Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cross-lingual Named Entity Corpus for Slavic Languages

Jakub Piskorski, Michał Marcińczuk, Roman Yangarber

TL;DR

The paper presents a large cross-lingual named-entity corpus for six Slavic languages, comprising 5,017 online-news documents annotated with five NE types, each surface form linked to a base form and a cross-lingual ID. It merges and curates data from four editions of Slavic NER shared tasks (2019–2023) and provides two train-tune splits (single-topic-out and cross topics) to benchmark cross-language generalization and in-domain performance. Baselines for NER, lemmatization, and linking are established using XLM-RoBERTa-large and mT5-large, with cross-topic NER F1 of 0.9222, lemmatization accuracy of 96.13% (Seq2seq), and linking accuracy of 87.84% (Seq2seq), illustrating strong cross-lingual transfer capabilities. The resource, including positional anchoring and cross-lingual identifiers, supports NE recognition, cross-language name linking, and lemmatization across Slavic languages, enabling broader cross-lingual information extraction research and practical applications. The corpus is publicly available for research, reflecting a significant contribution to Slavic NLP by providing a rare, integrated, cross-lingual NE resource.

Abstract

This paper presents a corpus manually annotated with named entities for six Slavic languages - Bulgarian, Czech, Polish, Slovenian, Russian, and Ukrainian. This work is the result of a series of shared tasks, conducted in 2017-2023 as a part of the Workshops on Slavic Natural Language Processing. The corpus consists of 5 017 documents on seven topics. The documents are annotated with five classes of named entities. Each entity is described by a category, a lemma, and a unique cross-lingual identifier. We provide two train-tune dataset splits - single topic out and cross topics. For each split, we set benchmarks using a transformer-based neural network architecture with the pre-trained multilingual models - XLM-RoBERTa-large for named entity mention recognition and categorization, and mT5-large for named entity lemmatization and linking.

Cross-lingual Named Entity Corpus for Slavic Languages

TL;DR

The paper presents a large cross-lingual named-entity corpus for six Slavic languages, comprising 5,017 online-news documents annotated with five NE types, each surface form linked to a base form and a cross-lingual ID. It merges and curates data from four editions of Slavic NER shared tasks (2019–2023) and provides two train-tune splits (single-topic-out and cross topics) to benchmark cross-language generalization and in-domain performance. Baselines for NER, lemmatization, and linking are established using XLM-RoBERTa-large and mT5-large, with cross-topic NER F1 of 0.9222, lemmatization accuracy of 96.13% (Seq2seq), and linking accuracy of 87.84% (Seq2seq), illustrating strong cross-lingual transfer capabilities. The resource, including positional anchoring and cross-lingual identifiers, supports NE recognition, cross-language name linking, and lemmatization across Slavic languages, enabling broader cross-lingual information extraction research and practical applications. The corpus is publicly available for research, reflecting a significant contribution to Slavic NLP by providing a rare, integrated, cross-lingual NE resource.

Abstract

This paper presents a corpus manually annotated with named entities for six Slavic languages - Bulgarian, Czech, Polish, Slovenian, Russian, and Ukrainian. This work is the result of a series of shared tasks, conducted in 2017-2023 as a part of the Workshops on Slavic Natural Language Processing. The corpus consists of 5 017 documents on seven topics. The documents are annotated with five classes of named entities. Each entity is described by a category, a lemma, and a unique cross-lingual identifier. We provide two train-tune dataset splits - single topic out and cross topics. For each split, we set benchmarks using a transformer-based neural network architecture with the pre-trained multilingual models - XLM-RoBERTa-large for named entity mention recognition and categorization, and mT5-large for named entity lemmatization and linking.
Paper Structure (25 sections, 4 figures, 18 tables)

This paper contains 25 sections, 4 figures, 18 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Named-Entity Annotation
  4. Taxonomy
  5. Document Acquisition
  6. Annotation Platform
  7. Annotation Process
  8. Annotation Format
  9. Statistics
  10. Baseline models
  11. Named-Entity Recognition
  12. Named-Entity Lemmatization
  13. Named-Entity Linking
  14. Access
  15. Conclusions
  16. ...and 10 more sections

Figures (4)

  • Figure 1: Distribution of the text length for all languages.
  • Figure 2: Screenshot of the Inforex Web interface, the tool used for data annotation.
  • Figure 3: An example of a raw document in Polish and a corresponding NE-annotated file.
  • Figure 4: Distribution of named-entity occurrences for all languages.