UD_Nenets-Tundra

Treebank card from UD release 2.18: metadata and texts come from UD_Nenets-Tundra/README.md; size is counted by streaming the .conllu files and checked against the release stats.xml. Generated by tools/treebanks build; do not edit by hand. Papers: Papers: UD_Nenets-Tundra; seeds: seeds/; converter: convert.md.

Annotation layers

UD scale: manual native — annotated by hand directly in UD; converted from manual — converted from manual annotation in another scheme; converted with corrections — converted and partly corrected by hand; automatic with corrections — automatic, with partial manual corrections; automatic — automatic; not available — the layer is absent.

Size

Tokens are surface tokens (Spanish vámonos is one), words are syntactic words (vamos + nos); multiword tokens are those split into several words; empty nodes are the N.M nodes of the enhanced graph. Splits train, dev are absent from the release.

Summary

The Tundra Nenets UD treebank is converted from the Tundra Nenets mSUD treebank. The conversion from mSUD to UD is performed automatically followed by a comprehensive manual revision to ensure compliance with the UD annotation standards.

Introduction

The treebank currently comprises 171 manually annotated sentences (approximately 11 minutes of recorded speech). The data were collected during a fieldwork session conducted in Moscow in 2017 with a native speaker of Tundra Nenets (Yamal dialect). The session elicited semi-spontaneous speech through visual stimulus–based tasks, including a modified version of the [HCRC Map Task] (groups.inf.ed.ac.uk/maptask/maptasknxt.html) and the the so-called Pear Story narrative task (Chafe, 1980).

The spoken data were transcribed by the native speaker using standard Tundra Nenets orthography rather than phonetic or IPA-based notation. Sentence segmentation builds on the original transcription and follows a combined prosodic and semantic approach, in which intonational boundaries are treated as sentence boundaries only when they correspond to semantically complete units; otherwise, adjacent material is merged. The recordings were manually time-aligned at both sentence and lexeme levels in Praat to support morphological and syntactic analysis. The transcription reflects normalized forms and does not capture phonetic variation or morphophonological processes (e.g. sandhi), although the audio data are available for reference.

Annotation focuses on spoken-language phenomena relevant to word-level analysis, especially word boundary identification. An inductive approach was adopted, identifying recurrent phenomena in the data and assigning dedicated tags, informed by spoken UD practices and adapted to the typological properties of Tundra Nenets. The dataset consists mainly of narrative monologues and lacks interactional features such as overlap. Spoken phenomena are divided into non-lexical items (e.g. noises, pauses, hesitation markers), assigned discourse relations, and lexical disruptions (e.g. unfinished words, false starts, repetitions), which affect syntactic structure; pauses are treated either as punctuation or discourse elements depending on function.

Lemmatization preserves dialectal variation due to the absence of a unified written standard. Only inflectional morphology is segmented from stems, while derivational morphology is retained; inflectional suffixes remain in their surface forms, and linking vowels are treated as part of the stem. POS tagging and morphological analysis were performed manually, with inflectional features segmented and glossed following established descriptive traditions and a tagset based on the Leipzig Glossing Rules, adapted to Tundra Nenets.

To account for its complex morphology and syntactic encoding, annotation is based on the morphologically enhanced Surface-Syntactic Universal Dependencies (mSUD) framework (Guillaume et al., 2024). Within the UniDive COST Action (CA21167), UD is extended with a layer for Information Structure roles (e.g. topic and focus). In Tundra Nenets, topicality is reflected in agreement: when a verb agrees with a topical object, the object is annotated as ISRole=Top and the corresponding verbal agreement as ISMarker[Top]=Agr.

The original Cyrillic transcription was transliterated into Latin script, taking into account language-specific properties. Conversion from mSUD to UD is performed in two stages, first from mSUD to SUD and then from SUD to UD, using iteratively applied Grew (Guillaume, 2021) rules.

This treebank is described in detail in Mus et al. (2025), where further information on annotation decisions and linguistic analysis can be found.

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/treebanks/yrk/tundra/