Rich Statistical Parsing and Literary Language

This thesisapplies the Data-Oriented Parsing framework in two areas:parsing & literature. The data-oriented approach rests on the assumptionthat re-use of chunks of training data can be detected and exploited attest time. Syntactic tree fragments form the common thread in the thesis.Chapter 2 presents a method to efficiently extract them from treebanks,based on heuristics of re-occurrence. This method is thus able to discoverthe potential building blocks of large corpora. Chapter 3 then develops amulti-lingual statistical parser based on tree-substitution grammar that handlesdiscontinuous constituents and function tags. We show how a mildly context-sensitive grammar can be employed to produce discontinuous constituents,and then compare this to an approximation that stays within the efficientlyparsable context-free framework. The conclusion from the empirical evaluationis that tree fragments allow the grammar to adequately capture the statisticalregularities of non-local relations, without the need for the increased generativecapacity of mildly context-sensitive grammar.The second part investigates what separates literary from other novels. Asidefrom an introduction in Chapter 4 to machine learning we discuss the differencebetween explanation and prediction.Chapter 5 discusses the data used for this investigation. We work with acorpus of novels and a reader survey with ratings of how literary novels areperceived to be. While considerable questions remain with respect to whether asurvey of the general public is an appropriate instrument to probe the conceptof literature, when viewed as a barometer of public opinion we may considerthe basic question of whether such opinions are at all predictable. The first goalis therefore to find out the extent to which the literary ratings can be predictedfrom the texts; the second, more challenging goal is to characterize the kind ofpatterns that are predictors of more or less literary texts.Chapter 6 establishes baselines for this question. We show that literary novelscontain less adjectives and adverbs than non-literary novels, and present severalsimple measures that are significantly correlated with the literary ratings, suchas vocabulary richness and text compressibility. Cliché expressions is establishedas a negative marker of literary language. A topic model is developed of the