Abstract not found

Prof. Aravind Joshi, my dissertation advisor has been my guide and mentor for the entire time that I spent at Penn. I thank him for all his academic help and personal kindness. The external member on my dissertation committee was Steven Abney, whose suggestions and advice have made the ideas presented here stronger. My dissertation committee members from Penn: Mitch Marcus, Mark Liberman and Martha Palmer provided questions whose answers shaped my dissertation proposal into the finished form in front of you. Many thanks to my academic collaborators; the work on prefix probabilities was done with Mark-Jan Nederhof and Giorgio Satta when they visited IRCS in 1998, the work on subcategorization frame learning was done in collaboration with Daniel Zeman when he visited IRCS in 2000. Thanks to B. Srinivas whose previous work provided the path to the experimental work in this dissertation. Thanks also to Paola Merlo and Suzanne Stevenson for discussions on their work on verb alternation classes. I also acknowledge the help of Woottiporn Tripasai in the extension of their work presented in this dissertation. Thanks to

Coordination of phrases of different syntactic categories has posed a problem for generative systems based only on syntactic categories. Although some prefer to treat them as exceptional cases that should require some extra mechanism (as for elliptical constructions), or to allow for unrestricted cross-category coordination, they can be naturally derived in a grammatic functional generative approach. In this paper we explore the ideia on how mixing syntactic categories and grammatical functions in the label set of a Tree Adjoining Grammar allows us to develop grammars that elegantly handle both the cases of same- and cross-category coordination in an uniform way. 1

This paper is concerned with the standardisation of evaluation metrics for lexical acquisition over precision grammars, which are attuned to actual parser performance. Specifically, we investigate the impact that lexicons at varying levels of lexical item precision and recall have on the performance of pre-existing broad-coverage precision grammars in parsing, i.e., on their coverage and accuracy. The grammars used for the experiments reported here are the LinGO English Resource Grammar (ERG;

ii Preface This workshop was conceived with the aim of bringing together the different computational linguistic subcommunities which model language predominantly by way of theoretical syntax, either in the form of a particular theory (e.g. CCG, HPSG, LFG, TAG or the Prague School) or a more general framework which draws on theoretical and descriptive linguistics. We characterise this style of computational linguistic research as deep linguistic processing, due to it aspiring to model the complexity of natural language in rich linguistic representations. Aspects of this research have in the past had their own separate fora, such as the ACL 2005 workshop on deep lexical acquisition, as well as TAG+, Alpino, ParGram and DELPH-IN meetings. However, since the fundamental approach of building a linguistically-founded system, as well as many of the techniques used to engineer efficient systems, are common across these projects and independent of the specific grammar formalism chosen, we felt the need for a common meeting in which experiences could be shared among a wider community. Deep linguistic processing has traditionally been concerned with grammar development for parsing and generation, with many deep processing systems using the same grammar for both directions. The linguistic precision and complexity of the grammars meant that they had to be manually developed

This dissertation deals with the robustness problem of deep linguistic processing. Hand-crafted deep linguistic grammars provide precise modeling of human languages, but are deficient in their capability of handling ill-formed or extra-grammatical inputs. In this dissertation, we argue that with a series of robust processing techniques, improved coverage can be achieved without sacrificing efficiency or specificity of deep linguistic processing. An overview of the robustness problem in state-of-the-art deep linguistic processing systems reveals that insufficient lexicon and overrestricted constructions are the major sources for the lack of robustness. Targeting both, several robust processing techniques are proposed as add-on modules to the existing deep processing systems. For the lexicon, we propose a deep lexical acquisition model to achieve automatic online detection and acquisition of missing lexical entries. The model is further extended for acquiring multiword expressions which are syntactically and/or semantically idiosyncratic. The evaluation shows that our lexical acquisition results significantly improved grammar coverage without noticeable degradation in accuracy. For the constructions, we propose the partial parsing strategy to maximally recover the intermediate results when the full analysis is not available. Partial parse selection models are proposed and evaluated. Experiment results show that the fragment semantic outputs recovered from the partial parses are of good quality and high value for practical usage. Also, the efficiency issues are carefully addressed with new extensions to the existing efficient processing algorithms. iii ivZusammenfassung Diese Dissertation befasst sich mit dem Robustheitsproblem tiefer Sprachverarbeitungssysteme. Manuell erstellte tiefe Grammatiken liefern

We present a computational model of grammar learning that combines domain-general learning mechanisms with rich representations of linguistic knowledge, world knowledge and situational and discourse context. These representations support processes of language understanding and inference (including both constructional analysis and reference resolution) that help the learner make sense of utterances in context. The learner then draws on generalization and statistical induction techniques to form new constructions that better capture correlations between linguistically identified and contextually inferred information. Our work is part of the larger Neural Theory of Language (NTL) project, whose goal is to build models of cognition and language that satisfy convergent constraints from biology, psychology, linguistics and computation (Chang,