Facts as Experts: Adaptable and Interpretable Neural Memory over Symbolic Knowledge
arXiv:2007.00849
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.
References in corpus (7)
- Language Models are Few-Shot Learners
- Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
- REALM: Retrieval-Augmented Language Model Pre-Training
- Neural-Symbolic Learning and Reasoning: A Survey and Interpretation
- Query2box: Reasoning over Knowledge Graphs in Vector Space using Box Embeddings
- Differentiable Reasoning over a Virtual Knowledge Base
- TensorLog: Deep Learning Meets Probabilistic DBs
Cited by in corpus (7)
- Modifying Memories in Transformer Models
- Knowledge Enhanced Pretrained Language Models: A Compreshensive Survey
- Neural Language Generation: Formulation, Methods, and Evaluation
- Iterative Hierarchical Attention for Answering Complex Questions over Long Documents
- Cross-Modal Retrieval Augmentation for Multi-Modal Classification
- Relation-Guided Pre-Training for Open-Domain Question Answering
- Highly Parallel Autoregressive Entity Linking with Discriminative Correction