Systematic Generalization: What Is Required and Can It Be Learned?

TL;DR

This paper investigates the ability of different grounded language understanding models to generalize systematically, revealing that modular models excel when properly structured, but end-to-end learning often struggles without explicit priors.

Contribution

It compares generic and modular models in systematic generalization, highlighting the importance of module layout and the challenges of end-to-end learning for this task.

Findings

Modular models exhibit more systematic generalization than generic models.

The layout of modules critically affects the generalization performance.

End-to-end learning often fails to produce appropriate module configurations for systematic generalization.

Abstract

Numerous models for grounded language understanding have been recently proposed, including (i) generic models that can be easily adapted to any given task and (ii) intuitively appealing modular models that require background knowledge to be instantiated. We compare both types of models in how much they lend themselves to a particular form of systematic generalization. Using a synthetic VQA test, we evaluate which models are capable of reasoning about all possible object pairs after training on only a small subset of them. Our findings show that the generalization of modular models is much more systematic and that it is highly sensitive to the module layout, i.e. to how exactly the modules are connected. We furthermore investigate if modular models that generalize well could be made more end-to-end by learning their layout and parametrization. We find that end-to-end methods from prior work often learn inappropriate layouts or parametrizations that do not facilitate systematic generalization. Our results suggest that, in addition to modularity, systematic generalization in language understanding may require explicit regularizers or priors.