UIMA Cas Editor OpenNLP Plugin

[Jörn Kottmann <ko...@gmail.com>]

Hi all,

as part of our annotations project I started to work on a plugin for the
UIMA Cas Editor (an eclipse based annotation tool), which I contributed 
years to the UIMA project.

The plugin currently only contains view for the name finder and can 
suggest names which
should be labeled. The view can be used by a user to quickly confirm names
in the shown document. It should also be possible to reject names, but 
that is not yet implemented.
Efficient keyboard support is also still missing.

For the described approach it is important that the name finder has a 
high recall, and suggest
at least almost every name in the text. After playing around a little, I 
now use the sequence validator
to force it to detect tokens which occurred in confirmed names always as 
names (start, cont).
That works great and boosts recall dramatically after the user confirmed 
the first few names,
but it still needs a little tweaking (e.g. filter stop words).

Anyway I hope I will have it soon in a state, that it can be tested by 
others also, it still has a couple
of rough edges.

Jörn

Re: UIMA Cas Editor OpenNLP Plugin

[Jörn Kottmann <ko...@gmail.com>]

On 8/29/11 8:39 PM, william.colen@gmail.com wrote:
> It is great, Jörn. I hope I can start contributing to the annotation project
> soon. I'll have some free time in a few weeks after I finish my master
> dissertation.

Sounds great, good luck with your thesis!

Jörn

Re: UIMA Cas Editor OpenNLP Plugin

["william.colen@gmail.com" <wi...@gmail.com>]

It is great, Jörn. I hope I can start contributing to the annotation project
soon. I'll have some free time in a few weeks after I finish my master
dissertation.


On Sat, Aug 27, 2011 at 10:59 AM, Jörn Kottmann <ko...@gmail.com> wrote:

> Hi all,
>
> as part of our annotations project I started to work on a plugin for the
> UIMA Cas Editor (an eclipse based annotation tool), which I contributed
> years to the UIMA project.
>
> The plugin currently only contains view for the name finder and can suggest
> names which
> should be labeled. The view can be used by a user to quickly confirm names
> in the shown document. It should also be possible to reject names, but that
> is not yet implemented.
> Efficient keyboard support is also still missing.
>
> For the described approach it is important that the name finder has a high
> recall, and suggest
> at least almost every name in the text. After playing around a little, I
> now use the sequence validator
> to force it to detect tokens which occurred in confirmed names always as
> names (start, cont).
> That works great and boosts recall dramatically after the user confirmed
> the first few names,
> but it still needs a little tweaking (e.g. filter stop words).
>
> Anyway I hope I will have it soon in a state, that it can be tested by
> others also, it still has a couple
> of rough edges.
>
> Jörn
>
>
>