v0.34

• require C++17
• require latest ticcutils and libfolia
• some code refactoring
• improved GitHub CI
• removed dependency on libtar

v0.33

• --KANON was not always hounored
• after a change in ucto, it was needed to reset the tokenizer more often
• in the lemmatizer, fuzzy matching of CGN tags is implemented

v0.32

[Ko van der Sloot]

• ignore but warn on empty derivations, a lame fix for #103

v0.31

[Ko van der Sloot]

• use ticcutils > 0.34. NFC normalizations is standard now
• use Tokenizer::config_prefix() instead of magic string 'tokconfig-'
• code cleanup and quality improvement (cppcheck is very useful)

[Maarten van Gompel]

• added frog demo gif

v0.30

• finally fixed a major memory-leak in MBMA which bothered me for months
• also some minor leaks are plugged

v0.29

[Ko van der Sloot]

• added a fix for #100
  (where Frog created invalid FoLiA in a corner-case)
• improved api_test
• small code refactoring
• require libfolia >= 2.15, for correct working of word correction
• improved MWU code. Using Unicode strings and detecting MWU's with a starting
  Capital.

[Maarten van Gompel]

• .gitignore: added build dir

v0.28

Minor bugfix release:

[Ko van der Sloot]

• We no longer accept FoLiA paragraphs with both Words and Sentences.

[Maarten van Gompel]

• Software metadata fix

v0.27.2

[Maarten van Gompel]

• Software metadata fix only, no functional changes

v0.27.1

[Maarten van Gompel]

• Software metadata update only, no functional changes

v0.27

[Ko van der Sloot]
Major Release.
Internally we always perform a 'deep' morphological analysis.
This information is used for XML and JSON output.
For the 'classic' Tabbed output, we maintain backward comptability.
You need to specify '--deep-morph' to get the deep analysis in the output.
You may also specify '--compounds' to get an extra column with compound
information.

Other changes:

• C++ code quality
• adapted to more recent Timbl implementations (Unicode awareness)
• Tokenizer:
  • Better handling of --languages option.
  • 'und' is now also acceptable as a "language"
  • Better debugging possibility
• Mbma: To many alternatives with Inverted Verbs were generated. As the
  Tagger doesn't help us directly, we filter on the person of the next
  word, and only return V/te2I when the next word is 2-nd person