Large neural language models, of the type discussed in this article, have existed for roughly 5 years. Since their advent, much work has attempted to understand the internal working of these models, rather than treatin… See more

A basic prerequisite of this definition of compositionality is the existence of discrete, modular representations of individual concepts, including representations of the roles that those concepts fill… See more

One way to show that neural networks contain more that just the basic prerequisites is to demonstrate a causalrelationship between the internal representation of the concepts (fillers or roles) and t… See more

A common criticism of neural networks as candidate models of the human mind is that neural networks lack the abstract symbolic structures and processing algorithms necessary to explain human language. Such claim… See more