Towards General Natural Language Understanding with Probabilistic Worldbuilding

TL;DR

The paper presents the Probabilistic Worldbuilding Model (PWM), a symbolic Bayesian approach to semantic parsing and reasoning aimed at achieving more general natural language understanding and AI capabilities.

Contribution

It introduces PWM, a fully-symbolic Bayesian model that encodes meanings and reasoning in human-readable formal language, enabling better generalization across domains and tasks.

Findings

PWM outperforms baselines on question-answering datasets

Demonstrates effective reasoning in out-of-domain tasks

Provides a proof-of-concept for domain-general NLU

Abstract

We introduce the Probabilistic Worldbuilding Model (PWM), a new fully-symbolic Bayesian model of semantic parsing and reasoning, as a first step in a research program toward more domain- and task-general NLU and AI. Humans create internal mental models of their observations which greatly aid in their ability to understand and reason about a large variety of problems. In PWM, the meanings of sentences, acquired facts about the world, and intermediate steps in reasoning are all expressed in a human-readable formal language, with the design goal of interpretability. PWM is Bayesian, designed specifically to be able to generalize to new domains and new tasks. We derive and implement an inference algorithm that reads sentences by parsing and abducing updates to its latent world model that capture the semantics of those sentences, and evaluate it on two out-of-domain question-answering datasets: (1) ProofWriter and (2) a new dataset we call FictionalGeoQA, designed to be more representative of real language but still simple enough to focus on evaluating reasoning ability, while being robust against heuristics. Our method outperforms baselines on both, thereby demonstrating its value as a proof-of-concept.