added code for training a simple rst parsing model using the perceptron (which does not allow k-best decoding, and so should probably be replaced by SGD for a log-linear model). also refactored some things to allow the code to be run as an end-to-end discourse parser.
added code for training a simple rst parsing model using the perceptron (which does not allow k-best decoding, and so should probably be replaced by SGD for a log-linear model). also refactored some things to allow the code to be run as an end-to-end discourse parser.