Unsupervised Learning of Verb Argument Structures

We present a statistical generative model for unsupervised learning of verb argument structures. The model was used to automatically induce the argument structures for the 1,500 most frequent verbs of English. In an evaluation carried out for a representative sample of verbs, more than 90% of the induced argument structures were judged correct by human subjects. The induced structures also overlap significantly with those in PropBank, exhibiting some correct patterns of usage that are not present in this manually developed semantic resource.

• Type: Conference
• Year: 2006
• Where: CICLING
• Authors: Thiago Alexandre Salgueiro Pardo, Daniel Marcu, Maria das Graças Volpe Nunes
• Keywords: Argument Structures, CICLING 2006, Natural Language Processing, Statistical Generative Model, Verb Argument Structures