Местами желтоват :) Местами в этом году вааще отсутствует :) Неудачное
сравнение :) Я бы сказал - чиста, как родниковая вода :)))))))))))))))
----- Original Message -----
From: "Michael Tanenblatt" <[email protected]>
To: <[email protected]>
Sent: Thursday, March 10, 2011 11:24 AM
Subject: Re: ConceptMapper and POS
I hope I get this right, as it's been a while since I've done this:
Assume you have a token annotation that has a feature containing a POS tag
as a string value, and that feature is called "posTag". If you want to use
that to limit your lookups, you would set the parameter
"TokenClassFeatureName" to "posTag" and then you would use the
"IncludedTokenClasses" parameter to list all of the values of that posTag
feature that would indicate the token should be included in the lookup
(conversely, you could use the "ExcludedTokenClasses" to indicate those to
ignore during lookup).
Specifying the POS tag in the dictionary is a way to override the tagger,
assuming your tagger doesn't overwrite existing tags, that's all.
ConceptMapper does not handle POS tags in any special way, they are just
features, like any other feature.
Does this make sense?
On Mar 9, 2011, at 5:39 PM, Chris wrote:
The documentation for the ConceptMapper annotator discusses the
possibility of
limiting matches to specific parts of speech. If I understand correctly,
by
specifying a POS value in a ConceptMapper dictionary entry, I can filter
matches to just those tokens that have been tagged with the specified POS
value (as supplied, in my case, by running the OpenNLP POS Tagger before
ConceptMapper). In practice, however, specifying POS attribute values for
my
dictionary entries does not appear to have the desired effect.
Am I misunderstanding what the POS attribute does in ConceptMapper?
Perhaps
it is meant only to use in the "write-back" capabilities of the annotator
to
populate/overwrite a POS attribute in a matched token instead of acting as
a
filter?
Any help is appreciated in better understanding what role POS plays in the
execution of ConceptMapper.
Thanks...
