Ambiguity is a notorious problem for Natural Language Processing. According to results obtained by Schmitz and Quantz I see disambiguation as a process in which contextual defaults are used to derive the most preferred interpretation of an expression. I show how contextual information comprising grammatical as well as conceptual knowledge can be modeled in a homogeneous manner using Terminological Logics (TL). I slightly modify the default extension to tl presented by Quantz and Royer to allow a relevance ordering between multisets of defaults. The preferred interpretation is the one containing the fewest exceptions with respect to such an ordering. Interpretation is thus achieved by exception minimization. I combine this idea with deductive and abductive approaches to interpretation and show how they can be formalized in terms of tl entailment. Furthermore, I obtain a variable depth of analysis by controling the granularity of interpretation via a set of relevant features.
展开▼
