Mid-Term Project: Semantic Role Labeling

---------------
Important dates
----------------------------------------------------------
    Train and devlopment data availible:    April 14, 2014
    Test data availible:                    May 5, 2014
    System submission due:                  May 12, 2014
    Report submission due:                  May 19, 2014
    
-----------------
Tasks Description
    Goal: Given a set of sentences and the semantic roles on them, build a semantic
    role labeler using sequence model.
    
    To learn more about semantic role labeling:
      http://www.lsi.upc.edu/~srlconll/st04/st04.html 
      http://www.lsi.upc.edu/~srlconll/st04/papers/intro.pdf
    


========================================================================================
The files under this directory are for the Mid-Term Project of the course EMNLP.
All files are encoded in UTF8.

This directory contains
    Directory:      data
    Perl script:    srl-eval.pl

The directory 'data' contains:
    Plain text files:
    dev.wrd     trn.wrd
    dev.pos-chk trn.pos-chk
    dev.props   trn.props


Train and Development Data
    .wrd:       One word per line;
                Sentences are seperated by empty lines.
    .pos-chk:   Corresponding POS and chunk;
                Two columns, the first of which is POSs, the second chunks;
                Chunks are in the format '([chunk]*' '*' ... '*' '*[chunk])'.
    .props:     Corresponding semantic roles;
                The words (not '-') in the first column are the target word in
                the sentence. Next come several columns, each corresponding
                to one target, in order. The chunks in those columns are the
                semantic roles of the corresponding target. (See the example
                below)

------------------
Example for .props
------------------------------------------------------
-	(A0*	*	*	*	*
-	*	*	*	*	*
-	*	*	*	*	*
-	*	*	*	*	*
-	*	*	*	*	*
-	*A0)	*	*	*	*
-	(AM-TMP*AM-TMP)	*	*	*	*
签署	(V*V)	*	*	*	*
-	*	*	*	*	*
-	(A1*	*	*	*	*
-	*	*	*	*	*
-	*	*	*	*	*
-	*	*	*	*	*
-	*	*	*	*	*
-	*	*	*	*	*
-	*	*	*	*	*
-	*A1)	*	*	*	*
-	*	*	*	*	*
-	*	*	*	(A0*	*
-	*	*	*	*	*
-	*	*	*	*	*
互利	*	(V*V)	*	*	*
互补	*	*	(V*V)	*	*
-	*	*	*	*	*
-	*	*	*	*	*
-	*	*	*	*A0)	*
-	*	*	*	(AM-DIR*	*
-	*	*	*	*AM-DIR)	*
进入	*	*	*	(V*V)	*
-	*	*	*	*	*
-	*	*	*	(A1*	*
-	*	*	*	*	*
新	*	*	*	*	(V*V)
-	*	*	*	*	*
-	*	*	*	*A1)	(A0*A0)
-	*	*	*	*	*

    In the above sentence, there are 5 targets, namely, '签署', '互利', '互补',
    '进入', '新', so there are 6 columns. Column 1 gives the targets, while
    column 2 is the semantic roles for '签署', column 3 for '互利', and so
    forth. In column 2, there are 4 chunks, namely, (A0******A0),
    (AM-TMP*AM-TMP), (V*V), (A1********A1), which are the semantic roles of
    '签署'. The rest columns give the semantci roles similarly.

Test Data
    Only .wrd file is provided;

Evaluation
    Precision, recall and F messure are considered.
    You can run srl-eval.pl to evaluate your result on the development data.

Requrements:
    1.  Build your own system. (Anyone caught cheating will not receive credit)
    2.  Submit your system and report in time. (Your credits will be penalized
        by 30% if your submission is late)
    3.  If you want to use POSs and chunks for the test data, build your own POS
        tagger and chunker. We strongly recommend you to do so.
    4.  Your submission of system is supposed to include source codes,
        excecutables, results in the format descripted above, and a simple
        readme file.
    5.  Your report is supposed to be a .pdf file, in formal paper format, 
        demonstrating your idea, algorithm, implementation, experiments,
        and other ralated information.