Latent Compositional Representations Improve Systematic Generalization in Grounded Question Answering

TL;DR

This paper introduces a compositional, tree-structured model for grounded question answering that improves systematic generalization to out-of-distribution examples by inducing latent trees with minimal supervision.

Contribution

The authors propose a novel bottom-up, CKY-style parser that induces latent tree structures to enhance systematic generalization in grounded question answering.

Findings

Significantly outperforms baselines on an arithmetic expressions benchmark.

Achieves 96.1% accuracy on CLOSURE dataset, surpassing prior models.

Latent tree induction improves out-of-distribution generalization.

Abstract

Answering questions that involve multi-step reasoning requires decomposing them and using the answers of intermediate steps to reach the final answer. However, state-of-the-art models in grounded question answering often do not explicitly perform decomposition, leading to difficulties in generalization to out-of-distribution examples. In this work, we propose a model that computes a representation and denotation for all question spans in a bottom-up, compositional manner using a CKY-style parser. Our model induces latent trees, driven by end-to-end (the answer) supervision only. We show that this inductive bias towards tree structures dramatically improves systematic generalization to out-of-distribution examples, compared to strong baselines on an arithmetic expressions benchmark as well as on CLOSURE, a dataset that focuses on systematic generalization for grounded question answering. On this challenging dataset, our model reaches an accuracy of 96.1%, significantly higher than prior models that almost perfectly solve the task on a random, in-distribution split.