Search

Technology for natural language analysis using linguistically precise grammars has matured to a level of coverage and efficiency that enables parsing of large amounts of running text. Research groups working within grammatical frameworks like Combinatory Categorial Grammar (CCG; Clark and Curran, 2004), Lexical-Functional Grammar (LFG; Riezler et al., 2002), and Head-Driven Phrase Structure Grammar HPSG (HPSG; Malouf and van Noord, 2004; Oepen et al., 2004; Miyao et al., 2005) have successfully integrated broad-coverage computational grammars with sophisticated statistical parse selection models. The former delineate the space of possible analyses, while the latter provide a probability distribution over competing hypotheses. Hypothesis Parse selection approaches for these frameworks often use discriminative Maximum Entropy (ME) models, where the probability of each parse tree, given an input string, is estimated on the basis of select properties (called features) of the tree (Abney, 1997; Johnson et al., 1999). Such features, in principle, are not restricted in their domain of locality, and enable the parse selection process to take into account properties that extend beyond local contexts (i.e. sub-trees of depth one). Named entity recognition@n-best parsing,||( [ABSTRACT FROM AUTHOR]

We give a detailed account of an algorithm for efficient tactical generation from underspecified logical-form semantics, using a wide-coverage grammar and a corpus of real-world target utterances. Some earlier claims about chart realization are critically reviewed and corrected in the light of a series of practical experiments. As well as a set of algorithmic refinements, we present two novel techniques: the integration of subsumption-based local ambiguity factoring, and a procedure to selectively unpack the generation forest according to a probability distribution given by a conditional, discriminative model. [ABSTRACT FROM AUTHOR]