Ray Jackendoff (e.g. jack:83,jack:90) develops a representation for conceptual structure, or entities which he argues reflect human knowledge of the world. His methodology for identifying the constituents of such structure includes analyzing alternations in the linguistic context in which particular words (mainly verbs and prepositions) can be used, to identify generalisations over relations between alternate uses of lexical items. These generalisations can be captured in regular lexical structure. Components of the lexical representation are assigned a consistent semantics and can be combined in constrained ways.
Given this methodology, it is clear that Jackendoff focuses primarily on syntactically-relevant lexical semantics, although he argues at length in favor of viewing conceptual structures as entities which capture concepts or thoughts. His analyses, then, can be used to address the problem of capturing regularities in the lexicon which have a direct influence on syntactic structure. Below, I will introduce some of these analyses and argue that current approaches in computational lexical semantics could benefit by incorporating them.