Regelbasierte und datenbasierte Pronomenresolution für das Deutsche: Vergleich und Bewertung von Datenquellen

DSpace Repositorium (Manakin basiert)


Dateien:

Zitierfähiger Link (URI): http://nbn-resolving.de/urn:nbn:de:bsz:21-opus-46044
http://hdl.handle.net/10900/46594
Dokumentart: Dissertation
Erscheinungsdatum: 2010
Sprache: Deutsch
Fakultät: 5 Philosophische Fakultät
Fachbereich: Allgemeine u. vergleichende Sprachwissenschaft
Gutachter: Hinrichs, Erhard (Prof. Dr.)
Tag der mündl. Prüfung: 2009-05-19
DDC-Klassifikation: 400 - Sprache, Linguistik
Schlagworte: Anapher <Syntax> , Regel , Linguistische Datenverarbeitung , Deutsch
Freie Schlagwörter: Selektionspräferenzen
Anaphora resolution , German , Machine-learning , Selectional preferences , Data-driven methods
Lizenz: http://tobias-lib.uni-tuebingen.de/doku/lic_mit_pod.php?la=de http://tobias-lib.uni-tuebingen.de/doku/lic_mit_pod.php?la=en
Gedruckte Kopie bestellen: Print-on-Demand
Zur Langanzeige

Inhaltszusammenfassung:

Thema dieser Dissertation ist die automatische Auflösung von Pronomen der dritten Person im Deutschen. Wie der Begriff "automatisch" impliziert, liegt der Schwerpunkt dabei in der Frage, wie diese Aufgabe praktisch auf einem Computer gelöst werden kann, und weniger in der Entwicklung einer normativen Theorie der Pronomenresolution. Im Zentrum dieser Dissertation steht der Vergleich zweier fundamental unterschiedlicher Systementwürfe für die automatische Auflösung. Der erste basiert auf einem linguistischen Regelwerk, wohingegen der zweite ein maschinelles Lernverfahren einsetzt. Wir evaluieren die Qualität beider Systeme und diskutieren ihre jeweiligen Eigenschaften, Vorteile und Nachteile. Von entscheidender Bedeutung für jedes automatische Verfahren für linguistische Verarbeitung ist die Art der linguistischen Information, auf die das System Zugriff hat. Für referentielle Relationen von Pronomen spielen semantische Gesichtspunkte eine große Rolle. Die Qualität automatischer Ansätze zur Pronomenresolution sollte daher von zusätzlicher semantischer Information profitieren. Wir untersuchen, wie zusätzliche Information über lexikalische Semantik in den Auflösungsprozess integriert werden kann und welchen Einfluss diese auf die Qualität des Auflösungsalgorithmus hat.

Abstract:

The topic of this dissertation is the automatic resolution of third person pronouns in German. As suggested by the term automatic, we will consider strategies of practically solving this problem on a computer, instead of developing a normative theory of pronoun resolution. Our main focus of interest will be on issues that directly relate to properties of automatic resolution processes. We will inspect the performance properties of two very different fundamental system designs. The first one is based on a set of linguistic rules, while the second has a machine learning approach at its core. We will further compare the results they produce and discuss their respective properties, advantages, and disadvantages. Central to any approach that solves a linguistic task is the kind of linguistic knowledge that it has access to. Referential relations of pronouns are strongly determined by semantics, therefore semantic information should be beneficial for the performance of an automatic approach to pronoun resolution. We will explore ways of how to incorporate knowledge drawn from lexical semantics and find out whether this knowledge can improve the performance of the resolver.

Das Dokument erscheint in: