opennlp-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jörn Kottmann <kottm...@gmail.com>
Subject Re: interest in new parser?
Date Mon, 23 Jun 2014 12:08:01 GMT
+1, it would be possible to include different styles and
implementations of parsers in OpenNLP.

Jörn

On 06/23/2014 01:12 PM, Rodrigo Agerri wrote:
> Hello,
>
> Ratnapharki's (1999) is a shift-reduced parser. Others like Stanford
> NLP are now releasing shift-reduced parsers. There are differences
> between them, though. For example, Zhang and Clark (2009)'s parser
> (cited by Stanford's new parser) is similar except that they use a
> global discriminative model applying Collins (2002) perceptron,
> whereas
> Ratnaparkhi’s parser has separate probabilities of actions chained
> together in a conditional model (based on ME).
>
> Perhaps that route, among others, would be an interesting one to have
> a new parser in opennlp.
>
> Cheers,
>
> Rodrigo
>
>
>
> On Sun, Jun 22, 2014 at 9:44 PM, Richard Eckart de Castilho
> <richard.eckart@gmail.com> wrote:
>> Some time ago I asked the mstparser developers if they would consider
>> contributing the parser to OpenNLP. They said that mstparser isn't
>> up-to-date anymore since better parsers are now available, but in
>> principle didn't reject the idea.
>>
>> If OpenNLP was interested in adopting the mstparser, that might
>> be something to follow up on.
>>
>> The mstparser is only a dependency parser, not a constituency parser
>> as the one currently included with OpenNLP.
>>
>> Cheers,
>>
>> -- Richard
>>
>> On 20.06.2014, at 09:33, Jörn Kottmann <kottmann@gmail.com> wrote:
>>
>>> On 06/19/2014 06:00 PM, Miller, Timothy wrote:
>>>> There is a paper at this year's ACL conference on a statistical parser
>>>> with some interesting properties [1]. I tracked down the software [2]
>>>> and it is apache-licensed (unlike most other high quality parsers such
>>>> as the Berkeley and Stanford parsers). It is written in Scala so in
>>>> theory it should be compatible. Most importantly it is about as accurate
>>>> as those state of the art parsers on English (about 33% error reduction
>>>> from the Ratnaparkhi parser that opennlp currently uses), and may be
>>>> superior for cross-language performance.
>>>>
>>>> I am going to play with it with some of our clinical data to get a feel
>>>> for speed/accuracy on clinical text. Just curious if there is any
>>>> interest in a wrapper for this parser in opennlp?
>>> I don't think a wrapper is interesting for us. If people want to use this parser
it is probably
>>> better if they integrate it directly or use a component framework like UIMA or
GATE.
>>>
>>> Anyway, getting a new parser as a contribution would be interesting.
>>>
>>> Jörn


Mime
View raw message