CzechTopic: A Benchmark for Zero-Shot Topic Localization in Historical Czech Documents

TL;DR

This paper introduces a new benchmark dataset for zero-shot topic localization in historical Czech documents, evaluating various language models and revealing significant variability in their performance.

Contribution

It provides the first human-annotated benchmark for topic localization in Czech historical texts, along with an evaluation framework for assessing model performance at multiple levels.

Findings

Large language models show wide performance variability.

Some models approach human-level accuracy in topic detection.

Smaller distilled models remain competitive despite their size.

Abstract

Topic localization aims to identify spans of text that express a given topic defined by a name and description. To study this task, we introduce a human-annotated benchmark based on Czech historical documents, containing human-defined topics together with manually annotated spans and supporting evaluation at both document and word levels. Evaluation is performed relative to human agreement rather than a single reference annotation. We evaluate a diverse range of large language models alongside BERT-based models fine-tuned on a distilled development dataset. Results reveal substantial variability among LLMs, with performance ranging from near-human topic detection to pronounced failures in span localization. While the strongest models approach human agreement, the distilled token embedding models remain competitive despite their smaller scale. The dataset and evaluation framework are publicly available at: https://github.com/dcgm/czechtopic.