> Tony is of course quite right. There was a lot of activity on part of speech
> tagging in the late 80's and several taggers from that period (including my
own)
> are still being used quite a bit. A number of additional taggers have become
> available since that time. It is hard to say that the newer taggers are
better
> than what came before. Evaluation is quite tricky. None of these taggers
work
> as well as we would like, but the standard evaluation methods are hard pressed
to
> say that one tagger is much better than another. Some taggers claim to be
more
> accurate than others and some don't. The jury is still out on the accuracy
> question.
We performed a pretty solid experimental comparison between
an HMM-based tagger and the English Constraint Grammar Parser
of Helsinki (EngCG-2), see
http://www.ling.helsinki.fi/~avoutila/cg/index.html
on a common disambiguation task. The latter outperformed the
former with a wide margin. The experiments are written up in
detail in Samuelsson & Voutilainen, (E)ACL'97, pages 246-253.
This article can also be obtained from the above web site.
Christer Samuelsson
Atro Voutilainen