GRMM is a powerful tool to construct and perform inference on graphical models. This tutorial is just the extension version of tutorial given by Charles Sutton in http://mallet.cs.umass.edu/grmm/general_crfs.php for ACRF. ACRF is an interface given by GRMM to train CRF with arbitrary graphical structure (Skip-Chain CRF, Factorial CRF, Hierarchical CRF, or other Dynamic CRF models). Unfortunately, GRMM (ACRF as well) is lack of documentation and available tutorial on how to use it directly.
In this tutorial, we present simple experiments using Dynamic CRF (i.e., Factorial CRF) for Noun-Phrase Chunking and POS Tagging at the same time. Of course, we will use ACRF in this experiment. You can refer to a paper authored by Charles Sutton (Dynamic Conditional Random Fields: Factorized Probabilistic Models for Labeling and Segmenting Sequence Data, ICML 2004) before following this tutorial.
