An implementation of the argument dependency model

Human sentence processing proceeds in a left-to-right incremental manner and is able to assign partial structural interpretations even if all the lexical items involved are not yet known. To mimic such a behaviour in a computational system is particularly difficult for head final languages where the rich information about the valency of the verb comes last. In such cases the Argument Dependency Model facilitates early hypotheses about the thematic status of a constituent by establishing direct argument-to-argument dependencies independent of the verb. An implementation of the model within the framework of Weighted Constraint Dependency Grammar is presented and evalated in a pseudo-incremental processing mode. Thanks to its non-monotonic nature the system can replicate the predictions of the Argument Dependency Model about the dynamic nature of revision processes during sentence comprehension with a very high degree of reliability. Menschliche Sprachverarbeitung erfolgt inkrementell von links nach rechts und ist in der Lage, partielle Strukturhypothesen aufzustellen, auch wenn die beteiligten lexikalischen Elemente erst teilweise bekannt sind. Die maschinelle Nachbildung eines solchen Verhaltens ist besonders problematisch fuer kopffinale Sprachen, in denen die reichhaltige Valenzinformation des Verbs erst sehr spaet verfuegbar ist. In solchen Faellen gestattet das Argument Dependency Model fruehzeitige Hypothesen ueber die thematischen Rollen einer Konstituente, indem es unabhaengig vom Verb, direkte Abhaengigkeitsbeziehungen zwischen seinen Argumenten aufbaut. Der Bericht stellt eine Implementation des Modells im Rahmen der Weighted Constraint Dependency Grammar vor und evaluiert diese in einem pseudo-inkrementellen Verarbeitungsmodus. Dank seiner nichtmonotonen Verarbeitungsweise kann das System die Vorhersagen des Argument Dependency Model ueber die dynamischen Revisionsprozesse beim Sprachverstehen mit einem sehr hohen Grad an Genauigkeit replizieren.

[1]  Joakim Nivre Incremental Non-Projective Dependency Parsing , 2007, HLT-NAACL.

[2]  Kilian A. Foth,et al.  Hybrid methods of natural language analysis , 2007, Ausgezeichnete Informatikdissertationen.

[3]  Angela D. Friederici,et al.  Who did what to whom? The neural basis of argument hierarchies during language comprehension , 2005, NeuroImage.

[4]  Beatrice Primus,et al.  Cases and thematic roles : ergative, accusative and active , 1999 .

[5]  Joakim Nivre,et al.  Labeled Pseudo-Projective Dependency Parsing with Support Vector Machines , 2006, CoNLL.

[6]  Matthias Schlesewsky,et al.  On incremental interpretation: Degrees of meaning accessed during sentence comprehension , 2004 .

[7]  Ina Bornkessel,et al.  The Argument Dependency Model: A neurocognitive approach to incremental interpretation , 2002 .

[8]  A. Friederici Towards a neural basis of auditory sentence processing , 2002, Trends in Cognitive Sciences.

[9]  Ingo Schröder,et al.  Natural language parsing with graded constraints , 2002 .

[10]  Randy J. LaPolla,et al.  Syntax: Structure, Meaning, and Function , 1999 .

[11]  Wolfgang Menzel,et al.  Hybrid Parsing: Using Probabilistic Models as Predictors for a Symbolic Parser , 2006, ACL.

[12]  Frank Keller,et al.  A Probabilistic Parser as a Model of Global Processing Difficulty , 2003 .

[13]  Yuji Matsumoto MaltParser: A language-independent system for data-driven dependency parsing , 2005 .

[14]  Horia F. Pop,et al.  An Experiment in Incremental Parsing Using Weighted Constraints , 2000 .

[15]  Angela D. Friederici,et al.  The Neurobiology of Language Comprehension , 1998 .

[16]  Wolfgang Menzel,et al.  A Transformation-based Parsing Technique With Anytime Properties , 2000, IWPT.

[17]  David R. Dowty Thematic proto-roles and argument selection , 1991 .

[18]  Joakim Nivre,et al.  MaltParser: A Language-Independent System for Data-Driven Dependency Parsing , 2007, Natural Language Engineering.