Erzya corpora

Welcome to the start page of Erzya language corpora: the Main corpus of literary Erzya (press, blogs, fiction and non-fiction) and the Corpus of Erzya-language social media.

Details To the main corpus To the social media corpus

Erzya corpora

Toggle navigation

This is the main page of the website where linguistic corpora of Erzya language are located. Currently, two corpora are available: the corpus of contemporary written literary Erzya (“the Main corpus”) and the corpus of Erzya-language social media and forums. They differ in what kind of texts the contain, but have mostly identical annotation and search capabilities. Here is a brief comparison:

Main corpus Social media corpus
Language Erzya Erzya and Russian
Size 2.3 million words 830 thousand words (the Erzya part)
5.23 million words (the Russian part)
Texts contemporary press (up to July 2018) — 67.4%, 20th century fiction — 14%; two translations of the New Testament — 6.7%; blogs — 6%; Wikipedia, non-fiction open posts and comments by Erzya-speaking vkontakte users (up to July 2018) — 48.5%; forum — 35%; forum — 16.5%
Language variety in most cases, standard written literary Erzya or close to it language of digital communication: closer to the spoken variety, influenced by the dialects and Russian language, contains numerous code switching instances
  • automatic morphological annotation (lemmatization, part of speech, all inflectional features), 93.6% words analyzedonly tokens that do not contain digits or Latin characters are taken into account
  • no disambiguation
  • annotation of Russian loanwords
  • annotation of several lexical/semantic classes: animate/human nouns, body parts, transport, different classes of proper names, diminutives
  • glossing
  • Russian translation of lemmata
  • automatic morphological annotation (lemmatization, part of speech, all inflectional features), 90.7% words analyzedonly tokens that do not contain digits or Latin characters are taken into account
  • no disambiguation
  • annotation of Russian loanwords
  • annotation of several lexical/semantic classes: animate/human nouns, body parts, transport, different classes of proper names, diminutives
  • glossing
  • Russian translation of lemmata
  • title of the text
  • author or title of the newspaper
  • creation year (exact date in the case of newspapers)
  • genre
  • group name (for groups)
  • publicly available user metadata: sex (for everyone); if available, also birth year (grouped in 5-year spans); real names and nicknames of the users are hidden
  • creation year
  • message type (post/comment)
  • language (tagged automatically, independently for each sentence)

Apart from the corpora available here, there exists another publicly available Erzya corpus developed by Jack Rueter. It contains 800,000 tokens of fiction, but has no morphological annotation.

You can find more detailed information about Erzya Social media corpus and its development in this paper. Please consider citing this paper if your research is based on this corpus:

Timofey Arkhangelskiy. 2019. Corpora of social media in minority Uralic languages. Proceedings of the fifth Workshop on Computational Linguistics for Uralic Languages, pages 125–140, Tartu, Estonia, January 7 - January 8, 2019.

What is a corpus?

A language corpus is a collection of texts in that language which has been enriched with additional linguistic information, called annotation, and, preferably, equipped with a search engine. Here you will find a short list of frequently asked questions about the Erzya corpora.

— Who needs corpora?

First of all, corpora are used by linguists. The search engine and annotation of corpora are designed in such a way that you can make linguistic queries such as “find all nouns in the genitive case” or “find all forms of the word катка followed by a verb”. Apart from linguists, corpus can be a useful tool for language teachers, language learners, and even the native speakers.

— Can I use the corpus as a library?

No, these corpora are not designed for that. When you work with a corpus, you make a query, i.e. search for a particular word, phrase or construction, and get back all sentences that contain what you searched for. By default, the sentences are showed in random order. You can expand the context of each of the sentences you get, i.e. look at their neighboring sentences. However, you may do so only a limited number of times for each sentence. Therefore, it is impossible to read an entire text in the corpus. This is done for copyright protection.

— Can I use the corpus as a dictionary?

Each Erzya word in the corpus has Russian translation (no English translations are available at the moment). However, they are only provided as auxiliary information for users who do not speak Erzya. The translations in the corpus are kept short and simple by design, they do not list all senses and do not provide usage examples like real dictionaries. If you want to know how to translate a word, the right way to do so is consulting a dictionary.

— What is morphological annotation and how do you get it?

The corpora located here are lemmatized and morphologically annotated. Lemmatization means that each word in the texts is annotated with its lemma, i.e. dictionary/citation form. Morphological annotation means that each word is annotated for its grammatical features, such as part of speech, number, case, tense, etc. Since the corpora in question are too large for manual annotation to be feasible, they were annotated automatically with a program called morphological analyzer. The analyzer uses a manually compiled grammatical dictionary and a formalized description of Erzya inflection. The analyzer together with the necessary materials is freely available in my bitbucket repository. Automatic annotation unfortunately means that, first, out-of-vocabulary words are not annotated, and, second, that some words have several ambiguous analyses. For example, confronted with the form валдо, the analyzer cannot determine whether it should be analyzed as the citation form of of валдо (“bright”), ablative of the word вал (“about a word”) or even a form of the verb валомс “pour”. Russian sentences in the social media corpus were annotated with the mystem analyzer.

