wordseg-0.6.2

• Improved documentation and algorithms description.

• Docker image now uses python-3.6 from anaconda,

• New tests to ensure replication of scores from CDSWordSeg <https://github.com/alecristia/CDSwordSeg>_ to wordseg for puddle,
  tp, dibs and dpseg.

• In wordseg-ag the <grammar> and <segment-category>
  parameters are now optional. When omitted a default colloc0 grammar
  is generated from the input text.

• In wordseg-dpseg

  • fixed forwarding of some arguments from Python to C++,
  • implementation of dpseg bugfix when single char on first line of
    a fold,
  • use the original random number generator to replicate exactly
    CDSWordSeg.
  • fixed default ngram to bigram (was already bigram but documented
    as unigram).

• In wordseg-dibs

  • fixed bug when loading train text at syllable level (new
    --unit* option)
  • safer use of train text (ensure there are word separators in
    it, ignore empty lines).

• In wordseg-eval

  • when called from bash, the scores are now displayed in a fixed
    order. New test to ensure bash and python calls to wordseg lead to
    identical results. See #31.
  • distinction between edge/no edge in boundary scoring. See #21.

• In wordseg-stats the scores are now displayed in a fixed order.

• In wordseg-syll

  • the --tolerant option allows to ignore utterances where the
    syllabification failed (the default is to exit the program on the
    first error). See #36.

wordseg-0.5

• Implementation of Adaptor Grammar as wordseg-ag,
• Installation now relies on cmake (was python setuptools),
• Improvements in tests and documentation,
• Various bugfixes.

wordseg-0.4.1

First public release, complete rewrite of the word segmentation pipeline in the CDSwordseg project.

• Four algorithms (tp, puddle, dpseg, dibs),
• Segmentation prepocessing and evaluation,
• Unit tests and documentation.