[Corpora-List] Named Entity Recognition Software

[Diana Maynard]

It would be helpful to know which system you've tried.....

Do be aware that most NE systems are going to need some kind of training 
or at least tweaking on different kinds of texts - at least if you want 
to get decent results. Literary texts tend to be a little different from 
news texts for the purposes of NER. So it's unrealistic to expect the 
advertised SOA 90%-ish accuracy with any system off the shelf on your texts.

You could try GATE (http://gate.ac.uk) which is freely available and 
comes with a large number of resources which you can mix and match or 
modify as you want, if you don't like the default NE system (ANNIE) that 
it comes with.

Diana





On 23/06/10 16:29, David L. Hoover wrote:
> Hello all,
>
> Apologies if this is a much answered question, but I am working on a 
> project that studies the use of names in literary texts, primarily 
> place names and person names, but also other names. The only named 
> entity recognition software I have tried is hopelessly inaccurate, 
> with a failure rate of more than 30%. At this rate, it is quicker and 
> more accurate to tag the text manually.
>
> What NER software do you all know of that is considered to be the most 
> accurate? And roughly what accuracy level could be expected on 
> literary texts (say, 20th century British and American novels)?
>
> Thanks,
> David
>

_______________________________________________
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora