ABSURDIST

Introduction

ABSURDIST (Aligning Between Systems Using Relations Derived Inside Systems for Translation) is a computational algorithm that uses within-system relations to find between-system translations. While illustrating the sufficiency of within-system relations to account for translating between systems, simulations of ABSURDIST also indicate synergistic interactions between intrinsic, within-system information and extrinsic information.

The first version of ABSURDIST exploits only one type of within-system relationship: the similarities between concepts. The semantic similarities between each concept pairs within a system are measured, and the two similarity matrices for the two concept systems are used as the input to generate the mappings between the concepts in the two systems.

Currently, we have extended and improved the original ABSURDIST algorithm. The current system, called ABSURDIST II, is an optimization algorithm using attributed graph matching to find translations between conceptual systems. It uses information about the internal structure of each
system by itself, or in combination with external information about similarities between concepts. It supports systems with multiple types of weighted or unweighted, directed or unidrected relations between concepts. The algorithm exploits graph sparsity to improve computational efficiency. Parameters for running the algorithm are optimized to achieve better mapping results, as well as improve noise tolerence, iteration speed, and scalability.

This project is partly funded by Lockheed Martin.

Download

ABSURDIST II is implemented in Java. The most up-to-date version of ABSURDIST II, ABS2.1, is available for download below. Besides the major source code implementing the algorithm, the zip file also includes packages for various testings which were used to produce the experimental results presented in our papers (see Publications). Instructions for how to use the code are provided in the readme file included in the download package.

Note: You will need Java 1.4 or above to compile and run the program.
Download ABS 2.1

Older versions of ABSURDIST are listed below:

ABS 2.0

Documentation

The online documentation for ABSURDIST II is available below:

ABS 2.1 Javadoc API

Demo

You can run the ABSURDIST II Applet online here: ABS 2.1 Applet. This version of the applet only allows you map randomly generated systems. To map systems from input files, you need to run the application.

Publications

Ying Feng, Robert Goldstone, Vladimir Menkov. A Graph Matching Algorithm and its Application to Conceptual System Translation. International Journal on Artificial Intelligence Tools, 14, 77-99, 2005. Download PDF file

Ying Feng, Robert Goldstone, Vladimir Menkov. ABSURDIST II: A Graph Matching Algorithm and its Application to Conceptual System Translation. Proceedings of FLAIRS, AAAI Press, Miami, FL, May 2004. Download PDF file

Goldstone, R. L., Feng, Y., & Rogosky, B. Connecting Concepts to each other and the world. In D. Pecher and R.Zuann (Eds.) Grounding cognition: The role of perception and action in memory, language, and thinking. Cambridge: Cambridge University Press, 292-314, 2005.

Goldstone, R. L. and Rogosky, B. Using relations within conceptual systems to Translate Across Conceptual Systems. Cognition, 2002. Vol. 84, pp 295-320.

ABSURDIST II Convergence Report

FLAIRS 2004 Presentation Slides

FLAIRS 2002 Presentation Slides