Induction of Treebank-Aligned Lexical Resources
Tejaswini Deoskar, Mats Rooth · 2008
We describe the induction of lexical resources from unannotated corpora that are aligned with treebank grammars, providing a systematic correspondence between features in the lexical resource and a treebank syntactic resource.We first describe a methodology based on parsing technology for augmenting a treebank database with linguistic features.A PCFG containing these features is created from the augmented treebank.We then use a procedure based on the inside-outside algorithm to learn lexical resources aligned with the treebank PCFG from large unannotated corpora.The method has been applied in creating a feature-annotated English treebank based on the Penn Treebank.The unsupervised estimation procedure gives a substantial error reduction (up to 31.6%) on the task of learning the subcategorization preference of novel verbs that are not present in the annotated training sample.t It 1 It 2 It 3 It4 PTB disagree VBP.n.-.-0.0014VBP.z.-.-2.01 VBP.z.-.-2.20 VBP.z.-.-2.22 VBP.z.-.-2.23 VBP.z.-.-1.0 VBP.t.-.-0.0012VBP.p.-.-0.98 VBP.p.-.-1.17 VBP.p.-.-1.20 VBP.p.-.-1.22 VBP.p.-.-1.0 VB.n.-.-0.0011VB.p.-.-0.64 VB.z.-.-0.60 VB.z.-.-0.61 VB.z.-.-0.61 VB.z.-.-1.0 VBP.aux.-.h 0.0009 VB.z.-.-0.56 VB.p.-.-0.54 VB.p.-.-0.53 VB.p.-.-0.53 VB.b.-.-1.0 VBP.b.-.-0.0008VBP.n.-.-0.27 VBP.n.-.-0.27 VBP.n.-.-0.25 VBP.n.-.-0.24 admit VB.n.-.-0.0040VB.n.