Term disambiguation techniques based on target document collection for cross-language information retrieval: An empirical comparison of performance between techniques

Translation approaches in Cross Language Information Retrieval

“…They are specialized terms may not be contained in a dictionary and its translation may not be upto date, translations stored in a dictionary could be inherently ambiguous, phrases may not be translated correctly if they are not covered by the dictionary. Kazuaki Kishida [9] suggested that in dictionary based translation for CLIR, the source term of a query may contain different translation candidates having different meanings. There are two methods for solving the ambiguity of translation, one is using term co-occurrence statistics in the collection and another is a technique based on pseudo-relevance feedback (PRF).…”

Double-pass clustering technique for multilingual document collections

“…where n(t, t′) is the number of English documents in which both terms t and t′ appear, and n(t) and n(t′) indicate the number of English documents including t and t′, respectively. Although other coefficients, e.g., the cosine coefficient, can be used as similarity measures, the results by using them were expected to be similar to that by the Dice coefficient from CLIR experiments in Kishida [26]. Therefore, to avoid unnecessary complexity of analysis, only the Dice coefficient was adopted in this experiment.…”

“…translation disambiguation) is important for executing CLIR. Indeed, retrieval experiments have shown that translation disambiguation often improves the search performance [26], so it is worth applying it to dictionary-based translation in MLDC.…”

“…A variety of methods of translation disambiguation such as techniques using parallel corpora, techniques based on term co-occurrence statistics, techniques based on query expansion (QE) and so on have been explored in the field of CLIR (see [9] and [26]). Among them, parallel corpus-based techniques would be inappropriate in a situation in which a pair of languages for which a parallel corpus is difficult to obtain is included in the document set.…”

“…Therefore, techniques based on term co-occurrence frequencies would be the most promising approach for disambiguation in practical MLDC. In particular, a version of this technique adopted by Adriani [27] and Gao et al [28] is selected in this article because a CLIR experiment by Kishida [26] showed that it is sufficiently effective and its computational complexity is relatively low.…”