Shakti Standard Format (SSF) is a representation for storing linguistic analysis of natural languages. Its widely being used for storing treebank annotations of Indian Languages. However, in order to train dependency parsers on treebank annotations, the annotations should be in CONLL format. The SSF-to-CONLL convertor facilitates this conversion.

How to use?

bash ssf2conll.sh <input (file|directory)> <output file> <log file> <annotation type (intra|inter)>

Input Data Format: Intra-Chunk vs Inter-Chunk

Inter-Chunk dependecies should be formated as in the sentence below:

<Sentence id='1'>
1       ((      NP      <fs name='NP' drel='k1:VGF'>
1.1     mEM     PRP     <fs af='mEM,pn,any,sg,1,d,0,0' name='mEM' posn='10'>
1.2     wo      RP      <fs af='wo,avy,,,,,,' name='wo' posn='20'>
        ))
2       ((      NP      <fs name='NP2' drel='k1s:VGF'>
2.1     axanA   JJ      <fs af='axanA,adj,m,sg,,d,,' name='axanA' posn='30'>
2.2     sA      RP      <fs af='sA,avy,m,sg,,d,,' name='sA' posn='40'>
2.3     iMsAna  NN      <fs af='iMsAna,n,m,sg,3,d,0,0' name='iMsAna' posn='50'>
        ))
3       ((      VGF     <fs name='VGF' stype='declarative' voicetype='active'>
3.1     hUM     VM      <fs af='hE,v,any,sg,1,,hE,hE' name='hUM' posn='60'>
        ))
4       ((      BLK     <fs name='BLK' drel='rsym:VGF'>
4.1     .       SYM     <fs af='.,punc,,,,,,' name='.' posn='70'>
        ))
</Sentence>

While Intra-Chunk dependencies should be in the expanded SSF format:

<Sentence id='2'>
1       Kusa    JJ      <fs af='Kusa,adj,any,any,,,,' drel='pof:raha' posn='10' name='Kusa' chunkId='JJP' chunkType='head:JJP'>
2       raha    VM      <fs af='raha,v,any,sg,2,,0,0' stype='declarative' posn='20' voicetype='active' name='raha' chunkId='VGF' chunkType='head:VGF'>
3       XUlIcanxa       NNP     <fs af='XUlIcanxa,n,m,sg,3,d,0,0' drel='rad:raha' posn='30' name='XUlIcanxa' chunkId='NP' chunkType='head:NP'>
4       .       SYM     <fs af='.,punc,,,,,,' drel='rsym:raha' posn='40' name='.' chunkId='BLK' chunkType='head:BLK'>
</Sentence>

Output:

Output of Sentence 2 in CONLL would look like:

1       Kusa    Kusa    adj     JJ      cat-adj|gen-any|num-any|pers-|case-|vib-|tam-|chunkId-JJP|chunkType-head|stype-|voicetype-      2       pof     _       _
2       raha    raha    v       VM      cat-v|gen-any|num-sg|pers-2|case-|vib-0|tam-0|chunkId-VGF|chunkType-head|stype-declarative|voicetype-active     0       main    _       _
3       XUlIcanxa       XUlIcanxa       n       NNP     cat-n|gen-m|num-sg|pers-3|case-d|vib-0|tam-0|chunkId-NP|chunkType-head|stype-|voicetype-        2       rad     _       _
4       .       .       punc    SYM     cat-punc|gen-|num-|pers-|case-|vib-|tam-|chunkId-BLK|chunkType-head|stype-|voicetype-   2       rsym    _       _

Dependencies:

Following are the dependencies of the convertor:

1. headcomputation
2. vibhakticomputation

Install:

Run the command in main directory:

make install