When automatically parsing natural language, a lot of ambiguities can arise, as the grammar is not always well defined as in computer languages.
Using a linguistic resources, we can remove a certain number of these ambiguities: the resource provides information on position of verbs in a sentence. e.g: the verb talk takes a human subject and can take a non-human complement.
We have created a constrained grammar based on rules specific to each type of verb. The specific parsing is initiated with different part of speech tags for each different type of verbs.
Evidently, this is not as obvious and never perfect. Semantic specificities are not easily detected and will always generate ambiguities.
