Corpus of spoken Slovenian ROG-Dialog 2.0

  Avtorji

Verdonik, Darinka ; et al.prikaži vse

Verdonik, Darinka ; Rupnik, Peter ; Vidinić, Jasna ; Ljubešić, Nikola ; Kranjc Ivič, Mira ; Ulčnik, Natalija ; Antloga, Špela ; Logar, Tamara

  Identifikator vnosa

http://hdl.handle.net/11356/2637

 URL projekta

https://govorjena-slovenscina.um.si/

 Datum objave

2026-11-09

 Vrsta

audio, corpus

 Velikost

50486 words, 5.167 hours

 Jezik(i)

Slovenian

 Opis

ROG-dialog 2.0 is an annotated corpus of spontaneous spoken Slovenian comprising 12 conversations (5 hours and 10 minutes, approximately 50,500 words) involving 25 adult speakers from all statistical regions of Slovenia. The recordings were contributed through the Govorjena slovenščina portal (https://govorjena-slovenscina.um.si/) and consist of informal conversations between people who know each other, recorded in private settings. They cover a range of conversational activities and topics, including everyday discussions, storytelling, interviews and instructions. The speakers represent diverse age groups, regional varieties, and urban and rural backgrounds. The corpus includes audio recordings, manually segmented and aligned colloquial and standardised orthographic transcriptions, and multiple layers of linguistic annotation: dialogue acts (ISO 24617-2), semantic content, sentiment and conversational humour. Morphological and syntactic annotations were generated automatically using CLASSLA-Stanza 2.2.3 (https://pypi.org/project/classla/2.2.3/ , https://github.com/clarinsi/classla/releases/tag/v2.2.3). The corpus is distributed in four separately downloadable packages: - DOC: Corpus documentation (PDF), README files in Slovenian and English, speaker and recording metadata (TSV), and information on the training, development and test splits. - EXMARaLDA: Audio recordings (WAV), aligned multilayer transcriptions and annotations (EXB), searchable EXS files, COMA metadata, and transcriptions in TRS and TXT formats. The TRS files are provided in Transcriber 1.5.1 format, with separate colloquial (-pog.trs) and standardised (-std.trs) versions. The TXT files contain plain-text transcriptions in colloquial (-pog.txt) and standardised (-std.txt) orthography. This package supports audio-aligned annotation analysis in EXMARaLDA and corpus searches using its EXAKT concordancer. - CoNLL-U: A single .conllu file containing token-level linguistic annotations, including morphological and syntactic dependency information. - VERT: A single .vert file containing linguistic and structural annotations, suitable for corpus indexing and concordance searches. ROG-dialog 2.0 is part of the ROG training corpus of spoken Slovenian and complements the ROG-Art subcorpus (part of ROG 1.1, http://hdl.handle.net/11356/2062). It supports research on spontaneous speech, dialogue structure, communicative functions, sentiment and humour, as well as the development and evaluation of spoken-language processing technologies. Detailed information on transcription conventions, annotation schemes, data formats and metadata is provided in the corpus documentation and README files. Version 2.0 retains the recordings from ROG-dialog 1.0 and introduces several improvements: - Revised and corrected dialogue-act annotations, with a hierarchical organisation of dimensions and communicative functions. - A new semantic-content annotation layer for dialogue acts labelled task/inform. - A new conversational humour annotation layer based on the five-category KROHOT scheme. - Sentence identifiers, corrections to transcriptions and segmentation, and technical harmonisation of annotation tiers and metadata. - Additional CoNLL-U and VERT formats containing automatically generated morphological and syntactic annotations.

 Izdajatelj

Faculty of Electrical Engineering and Computer Science, University of Maribor

 

Jožef Stefan Institute

 Zahvala

ARIS (Slovenian Research and Innovation Agency) GC-0002 "LLM4DH: Large Language Models for Digital Humanities"

Ministry of Culture C3340-26-144004 "Spoken Slovenian 2.0: Portal upgrade and ROG training spoken corpus upgrade"

ARIS (Slovenian Research Agency) P2-0069 "Advanced methods of interaction in telecommuncations"

 Ključne besede

speech transcription speech recordings dialogue act spoken corpus spoken language sentiment classification humour

 Zbirke

CLARIN.SI data & tools

 Druge različice

Seznam vseh različic  

Prikaži polni zapis vnosa