Morpho-syntactic annotation and dependency parsing of German

DSpace Repositorium (Manakin basiert)


Dateien:

Zitierfähiger Link (URI): http://nbn-resolving.de/urn:nbn:de:bsz:21-opus-15239
http://hdl.handle.net/10900/46243
Dokumentart: Dissertation
Erscheinungsdatum: 2004
Sprache: Englisch
Fakultät: 5 Philosophische Fakultät
Fachbereich: Sonstige - Neuphilologie
Gutachter: Hinrichs Erhard
Tag der mündl. Prüfung: 2004-12-13
DDC-Klassifikation: 430 - Deutsch
Schlagworte: Morphosyntax , Syntaktische Analyse , Ambiguität
Freie Schlagwörter: Morphosyntaktische Annotation , Dependenzparsing , morphologische Ambiguität
probabilistic phrase-structure grammars , morpho-syntactic annotation , tagging , morphological ambiguity , dependency parsing
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:

Das Parsing natürlicher Sprache hängt von den syntaktischen Kategorien der Wörter ab: Die POS-Kategorie ist eine der am häufigsten verwendeten Informationsquelle für das Parsing. Beim Parsing stark flektierender Sprachen spielt morphologische Information, wie Kasus, Numerus und Genus, ein wichtige Rolle. Sie hilft dabei, syntaktische Ambiguität beim Shallow Parsing aufzulösen und stellt sich als besonders nützlich heraus, wenn sie auf Sprachen mit relativ freier Wortfolge angewandt wird, da sie die Argumentenstruktur eines Satzes teilweise mitbestimmt. Im Deutschen, einer stark flektierenden Sprache mit teilweise freier Wortfolge, ist das Problem der Zuordung morphosyntaktischer Kategorien, wie POS, Kasus, Numerus, Genus, Person, Tempus und Modus, schwierig, da die Tokens eine hohe Ambiguität besitzen. Zusätzlich verkompliziert wird das Problem durch einen teilweise paradigmaabhängigen Synkretismus im Kasus, der dieser Sprache eigen ist. Diese Arbeit beschäftigt sich mit der automatischen morphosyntaktischen Annotation im Deutschen. Verschiedene Ansätze, diese Aufgabe zu bewältigen, wurden erarbeitet und ein hybrides System mit einem regelbasierten und einem statistischen Modul wird vorgestellt, das die Stärken regelbasierter und statistischer Methoden vereint. Das regelbasierte Modul basiert auf dem Xerox Incremental Deep Parsing System und bildet ein neues constraint-basiertes System, das phraseninterne Kongruenzregeln und phrasenexterne syntaktische Heuristiken in eine einheitliche Architektur integriert. Das regelbasierte Modul reduziert die von der morphologischen Analyse gelieferten möglichen Analysen erfolgreich. Das statistische Modul basiert auf einer neuartigen Nutzung probabilistischer Phrasenstrukturgrammatiken zur morphosyntaktischen Annotation. Es löst die verbleibenden Fälle von Ambiguität und liefert präzise und vollständig desambiguierte Analysen. Der Nutzen morphosyntaktischer Information wird durch den Aufbau eines Dependenz-Parsers für das Deutsche empirisch evaluiert. Die Eingabe für den Parser ist auf die Tokens und deren morphosyntaktische Eigeschaften beschränkt. Der Paser erreicht eine State-Of-The-Art-Performanz.

Abstract:

The parsing of natural language relies on the syntactic characteristics of words. The part of speech category is one of the most common sources of information in parsing. In the parsing of highly inflectional languages, morphological information, such as case, number and gender, also plays an important role. It helps to resolve syntactic ambiguity in shallow parsing and is particularly useful in dependency parsing of languages with free word order, since it partly determines the argument structure of the sentence. For German, a highly inflectional language with partially free word order, the problem of assigning morpho-syntactic categories, such as part of speech, case, number, gender, person, tense} and mood, i.e. the problem of morpho-syntactic annotation, is complicated by the high ambiguity inherent in tokens. Moreover, the partially paradigm-dependent case syncretism of this language makes the problem particularly intricate. This thesis is concerned with the automatic morpho-syntactic annotation of German. Different approaches to the task are investigated in this thesis. A hybrid system with rule-based and statistical modules that combines the relative strengths of the rule-based and statistical methods involved is presented. The rule-based module is based on the Xerox Incremental Deep Parsing System and provides a novel constraint-based framework that integrates phrase-internal concord rules and phrase-external syntactic heuristics into one uniform architecture. The rule-based module successfully reduces the candidate analyses provided by a morphological analyzer. The statistical module is based on a novel use of probabilistic phrase-structure grammars for morpho-syntactic annotation. The module resolves the remaining cases of ambiguity, providing unambiguous and highly accurate output. The usefulness of morpho-syntactic information is evaluated empirically in the creation of a dependency parser for German. The input to the parser is limited to tokens and their morpho-syntactic characteristics. The parser reaches state-of-the-art performance.

Das Dokument erscheint in: