Machine Learning for Natural Languages Processing

Probabilistic Context Free Grammars (PCFG) is a powerful formalism that has been used for several applications in Computational Linguistics. One important problem of these models is the probabilistic estimation of the probabilistic part of the models. This probabilistic estimation is based on tabular algorithms similar to the CKY algorithm. We review these estimation algorithms and their properties. The use of these models for Language Modeling and Machine Translation is also introduced. Finally, an interactive-predictive framework for parsing is explained, that can be used for developing both on-line learning techniques and active learning techniques.