GF 3.6 Release Notes

June 2014

Installation

See the download page.

What's new

Over 1300 changes have been pushed to the source repository since the release of GF 3.5 in August 2013.

Closed issues: 25, 30, 41, 57, 60, 61, 68, 80, 81, 82.

GF compiler and run-time library

New features and notable changes

• The default character encoding in grammar files has been changed from Latin-1 to UTF-8. Also, alternate character encodings should now be specified as --# -coding=enc, instead of flags coding=enc. See the separate document GF character encoding changes for more details.
• Record update: in record objects (but not types) of form r ** s, the values assigned in s now overwrite those in r. In previous versions, record extensions with overlapping assignments in r and s were not supported, and their behaviour was unpredictable.
• The PGF file format has changed in minor but backward incompatible ways. GF can still read the old format, thanks to a backward compatibility module. The C run-time system supports only the new format. Old PGF files can be converted to the new format by loading them in the GF shell and writing out a new PGF file with the pg -pgf command.
• The PGF run-time Haskell library includes some new modules, e.g. PGF.Lexing which contains lexing and unlexing functions, and PGF2, which is a binding to the C run-time system. The latter is included only if GF is configured with -fc-runtime, which is not the default, since the C run-time system still has to be compiled and installed separately before installing GF. The binary packages for GF 3.6 available from the download page will include C run-time support.
• nonExist now exists. It can be used in concrete syntax to mark non-existing morphological forms.
• Token gluing: BIND and SOFT_BIND can be used in concrete syntax to mark that the surrounding tokens should not be separated by space.

Bug fixes and minor changes

• Nonlinear patterns (i.e., patterns where the same variable appears more than once) in concrete syntax are now detected and reported as errors. (Section C.4.13 in the GF book states that patterns must be linear.)
• A pattern matching bug which prevented variables in patterns to be bound to table values has been fixed.
• A bug with local variable name shadowing has been fixed.
• Type checking fixes: correction of subtyping check for Ints n<=Ints m and contravariance in table types.
• The refresh pass had become obsolete and has been removed.
• The old partial evaluator (and thus the command line options --new-comp and --old-comp) has been removed.
• A missing rewrite rule has been added to the partial evaluator. It can significantly reduce the compilation time for certain grammars.
• Some additional datestamp checks have been added to gf -make to avoid recreating a PGF file when an uptodate PGF file already exists.
• The --output-dir option now applies also to PGF files.

Resource Grammar Library

• New languages in the Resource Grammar Library: Estonian (see the updated synopsis).
• Grammars and dictionaries for wide coverage translation (source).
• Bug fixes and some new inflection paradigms in many languages.

GF Cloud services

• There is a preliminary new Wide Coverage Translation Demo web app. The same wide coverage translation method has also been added to the Simple Translation Tool.
• PGF web service API:
  • the lookupmorpho request has been added to lookup the morphological analysis of a word.
  • Requests like parse and translate that parse text input now accept a lexer parameter to apply a lexer to the input text before parsing. The lexer can be text, code or mixed.
  • Requests like linearize and translate that produce linearizations now accept an unlexer parameter to apply an unlexer to the linearizations. The unlexer can be text, code or mixed.
  • Preliminary new requests to access the C run-time system have been added:
    • c-parse, c-linearize, c-translate, c-lookupmorpho, c-grammar. These work in the same way as the corresponding requests without the c- prefix. Since the c-parse and c-translate can produce very many (even infinitely many) results when used with large ambiguous grammars, using the limit parameter to restrict number of results, e.g. limit=10, is recommended.
    • c-wordforword: this works as c-translate but does a word-for-word lookup to create a (potentially very low quality) translation that can be used if all else fails.