Evaluating a probabilistic model for cross-lingual information retrieval

Probabilistic Structured Query Methods

“…For example, the Arabic word "£ Y ¤¥ " can be translated as "bread" or "bake," and equation (1) would (with proper stemming) reward an occurrence of "baking bread." Corpus-based approaches to CLIR have generally developed within a framework based on language modeling rather than vector space models, at least in part because modern statistical translation frameworks offer a natural way of integrating translation and language models [18]. In general, language modeling approaches to retrieval rely on collection frequency (CF) in place of DF: 2 …”

Structured queries, language modeling, and relevance modeling in cross-language information retrieval

“…(Hiemstra and de Jong, 1999;Berger and Lafferty, 1999;Xu, et al, 2001;Federico and Bertoldi, 2002). Our crosslingual work is based on a widely-used extension of the above monolingual model (Xu et al, 2001). To generate a query from a document in a different language, say Arabic, one samples either the Arabic document, or the English background model.…”

“…The cross-language relevance model is given in Equation (12), which is the same as Equation (10), above, but now subscripts indicate that the documents D a and terms w a are in Arabic, and the query terms e i are English: (12) P(w a |D a ) can still be estimated as in Equation (11), with the document and background models in Arabic. P(e i |D a ) is estimated as in Berger and Lafferty (1999) and Xu, et al, (2001):…”

“…One other study has compared structured query translation with LM for cross-language retrieval (Xu, et al, 2001). They found that language modeling performed better than structured query translation when used with a dictionary derived from a parallel corpus, or a combined dictionary derived from parallel and nonparallel sources.…”

“…Lavrenko et al (2002) compared relevance modeling with cross-lingual language modeling based on unexpanded queries, on the same Chinese data set used in the Xu, et al (2001) experiments, but given that relevance modeling is effectively a query expansion technique, we feel that it is fairer to compare it with another expansion technique.…”