In-context learning (ICL) has emerged as a powerful paradigm for task adaptation in large language models (LLMs), where models infer underlying task structures from a few demonstrations. However, ICL remains susceptible to biases that arise from prior knowledge and contextual demonstrations, which can degrade the performance of LLMs. Existing bias calibration methods typically apply fixed class priors across all inputs, limiting their efficacy in dynamic ICL settings where the context for each query differs. To address these limitations, we adopt implicit sequential Bayesian inference as a framework for interpreting ICL, identify “surprise” as an informative signal for class prior shift, and introduce a novel method—Surprise Calibration (SC). SC leverages the notion of surprise to capture the temporal dynamics of class priors, providing a more adaptive and computationally efficient solution for in-context learning. We empirically demonstrate the superiority of SC over existing bias calibration techniques across a range of benchmark natural language processing tasks.

In recent years, new deep learning methods and pre-training language models have been emerging in the field of natural language processing (NLP). These methods and models can greatly improve the accuracy of automatic word segmentation and part-of-speech tagging in the field of ancient Chinese research. In these models, the BERT model has made amazing achievements in the top-level test of machine reading comprehension SQuAD-1.1. In addition, it also showed better results than other models in 11 different NLP tests. In this paper, SIKU-RoBERTa pre-training language model based on the high-quality full-text corpus of SiKuQuanShu have been adopted, and part corpus of ZuoZhuan that has been word segmented and part-of-speech tagged is used as training sets to build a deep network model based on BERT for word segmentation and POS tagging experiments. In addition, we also use other classical NLP network models for comparative experiments. The results show that using SIKU-RoBERTa pre-training language model, the overall prediction accuracy of word segmentation and part-of-speech tagging of this model can reach 93.87% and 88.97%, with excellent overall performance.