MaltParser

One of the most commonly used techniques in natural language processing is dependency parsing, which is a specialized analysis process that can provide a way of uncovering the grammatical relationship between words in given sentences. When dealing with treebank data and wanting to integrate it with such processing, users can rely on specialized tools, such as MaltParser.

This Java-based application was designed specifically in order to provide the means to perform data-driven dependency parsing, when dealing with treebank data. By using it, one will be able to induce parsing models from the involved treebanks and parse the new data with induced models.

This data-driven parser-generator features a different approach. While traditional parser-generators rely on constructing parsers by using grammar, it relies on treebanks. Its implementation of the inductive dependency parsing will involve the syntactic analysis of sentences will yield derivations of dependency structures.

In the process, machine learning is used to guide the parser at nondeterministic choice points.

The main working principle of MaltParser relies on three components: deterministic parsing algorithms for creating labeled dependency graphs, history-based models for predicting the next parser actions when encountering nondeterministic choice points, and finally, discriminative learning for mapping histories to parser actions.