bb-ai.net/KARaML

KARaML Tools

KARaML is an acronym standing for Knowledge Assimilation using Reasoning and Machine Learning and refers to a methodology for deriving information from natural language text using a combination of symbolic Knolwedge Representation techniques and Machine Learning based technologies (such as language models).

The KARaML Network is a group of researchers interested in exploring and applying this methodology, especially to problems of Natural Langauge Understading.

This KARaML Tools page contains links to some software interfaces that are intended to provide accessible, easy-to-use functionality that may be useful for research in this area.

A pip pakage, karaml , is now available, from which all the tool packages can be accessed. This is now the recommended method for using the KARaML Tools modules as it enables easy access and combination of multiple tools and does not require the use of long ID codes to download module files. (Some of links below are to notebooks using module-loading by file ID and will be udpdated to the new method in due course.)
KARaML Tools top-level entry point Notebook file on Colab
The current version just loads up the Kocijan BERT testing system, which enables easy testing of Winograd schemas using the same model and method described in the paper A Surprisingly Robust Trick for Winograd Schema Challenge .
K-Parser Knowledge Parsing system developed by Arpit Sharma.
Clingo, Answer Set Programming.
FCFG Generator.
A tool for generating sentences using a Context Free Grammar augmented with feature attributes and feature variables