Enriching a Model's Notion of Belief using a Persistent Memory
arXiv:2104.08401
Abstract
Although pretrained language models (PTLMs) have been shown to contain significant amounts of world knowledge, they can still produce inconsistent answers to questions when probed, even after using specialized training techniques to reduce inconsistency. As a result, it can be hard to identify what the model actually "believes" about the world. Our goal is to reduce this problem, so systems are more globally consistent and accurate in their answers. Our approach is to add a memory component -- a BeliefBank -- that records a model's answers, and two mechanisms that use it to improve consistency among beliefs. First, a reasoning component -- a weighted SAT solver -- improves consistency by flipping answers that significantly clash with others. Second, a feedback component re-queries the model but using known beliefs as context. We show that, in a controlled experimental setting, these two mechanisms improve both accuracy and consistency. This is significant as it is a first step towards endowing models with an evolving memory, allowing them to construct a more coherent picture of the world.
arXiv commentThis is an old and now obsolete draft. See arXiv:2109.14723 ("BeliefBank: Adding Memory to a Pre-Trained Language Model for a Systematic Notion of Belief") for the final paper
References in corpus (10)
- Bidirectional Attention Flow for Machine Comprehension
- ConceptNet 5.5: An Open Multilingual Graph of General Knowledge
- End-To-End Memory Networks
- Tracking the World State with Recurrent Entity Networks
- Probabilistic Similarity Logic
- Language Models as Knowledge Bases?
- Commonsense Knowledge Mining from Pretrained Models
- Measuring Inconsistency in Probabilistic Knowledge Bases
- Measuring and Improving Consistency in Pretrained Language Models
- Be Consistent! Improving Procedural Text Comprehension using Label Consistency