684.02 Assignment 5

Author:

Chris Brew

Date set:

Thursday February 8th 2002

Due Date:

5pm Friday February 15th,2002

Length limit: no more than 3 sides of A4

The assignment

We provide two papers about part-of-speech tagging. Both are available on-line.

The first uses a technique called Error-Driven Learning:

http:/www.ling.osu.edu/~cbrew/papers/brill-aaai94.ps.gz

and the second is a more conventional HMM approach:

http:/www.ling.osu.edu/~cbrew/papers/cutting92.ps.gz

You have been asked to prepare a report on these technologies for the CEO of a company working in language technology. They apparently need to build a part-of-speech tagger for some language. They won't reveal to you which language. Your task is to explain the differences between these technologies. You should say something about each of:

error rate.
amount and type of training needed.
each method's requirements for expertise (especially linguistic expertise, we assume that programmers, if needed are available).

You should express yourself in CEO-friendly terms. The tough length limit is there to both make sure that you stick to the essentials of the task, and to ensure that you fit into the CEO's limited attention span.

This document was translated from L^AT_EX by H^EV^EA.