In-context learning can improve the performances of knowledge-rich tasks such as question answering. In such scenarios, in-context examples trigger a language model (LM) to surface information stored in its parametric knowledge. We study how to better construct in-context example sets, based on whether the model is aware of the in-context examples. We identify ‘known’ examples, where models can correctly answer from their parametric knowledge, and ‘unknown’ ones. Our experiments show that prompting with ‘unknown’ examples decreases the performance, potentially as it encourages hallucination rather than searching for its parametric knowledge. Constructing an in-context example set that presents both known and unknown information performs the best across diverse settings. We perform analysis on three multi-answer question answering datasets, which allows us to further study answer set ordering strategies based on the LM’s knowledge of each answer. Together, our study sheds light on how to best construct in-context example sets for knowledge-rich tasks.

This paper describes our submission to the PragTag task, which aims to categorize each sentence from peer reviews into one of the six distinct pragmatic tags. The task consists of three conditions: full, low, and zero, each distinguished by the number of training data and further categorized into five distinct domains. The main challenge of this task is the domain shift, which is exacerbated by non-uniform distribution and the limited availability of data across the six pragmatic tags and their respective domains. To address this issue, we predominantly employ two data augmentation techniques designed to mitigate data imbalance and scarcity: pseudo-labeling and synonym generation. We experimentally demonstrate the effectiveness of our approaches, achieving the first rank under the zero condition and the third in the full and low conditions.