Compositionality Decomposed: How do Neural Networks Generalise?

TLDR

Despite many empirical studies, consensus on neural networks’ ability to generalise compositionally remains elusive, partly because the definition of compositionality is unclear. The authors introduce a suite of tests that bridge linguistic theory of compositionality with practical neural language models. They formalise five task‑independent tests—recombination, extrapolation, locality, synonym robustness, and rule‑vs‑exception preference—apply them to a PCFG‑based dataset, and evaluate three sequence‑to‑sequence architectures (RNN, CNN, transformer). Analysis shows distinct strengths and weaknesses across the three models, highlighting specific areas where compositional generalisation can be improved.

Abstract

Despite a multitude of empirical studies, little consensus exists on whether neural networks are able to generalise compositionally, a controversy that, in part, stems from a lack of agreement about what it means for a neural model to be compositional. As a response to this controversy, we present a set of tests that provide a bridge between, on the one hand, the vast amount of linguistic and philosophical theory about compositionality of language and, on the other, the successful neural models of language. We collect different interpretations of compositionality and translate them into five theoretically grounded tests for models that are formulated on a task-independent level. In particular, we provide tests to investigate (i) if models systematically recombine known parts and rules (ii) if models can extend their predictions beyond the length they have seen in the training data (iii) if models’ composition operations are local or global (iv) if models’ predictions are robust to synonym substitutions and (v) if models favour rules or exceptions during training. To demonstrate the usefulness of this evaluation paradigm, we instantiate these five tests on a highly compositional data set which we dub PCFG SET and apply the resulting tests to three popular sequence-to-sequence models: a recurrent, a convolution-based and a transformer model. We provide an in-depth analysis of the results, which uncover the strengths and weaknesses of these three architectures and point to potential areas of improvement.