Detecting compositionality using semantic vector space models based on syntactic context: shared task system description

This paper reports on the participation of the NLP Group at UNED in the DiSCo'2011 compositionality evaluation task. The aim of the task is to predict compositionality judgements assigned by human raters to candidate phrases, in English and German, from three common grammatical relations: adjective-noun, subject-verb and subject-object. Our participation is restricted to adjective-noun relations in English. We explore the use of syntactic-based contexts obtained from large corpora to build classifiers that model the compositionality of the semantics of such pairs.