Towards Understanding In-Context Learning with Contrastive Demonstrations and Saliency Maps

TL;DR

This paper examines how different demonstration components like labels, input distribution, and explanations influence in-context learning in large language models, using explainable NLP methods and saliency maps.

Contribution

It provides a detailed analysis of the effects of label flipping, input changes, and explanations on ICL performance, offering insights into LLM interpretability and demonstration design.

Findings

Flipping ground-truth labels significantly impacts saliency, especially in larger LLMs.

Changing sentiment-indicative terms has less effect than label alterations.

Explanations improve ICL performance mainly in symbolic reasoning tasks, not sentiment analysis.

Abstract

We investigate the role of various demonstration components in the in-context learning (ICL) performance of large language models (LLMs). Specifically, we explore the impacts of ground-truth labels, input distribution, and complementary explanations, particularly when these are altered or perturbed. We build on previous work, which offers mixed findings on how these elements influence ICL. To probe these questions, we employ explainable NLP (XNLP) methods and utilize saliency maps of contrastive demonstrations for both qualitative and quantitative analysis. Our findings reveal that flipping ground-truth labels significantly affects the saliency, though it's more noticeable in larger LLMs. Our analysis of the input distribution at a granular level reveals that changing sentiment-indicative terms in a sentiment analysis task to neutral ones does not have as substantial an impact as altering ground-truth labels. Finally, we find that the effectiveness of complementary explanations in boosting ICL performance is task-dependent, with limited benefits seen in sentiment analysis tasks compared to symbolic reasoning tasks. These insights are critical for understanding the functionality of LLMs and guiding the development of effective demonstrations, which is increasingly relevant in light of the growing use of LLMs in applications such as ChatGPT. Our research code is publicly available at https://github.com/paihengxu/XICL.