Old-style computational semantics at the time of Python 3.6
This repo has been developed together with the Medium post The meaning of "life" and other NLP stories:
it contains classes and data structures to parse formulas in first-order logic and check their truth in a user supplied model.
The code is commended and written striving for clarity, and mostly for pedagogical purposes (please see the relevant post for more details): the idea is to provide a simple-to-follow working project for i) people familiar with model theory that would like to see it at work in a modern programming language, ii) people familiar with Python, programming and possibly NLP that would like to see a fully spelled-out model in formal semantics.
The code has been written and tested in Python 3.6.
To run the program, make sure to install the prerequisite dependencies first (we used lark to quickly build an FOL parser). Then you can run:
python fol_main.py
specifying in the main which formula and which model you would like to use. Please see the grammar specs in fol_grammar.py to get
a sense of the symbols supported by the implementation.
The notebook folder contains a sketch for a pandas-based semantics (see the blog post for details).
Tests built with pytest. Use
pytest test_fol_semantics.py
to run some basic tests (or something like VIRTUAL_ENV/python -m pytest test_fol_semantics.py from a virtualenv).
Satisfaction in a model with partial assignments follows closely the exposition of Language, Proof, Logic (by David Barker-Plummer, Jon Barwise and John Etchemendy), which is also an heavily recommended book to learn first-order logic. A good book on computational semantics is Representation and Inference for Natural Language (by Johan Bos and Patrick Blackburn), which contains a detailed discussion on how to build a "model checker" in Prolog (interestingly, the authors lament that faithful implementations of "vanilla" first-order checkers are very hard to come by ;-)).
Please see the blog post for more references.
All the code in this repo is provided "AS IS" and it is freely available under the Apache License Version 2.0.