about

Contributors

Language

Genetic affiliation: Nakh-Daghestanian/East Caucasian, Nakh
Place: Georgia
Language Code: bbl
Population: below 500
Endangerment: severely endangered (Endangered Languages Project)

Open map/tree (glottolog)

data

Instructions

The speakers produced the narratives as an answer to the following instructions. Instructions were presented in Georgian.

Recordings

• Recordings took generally place in quite places (not laboratory conditions),
• with a ZOOM H6 recorder, interchangeable clip-on microphones, at a sampling frequency of 96000 kHz, audio files were saved in wav format.
• Total soundfiles: 80;
• Total duration: 138 min.

Duration of our recordings (in seconds):

File names

• Example: BBL-TXT-AN-00000-01
• BBL- language code: BBL=“Batsbi"
• TXT- subcollection code: TXT=“Texts"
• AN- instruction code; 5 options: AN="Ancestors", AC="Activity", CD="Comparison", EV="Event", PA="Path" (see instructions above)
• 00000- does not play a role in this subcollection
• 01 Speaker identifier, ranges from 01 to 16

Annotations

Layers

The annotation layers of the corpus files follow the conventions used in ELAN. "A_" stands for the speaker, the further layers are associated with phrase/word/morph intervals, and either contain text (txt) in Tush (bbl) or translations (gls)/part of speech (pos) in English (en). Precisely:

Transcriptions

Categories

The abbreviations for glosses follow the Leipzig Glossing Rules.

Noun classes are annotated as lexical property of nouns or as a property of the agreement prefixes.

Texts

The following texts illustrate the types of elicited narratives of the present corpus. The entire corpus (sound files in .wav and annotations in ELAN) are archived and available to download in Zenodo:
DOI: 10.5281/zenodo.15863417.

The corpus is also accessible for queries online through the ANNIS database; see below.

syntax

Syntactic annotations of the Tush corpus are based on SUD (=Surface Universal Dependencies), see Kim Gerdes, Bruno Guillaume, Sylvain Kahane, and Guy Perrier. 2018. SUD or Surface-Syntactic Universal Dependencies: An annotation scheme near-isomorphic to UD. In Proceedings of the Second Workshop on Universal Dependencies (UDW 2018), pages 66–74, Brussels, Belgium. Association for Computational Linguistics. The basic properties of our corpus are summarized In the following. For more elaborated discussion, we refer to the SUD website.

You can visit and download the Tush Treebanks in ArboratorGrew (Treebank BBL-SUD). The current version of the Treebanks was created semi-automatically with a parser written in R. The data contains errors that will be manually corrected. Next release scheduled for 1.11.2025.

Part of Speech

Part of speech information was converted to the UPOS categories following UD. The XPOS layer contains the abbreviations used in FLEx and are sometimes more detailed (e.g., they are informative for subclasses of pronouns). The inflectional features (FEATS) are only used for the inflectional categories of nouns and the different verb forms.

Root

The root of the dependency tree is the finite verb of a clause or – in the absence of a finite verb – the highest head of the annotated unit.

Root examples in CoNNLU Viewer

Modifiers

SOURCE: head, TARGET: modifier

• adjective/quantifier [mod]← noun
• adverbs [mod]← verb
• commitative, instrumental, locative, adverbial cases or ablative, superlative, etc. postpositions [mod]← verb (occasionally also other)
• degree adverb [mod]← adverb or adjective
• subordinators of adverbial clauses [mod]← element in matrix clause

• Genitive noun [udep]← noun
• "possessive pronouns" (=genitive of personal pronouns) [udep]← noun

Modifier examples in CoNNLU Viewer

Complements

SOURCE: head, TARGET: complement

• absolutive argument [comp:obj]← transitive verb
• infinitival complement [comp:obj]← transitive verb
• nominal complement [comp:obj]← postposition
• highest verb of embedded clause [comp:obj]← subordinating conjunction
• comparative conjunction [comp:obj]← comparative marker (degree head) (see comparative constructions)
• standard of comparison [comp:obj]← comparative conjunction
• reported speech FEATS [Reported=Yes] [comp:obj]← verb of saying (see reported speech)

In our annotation, "obliques" refer to nominal arguments that are selected by certain (classes of) verbs, such as datives/allatives as well as absolutives as goals of motion verbs.

• dative/allative of indirect object [comp:obl]← verb
• experiencer dative [comp:obl]← psych verb
• controller dative [comp:obl]← modal verb
• absolutive of goal of motion [comp:obl]← motion verb
• allative/directional case [comp:obl]← motion verb
• causee dative/allative [comp:obl]← causative verb

• predicative adjective [comp:pred]← copula

Note: The subject is a dependent of the copula, not of the predicative adjective.

Complement examples in CoNLLU Viewer

Specifiers

SOURCE: head, TARGET: specifier

Subjects and determiners have separate labels in SUD. Note that subjects (target) are dependents of verbs (source), and determiners (target) are dependents of nouns (source).

• demonstrative [det]← noun (if the demonstrative pronoun is used attributively)

• ergative argument [subj]← transitive verb
• single argument (ergative or absolutive) [subj]← intransitive verb
• infinitive as subject [subj]← transitive/intransitive verb

Specifier examples in CoNNLU Viewer

Other

• conjunction [cc]← last conjunct
• preceding conjunct [conj:coord]← subsequent conjunct

Note: This implies that the syntactic relation of the coordinated phrase to the head is only annotated at the leftmost conjunct.

• target [conj:appos]← apposition

The first expression (of disfluencies, repetitions, reformulations) is treated as head, which means that it hosts the dependency relation of the entire unit to the head in the clause or is the root.

• false start [conj:dicto]← revised unit
• expression [conj:dicto]← repeated or reformulated unit

• punctuation [punct]← highest element of the preceding punctuation domain

• Vocative [vocative]← root
• Interjections/discourse markers [discourse]← root

Other examples in CoNNLU Viewer

queries

The Tush corpus is online available in ANNIS that allows for visualizations and queries in multimodal annotations. It comes with a powerful query language (AQL=ANNIS query language) that allows to retrieve complex data patterns in multilayered annotations (Krause, Thomas & Zeldes, Amir 2016: ANNIS3: A new architecture for generic corpus query and visualization. in: Digital Scholarship in the Humanities 2016 (31). http://dsh.oxfordjournals.org/content/31/1/118).

You can access the corpus in the SPW installation at: https://spw.uni-goettingen.de/annis/.

Before the query, you need to select the corpus "BBL-0.1-mp3". You write your query in the query window:

Plain-text queries in AQL

In your queries, you need to specify the annotation layer and define the expression that you are looking for. Notice that the query tool only retrieves sentences that equal the queried expression (and not sentences that contain the queried expression).

Regular expressions in AQL

Regular expressions are included in slashes. You find some illustrative examples below. More details about the regular expressions in AQL are found here.

More about AQL: AQL documentation site.

sources

citation

Kakashvili, D., & Skopeteas, S. (2025). Tsova-Tush (Bats/bi) spoken data corpus (1.0.0) [Data set]. Zenodo. (doi: 10.5281/zenodo.15863418)