‹header›
‹date/time›
Click to edit Master text styles
Second level
Third level
Fourth level
Fifth level
‹footer›
‹#›
Problem:  Lexical entries in WordNet can dozens of different senses. 
In order to make a semantic representation, we need to decide which meaning is most appropriate.  This task is called Word Sense Diambiguation – but WordNet’s senses are too-fined grained to make this a plausible task for untrained native speakers, and too difficult of a task to model in one step.  So we are working on an intermediate goal: Semantic Class Disambiguation. 
(This is the basic case: most frequent senses succeed.)
This sentence is unambiguous syntactically.
So there’s only one part of speech combination to consider.
noun(“woman”) – verb(“yawn”)
The most common sense of yawned is v-body,
And most common sense of woman is n-person,
So they get tried first.
V-body allows n-person as a subject, so it passes constraint.
And we get the parse on the right.
(Most frequent noun sense inappropriate)
“Chair” is morphologically ambiguous –
can be either a noun or a verb.  NL-Soar tries both trees,
But the syntax module rejects “chair” as a verb (green X on left).
This causes semantic module to throw out all the verb-classes which include a sense of the word “chair” (green X on right through v-social)
This still leaves two noun-classes: n-artifact and n-person.
N-artifact is more common, and is tried first. 
But v-body rejects n-artifact as a subject, so it fails constraint.
NL-Soar then tries n-person (e.g., dept. chair), which succeeds.
Example #3:  The crevasse yawned.
(Most frequent verb sense inappropriate)
There is only one sense for crevasse in WordNet, and it belongs to n-object.  This doesn’t work as a subject of body verbs, so the verb sense fails (matching with no subjects and failing to fill its external role).
There is another sense of yawn (“be open, gape, yaw”) which does fit with natural objects, and this sense matches.