PluG: A Corpus of 19th- and Early 20th-Century Ukrainian Texts

1 Overview

The Pluperfect GRAC (PluG) corpus draws from the General Regionally Annotated Corpus of Ukrainian (GRAC) (Shvedova, 2020, 2026; Shvedova et al., 2017), a broad-coverage corpus of Ukrainian texts maintained and expanded by a community of linguists. PluG represents the historical part of GRAC, focusing on works published before 1955 that are now in the public domain. With the rapid development of large language models, we decided to release the historical portion of GRAC in full, with the aim of providing high-quality Ukrainian material for model training and research.

The corpus covers a wide range of historically significant Ukrainian texts. It includes extensive collections of works by the most famous authors such as Taras Shevchenko and Lesia Ukrainka, the first Ukrainian translation of the Bible, and a substantial collection of historical press (Shvedova & Lukashevskyi, 2026). Texts span fiction, journalism, legal texts, original and translated texts, authored by 7,590 unique writers, and drawn from publishing centers across Ukraine, most prominently Kyiv, Lviv, and Kharkiv. Each text is accompanied by rich metadata following the GRAC annotation scheme, including style, genre, orthographic system, date, region, author(s), and publication details¹ (see Table 1). The corpus provides broad coverage across time periods and functional styles, as illustrated in Figure 1 and Table 2: predominantly fictional and poetic in the early period, with the diversity of text types increasing over time.

Figure 1

In addition to standard Ukrainian, PluG includes texts reflecting language variation of the late 19th and early 20th centuries, which is underrepresented in existing NLP resources. Exposing a model to non-standard spelling, grammar, and vocabulary improves its ability to handle the full range of variation it may encounter, including in contemporary non-standard usage.

PluG2 extends the original corpus by incorporating a substantially larger collection of Western Ukrainian texts from the 1880s to the 1920s written in the Zhelekhivka orthography. These texts do not merely exhibit a distinct orthographic system: they represent a historically and regionally specific literary standard with its own grammatical patterns and vocabulary (Franko, 1995), making them a distinct and valuable component for language modeling and historical linguistic analysis.

Existing large-scale Ukrainian language resources, such as UberText 2.0 (3.27 billion tokens) (Chaplynskyi, 2023) and Kobza (60 billion tokens) (Haltiuk & Smywiński-Pohl, 2025), focus predominantly on modern Ukrainian. For historical Ukrainian specifically, the closest comparable resource is the Ukrainian Cultural Heritage Books collection (462 million words from 19,574 digitized Internet Archive files) (Langlais & PleIAs, 2024); however, that corpus has not undergone manual OCR correction and lacks the structured metadata that PluG provides for each document. PluG thus fills a distinct niche as a manually curated, richly annotated historical corpus suitable for both linguistic research and language model training.

2 Method

3 Dataset Description

4 Reuse Potential

PluG and PluG2 serve a broad range of research and applied purposes. For NLP and machine learning, the corpora provide copyright-free historical Ukrainian text at scale suitable for pre-training or fine-tuning language models. The inclusion of texts in multiple historical orthographic systems makes the corpora especially useful for developing models capable of handling historical spelling variation.

For historical and corpus linguistics, PluG enables diachronic analysis of lexical, grammatical, and orthographic change in Ukrainian over more than a century. The corpus spans multiple genres and regional traditions, allowing researchers to study register variation, regional features, and the gradual emergence of the modern standard. The structured metadata supports systematic filtering and subcorpus construction for targeted studies.

In literary and cultural studies, the corpus provides machine-readable access to a large body of Ukrainian fiction, journalism, and other writing from the 19th and early 20th centuries, supporting distant reading approaches and investigations of thematic and stylistic patterns across the period. For educators, the corpora offer freely accessible authentic historical Ukrainian texts useful for language instruction, translation studies, and the history of the Ukrainian literary language.

5 Limitations

Potential limitations include the following. First, as the texts were digitized via OCR, some transcription errors may persist despite manual correction. Second, the historical orthographic systems present in PluG2 in particular may require additional preprocessing for applications targeting the contemporary standard. Third, metadata completeness varies across texts, as not all source materials provide full publication details.

AI Declaration

Generative AI tools were used for language editing and grammatical checking. The authors are fully responsible for the final content.

Author Contributions

Maria Shvedova: Conceptualization; Data curation; Methodology; Supervision; Validation; Writing – original draft; Writing – review & editing.

Arsenii Lukashevskyi: Data curation; Software; Formal analysis; Visualization; Writing – original draft; Writing – review & editing.