|<< Prev||- Up -|
Another obvious constraint on resolving pronouns is of semantic nature.
Another obvious constraint on resolving pronouns is of semantic nature. The English pronouns ``he'' and ``him'' refer to male objects, ``she'' and ``her'' to female objects, and ``it'' normally to non-human objects. Our basic implementation ignores these kinds of sortal constraints, and happily identifies ``John'' as an antecedent for the pronoun ``she'':
> John walks. She smokes.
1 exists(A,smoke(A) & (female(A) & (walk(A) & A=john))
As our computer program doesn't have any information as to which John is male or female, it won't stumble over the fact that we refer to ``John'' with the pronoun ``she''. So what would be a good way of introducing these sortal constraints in our DRT implementation?
First of all we need the sortal information of the pronoun at our disposal. Actually, this information is already present in the DRS, as we have specified this in the lexicon. For the example above, the pronoun ``she'' introduced the DRS-condition
Secondly, we need more information about other entities. We need to know, for instance, that John is a male, that women are female, the females are disjoint from males, and so on. In other words, we need a semantic ontology stipulating semantic relations between entities.
Finally, we need to use this ontology to filter out DRSs that violate sortal constraints. An interesting way of doing this is to translate the ontology into a first-order theory, use the first-order translation of the DRSs, and use a theorem prover (for instance the tableaux-based theorem prover we developed earlier in the course) to check for inconsistent information.
Extend the grammar and lexicon with the possessive pronouns ``his'', ``her'' and ``its''. Hint: think of these pronouns as determiners.
|<< Prev||- Up -|