Iteratively Prompt Pre-trained Language Models for Chain of Thought

TL;DR

This paper introduces an iterative, context-aware prompting method for pre-trained language models to improve multi-step reasoning by dynamically generating prompts based on current inference contexts.

Contribution

It proposes a novel iterative, context-aware prompting framework that enhances PLMs' multi-step reasoning capabilities by dynamically synthesizing prompts conditioned on inference context.

Findings

Improved multi-step reasoning performance on three datasets.

Iterative prompting outperforms traditional single-shot prompts.

Context-aware prompter adapts to different inference steps effectively.

Abstract

While Pre-trained Language Models (PLMs) internalize a great amount of world knowledge, they have been shown incapable of recalling these knowledge to solve tasks requiring complex & multi-step reasoning. Similar to how humans develop a "chain of thought" for these tasks, how can we equip PLMs with such abilities? In this work, we explore an iterative prompting framework, a new prompting paradigm which progressively elicits relevant knowledge from PLMs for multi-step inference. We identify key limitations of existing prompting methods, namely they are either restricted to queries with a single identifiable relation/predicate, or being agnostic to input contexts, which makes it difficult to capture variabilities across different inference steps. We propose an iterative context-aware prompter, which addresses these limitations by learning to dynamically synthesize prompts conditioned on the current step's contexts. Experiments on three datasets involving multi-step reasoning show the effectiveness of the iterative scheme and the context-aware prompter design.