CL-NLP -- a Lisp NLP toolkit
Eventually, CL-NLP will provide a comprehensive and extensible set of tools to solve natural language processing problems in Common Lisp.
The goals of the project include the following:
- support for constructing arbitrary NLP pipelines on top of it
- support for easy and fast experimentation and development of new models and approaches
- serve as a good framework for teaching NLP concepts
It comprises of a number of utility/horizontal and end-user/vertical modules that implement the basic functions and provide a way to add own extensions and models.
The utility layer includes:
- tools for transforming raw natural language text, as well as various corpora into a form suitable for further processing
- basic support for language modelling
- support for a number of linguistic concepts
- support for working with machine learning models and a number of training algorithms
The end-user layer will provide:
- POS taggers
- constituency parsers
- dependency parsers
- other stuff (will be added step-by-step, suggestions are welcome)
How to start working with CL-NLP
The project has already reached a stage of usefulness for the primary author: for instance, it supports my current language modelling experiments by providing easy access to treebanks and other utilities.
Yet, it is far from being production-ready. So, if you want to use it for production tasks, expect to bleed on the bleeding edge.
Otherwise, if you want to contribute to developing the toolkit, you're very welcome. Here are a few write-ups to give you the sense of the project and to help get started:
You'll also, probably, need to track the latest version of RUTILS from git.
For CL-NLP to reach v.0.1 that may be considered suitable for limited use by non-contributors, the following things should be finished (work-in-progress):
- implement a comprehensive test-suite and fix all bugs encountered in the process
- describe available models and their quality metrics
The license of CL-NLP is Apache 2.0.
Specific models may have different license due to the limitations of the dataset
they are built with. Please see a
<model>.license file accompanying each model for details.
(c) 2013-2014, Vsevolod Dyomkin firstname.lastname@example.org