A Natural Proof System for Natural Language

Venue:
ESSLLI 2019 in Riga, Latvia

Abstract

Natural language inference requires high reasoning capacity in order to recognize semantic
relations between natural language expressions.
The course is built around this highly interdisciplinary problem.
It introduces a computational theory of reasoning, called Natural Tableau,
that combines the idea of a Natural Logic, (logic using natural language as its vehicle of reasoning),
with the semantic tableau method, a proof calculus that can be interpreted as search for a certain situation.
The course aims at not only introducing the theory of Natural Tableau
but also at showing how it can be put to use in a downstream application such as Recognizing Textual Entailment.
During the course attendees will have the opportunity to run an implemented theorem prover
on natural language sentences and examine the resulting reasoning procedures based on natural language phrases.

Schedule

Day 1 - Natural Language Inference

Day 2 - Semantic Tableau Method

Day 3 - Natural Tableau System

Day 4 - Wide-Coverage Theorem Prover for Natural Language

Day 5 - Natural Language Inference with Natural Theorem Prover

The SICK and FraCaS data; data-driven adaptation and development of LangPro; Evaluation and analyses; demo of adding a new tableau rule to the prover.

References

- Abzianidze, L. (2017):
*LangPro: Natural Language Theorem Prover*. EMNLP. - Abzianidze, L. (2016):
*Natural Solution to FraCaS Entailment Problems*. *SEM. - Abzianidze, L. (2015):
*A Tableau Prover for Natural Logic and Language*. EMNLP. - Muskens, R. (2010):
*An Analytic Tableau System for Natural Logic*. LNCS, Springer. - LangPro: GitHub repository; online demo