Web-based and combined language models: a case study on noun compound identiï¬cation
Carlos Ramisch, Aline Villavicencio, Christian Boitet · International Conference on Computational Linguistics · 2010
This paper looks at the web as a corpus and at the effects of using web counts to model language, particularly when we consider them as a domain-specific versus a general-purpose resource. We first compare three vocabularies that were ranked according to frequencies drawn from general-purpose, specialised and web corpora. Then, we look at methods to combine heterogeneous corpora and evaluate the individual and combined counts in the automatic extraction of noun compounds from English general-purpose and specialised texts. Better n-gram counts can help improve the performance of empirical NLP systems that rely on n-gram language models.