|Jacob Sparre Andersen: Computing: Translation precision||[ Content ]|
Translation precisionfor search purposes
If you are looking for an object which you in one language might describe with the term A, how likely is it that an object described with the term B in another language is what you asked for? (thought you were asking for?)
Notice that the translation precision does not have to be symmetric. Actually it is possible that it will mostly be assymmetric.
Once we have translations with a defined translation precision, we can search for documents in multiple languages, even if the user only writes the search criteria in his/her primary language. Searches on translated terms (i.e.
bluit instead of
blåda) will then be discounted using the translation precision (i.e. 0.7 in the current example). When one word has multiple translations the search should be performed on all the translations, and the search score should somehow add up the scores for the different translations.