A survey of machine learning approaches to analysis of large corpora

Corpus-based Machine Learning of linguistic annotations has been a key topic for all areas of Natural Language Processing. This paper presents a survey, along three dimensions of classification. First we outline different linguistic level of analysis: Tokenisation, Part-of-Speech tagging, Parsing, Semantic analysis and Discourse annotation. Secondly, we introduce alternative approaches to Machine Learning applicable to linguistic annotation of corpora: N-gram and Markov models, Neural Networks, Transformation-Based Learning, Decision Tree learning, and Vector-based classification. Thirdly, weexamine a range of Machine Learning systems for the most challenging level of linguistic annotation, discourse analysis; these illustrate the various Machine Learning approaches. Our overall aim is to provide an ontology or framework for further development of our research.