Il Corpus dei Galatei Italiani dell’Ottocento (CGIO) è un corpus specialistico compilato da Annick Paternoster e Francesca Saltamacchia (Università della Svizzera italiana). Il CGIO contiene un campione rappresentativo della vasta produzione di galatei e manuali di etichetta in Italia durante l’Ottocento lungo, 1800-1920.

The primary source of data is a corpus of Swiss WhatsApp conversations: "What's up Switzerland?" (cf. https://www.whatsup-switzerland.ch/index.php/en/). The released data contains function- and emotion-based annotations obtained via a Citizen Science campaign for a number of emojis in the context of the Swiss German and French chats where they occurred, as well as demographic information about the citizen science annotators.

Sample of the WoPoss corpus annotated by the annotators of the WoPoss team (for a more precise responsibility description, see each file).

The sample contains the following works:

Octavius by M. Minucius Felix:
Text provided by Perseus Digital Library, with funding from The Mellon Foundation. Original version available for viewing and download at http://www.perseus.tufts.edu/hopper/
Text retrieved from: http://data.perseus.org/texts/urn:cts:latinLit:stoa0203.stoa001.perseus-lat1
Original printed edition: Tertullian Apology - De spectaculis - Minucius Felix Octavius. Translated by T.R. Glover & William Heinemann. Harvard University Press: London; Cambridge, Massachusetts. 1931.
Reference edition: Minucius Felix, Marcus. Octavius. Edited by Bernhard Kytzler. Bibliotheca scriptorum Graecorum et Romanorum Teubneriana. Berlin, Boston: De Gruyter. 1992.

Metamorphoses (First book) by P. Ovidius Naso:
Text provided by Perseus Digital Library, with funding from The Mellon Foundation. Original version available for viewing and download at http://www.perseus.tufts.edu/hopper/
Text retrieved from: http://data.perseus.org/texts/urn:cts:latinLit:phi0959.phi006.perseus-lat1
Original printed edition: Ovid. Metamorphoses. Edited by Hugo Magnus. Gotha, Germany: F.A. Perthes. 1892.
Reference edition: P. Ovidi Nasonis Metamorphoses. Edited by Richard Tarrant. Oxford Classical Texts. Oxford: Clarendon Press. 2004.

Satyricon by G. Petronius Arbiter:
Text provided by Perseus Digital Library, with funding from The Mellon Foundation. Original version available for viewing and download at http://www.perseus.tufts.edu/hopper/
Text retrieved from: http://data.perseus.org/citations/urn:cts:latinLit:phi0972.phi001.perseus-lat1:1
Original printed edition: Petronius Arbiter. Petronius. Edited by Michael Heseltine. London: William Heinemann. 1913.
Reference edition: Petronius Arbitrer. Satyricon reliquiae. Edited by Konrad Müller. Bibliotheca scriptorum Graecorum et Romanorum Teubneriana. Berlin, Boston: De Gruyter. 2004.

De spectaculis by Q. S. F. Tertullianus:
Text provided by Perseus Digital Library, with funding from The Mellon Foundation. Original version available for viewing and download at http://www.perseus.tufts.edu/hopper/
Text retrieved from: http://data.perseus.org/texts/urn:cts:latinLit:stoa0275.stoa027
Original printed edition: Tertullian Apology - De spectaculis - Minucius Felix Octavius. Translated by T.R. Glover & William Heinemann. Harvard University Press: London; Cambridge, Massachusetts. 1931.
Reference edition: Tertullianus, Quintus Septimius Florens. Les spectacles = (De spectaculis). Sources chrétiennes no 332. Paris: Les Editions du Cerf. 1986.

