next up previous contents
Next: Conclusions Up: Jackendoff and inferences Previous: Functions relating entities

Selectional restrictions

Although it is clearly impossible to attempt to rule out all semantically anomalous combinations of lexical items at the lexical level (not to mention undesirable due the fact that arbitrary bits of world knowledge can affect the felicity of a sentence -- see copestake:92), the definition of function-argument structures in terms of ontological categories does rule out certain combinations. For example, the sentence in ji22 would be ruled out due to the fact that the go function underlying the semantics of walk requires its first argument to be a thing, not a property.gif

  *Happiness walked down the street.

Similarly, some selectional restrictions are captured by explicitly specifying function values within a verb's lexical structure. For example, the verb pass may only appear with a PP complement specifying a route (e.g. John passed by the office). Goals, directions, etc. are incompatible with the meaning of the verb. Therefore, the lexical semantics of pass is specified as in ji23. The semantics of its PP complement must be compatible with the Path type specified therein.   tex2html_wrap_inline31382

Jackendoff also attempts to capture more fine-grained selectional restrictions, such as the fact that the argument of drink must be a liquid. He accomplishes this by specifying the value of the semantic argument at the appropriate argument position within the lexical semantics of the verb, as liquid. He further assumes an operation of fusion which merges the value of the verb semantic argument with the semantic value of the argument only if the two are compatible.

A similar approach is used to capture some lexically-specific inferences, such as the inference associated with the sentence ji9a that it is butter (and not anything else) which goes onto the bread. In this case, the first argument of the go function is specified to be butter in the lexical semantics of the verb butter (e.g. tex2html_wrap_inline31384 ).


next up previous contents
Next: Conclusions Up: Jackendoff and inferences Previous: Functions relating entities