[Corpora-List] "Cargo cult" NLP?

[Noah A Smith]

Even in engineering, it is helpful to distinguish between client-driven
problem solving and basic research.  Those of us who consider ourselves
researchers need to know how to appropriately use metrics in service of the
larger goal (e.g., knowing that achieving better scores isn't an end in
itself, but that better scores *can* be very compelling evidence), but we
also don't have "clients."  As a researcher, I would never tell a
practitioner not to use something that works well for his/her problem, but
practitioners should keep in mind that the usefulness horizon of basic
research is much, much longer, and indeed much research is (and should be)
done without worrying about eventual usefulness in practice.  Don't like my
research result?  Okay, don't use it.

--
Noah Smith
Associate Professor
School of Computer Science
Carnegie Mellon University


On Thu, Apr 10, 2014 at 2:00 AM, Phil Gooch <philgooch at gmail.com> wrote:

>
> On Wed, Apr 9, 2014 at 11:01 PM, John F Sowa <sowa at bestweb.net> wrote:
>
>> NLP is primarily a branch of engineering.  The challenge for
>> an engineer is not to get a better score on some "metric", but
>> to solve a client's problem within the constraints of budgets,
>> deadlines, and available resources.
>>
>
> I think this is very true. Some of the most successful NLP companies use
> quite simple techniques under the hood. Rule-based approaches (if well
> designed and modular) are popular with clients as they are easily
> understood and can be rapidly customised for a specific domain.
>
> Phil
>
> _______________________________________________
> UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
> Corpora mailing list
> Corpora at uib.no
> http://mailman.uib.no/listinfo/corpora
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/corpora/attachments/20140410/8f4d19dd/attachment.htm>
-------------- next part --------------
_______________________________________________
UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora