In this paper, we introduce a learning analytics framework to analyze the in-context learning (ICL) behavior of large language models (LLMs) through the lens of the Zone of Proximal Development (ZPD), an established theory in educational psychology. ZPD delineates the space between what a learner is capable of doing unsupported and what the learner cannot do even with support. We adapt this concept to ICL, measuring the ZPD of LLMs based on model performance on individual examples with and without ICL. Furthermore, we propose an item response theory (IRT) model to predict the distribution of zones for LLMs. Our findings reveal a series of intricate and multifaceted behaviors of ICL, providing new insights into understanding and leveraging this technique. Finally, we demonstrate how our framework can enhance LLM in both inference and fine-tuning scenarios: (1) By predicting a model's zone of proximal development, we selectively apply ICL to queries that are most likely to benefit from demonstrations, achieving a better balance between inference cost and performance; (2) We propose a human-like curriculum for fine-tuning, which prioritizes examples within the model's ZPD. The curriculum results in improved performance, and we explain its effectiveness through an analysis of the training dynamics of LLMs.

Large Language Models (LLMs) have become ubiquitous in NLP and deep learning. In-Context Learning (ICL) has been suggested as a bridging paradigm between the training-free and fine-tuning LLMs settings. In ICL, an LLM is conditioned to solve tasks by means of a few solved demonstration examples included as prompt. Argument Mining (AM) aims to extract the complex argumentative structure of a text, and Argument Type Classification (ATC) is an essential sub-task of AM. We introduce an ICL strategy for ATC combining kNN-based examples selection and majority vote ensembling. In the training-free ICL setting, we show that GPT-4 is able to leverage relevant information from only a few demonstration examples and achieve very competitive classification accuracy on ATC. We further set up a fine-tuning strategy incorporating well-crafted structural features given directly in textual form. In this setting, GPT-3.5 achieves state-of-the-art performance on ATC. Overall, these results emphasize the emergent ability of LLMs to grasp global discursive flow in raw text in both off-the-shelf and fine-tuned setups.