Senatus consultum de Bacchanalibus:
Text provided by Epigraphic Database Rome (EDR). Record: EDR169402, 24-04-2021 (Annarosa GALLO) http://www.edr-edr.it/edr_programmi/res_complex_comune.php?do=book&id_nr=EDR169402
Based on the edition: Gallo, Annarosa, “Senatus consulta de Bacchanalibus. Normenpluralität in der Tafel von Tiriolo und in der livianischen Überlieferung” in Die senatus consulta in den epigraphischen Quellen. Texte und Bezeugungen edited by Pierangelo Buongiorno and Giuseppe Camodeca. Stuttgart: Franz Steiner Verlag. 2021, pp. 107-145.


To cite single files or the results related to single files, please use the following format:

Octavius by M. Minucius Felix. How to cite: Dell’Oro, Francesca; Reymond, Séverine; Bermúdez Sabel, Helena (2022). Octavius by M. Minucius Felix annotated according to the WoPoss guidelines. Swiss National Science Foundation.
Metamorphoses (First book) by P. Ovidius Naso. How to cite: Dell’Oro, Francesca; Bermúdez Sabel, Helena (2022). First book of the Metamorphoses by P. Ovidius Naso annotated according to the WoPoss guidelines. Swiss National Science Foundation.
Satyricon by G. Petronius Arbiter. How to cite: Dell’Oro, Francesca; Marongiu, Paola; Bermúdez Sabel, Helena (2022). Satyricon by G. Petronius Arbiter annotated according to the WoPoss guidelines. Swiss National Science Foundation.
De spectaculis by Q. S. F. Tertullianus. How to cite: Dell’Oro, Francesca; Reymond, Séverine; Bermúdez Sabel, Helena (2022). De spectaculis by Q. S. F. Tertullianus annotated according to the WoPoss guidelines. Swiss National Science Foundation.
Senatus consultum de Bacchanalibus. How to cite: Dell’Oro, Francesca; Bermúdez Sabel, Helena (2022). Senatus consultum de Bacchanalibus annotated according to the WoPoss guidelines. Swiss National Science Foundation.

To cite the project and/or website as a whole please use the following format:

Francesca Dell’Oro, dir., WoPoss. A corpus to analyse the evolution of modality in the diachrony of the Latin language, Swiss National Science Foundation [https://woposs.unine.ch/].

In this dataset the user will find a collection of 76 interactive diachronic maps of the lexical modal markers analyzed in the framework of the WoPoss project (https://woposs.unine.ch/). The drawing of the maps is based on the descriptions of the lemmas in the Thesaurus Linguae Latinae, combined with the personal analysis of the relevant Latin passages carried out by the authors. As the Thesaurus Linguae Latinae is still under construction, for the lemmas nolo, (ne)queo, usus, valeo, volo and voluntas we relied on the Oxford Latin Dictionary.

The maps are in json format and can be visualized on the Pygmalion platform at https://woposs.unine.ch/map.html.

The dataset consists of audio recordings with metadata and content descriptions. Some recordings include transcripts with translations in Burmese and English. Part of the data has been glossed and annotated. The project is ongoing and more recordings are being transcribed, translated, and annotated continuously.
Initial data was collected in Shan State, Myanmar in 2019 and 2020 by Mathias Jenny, Rachel Weymuth, and Alexandra Herdeg, in collaboration with the Department of Anthropology, Mandalay University. The data is being edited and transcribed by Han Tin, and additional data is being collected online.

The dataset includes 198 audio files from 77 recording sessions. For each of the 198 sentences, a wav and a TexGrid are provided. The fricative data set consists of 293 tokens for each of the 77 recording sessions.

All audio files are automatically pre-processed using the Munich Automatic Segmentation System, MAUS (Kisler et al., 2017; Schiel, 1999) available online at https://www.bas.uni-muenchen.de/Bas/BasMAUS.html. During this process, TextGrid annotation files are generated.
The TextGrid output contains three tiers. The first tier shows the sentence segmented by word in Russian. In the second tier, the sentence is transcribed in Sampa. The third tier provides time-aligned word segmentation at a phonemic level. This tier served for the manual correction of the boundaries of the target fricatives using Praat version 3.9.
The audio files (Recording_1) were filtered below 80 and above 20050Hz with a smoothing of 80HZ to get rid of parasite noise.
It is to note that only the first recordings are so far manually corrected, and the 18 second recordings (Recording_2) are only automatically reprocessed.
In total there are 22561 fricatives in the database, of which 17287 are extracted and are available as individual wav files.

