Am 02.05.2013 um 08:57 schrieb swirl <[email protected]>:

> Thanks Richard for your informative reply.
> You mentioned you able to use DKPro Core and ClearTK together.
> a. Do you use them as PEAR or do you simply download (manually  or via 
> Maven) the JARs and use the annotators like a library?

We add the JARs as Maven dependencies and build our processing pipelines
dynamically using uimaFIT.

> b. Your post imply that DKPro Core and ClearTK has some sort of configurable  
> type system or are type agnostic. So far I have seen that OpenNLP are able 
> to load whatever type in the application via the parameter configuration 
> settings like "opennlp.uima.SentenceType". Are DKPro Core and ClearTK simple 
> to OpenNLP annotators in this aspect?

Most of the pre-processing components of DKPro Core and ClearTK are not 
configurable
with respect to types. We use the pre-processing component from DKPro Core,
(tokenizers, taggers, parsers, etc.) and combine that with the machine learning 
part
of ClearTK (cleartk-ml), mainly with the feature extractor framework, which is 
largely
type agnostic, but requires implementing a thin UIMA component deciding which 
features
are extracted from the CAS. This thin component contains the knowledge about 
the DKPro
Core type system (or whatever other type system we use/custom types for an 
experiment).

Cheers,

-- Richard

Reply via email to