Analogs of Linguistic Structure in Deep Representations
This work addresses the problem of understanding emergent linguistic structures in AI for researchers in machine learning and cognitive science, though it is incremental as it builds on prior studies of neural representations.
The study investigated whether deep networks trained on a communication game develop compositional structures in their message vectors, finding that these vectors spontaneously form transformations analogous to negation, conjunction, and disjunction, similar to natural language syntax.
We investigate the compositional structure of message vectors computed by a deep network trained on a communication game. By comparing truth-conditional representations of encoder-produced message vectors to human-produced referring expressions, we are able to identify aligned (vector, utterance) pairs with the same meaning. We then search for structured relationships among these aligned pairs to discover simple vector space transformations corresponding to negation, conjunction, and disjunction. Our results suggest that neural representations are capable of spontaneously developing a "syntax" with functional analogues to qualitative properties of natural language.