|M.Sc Student||Rotstein Carmen|
|Subject||A Semantic Framework for Total and Partial Adjectives|
|Department||Department of Computer Science||Supervisor||MR Yoad Winter|
This thesis studies the distinction that was proposed in previous works between total and partial adjectives. In pairs of adjectives such as safe-dangerous, clean-dirty and healthy-sick, the first (``total'') adjective describes lack of danger, dirt, malady etc., while the second (``partial'') adjective describes the existence of such properties. It is shown that the semantics of adjective phrases with modifiers such as almost, slightly and completely is sensitive to whether the adjective is total or partial. The interpretation of such modified constructions is accounted for using a novel scale structure for total and partial adjectives. It is proposed that the standard value of a total adjective is always fixed as the lower bound of the corresponding partial adjective. By contrast, the standard value of partial adjectives can take any point on the relative scale. The effects of this theoretical distinction on the behavior of modified constructions are studied in detail. Some other phenomena that are explored show evidence for total and partial adjectival compounds, including comparatives and exceptive constructions. The formal theory that is developed in this thesis involves high order logic. However, it is possible to incorporate some of the linguistic facts into a computational system that is based on a first order logic theorem prover. The system extended here is the CURT inference system by Blackburn and Bos (2000). This system builds first order logic representations for discourses, and decides whether the discourse is consistent and informative. The system is extended to include adjectival modifiers and to capture the differences between total and partial adjectives.