Skip to content
This repository has been archived by the owner on Aug 9, 2023. It is now read-only.


Repository files navigation

Abduction-based Discourse Processing System

This is a repository of the ISI metaphor project team, in which we store all the resources and tools constituting our natural language understanding system based on abductive reasoning, implemented for four languages:

  • English
  • Spanish
  • Russian
  • Farsi

The system is largely based on ideas summarized in [Hobbs, 1993].

Our abductive Natural Language Understanding pipeline is shown below.


Text fragments are given as input to the pipeline. The text fragments are parsed. For Russian and Spanish tagging, we use TreeTagger. For Farsi tagging, we use the Stanford NLP tagger. For parsing, we use the dependency parser Malt for Spanish, Russian, and Farsi. For English, the whole processing is performed by the Boxer semantic parser.

The parses are input to the module converting them into logical forms. A logical form (LF) is a conjunction of propositions, which have generalized eventuality arguments that can be used for showing relationships among the propositions. We use logical representations of natural language texts as described in [Hobbs, 1995]. For Spanish, Russian, and Farsi, we have developed logical form converters. For English, we use the LF converter built in the Boxer semantic parser.

Logical forms and a knowledge base are input to the abductive reasoner based on Integer Linear Programming [Inoue et al., 2012]. The reasoner produces flat first order logic interpretations in the textual format and proof graphs in the PDF format.

More details about each component can be found here.

Installation and running

  1. Clone Metaphor-ADP repository:
git clone
  1. Build Docker image:
  1. Run Docker image:
  1. Run the system. See instructions

Note: If you prefer to run natively rather than using Docker, see instructions

System Requirements

  • Linux or macOS
  • at least 4 cores CPU
  • at least 8GB RAM

Related Publications


This work was supported by the Intelligence Advanced Research Projects Activity (IARPA) via Department of Defense US Army Research Lab contract W911NF-12-C-0025. The US Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright annotation thereon. Disclaimer: The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of IARPA, DoD/ARL, or the US Government.