Ossetic-COT: Designing a morphologically annotated corpus and morphological analyzer for Ossetic

Jul 16, 2026

Upload by Noah

First morphologically annotated corpus for Iron Ossetic is introduced, built to conform to the Universal Dependencies (UD) schema. The corpus contains 5,454 manually annotated sentences with 74,032 tokens, converted from the Iron Ossetic Corpus of Oral Texts. The resulting UD-compliant dataset is then used to train a BERT-based morphological analyzer, reaching per-token tag accuracy of 95.60%. The work motivates the need for disambiguated, UD-standard annotation for education and research.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

Introduction

Background

Data sources

Corpus and annotation methodology

Morphological properties and annotation approach

Training and evaluation of the BERT analyzer

Limitations

Conclusion

Ossetic-COT: Designing a morphologically annotated corpus and morphological analyzer for Ossetic arXiv :2607 .04895v2 [ cs .CL] 7 Jul 2026 Anna Shatskikh Lomonosov Moscow State University [anapriselec@gmail. com](anapriselec@gmail. com) Alexey Sorokin Lomonosov Moscow State University Yandex [alexey. sorokin@list. ru](alexey. sorokin@list. ru) Abstract In this work we present the first morphologically annotated corpus for Iron Ossetic that conforms to the Universal Dependencies schema. The corpus includes 5454 manually annotated sentences from the Iron Ossetic Corpus of Oral Texts, containing 74032 tokens. We use this corpus to train a BERT-based morphological analyzer. The analyzer achieves tag accuracy of 95.60% . Keywords: Iron Ossetic, Universal Dependencies, Morphological Tagging, Morphological Analysis DOI: 10.29003/2075-7182-2026-24-544-555 Ossetic-COT: Создание корпуса с морфологической разметкой иморфологического анализатора для осетинского языка Анна Шатских Алексей Сорокин МГУ им . Ломоносова МГУ им . Ломоносова, Яндекс [anapriselec@gmail.com](anapriselec@gmail.com) [alexey.sorokin@list.ru](alexey.sorokin@list.ru) Аннотация Работа посвящена созданию первого корпуса для осетинского языка с морфологической разметкой в формате Universal Dependencies. Корпус состоит из размеченных вручную текстов Устного корпуса осетинского языка и содержит 5454 предложения суммарной длиной 73 042 токена .Также в работе представлен морфологический анализатор, основанный на модели BERT и достигающий пословной точности 95,60% . Ключевые слова: осетинский язык, иронский диалект, Universal Dependencies, морфологический анализ 1 Introduction Ossetic is an Iranic language from the Indo-European family spoken by up to 641,450 speakers mostly in North Causasus region 1. There exist two main corpora of Ossetic: Ossetic National Corpus (ONC) 2 and the Corpus of Oral Texts (COT) 3. ONC contains Ossetic literary texts, mostly written by contemporary authors (1999-2014), as well as the works by the most celebrated writers of the 20th century. It also includes texts from «Махдуг» («Our epoch») and «Ногдзау» («Pioneer») literary magazines and «Спутник» («Satellite») online newspaper, as well as the Nart Epic collection. The size of ONC is approximately 12 millions of tokens. The corpus is automatically annotated for parts of speech, morphological features and lemmas via Uniparser-Morph (Arkhangelskiy et al., 2012) . Because the parser is context-free, the corpus is not disambiguated, providing multiple possible tags for a given word. 1[http://www.ethnologue.com/show_language.asp?code=oss](http://www.ethnologue.com/show_language.asp?code=oss) 2[http://corpus.ossetic-studies.org](http://corpus.ossetic-studies.org) 3[https://www.ossetic-studies.org/ru/texts/iron](https://www.ossetic-studies.org/ru/texts/iron) COT includes Ossetic oral texts collected during linguistic expeditions in 2007-2013 in various regions of North Ossetia. It also contains a number of issues of «Одноклассники» («Classmates») radio program. The size of COT is 60,000 tokens. The corpus is provided with manual interlinear glossing done using SIL FieldWorks Language Explorer 4. Although it is possible to extract part-of-speech tags and morphological features from SIL format, the corpus has only incomplete coarse-grained annotation (see Section 4) . This not only complicates automated conversion to the generally accepted Universal Dependencies annotation schema (Nivre et al., 2020) but also limits the applicability of the corpus for educational or research purposes. These factors necessitate the development of a morphologically disambiguated Ossetic corpus that adheres to Universal Dependencies (UD) standards. The present work fulfills this objective. The remainder of the paper is organized as follows. Section 2 provides necessary background linguistic information about Ossetic. In Section 3 we describe the corpus and annotation methodol

FAQs

What is Ossetic-COT, and what standard does it follow?

Ossetic-COT is the first morphologically annotated corpus for Iron Ossetic designed to follow the Universal Dependencies (UD) schema.

What data size and annotation does the corpus include?

It contains 5,454 manually annotated sentences totaling 74,032 tokens, with morphology provided in a UD-compatible format.

How accurate is the BERT-based morphological analyzer trained on the corpus?

The trained analyzer achieves 95.60% per-token tag accuracy.

Jul 16, 2026

Upload by Noah