Benchmarking Compositionality with Formal Languages
This addresses the challenge of compositional generalization in neural models for NLP researchers, but it is incremental as it builds on existing formal language methods.
The paper tackled the problem of whether neural NLP models can learn compositional generalization by using deterministic finite-state transducers to generate datasets with controllable properties, finding that models either learn completely or not at all, with a key limit of 400 examples per transition.
Recombining known primitive concepts into larger novel combinations is a quintessentially human cognitive capability. Whether large neural models in NLP can acquire this ability while learning from data is an open question. In this paper, we investigate this problem from the perspective of formal languages. We use deterministic finite-state transducers to make an unbounded number of datasets with controllable properties governing compositionality. By randomly sampling over many transducers, we explore which of their properties contribute to learnability of a compositional relation by a neural network. We find that the models either learn the relations completely or not at all. The key is transition coverage, setting a soft learnability limit at 400 examples per transition.