Hypothesis
The
faculty member participating in this project has developed an Arabic POS tagger
using a combination of language rules and statistics (Khoja 2001). The Arabic
tagger uses
Hidden Markov Models (HMM’s) along with the Viterbi algorithm. This tagger
achieves
an accuracy of 94%.
Current research has shown that neural network taggers may achieve a better
accuracy
rate than HMM taggers (Schmid, 99). We want to test this hypothesis on the Arabic
language.
It has also been shown that tagging some text with multiple taggers improves
the tagging
results (Wu, 97). We will be testing this theory on Arabic by comparing the
results of
tagging the same text with both the statistical POS tagger and the neural network
tagger
that we will develop for this project.