Accurate recovery of predicate-argument dependencies is vital for interpretation tasks like information extraction and question answering, and unbounded dependencies may account for a significant portion of the dependencies in any given text. This thesis describes a Generalized Categorial Grammar (GCG) which, like other categorial grammars, imposes a small, uniform, and easily learnable set of semantic composition operations based on functor-argument relations, but like HPSG, is generalized to limit the number of categories used to those needed to enforce grammatical constraints.The thesis also describes a system for automatically reannotating syntactically-annotated corpora for the purpose of refining linguistically-informed phrase structure analyses of various phenomena. In particular, it describes a method for implementing syntactic analyses of various phenomena through automatic reannotation rules, which operate deterministically on a corpus like the Penn Treebank (Marcus et al., 1993) to produce a corpus with desired syntactic analyses. This reannotated corpus is then used to define a probabilistic grammar which is automatically annotated with additional latent variable values (Petrov and Klein, 2007) and used to parse the constituent and syntactic dependencies from input sentences of the Wall Street Journal and from a minimal but special corpus introduced by (Rimell et al., 2009) that contains only sentences having Object extraction from a relative clause, Object extraction from a reduced relative clause, Subject extraction from a relative clause, Free relatives, Object wh-questions, Right node raising, and Subject extraction from an embedded clause. This corpus was designed specifically to test various parsers on their capability to recover these unbounded dependencies as studied by (Rimell et al., 2009, Nivre et al., 2010). Our system achieves the best result with noticeable margin on unbounded dependency recovery task compared to the results of all 7 other major systems studied by (Rimell et al., 2009, Nivre et al., 2010). The first paper describing this system earned the attention from the NLP research community after it won the Best Paper Award at the international conference COLING 2012.