Morhological Mesure
The idea behind the Morphological Mesure (M Mesure thereafter) is to evaluate a candidate agains a reference on the basic of its morphological attributes. The principes are the following:
- the more specified is the candidate the higher valuation it is likely to get BUT the risk of over-specification comes along.
- the more specified is the reference the lower valuation we may get BUT the risk of under-specification comes along
- the basic score if 1 (both candidate and reference are empty)
- when an attribute is present (resp. abscent) it multiply (resp. divide) the score by its attribute weight.
With the following weights:
- Level 1 attributes (attribute weight = 2) : V (verb), N(noun), ADJ (adjectif), ADV (adverb)
- Level 2 attributes (attribute weight = 1.5) : VI (intransitive verb), VT (transitive verb), VP (pronominal verb), MASC(masculine), FEM (feminine)
- Level 3 (attribute weight = 1.25) : all others
The morphological mesure is used when fusionning two vectors as a relative weight with a morphological reference. For example, if you are asked for a femine noun (N FEM), and you have (1) a vector for a noun (N) and (2) for a verb (V): then the weight of (1) is 1.67 and the weight of (2) is 0.42.
Morhological distance
The idea behind the Morphological Distance (M Distance thereafter) is to compare two morphological sets. If identical, the the distance is 0, and as a real distance function the other commons distance properties are verfied. Any difference in the two morphological set will increase the distance. The function is defined as follows:
DMorpho(A,B) = Weight-Sum ((MorphoSet(A) + MorphoSet(B)) - Intersection(MorphoSet(A) + MorphoSet(B)) )
With the following weights:
- Level 1 attributes (attribute weight = 2) : V (verb), N(noun), ADJ (adjectif), ADV (adverb)
- Level 2 attributes (attribute weight = 1) : VI (intransitive verb), VT (transitive verb), VP (pronominal verb), MASC(masculine), FEM (feminine)
- Level 3 (attribute weight = 0.5) : all others
The morphological distance is often used in conjunction with the angular distance. A common fonction is the following :
Distance(A, B) = max (DAngular(A,B), (DAngular(A,B) + DMorpho(A,B))/2)
In the above function, the morphological distance can only degrade de angular distance if the two morphological sets are not identical.