When people communicate, they rely on a large body of shared common sense knowledge in order to understand each other. Many barriers we face today in artificial intelligence and user interface design are due to the fact that computers do not share this knowledge. To improve computers' understanding of the world that people live in and talk about, we need to provide them with usable knowledge about the basic relationships between things that nearly every person knows.
In 1999, we began a project at the MIT Media Lab to collect common sense from volunteers on the internet. Nearly ten years later our project has expanded to encompass many different areas, languages, and problems. Currently, our English dataset has over a million sentences from over 15,000 contributors. We have expanded far beyond the original Web site, but we are still collecting knowledge at http://openmind.media.mit.edu.
OMCS includes the following subprojects, among others:
- conceptnet: A semantic network of the knowledge we have collected.
- simplenlp: Lightweight natural language processing tools.
- divisi2: A library for learning from dimensionality reduction of a semantic network.
- csc-utils: Useful tools shared between our projects.
- openmind-commons: The code of our Web site for browsing and collecting knowledge.
The following projects are also in the OMCS namespace but are not currently managed as subprojects:
- divisi: The old version of Divisi, no longer supported.
- LexiconLinking: A project to learn a lexicon of verb classes and the nouns that they relate to.
This is the top-level project for Open Mind Common Sense. All of the
actual code is in submodules. To check out or update its contents, run
./update
(requires a recent version of git). To install everything,
decide if you want the equivalent of python setup.py develop
or
python setup.py install
, and run one of ./develop
or ./install
.
The git submodule system is a bit strange in that it checks out commits instead of branches. This means checkouts start in "disconnected head" mode, which makes it too easy to lose work. You may want to do something like the following:
git submodule foreach git checkout master
If not, at least make a branch for your own work first:
git checkout -b universal_semantics
Alternatively, you can install the submodules into your Python environment
using pip
. See requirements.txt
or devel_requirements.txt
.