uima-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Marshall Schor <...@schor.com>
Subject Re: create complex Type System
Date Thu, 02 Dec 2010 17:38:53 GMT

On 12/2/2010 11:17 AM, Andrea Turbati wrote:
>  Il 02/12/2010 16:33, Marshall Schor ha scritto:
>> On 11/30/2010 9:44 AM, Andrea Turbati wrote:
>>>   Hello,
>>> I'm new to the UIMA AAE and I was able to create a "simple" AAE by putting
>>> different AE together without any real interaction between them. I'd like to
>>> go further and compose an AAE in which the type returned is not just the set
>>> of simple types from the AEs, but it is a new Type in which the elements are
>>> connected and interact in some way.
>>> So for example if I have 3 simple AEs:
>>>     * The first recognises *first name*
>>>     * The second the* last name*
>>>     * The third  the *title *(Mr. Miss, Mrs, etc.)
>>> I'd like to create an AAE in which I have as an output just one type that is
>>> able to identify for example "Mr. Andrea Turbati"as one annotation and not as
>>> three different ones.
>>> I'd like to do the composition not manually but from a java application at
>>> runtime.
>> I'm not certain what your question is, but this sounds similar to one of the
>> example annotators: MeetingAnnotator - it is an annotator that takes the results
>> of 2 previous annotators, "TutorialDateTime" and RoomNumberAnnotator, and makes
>> a new Meeting annotation, whose span is from the beginning of the earliest
>> previous annotation to the end of the latest one.
>> You can see the source here:
>> http://svn.apache.org/viewvc/uima/uimaj/trunk/uimaj-examples/src/main/java/org/apache/uima/tutorial/ex4/MeetingAnnotator.java?revision=933273&view=markup
>> <http://svn.apache.org/viewvc/uima/uimaj/trunk/uimaj-examples/src/main/java/org/apache/uima/tutorial/ex4/MeetingAnnotator.java?revision=933273&view=markup>
> Thank you for the answer.
> I thought that in UIMA you could compose Type System using only an XML
> descriptor without having to write any new code. Because writing code is a
> "human task" and in my case I'm trying to create a tool that composes two or
> more AE and gives as output just an XML describing the AAE. Then I should be
> able to pass the single AEs and the descriptor of the AAE to another module
> which will annotate a text using the information from the AAE descriptor and
> the code of the AEs.
> So for example if I have an AE which is able to annotate last Name and another
> one that annotate Government title and in a text I find "President Obama", I'd
> like to have a single annotation for "President Obama" and not two separate
> one for each word.

You might be able to make such an "annotator".  Here's some thoughts:

I guess it would need to be configured with some parameters: specifying the two
types you wanted to "combine" ("LastName" and "GovernmentTitle"), and maybe some
information about a "window" - a window of size 1 would get "President Obama"
but not "President Barack Obama".  I think you need a window to disallow, for
instance, having "President" somewhere in the document, and Obama somewhere
completely unrelated to the first annotation.

You could probably make such a combining annotator, by generalizing the

Would this do what you want?


View raw message