token number error when 1st sentence starts with PONCT

[enavarro222]

Two tokens with number 1 :

$ echo '" Un exemple de phrase "' | java -Xmx1024M -jar talismane-fr-1.8.2b-allDeps.jar command=analyse 
1   "   "   PONCT   PONCT   _   0   _   _   _

1   Un  un  DET DET g=m|n=s 2   det _   _
2   exemple exemple NC  nc  g=m|n=s 0   _   _   _
3   de  de  P   P   _   2   dep _   _
4   phrase  phrase  NC  nc  g=f|n=s 3   prep    _   _
5   "   "   PONCT   PONCT   _   4   ponct   _   _

But here is a bunch of similar examples that work well :

$ echo " ' Un exemple de phrase ' " | java -Xmx1024M -jar talismane-fr-1.8.2b-allDeps.jar command=analyse 
1   '   '   PONCT   PONCT   _   0   _   _   _
2   Un  _   NPP _   _   0   _   _   _
3   exemple exemple NC  nc  g=m|n=s 2   mod _   _
4   de  de  P   P   _   3   dep _   _
5   phrase  phrase  NC  nc  g=f|n=s 4   prep    _   _
6   _'  _   ADJ _   _   5   mod _   _

$ echo ' " Un exemple de phrase " ' | java -Xmx1024M -jar talismane-fr-1.8.2b-allDeps.jar command=analyse 
1   "   "   PONCT   PONCT   _   0   _   _   _
2   Un  un  DET DET g=m|n=s 3   det _   _
3   exemple exemple NC  nc  g=m|n=s 0   _   _   _
4   de  de  P   P   _   3   dep _   _
5   phrase  phrase  NC  nc  g=f|n=s 4   prep    _   _
6   "   "   PONCT   PONCT   _   0   _   _   _

$ echo '" un exemple de phrase "' | java -Xmx1024M -jar talismane-fr-1.7.4b-allDeps.jar command=analyse 
1   "   "   PONCT   PONCT   _   0   _   _   _
2   un  un  DET DET g=m|n=s 3   det _   _
3   exemple exemple NC  nc  g=m|n=s 0   _   _   _
4   de  de  P   P   _   3   dep _   _
5   phrase  phrase  NC  nc  g=f|n=s 4   prep    _   _
6   "   "   PONCT   PONCT   _   0   _   _   _

[urieli]

This is an analysis error, rather than a programming error: the Talismane sentence detector has decided to place a sentence break after the first double quote, therefore breaking the input up into two sentences (hence the repeated index 1). It is presumably getting confused by the space between the quotation and the sentence start (but strangey, only when preceded by another space). I'll have to look into this when I have some time. By the way: if you're only sending one sentence at a time, it is simpler to skip the sentence detector stage entirely and start with the tokeniser.