I am thinking about the following experiment and was wondering if it was already done before and what the results were.
I have a noun phrase, in language L1, composed of adjective A1 and noun N1 and I want to automatically translate it (create a semantic sense link) into a corresponding noun phrase in language L2.
I do have a fairly complete dictionary of translations from L1 to L2.
If there is only one sense, there is no problem.
N1 has n >1 senses in L2.
Can I determine from the junction of the sets of translations of the adjectives in L2 and L1 which sense of N1 is selected through A1?
There are adjectives (category 1) that are universal in the sense that they can be attached to anything (e.g. ‘big’). Obviously these are not useful if they appear in all senses.
There are adjectives (category 2) that specifically appear only in one of the senses. These should be recorded with the sense as indicators for the sense.
Anything in between (several but not all senses have that attribute): investigate if it is an attribute of category 1 or a mis-attribution? If not these attributes (category 3) should indicate the closeness of the senses that contain the adjectives.
Implementation plan example:
Access 2-grams of large English and e.g. German corpus,
sort out the adjective noun combinations for both.
Pick a word in English, determine the senses in German.
Generate the adjective sets for each and execute a junction operation over the sets.
Determine adjectives of categories 1-3.
Can the adjective subsets be used to determine ISA hierarchies?
Are there additional adjective categories that could be useful?