Erzya language

Erzya is one of the two Mordvinic languages, which belong to the Uralic family. The number of speakers is unknown due to the fact that in the censuses, most Erzya and Moksha speakers indicate “Mordvin” as their language; it can be very roughly estimated at 400,000. Erzya uses Cyrillic orthography based on the Russian alphabet. All morphological markers are suffixes that mostly attach to the stem agglutinatively. Most suffixes have two vowel-harmony versions (containig o/e vowels and palatalized/non-palatalized consonants). Nominal grammatical categories are number, case, definiteness and possessiveness. Transitive verbs can index person and number of the subject and the direct object. The direct object can be marked either in the nominative or in the genitive (DOM). The word order in the sentence is free, with SVO (subject – verb – object) being the default.


The grammatical features of the words in the corpora are marked with short tags. Here is the full list of tags used in Erzya corpora. Both corpora have identical set of tags.

  • A — adjective
  • APRO — adjectival pronoun
  • ADV — adverb
  • ADVPRO — adverbial pronoun
  • CONJ — conjunction
  • IMIT — ideophone
  • INTRJ — interjection
  • N — noun
  • NUM — numeral
  • PARENTH — parenthetic word
  • PART — particle
  • PN — proper noun (subtype of nouns)
  • POST — postposition
  • PREDIC — predicative
  • PRO — pronoun
  • V — verb
  • 1.o — 1st person of the object
  • 1.s — 1st person of the subject
  • 1pl — 1st person plural possessive
  • 1sg — 1st person singular possessive
  • 2.o — 2nd person of the object
  • 2.s — 2nd person of the subject
  • 2pl — 2nd person plural possessive
  • 2sg — 2nd person singular possessive
  • 3.o — 3rd person of the object
  • 3.s — 3rd person of the subject
  • 3pl — 3rd person plural possessive
  • 3sg — 3rd person singular possessive
  • abbr — abbreviation
  • abl — ablative
  • add — additive clitic
  • all — allative
  • anim — animate noun
  • body — body part
  • car — caritive (abessive)
  • case_comp — case compounding
  • caus — causative (-vt-)
  • coll — collective numeral
  • com — comitative (unproductive)
  • comp — comparative
  • cond — conditional mood
  • cvb.caus — converb of causation (-mga)
  • cvb.simult — converb of simultaneity (-msto/-mste)
  • dat — dative
  • def — definite declension
  • desid — desiderative mood
  • dim — diminutive
  • distr — distributive numeral
  • el — elative
  • famn — family name
  • gen — genitive
  • hum — human
  • ill — illative
  • imp — imperative
  • inch — inchoative (-źev-)
  • inf — infinitive (-ms)
  • iter — iterative (-kšn-)
  • loc — locative/inessive
  • missp — typo
  • mult — multiplicative (-ńe-, -śe-)
  • neg — negative form
  • nmlz — any nominalization
  • nmlz_ma — nominalization in -ma
  • nmlz_mka — nominalization in -mka
  • nmlz_mo — nominalization in -mo/-me
  • nom — nominative (unmarked form)
  • non_obj — objectless conjugation
  • npst — non-past tense
  • num_approx — approximative numeral
  • opt — optative mood
  • ord — ordinal numeral
  • pair — pair numeral
  • pass — passive (-v-)
  • patrn — patronymic
  • persn — personal (given) name
  • pl — plural
  • pl.o — plural object
  • pl.s — plural subject
  • pl_comp — plural in case compounding
  • prol — prolative
  • pst — first past tense
  • pst2 — second past tense
  • ptcp.hab — active present (habitual) participle
  • ptcp.prs.pass — present passive participle
  • ptcp.pst — past participle
  • rel_n — relational noun (inflected postposition)
  • rus — Russian borrowing (or borrowing through Russian)
  • sg — singular
  • sg.o — singular object
  • sg.s — singular subject
  • subj — subjunctive mood
  • supernat — noun that denotes a supernatural beingThis category is a byproduct of animacy/humanness annotation. Since it is not clear whether these cases should be classified as human, we put them in a separate box, so that the user can decide that for themselves.
  • time_meas — time measurement unit
  • topn — toponym (geographical name)
  • trans — translative
  • transport — transport

The tagset for the Russian-language part (Russian sentences in the social media corpus) can be found in the Russian National Corpus.


The corpora and morphological analyzer are developed and maintained by Timofey Arkhangelskiy. The first versions of the corpora were released in 2018 as part of his postdoctoral project supported by Alexander von Humboldt Foundation. The background picture was kindly provided by Polina Pleshak. The corpora are hosted by the School of linguistics at HSE, Moscow.


If you have questions, would like to propose collaboration, or noticed an error in the corpusexcept typos in blogs and social media: these text are left "as is", please contact Timofey Arkhangelskiy. You can also use the Erzya morphological analyzer and the tsakorpus corpus platform, which are open source and freely available.