LaSSI

LaSSI stands for LogicAl, Semantic, and Structure text Interpretation. This pipeline challenges the usual learning-based approaches by making the machine directly interpret the text through logical and verifiable steps in light of recent insights on Verified Artificial Intelligence. This project originated in 2020 for a web-crawling app that was later incorporated into a broader pipeline for textual data analysis.

Authors

• Giacomo Bergami (2020 -)
• Franco O. Saez Vander Linder (2025)
• Oliver Robert Fox (2023 - 2025)

Installing

First, please install the dependencies required for building some of the external packages:

pip install requests==2.31.0
pip install setuptools==66.1.1
sudo apt-get install libpq-dev

Afterwards, you can use the usual way to install packages:

pip install .

Also ensure Java is installed.

PostgreSQL

PostgreSQL must be installed, along with a database and user:

sudo apt install postgresql -y
sudo -u postgres psql

create database conceptnet;
create user lassi with encrypted password 'drowssap';
grant all privileges on database conceptnet to lassi;
\c conceptnet postgres
grant all on schema public to lassi;
exit

Known Issues

• If PyDatagramDB fails to install, make sure all other dependencies are installed before installing.
• Currently, LaSSI has been optimised for Ubuntu, we are aware of issues with macOS which are to be fixed in the future.

Running

The main pipeline can be run using main.py. The main datasets with annotations are provided in the orig folder. The benchmarking sentences are given in this other folder.