Sciences (hsv//eng)

Keyword

Amharic

Cross language information retrieval

Query analysis

Publication and Content Type

ref (subject category)

kon (subject category)