The GLF kernel has been extended by an inference component, and is now called GLIF kernel. You may want to check out the GLIF kernel's repository, as its functionality subsumes the GLF kernel.
You can test the GLIF kernel online at at mybinder.org: https://mybinder.org/v2/gh/jfschaefer/GlifBinder/master. Note that it may take a minute to launch.
You can also try to run the kernel locally via docker. By default, you get the Jupyter Lab interface.
docker run -p 8888:8888 jfschaefer/glif:1.0
glf_kernel
is a kernel for Jupyter that combines functionalities of GF and MMT.
It is based on the GF Kernel, which only supports GF.
GLF is intended as a framework for natural language understanding (NLU) experiments. GF can be used to quickly write grammars for natural language fragments. MMT can be used to describe a logic and the translation from GF's parse trees into that logic, which is called semantics construction. This way, the GLF kernel can be used to quickly implement the entire pipeline from natural language strings to a logical expression.
If you want to read more about GLF, you may be interested in this paper.
You can also take a look at the tutorial notebook.
This package requires Python 3.8 (or newer) and this README assumes that this is the default python version.
If in doubt, use python3
instead of python
and python3 -m pip
instead of pip
.
Additionally, setuptools
is required to install this package.
This likely came with your Python distribution, but in case it did not use pip install setuptools
to install it.
If you haven't installed GF already, get it from the official download website.
You can find installation instructions for MMT here.
Please note that as of now (November 2019), GLF uses some not-yet released features of MMT.
So you will have to either talk to us (probably the best idea) or try building MMT yourself from the devel
branch of the git repository.
If you only want to use MMT for the glf_kernel
, you don't have to install an MMT development IDE (in a way, the notebooks will be your IDE).
For graph visualization GF uses Graphviz. Under Ubuntu etc. you can install it with
sudo apt install graphviz
or under Mac OS X with homebrew:
brew install graphviz
The glf_kernel
has been succesfully installed on Windows as well.
There are two ways you can go about it, and it is not clear which one is better/easier/more likely to work:
In this case you can basically follow the installation instructions for linux. Since Jupyter simply runs a server you can use your notebooks from a browser that is not part of the WSL. These remarks on Jupyter notebooks in WSL may be helpful.
In this case you have to add GF and Graphviz (in particular dot
) to the PATH variable.
You can either install the kernel from the source repository:
git clone 'insert right URL here'
cd glf_kernel
pip install .
or install it directly from pip:
pip install glf-kernel
Afterwards, install the kernel module with:
python -m glf_kernel.install
You're now ready to go and can start a Jupyter notebook with:
jupyter notebook
Alternatively you can also use this Kernel in Jupyter Lab
jupyter lab
If you intend on using the visulaization capabilities of the kernel please make sure you have widgets enabled in your Jupyter Lab.
conda install -c conda-forge nodejs
jupyter labextension install @jupyter-widgets/jupyterlab-manager
Select the glf_kernel
as kernel in your notebook.
The kernel supports all of the GF shell commands.
Output files produced by these commands will be placed into the current directory.
The kernel can also be used to define new grammars, which are immediately imported for usage upon defining. If you would like to have line numbers for editing your code you can use the Juypter shortcut Esc+L to enable them.
In addition to the GF shell commands the kernel supports the following commands:
show
: show the graph(s). Usage: [graph | graph generating command] | show. (e.g. parse "John loves Mary" | vt | show)help
: shows the help messageexport
: saves the specified grammar in the current diretoryclean
: removes all.dot
,.png
and.gfo
files from the current directory
It also supports MMT specific functionalities like:
archive
: creates the specified archive. If it already exists this archive will be set as the working archive. If no arguments are supplied this will display the current working archive. Also also allows creation of nested archives. (e.g.archive comma/jupyter
)subdir
: used to create or switch subfolders in the current working archives source folder. It is used just like thearchive
command.construct
: sends a construct request to MMT and displays the result.- You can use the kernel to define new theories or views just like you would do with grammars. For this the kernel also supports Tab completion of some Unicode characters. (e.g. \rightarrow + Tab will give you โ. \MD + Tab, \OD + Tab and \DD + Tab will give you module, object and declaration delimiters respectively).
- experimental: you can generate stubs for concrete syntaxes and semantics construction views from abstract syntax. Let's say you have defined an abstract syntax
MyGrammar
. Then you can generate a concrete syntax for e.g. English by enteringMyGrammarEng
and pressing Tab (stub-generation is autocompletion). Similarly, you can enterMyGrammarSemantics
to generate a stub for the semantics-construction view. This only works, if our Python parse for GF can handle the abstract syntax (and it's still more of a prototype, so it can't handle e.g. dependent types).