[Corpora-List] SVMTool v1.2.1

Jesus Angel Gimenez Linares jgimenez at lsi.upc.es
Fri Oct 8 09:20:40 UTC 2004


Dear all,

  please check SVMTool v1.2.1 at http://www.lsi.upc.es/~nlp/SVMTool
<http://www.lsi.upc.es/%7Enlp/SVMTool>,
developed a the TALP Research Center in  Universitat Politècnica de
Catalunya
(http://www.lsi.upc.es/~nlp) and released under LGPL.

  In this version some new features have been incorporated, specially:

    * *Sentence-level tagging* improvement (faster decoding, beam serch...)
    * *C parameter tuning* algorithm (against a validation set or
      through /cross-validation/)
    * *Train/Test* options (test against a test set or through
      /cross-validation/)

  Also available now:

    * state-of-the-art pos-tagging models for Catalan, Spanish and
      English (trained on 3LB
      <http://www.dlsi.ua.es/projectes/3lb/index_en.html> and Penn
      Treebank <http://www.cis.upenn.edu/%7Etreebank/home.html>,
      respectively)
    * weight filtering (gain in both speed and accuracy)
    * unknown word features (greater flexibility)
    * use of a softmax function to transform scores into probabilities
      (applicable for sentence-level tagging and LRL tagging)
    * get all predictions (not only the winner)
    * lemmatization (given a lemmae dictionary in a convenient format)
    * very  high verbosity

  finally, a memory leak has been fixed.

  do not hesitate to contact me for any comment, suggestion, bug reporting.
of course, any feedback will be highly appreciated,

  best,

  jesus gimenez

--
Jesus Gimenez - PhD student at Universitat Politècnica de Catalunya
Address: C/Jordi Girona 1-3, edifici Omega, 305. Barcelona (08034).
phone: (+34) 93 4137950
fax:   (+34) 93 4017014             http://www.lsi.upc.es/~jgimenez

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/corpora/attachments/20041008/627eac35/attachment.htm>


More information about the Corpora mailing list