Dyna-bAbI: unlocking bAbI's potential with dynamic synthetic benchmarking

TL;DR

This paper introduces Dyna-bAbI, a flexible synthetic benchmarking framework for story understanding, revealing limitations of current models in compositional generalization and emphasizing the need for controllable task generation.

Contribution

We developed Dyna-bAbI, enabling systematic control over bAbI tasks, and demonstrated its effectiveness by creating new compositional tasks that expose current models' limitations.

Findings

State-of-the-art models fail in compositional generalization tasks

Data diversification improves robustness but is insufficient alone

Controllable synthetic benchmarks are crucial for advancing NLU

Abstract

While neural language models often perform surprisingly well on natural language understanding (NLU) tasks, their strengths and limitations remain poorly understood. Controlled synthetic tasks are thus an increasingly important resource for diagnosing model behavior. In this work we focus on story understanding, a core competency for NLU systems. However, the main synthetic resource for story understanding, the bAbI benchmark, lacks such a systematic mechanism for controllable task generation. We develop Dyna-bAbI, a dynamic framework providing fine-grained control over task generation in bAbI. We demonstrate our ideas by constructing three new tasks requiring compositional generalization, an important evaluation setting absent from the original benchmark. We tested both special-purpose models developed for bAbI as well as state-of-the-art pre-trained methods, and found that while both approaches solve the original tasks (>99% accuracy), neither approach succeeded in the compositional generalization setting, indicating the limitations of the original training data. We explored ways to augment the original data, and found that though diversifying training data was far more useful than simply increasing dataset size, it was still insufficient for driving robust compositional generalization (with <70% accuracy for complex compositions). Our results underscore the importance of highly controllable task generators for creating robust NLU systems through a virtuous cycle of model and data development.