Language Independent Probabilistic Context-Free Parsing Bolstered by Machine Learning Michael Schiehlen & Kristina Spranger Institut f¨ ur Maschinelle Sprachverarbeitung Universit¨ at Stuttgart Language Independent Probabilistic Context-Free Parsing Bolstered by Machine Learning – p.1/7
Outline General approach: convert dependency structure to constituency structure and use plain PCFG insert information on subcategorisation into the grammar (automatically from dependency relations) which names for phrasal categories? Language Independent Probabilistic Context-Free Parsing Bolstered by Machine Learning – p.2/7
� ✁ � ✁ � ✁ From Dep. to Const. Structure ROOT.P ROOT.P ROOT.P pc.P su.P obj1.P det su R su,vc vc KON R pc pc obj1 det mod obj1 Haar neus werd platgedrukt en leek op een jonge champignon vc mod su det cnj pc det cnj obj1 Language Independent Probabilistic Context-Free Parsing Bolstered by Machine Learning – p.3/7
� � � � � � � � � � � � Performance in CONLL AR CH CZ DA DU GE JA PO SL SP SW TU Language Independent Probabilistic Context-Free Parsing Bolstered by Machine Learning – p.4/7
Improvements after Submission Markovization of PCFG rules (minor improvements) language-dependent manual determination of phrasal categories for Chinese, Czech, German, Slovene, Spanish (major improvements) Language Independent Probabilistic Context-Free Parsing Bolstered by Machine Learning – p.5/7
� ✁ ✁ Tagging Approach Dependency Parsing as Tagging: use MaxEnt-tagger to assign head–relation pairs to individual tokens heads in ‘nth-tag’ representation, e.g. for the last token with POS tag for the second to the right Combination of PCFG-Parsing and Tagging: use parser output as an additional feature Language Independent Probabilistic Context-Free Parsing Bolstered by Machine Learning – p.6/7
� � � � � � � � � � � � � � � � � � � � � � � � Performance of Combination AR CH CZ DA DU GE JA PO SL SP SW TU Language Independent Probabilistic Context-Free Parsing Bolstered by Machine Learning – p.7/7
Recommend
More recommend
