bootphon/wordseg: wordseg-0.7

Added tools/wordseg-qsub.sh, a script to schedule a list of segmentation jobs to a cluster running Sun Grid Engine and the qsub scheduler.

Added example phonological rules and updated contributong guide in documentation.

In wordseg-prep ignore empty lines in both gold and segmented texts.

In wordseg-syll the syllabification is improved: syllabification of words with no vowel, better error messages (see #35, #36).

In wordseg-tp add of the mutual information dependancy measure. In the bash command, the argument --probability {forward,backward} is replaced by --dependency {ftp,btp,mi} (maintained for backward compatibility). See #40.

In wordseg-ag:

• niteration is now 2000 by default (was 100),
• improved log of iterations with -vv,

• refactored postprocessing code:

  • parallelized
  • constant memory usage (was linear wrt niterations*nutts)
  • tree to words conversion in C++ instead of Python
  • temporary parses file is now gziped (gains a factor of 20 in disk usage)
  • new --temdir option to specify another path for tempfile (default is /tmp)
  • detection of incomplete parses (if any issues a warning)
  • better comments in code, more unit tests