[Corpora-List] For summarization community: JRouge tool released

[Marina Litvak]

The following announcement may be of interest for summarization community.

Released: JRouge - Java ROUGE Implementation:
https://bitbucket.org/nocgod/jrouge/wiki/Home.
JRouge implements the N-gram based ROUGE metrics in the Java language and
resolves multiple limitations of the original ROUGE toolkit. Namely:

   - JRouge uses Unicode regular expressions to match characters and
   numerals of any Unicode text, making JRouge work on a variety of languages.
   - JRouge calculates scores more precisely than the original ROUGE since
   it does not truncate and round the results to the fifth digit after the
   decimal point. Instead, JRouge uses the whole extent of the Double
   primitive type of Java to provide precise scores.
   - The JRouge implementation of the ROUGE metric is written in the Java
   language. It runs evaluations of summaries on the fly without writing them
   to the HDD and creating the configuration file. Such implementation reduces
   the number of I/O accesses and the tool’s run time significantly.
   - Since JRouge does not use a configuration file, and gets input
   summaries by their location (paths), it requires special naming conventions
   for the input files. These conventions are described in the JRouge
   documentation.
   - For clean texts (no annotations for sentence splitting) JRouge uses
   the Stanford NLP toolkit to split the sentences. Currently this step is
   redundant for n-grams computation in JRouge (JRouge, like an original ROUGE
   toolkit, glues sentences together and produces cross-sentence n-grams).
   However, sentence splitting is required for creating SEE format files, in
   order to provide system/model and configuration (settings.xml) files for
   the original ROUGE. This functionality is required if one needs to compare
   the scores of JRouge and ROUGE toolkits on the same input (see more about
   the utility services in the documentation).
   - JRouge API has an interface adapter for creating delegation of method
   calls [6] from the JRouge library to the data structures used by the
   software. For further information, please consult the JRouge documentation.
   - JRouge provides consistent results in different OSes (contrary to the
   original ROUGE where different results were produced for Windows
and Linux-based
   OSes).


For the details please visit JRouge homepage:
https://bitbucket.org/nocgod/jrouge/wiki/Home
For any questions/comments please contact Dr. Marina Litvak (email:
litvak.marina at gmail.com)

-- 
Best regards,
Marina Litvak.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/corpora/attachments/20140730/d8747106/attachment.htm>
-------------- next part --------------
_______________________________________________
UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora