Discriminative Reranker for a Probabilistic Parser

This repository contains an implementation of a parsing reranker trained with two learning algorithms: Primal SVM and Perceptron. The base parser produces a set of candidate parses for each sentence, with associated log probabilities that define the initial ranking of these parses. A second model attempts to improve on this initial ranking, using additional features extracted from the trees in the training set. (Assignment for Algorithms for NLP course @CMU (11-711))

For details on the results and implementation please see the writeup: Result_Writeup.pdf