PolQA: Polish Question Answering Dataset

TL;DR

PolQA delivers the first Polish OpenQA dataset (7,000 questions, 87,525 evidence passages, and a 7,097,322-candidate passage corpus) and rigorously analyzes annotation strategies for retriever–reader QA pipelines. The work compares Standard manual and Efficient sampling strategies, showing that Efficient sampling with verification dramatically reduces annotation cost (≈82%) while achieving competitive or superior retrieval performance (≈+10 p.p. in top-10 accuracy). Key findings reveal that unbiased passage collection does not always improve end-to-end QA performance, and that including hard negatives has nuanced effects across retriever and reader components. Overall, PolQA provides practical guidelines for building OpenQA datasets in Polish and other languages, with implications for cross-lingual OpenQA research and efficient dataset construction.

Abstract

Recently proposed systems for open-domain question answering (OpenQA) require large amounts of training data to achieve state-of-the-art performance. However, data annotation is known to be time-consuming and therefore expensive to acquire. As a result, the appropriate datasets are available only for a handful of languages (mainly English and Chinese). In this work, we introduce and publicly release PolQA, the first Polish dataset for OpenQA. It consists of 7,000 questions, 87,525 manually labeled evidence passages, and a corpus of over 7,097,322 candidate passages. Each question is classified according to its formulation, type, as well as entity type of the answer. This resource allows us to evaluate the impact of different annotation choices on the performance of the QA system and propose an efficient annotation strategy that increases the passage retrieval accuracy@10 by 10.55 p.p. while reducing the annotation cost by 82%.