Facts as Experts: Adaptable and Interpretable Neural Memory over Symbolic Knowledge

TL;DR

This paper introduces a neural language model with an explicit symbolic interface that enhances interpretability, allows dynamic updates of factual knowledge, and improves performance on knowledge-intensive tasks.

Contribution

The authors develop a neural model integrating symbolic knowledge, enabling interpretability and dynamic updates without retraining, addressing limitations of traditional parameter-based knowledge storage.

Findings

Significantly improves question-answering performance on knowledge tasks.

Allows updating facts without retraining by manipulating symbolic representations.

Enhances interpretability of factual knowledge in neural models.

Abstract

Massive language models are the core of modern NLP modeling and have been shown to encode impressive amounts of commonsense and factual information. However, that knowledge exists only within the latent parameters of the model, inaccessible to inspection and interpretation, and even worse, factual information memorized from the training corpora is likely to become stale as the world changes. Knowledge stored as parameters will also inevitably exhibit all of the biases inherent in the source materials. To address these problems, we develop a neural language model that includes an explicit interface between symbolically interpretable factual information and subsymbolic neural knowledge. We show that this model dramatically improves performance on two knowledge-intensive question-answering tasks. More interestingly, the model can be updated without re-training by manipulating its symbolic representations. In particular this model allows us to add new facts and overwrite existing ones in ways that are not possible for earlier models.