LOGEN: Few-shot Logical Knowledge-Conditioned Text Generation with Self-training

TL;DR

This paper introduces LOGEN, a framework for logical knowledge-conditioned text generation in few-shot settings, utilizing self-training with pseudo logical forms to improve performance with limited data.

Contribution

It presents a unified approach that effectively leverages self-training and pseudo logical forms to enhance few-shot logical knowledge-conditioned text generation.

Findings

Outperforms baselines in few-shot scenarios

Effective with as few as 20 seed logical forms

Improves content and structure consistency

Abstract

Natural language generation from structured data mainly focuses on surface-level descriptions, suffering from uncontrollable content selection and low fidelity. Previous works leverage logical forms to facilitate logical knowledge-conditioned text generation. Though achieving remarkable progress, they are data-hungry, which makes the adoption for real-world applications challenging with limited data. To this end, this paper proposes a unified framework for logical knowledge-conditioned text generation in the few-shot setting. With only a few seeds logical forms (e.g., 20/100 shot), our approach leverages self-training and samples pseudo logical forms based on content and structure consistency. Experimental results demonstrate that our approach can obtain better few-shot performance than baselines.