Title: Natural Language Processing

Required reading: (17 pages)
Chapter 1 {PDF} of Speech and Language Processing (by Daniel Jurafsky and James H. Martin), 2000. Prentice Hall, Upper Saddle River, NJ.

Optional additional reading: (32 pages)
Chapter 1 {PDF} of Foundations of Statistical Natural Language Processing (by Christopher D. Manning and Hinrich Schütze), 1999. MIT Press, Cambridge, MA. © 1999, Massachusetts Institute of Technology.

Optional free book on the web:
The book Survey of the State of the Art in Human Language Technology that reviewed the whole field of Natural Language Processing in 1996 (editors Ron Cole, Joseph Mariani, Hans Uszkoreit, Annie Zaenen, and Victor Zue) can be found at http://cslu.cse.ogi.edu/HLTsurvey/.

The Questions (Think about them, you are not required to hand in any essay for this lecture. Since this one does not count, our course will have only 11 HWs total. Keep in mind you need 12 to pass!)

1. Natural Language Processing is both a theoretical field (in that it relates to theories of language and thought as studied in Linguistics, Psychology, Neuroscience, and Philosophy) and a practical field (in that its applications include speech recognition, language translation, text summarization, and web search). Sometimes, the requirements of theory and practise diverge. Which do you think is more important? That is, which aspects should be focused on in teaching and in research? Why?

2. Choose one application of NLP that you think could have a major social impact and explain (a) why, (b) what is the reason the impact has not yet been achieved, (c) what are the minimum conditions required for the impact to be achieved and (d) what are the main impediments to reaching these conditions.

3. Design a system that uses the Web as its information source to answer questions of the following type: "what is the difference between capitalism and communism?" and "has the United Nations achieved anything?". Be as specific as you can in describing how the system would collect, select, and categorize relevant information. Extra points for an implemented system!

4. "You can know a word by the company it keeps". Related words cluster together: if you listen to the news or read a newspaper, one news item or article will have a high frequency of some words that never appear anywhere else, and the next item or article will feature other words. Zipf's Law states that when you list all words in order of frequency of average occurrence, then the rank of a word is inversely directly proportional to its frequency. That is, F = k.R for some constant k. This has immediate and obvious uses: you can for example create a system to categorize input texts into topics, simply by looking at unusual word frequencies in the texts. Can you think of other uses of Zipf's Law in NLP? Describe them, motivate their importance, and design a system to perform the task(s).