On 3/15/11 9:39 PM, Grant Ingersoll wrote:
This change isn't about performance, it's about thread safety. Like I said,
instead of talking on it, I'll put up a patch as soon as I get some spare time.
Just got a little confused, because you started speaking about the
performance, sorry.
Anyway, the approach you are proposing could be taken for the tokenizer,
sentence detector,
pos tagger, chunker. I believe it should also be possible for the
parser. For coref I would have
to take a look at the code. The name finder is state-full by design and
there it would need
more redesign.
The biggest issue I see for POS Tagger, Chunker and Coref is that we
planned to use common
feature generation code which is not thread safe. At least not the way
it is implemented right now.
There we have to discuss how that could be handled.
Jörn
