Turning Tables: Generating Examples from Semi-structured Tables for Endowing Language Models with Reasoning Skills

TL;DR

This paper introduces a method to enhance language models' reasoning abilities by generating synthetic question-answer pairs from semi-structured tables, focusing training on specific reasoning skills for improved performance.

Contribution

The authors propose a novel pre-training approach using automatically generated reasoning questions from tables, with targeted sampling strategies to improve reasoning skills in language models.

Findings

Model PReasM outperforms T5 on reasoning tasks.

Sampling based on model errors accelerates training.

Synthetic data improves reasoning skills in language models.

Abstract

Models pre-trained with a language modeling objective possess ample world knowledge and language skills, but are known to struggle in tasks that require reasoning. In this work, we propose to leverage semi-structured tables, and automatically generate at scale question-paragraph pairs, where answering the question requires reasoning over multiple facts in the paragraph. We add a pre-training step over this synthetic data, which includes examples that require 16 different reasoning skills such as number comparison, conjunction, and fact composition. To improve data efficiency, we propose sampling strategies that focus training on reasoning skills the model is currently lacking. We evaluate our approach on three reading comprehension datasets that are focused on reasoning, and show that our model, PReasM, substantially outperforms T5, a popular pre-trained encoder-decoder model. Moreover, sampling examples based on current model errors leads to faster training and higher overall performance.