UIT The arctic university of Norway > Giellatekno

Newinfra Catalogues

This is an overview of the file structure for each language catalogue found in our (new) infrastructure, i.e. over the directories giella-core, langs located under our main catalogue. Also the file README in $GTHOME describes some basic properties of the new infrastructure.

Catalogue structure

The starting point is the mother catalogue $GTHOME (called main if you follow the standard checkout procedure. It contains (a.o.) the catalogues giella-core, langs, the former is for language-independent files, and the latter is for language-specific files, one subfolder for each language.


The catalogue giella-core contains scripts used for all languages.

  • gtdshared
    • src
      • filters (regex files for removing tags or strings, for extracting strings or making them optional)
  • scripts (scripts used by the GT infrastructure)
  • schemas (for languages having their lexica in xml)
  • langs-templates
    • plxtools (speller tools)
    • und (= «undefined language») contans dummy files for new languages, files governing language independent fst compilation, and und.timestamp for logging updates.
    • smi (common source files for the Saami languages)
  • prooftesting-templates
  • schemas (for Uralic dictionaries writing lexc in xml)
  • scripts


Each language has its own folder, the naming convention is from ISO 639-3.

Each folder contains the following folders:

  • am-shared ( automake shared, common commands for fst compilation, never changed here, only in giella-core/langs-templates)
  • doc (folder where we write language-specific documentation)
  • m4 (support files for compilation)
  • misc (a grab bag dir for anything you don't want in svn - all files are ignored)
  • src (this is where the linguistic source files are)
    • filters (language-specific filters)
    • hyphenation (each phenomenon its folder)
    • morphology (see below)
    • orthography (capital letters, spellrelax)
    • phonetics (for different scripts for phonetic transcription)
    • phonology (morphophonology: the twolc files)
    • syntax (see below)
  • test (see below)
  • tools (tools we build, both proofing tools and other tools)

Here's a tree view of the und template dir (and thus the dir structure of all languages):

und$ tree -d
├── am-shared
├── doc
│   └── resources
│       └── images
├── m4
├── misc
├── src
│   ├── filters
│   ├── hyphenation
│   ├── morphology
│   │   ├── affixes
│   │   └── stems
│   ├── orthography
│   ├── phonetics
│   ├── phonology
│   ├── syntax
│   ├── tagsets
│   └── transcriptions
├── test
│   ├── data
│   ├── src
│   │   ├── morphology
│   │   ├── phonology
│   │   └── syntax
│   └── tools
│       └── spellcheckers
└── tools
    ├── grammarcheckers
    ├── shellscripts
    └── spellcheckers
        ├── fstbased
        │   ├── foma
        │   └── hfst
        └── listbased
            └── hunspell

Some directories are described in further detail below.

The morphology folder under langs/$LANG/src/

  • root.lexc (defines tags and basic parts of speech)
  • the folder affixes (here we find the lexc morphology files)
  • the folder stems (lexc stem files, the lists of words)

The folder might also contain some lexc files, like:


The morphology Makefile.am

The makefile defines two, perhaps four variables (the two first must be defined)

  1. GT_LEXC_ROOT=root.lexc (defining the root of lexc)
  2. GT_LEXC_SRC= (listing all the other source files, except generated ones)
  3. GENERATED_LEXC_SRCS= (listing lexc source files that originate as something else, e.g. xml)
  4. GT_XML_SRC= (defining possible xml source files)

We define all the source files we need to build the transducers. The build system will take care of putting them together and compiling them.

include $(top_srcdir)/am-shared/lesc-include.am

This statement includes the majority of the build instructions. You should never need to touch the included file.

The orthography folder

Here we take care of initial capitalisation, and of spellrelax. Note that spellrelax now will be language-specific.

The phonetics folder under langs/$LANG/src/

This folder contains files for conversion to IPA.

The syntax folder under langs/$LANG/src/

This folder contains disambiguation.cg3.

The files functions.cg3, dependency.cg3 for for sma/sme/smj are in giella-shared/smi/src/syntax/. Faroese also uses the common dependency.cg3, but has its own functions.cg3.

The test catalogue under langs/$LANG

Within it there are several subdirs for different kind of tests. Each test is wrapped in a shell script that emits one of the following values, depending on the outcome of the test:

  • 0 = success
  • 77 = skipped test
  • 99 = hard fail (e.g. out-of-memory crashes etc.)
  • any other value is a regular fail

If you need a new test, just write a shell script that follows this convention, and add it to the TESTS variable in the Makefile.am file. That's it.

Also take care that the shell script uses AM variables for all references to files and commands outside the script which can not be assumed to be universally available. This will make the test scripts portable.

So far we have morphology tests only, but there is a setup for syntax and phonology as well.


What kind of and how many transducers we produce varies from language to language. The binary files are stored in $LANG/src, with the following name conventions.

For some languages (at least sme and sma) the content of and difference between the transducers is explained on the documentation page of each language