GiellaLT provides an infrastructure for rule-based language technology aimed at minority and indigenous languages, and streamlines building anything from keyboards to speech technology.
When you have set up the prerequisites in the Getting Started documentation, you want to compile the linguistic analysers and use them.
You build the analysers in the ‘language folder’. Before you do so you must set up a variable $GTLANGS in your ~/.profile file, pointing to the directories where your language directories are stored. Open the file, e.g. as follows cd && open .profile. (some unix systems use .bashrc or other files, ask your local unix expert). In this file, add the string:
export GTLANGS="$HOME/path/to/directory"
where path/to/folder is the path to the directory where you have stored the language directory.
If you have done that, you may go to this language directory. We use Pite Saami as our example (lang-sje), replace sje with the language code of the language you want to compile:
cd $GTLANGS/lang-sje
First set up the files required to build the analysers (if you get error messages saying that some required files are missing you have probably skipped some steps on the Getting started pages):
./autogen.sh
For the first language you compile, you will be asked to go to giella-core and giella-shared to run some commands there. Do as the system tells you, and return to lang-sje.
Now, you must decied what analysers to build. Setup for the core ones you get with the command (for more options, see below)
./configure
You then build the analysers with the command
make -j
This command may take a couple of minutes to run. For our most developed languages (like North Saami) on a not too fast machine it may take half an hour or more.
When the process is done you should find a new-built analyser file: src/analyser-gt-desc.hfstol and several more like it in the src catalogue.
For more advanced build options, see the last section below.
The following pages give more information on how to use the analysers.
The Giella infrastructure can build scores of different linguistic analysers and genrators, taylored for different purposes and using different compilers. The ./configure command has a wide range of options for that. Different compilers are turned on and off by adding e.g. --with-xfst (compiles by using the xfst compiler instead of the default hfst). To turn off hfst and compile with xfst only, write e.g. --with-xfst --without-hfst.
Different analysers can then be built by adding the --enable option (--disable turns off default options). To take an example: In order to enable your system to turn your language model into a spellchecker, add the following to the ./configure option:
./configure --enable-spellers
A full list of the options is given by writing
./configure --help
The list of programs that are build (or not) is found in the last half of the help text.
Your current ./configure setting (which is valid until you change it) is shown by writing
head config.log
After you have (re) set your ./configure option, you must recompile, by writing make -j again.
It is possible to have multiple configurations in parallel. Follow these instructions if you want that.
For the full range of possibilities, there is an overview of the technical documentation for details on how to use our infrastructure to develop your morphologies, lexicons and more to create tools for yourself and your language community.
You may encounter troubles with your CLASSPATH. We are working on it, here is a fix if the compiler complains it is not set: In the langs directory, write:
export CLASSPATH=