Description

This course will explore current statistical techniques for the automatic analysis of natural (human) language data. The dominant modeling paradigm is corpus-driven statistical learning, with a split focus between supervised and unsupervised methods.

In the first part of the course, we will examine the core tasks in natural language processing, including language modeling, word-sense disambiguation, morphological analysis, part-of-speech tagging, syntactic parsing, semantic interpretation, coreference resolution, and discourse analysis. In each case, we will discuss which linguistic features are relevant to the task, how to design efficient models which can accommodate those features, and how to learn with such models in data-sparse contexts. In the second part of the course, we will explore how these core techniques can be applied to user applications such as information extraction, question answering, speech recognition, machine translation, and interactive dialog systems.

Course assignments will highlight several core NLP tasks. For each task, you will construct a basic system, then improve it through a cycle of linguistic error analysis and model redesign. There will also be a final project, which will investigate a single topic or application in greater depth. This course assumes a good background in basic probability and a strong ability to program in Java. Prior experience with linguistics or natural languages is helpful, but not required.  Disclaimer: there will be a lot of statistics, algorithms, and coding in this class.

Readings

The primary texts for this course are:

• Jurafsky and Martin, Speech and Language Processing, 2nd edition ONLY [amazon]
• Manning and Shuetze, Foundations of Statistical Natural Language Processing [amazon] [online]

Note that M+S is free online.  Also, make sure you get the purple 2nd edition of J+M, not the white 1st edition.

Syllabus [subject to substantial change!]