This repo provides an Atomese API for Link Grammar, allowing Link Grammar dictionaries and parse results to be accessed from the AtomSpace.

Link Grammar has its roots in Natural Language Processing, but appears to be more general, and seems capable of assigning structural relationships to items in arbitrary configurations. That is, not just words in sentences, but linear streams, and even non-planar relationship graphs. That is, it is a kind of "graph parser". Exactly where the theoretical limits are is unclear; its usage in a general setting continues to be explored, as of 2025.

Traditional LG worked with language description files, aka "dictionaries" providing the "lexis" from which parsing is done. The original natural language dictionaries were created by hand. Current versions of Link Grammar can pull dictionary entries out of the AtomSpace, thus allowing the dictionary to be maintained dynamically, using learning algorithms of any desired sort. In particular, learning and parsing can be interleaved, sentence by sentence, instead of considned to batch processing.

The code here has been used in production for many years. Version 2.0 removes several older formats that were provided in version 1.0. See the documentation for details:

• LgParseLink
• LgParseBonds
• LgParseDisjuncts
• LgParseSections
• LgDictEntry
• LgConnExpand

However, there are still some things that would be nice to have. This includes:

• Placing costs (aka entrpy) on the parse results.
• Reporting costs (aka entropy) in dictionary lookups.

The following packages must first be built and installed.

• Common OpenCog C++ utilities
• https://github.com/opencog/cogutil
• It uses exactly the same build procedure as this package. Be sure to sudo make install at the end.

• OpenCog Atomspace hypergraph database
• https://github.com/opencog/atomspace
• It uses exactly the same build procedure as this package. Be sure to sudo make install at the end.

• OpenCog Atomspace Storage subsystem
• https://github.com/opencog/atomspace-storage
• It uses exactly the same build procedure as this package. Be sure to sudo make install at the end.

• Natural Language Parser for English, Russian, other languages.
• https://opencog.github.io/link-grammar-website/
• Git repo: https://github.com/opencog/link-grammar
• This uses the autoconf toolset. When building from the tarball, say mkdir build; cd build; ..configure; make -j; sudo make install When building from the git repo, you must first say ./autogen.sh --no-configure. There are various prereqs to install, first, including autotools, autotools-archive, libpcre-dev, aspell, and libeditline-dev

Perform the following steps at the shell prompt:

    cd to project root dir
    mkdir build
    cd build
    cmake ..
    make
    sudo make install

Libraries will be built into subdirectories within build, mirroring the structure of the source directory root.

To build and run the unit tests, from the ./build directory enter (after building opencog as above):

    make check

See the examples directory for examples of how to use this Atomese API to Link Grammar. There are three basic functions provided:

• Parsing sentences to get links (bonds), disjuncts and sections.
• Looking up words in a dictionary to get disjunt choices.
• Analyzing connectors, expanding them into types and subtypes.