next up previous
Next: Punctuation, Content Words and Up: pos_phrase_nice Previous: Part-of-Speech Tagging

Part-of-Speech Sequence Models

This section describes experiments which investigate the effect of varying the parameters relating to equation 2, namely the size of the POS sequence window, L; the number of tags before the juncture, M and the size, K and composition, V of the tageset. In these experiments, for compactness we use two phrase break models representing n-grams of order 1 and 6. A 1-gram represents the simplest case, and as explained later the 6-gram is the best performing phrase-break model.





Alan W Black
1999-03-20