Dependency relations between words are increasingly recognized as an important level of linguistic representation that is close to the data and at the same time to the semantic functor-argument structure as a target of syntactic analysis and processing. Correspondingly, dependency structures play an important role in parser evaluation and for the training and evaluation of tools based on dependency treebanks. Gold standard dependency treebanks have been created for some languages, most notably Czech, and annotation efforts for other languages are under way. At the same time, general techniques for detecting errors in dependency annotation have not yet been developed. We address this gap by exploring how a technique proposed for detecting errors in constituency-based syntactic annotation can be adapted to systematically detect errors in dependency annotation. Building on an analysis of key properties and differences between constituency and dependency annotation, we discuss results for dependency treebanks for Swedish, Czech, and German. Complementing the focus on detecting errors in dependency treebanks to improve these gold standard resources, the discussion of dependency error detection for different languages and annotation schemes also raises questions of standardization for some aspects of dependency annotation, in particular regarding the locality of annotation, the assumption of a single head for each dependency relation, and phenomena such as coordination. 1

In this thesis we develop a discriminative learning method for dependency parsing using online large-margin training combined with spanning tree inference algorithms. We will show that this method provides state-of-the-art accuracy, is extensible through the feature set and can be implemented efficiently. Furthermore, we display the language independent nature of the method by evaluating it on over a dozen diverse languages as well as show its practical applicability through integration into a sentence compression system. We start by presenting an online large-margin learning framework that is a generaliza- tion of the work of Crammer and Singer [34, 37] to structured outputs, such as sequences and parse trees. This will lead to the heart of this thesis – discriminative dependency pars- ing. Here we will formulate dependency parsing in a spanning tree framework, yielding efficient parsing algorithms for both projective and non-projective tree structures. We will then extend the parsing algorithm to incorporate features over larger substructures with- out an increase in computational complexity for the projective case. Unfortunately, the non-projective problem then becomes NP-hard so we provide structurally motivated ap- proximate algorithms. Having defined a set of parsing algorithms, we will also define a rich feature set and train various parsers using the online large-margin learning framework. We then compare our trained dependency parsers to other state-of-the-art parsers on 14 diverse languages: Arabic, Bulgarian, Chinese, Czech, Danish, Dutch, English, German, Japanese, Portuguese, Slovene, Spanish, Swedish and Turkish. Having built an efficient and accurate discriminative dependency parser, this thesis will then turn to improving and applying the parser. First we will show how additional re- sources can provide useful features to increase parsing accuracy and to adapt parsers to new domains. We will also argue that the robustness of discriminative inference-based learning algorithms lend themselves well to dependency parsing when feature representa- tions or structural constraints do not allow for tractable parsing algorithms. Finally, we integrate our parsing models into a state-of-the-art sentence compression system to show its applicability to a real world problem.

Abstract. Interaction Grammars (IGs) are a grammatical formalism based on the notion of polarity. Polarities express the resource sensitivity of natural languages by modelling the distinction between saturated and unsaturated syntactic structures. Syntactic composition is represented as a chemical reaction guided by the saturation of polarities. It is expressed in a model-theoretic framework where grammars are constraint systems using the notion of tree description and parsing appears as a process of building tree description models satisfying criteria of saturation and minimality.

The aim of this study is to assemble and deploy various NLP components and resources in order to parse scientific medical text data and evaluate the degree in which these resources contribute to the overall parsing performance. With parsing we limit our efforts to the identification of unrestricted noun phrases with full phrase structure and investigate the effects of using layers of semantic annotations prior to parsing. Scientific medical texts exhibit complex linguistic structure but also regularities that can be captured by pre-processing the texts with specialized semantically-aware tools. Our results show evidence of improved performance while the complexity of parsing is reduced. Parsed scientific texts and inferred syntactic information can be leveraged to improve the accuracy of higher-level tasks such as information extraction and enhance the acquisition of semantic relations and events. 1

Abstract: Interaction Grammar (IG) is a grammatical formalism based on the notion of polarity. Polarities express the resource sensitivity of natural languages by modelling the distinction between saturated and unsaturated syntactic structures. Syntactic composition is represented as a chemical reaction guided by the saturation of polarities. It is expressed in a model-theoretic framework where grammars are constraint systems using the notion of tree description and parsing appears as a process of building tree description models satisfying criteria of saturation and minimality.