Hi
(First of all a BIG THANKS to ALL open source developers at UIMA and the
other projects I mention below whom I am now relying on :-) ).
I am looking at researching a particular knowledge base extraction task
using UIMA components as part of the solution. To do this work I need
UIMA output persistence and to be able to visualise this output as
multiple annotation layers on the same text. Ultimately I want my
automated annotations and visualisations to be web based and allow me to
make additional manual annotations if required. Once I have my multiple
annotations made on a text I will then be able to apply my new knowledge
extraction logic.
I have looked at webanno (which incorporates Brat for its UI) and
U-Compare as well as Argo (See https://code.google.com/p/webanno/,
http://brat.nlplab.org/, http://u-compare.org/,
http://nactem.ac.u/ucompare/downloads/,
http://argo.nactem.ac.uk/about-argo/). I had hoped that I could use
webanno for this task however webanno does not allow the direct import
of UIMA components or UIMA output. I found that I could get U-Compare
to work as I wanted and it shows promise however if I get my any
configuration wrong between any UIMA components it crashes out. I got
the software to work for me after I spent more time reading the manual.
I found I needed to manually configure the input types for each
component in the pipeline. The software recognises subsequent pipeline
component compatibility when a new component is added to a work flow.
My initial errors came as I had initially expected subsequent U-Compare
components to automatically pick up their input from the output from
previous workflow components. Whilst the U-compare software does
support the saving of previous session data the software is not fully
open source so I do not have easy access to this data. I have not
looked at the webservice pipeline generation fetaures of U-Compare as
yet; this might hold promise if it gives me a download configuration
rather than a hosted solution. When I looked at the argo tool I had
similar problems with a lack of output. I would assume for the same
reasons. Again Argo is not fully open source so I cannot work on
modifying this tool to my own ends. Are there any other better tools
available that support web based UIMA layered visualisation and output
persistence?
Currently I plan to continue to experiment with UIMA components using
U-compare however I am looking to implement persistence and
visualisation in a production tool. If someone already has a good open
source implementation of this need I would prefer not to spend time
reinventing this particular wheel.
I would be very happy if the U-compare and webanno teams would work
together and get their software integrated. I will pass this mail onto
these teams as a suggestion.
The particular data extraction task I am interested in is different to
the current popular research shared task
(http://www.nist.gov/tac/2014/KBP/) and one which I plan to share once
have made some progress.
Thanks in advance.
Further information about me and my project can be found at www.hemseye.org
James Kitching
