uima-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Thilo Goetz <twgo...@gmx.de>
Subject Re: Parameters in uima descriptors
Date Thu, 04 Jun 2009 15:26:47 GMT
Steven Bethard wrote:
> On Thu, Jun 4, 2009 at 2:24 AM, Thilo Goetz <twgoetz@gmx.de> wrote:
>> Forwarded with permission by Terry...
>>
>> Terry Heinze wrote:
>>> We use UIMA fairly extensively as an underlying framework for supporting NLP
>>> solutions for various Thomson Reuters divisions. Regrettably, I don't follow
>>> the discussion threads as religiously as I should. Can you point me to the
>>> discussion/proposal on using Java descriptors vs. xml descriptors. Our
>>> current deployment strategy depends on the ability to read in and
>>> dynamically alter the descriptors prior to creating the corresponding UIMA
>>> components.
>> [...]
> 
> It's certainly possible to dynamically alter AnalysisEngineDescription
> objects - if it weren't, we couldn't create them in the first place.
> ;-)  Of course, the way you modify an AnalysisEngineDescription Java
> object, and the way you modify an XML descriptor object are different.
> 
>> There was no proposal, really.  The ClearTK folks said (if I understood
>> them correctly) that they don't keep any descriptors on disk, but generated
>> them dynamically at runtime.  To which I replied that this essentially
>> makes the components unusable to anybody else, since they're missing the
>> descriptors.  I'm still not sure though that there wasn't a misunderstanding
>> on my part somewhere...
> 
> At the moment, we include XML descriptors for our components as well
> as the AnalysisEngineDescriptor factory functions. But I'm still
> unconvinced that not having those XML descriptors makes our components
> "unusable". The UIMA framework APIs work on AnalysisEngineDescription
> objects, so components that provide such objects are still fully
> usable within the UIMA framework. But I think I probably just don't
> understand what you mean by "unusable".

The most common way to reuse someone elses component is to embed
said component in your own UIMA processing pipeline.  This is done
by referencing the component's XML descriptor in your own aggregate
descriptor.  No XML descriptor, no referencing, no reuse.  That's
what I mean by "unusable".  Does that make sense?

--Thilo

> 
> Steve

Mime
View raw message