The Manual's description of it's POS tagger?

[Joseph M]

The Festival manual describes its POS tagger:

“Festival includes a part of speech tagger following the HMM-type taggers as found in the Xerox tagger and others (e.g. DeRose88). Part of speech tags are assigned, based on the probability distribution of TAGS given a WORD, and from ngrams of tags”

But don’t most (all?) HMM taggers use the probability of the WORD given the TAG, for the ‘likelihood’ part of the equation?? (the opposite of what the manual says)?
J&Mm 2009, Chapter5.5

Is this a typo, or is it really using these probabilities? (I added caps for clarity)

[Simon]

Hmm – that’s a good point, and not something I’d spotted before. It is almost certainly a typo.

Anyway, for the purposes of understanding, it’s fine to assume that Festival performs POS tagging using precisely the method described in Jurafsky & Martin.

[Author]

Posts