Home | 1 | 2 | 3

Statistical NLP: Exercise 3

This is the third of a series of exercises on statistical natural language processing. In this exercise you examine a statistical model for spelling correction.

This exercise has been created by Erik Tjong Kim Sang, University of Antwerp, Campus Drie Eiken, room J0.07, phone 03-8202793, e-mail erikt@uia.ua.ac.be

Assignments

Use the available spelling corrector for making the following assignments:

1. Make a multi-word sentence with exactly one spelling error that is corrected in the right way by the spelling corrector. You may choose yourself whether you want to work with Dutch or English. What is the sentence that you have presented to the corrector? What output did it generate?
2. Make a multi-word sentence with exactly one word that the corrector does not know and for which it cannot generate an alternative. What is the sentence that you have presented to the corrector? What is the unknown word?
3. Make a multi-word sentence with exactly one spelling error that is corrected in the wrong way by the spelling corrector although the correct word is in its dictionary. What is the sentence that you have presented to the corrector? What should have been the response of the program? And what output did it generate instead?
4. Explain the error made by the spelling corrector in the previous assignment by computing the score it assigns to the correct word and its incorrect suggestion.
   Hint: for this assignment you need to estimate the priors and likelihoods which the search opportunities in the background information of the spelling corrector.
5. The spelling corrector does not perform perfectly. Give a non-technical suggestion for how its performance can be improved.