I think you can certainly do this, although you would very likely need to write your own measure using the WordNet::Similarity API (which is very possible to do).
If most of the compound words you are dealing with (like administrative office) are compositional (meaning that you can get a good idea of their meaning by looking at the meanings of the individual words and combining them) then you could get the glosses for the two individual words from WordNet and then combine them. You could do the same for the other word/compound if necessary, and then measure their overlaps much like the lesk measure does. As an idea of what I mean - the following method will let you take two words (with senses indicated) and combine their glosses. http://search.cpan.org/~tpederse/WordNet-Similarity/lib/WordNet/Similarity/GlossFinder.pm You could then measure the overlap in your glosses using the following : http://search.cpan.org/~tpederse/WordNet-Similarity/lib/WordNet/Similarity/GlossFinder.pm This is probably a fairly rough solution, but seems like it might at least be an interesting starting point. Let us know how it goes, or if you figure out something more successful! Cordially, Ted On Fri, Feb 14, 2014 at 1:45 AM, Amir H. Jadidinejad <[email protected]>wrote: > > > > > I'm going to measure the semantic relatedness/similarity of > "administrative office" (a term which is not mapped to a specific WNsynset > but also a bunch of WNsynset: office#n#1,administration#n#2) as the first > terms to another word. As you mentioned, the second word could have a > specific WNsynset or haven't. For example: > > - "administrative office", "town hall": The first term doesn't map to > a specific WNsynset ("office#n#1,administration#n#2") but the second one > have no problem (town hall#n#1). > > Kind regards, > Amir H. Jadidinejad > > ------------------------------ > *From:* Ted Pedersen <[email protected]> > *To:* "[email protected]" <[email protected]> > *Sent:* Friday, February 14, 2014 6:58 AM > *Subject:* Re: [wn-similarity] Fw: A question about wn-similarity > > > Just to clarify your question - are you wondering if you could measure > the semantic relatedness of "administrative office" to another word (or > compound) X? Or are you looking for a value that would tell you how > related the terms "administrative" and "office" are to each other in this > compound...? > > An example of what you would like to measure might be helpful... > > Cordially, > Ted > > > On Thu, Feb 13, 2014 at 2:27 PM, Amir H. Jadidinejad < > [email protected]> wrote: > > > > Hi, > Some terms don't map to a specific WordNet synset. For example, > "administrative office". In this situations, we need to map a term to a > bunch of synsets. For example: office#n#1,administration#n#2. > Is it possible to calculate the semantic relatedness in this situation? > Kind regards, > Amir H. Jadidinejad > > > > > > > -- > Ted Pedersen > http://www.d.umn.edu/~tpederse > > > > > > -- Ted Pedersen http://www.d.umn.edu/~tpederse