The dataset consists of original audio recordings, partly with transcripts (in Mon and/or IPA) and translations, as well as a few published texts in IPA transcription. The data was collected for descriptive linguistic purposes during a PhD project at the University of Zurich 2001-2005. This dataset contains the main recordings and text used in the PhD project. Additional recordings that were collected during and after the project time have not been systematically analyzed but are available on request.

In Experiment 1, two variables were manipulated: the distributive pattern of the matrix verb, with three levels: predicates selecting only the subjunctive, predicates selecting only the indicative and predicates accepting both moods, and the tense of the matrix verb, with two levels: present and past time. Experiment 1 had 30 experimental items.

In Experiment 2, two variables were manipulated: the distributive pattern of the matrix verb as in the first experiment, and the grammatical aspect of matrix verb, with two levels: perfective and imperfective aspects. Experiment 2 had 42 experimental items.

A multilingual corpus of movie subtitles aligned on the sentence-level. Contains data on more than 50 languages with a focus on the Indo-European language family. Morphosyntactic annotation (part-of-speech, features, dependencies) in Universal Dependency-style is available for 47 languages.

The Ruching Palaung dataset is the first of a series of original audio and video recordings of Austroasiatic languages of SA and SEA, partly with transcripts and annotations. The present dataset includes a selection of audio recordings and transcripts collected in Shan State, Myanmar.
The data was collected and processed mainly for linguistic purposes, but also contains cultural and social information in the form of interviews and descriptions of everyday life.
Data was collected by the team Mathias Jenny, Hiram Ring, We-Wei Lee, Rachel Weymuth, in cooperation with the Department of Anthropology, Mandalay University, and the Yangon University of Foreign Languages and supported by the SNSF. Data processing was assisted by Alexandra Herdeg.

Tableau comparatif des manuels étudiés selon les critères suivants:
-Titre
-Auteur(s)
-Années parution
-Pays édition
-Ville et maison d'édition
-Rééditions
-Changements dans les rééditions
-Spécialisation auteur
-Auteur-s russophone
-Méthode d'apprentissage
-Causes de création et parution
-Présentation du russe

These are Standard German daily news (Tagesschau) and Swiss German weather forecast (Meteo) episodes broadcast and interpreted into Swiss German Sign Language by hearing interpreters (among them, children of Deaf adults, CODA) via Swiss National TV (Schweizerisches Radio und Fernsehen, SRF) (https://www.srf.ch/play/tv/sendung/tagesschau-in-gebaerdensprache?id=c40bed81-b150-0001-2b5a-1e90e100c1c0). For a more extended description of the data, visit https://www.wmt-slt.com/data.

These are Standard German daily news (Tagesschau) and Swiss German weather forecast (Meteo) episodes broadcast and interpreted into Swiss German Sign Language by hearing interpreters (among them, children of Deaf adults, CODA) via Swiss National TV (Schweizerisches Radio und Fernsehen, SRF) (https://www.srf.ch/play/tv/sendung/tagesschau-in-gebaerdensprache?id=c40bed81-b150-0001-2b5a-1e90e100c1c0). For a more extended description of the data, visit https://www.wmt-slt.com/data.

This data collection consists of interviews in Hebrew that were recorded by Philipp Striedl between summer 2018 and early 2020 for the dissertation "Representations of Variation in Modern Hebrew in Israel: Cognitive Processes of Social and Linguistic Categorization" (Striedl 2022).
Process and structure of the data collection are described in detail in Chapter 3 of the dissertation.
Further notes about transcription conventions are included at the beginning of the text on pages xxi and xxii.

Striedl, P. (2022). Representations of variation in Modern Hebrew in Israel: Cognitive processes of social and linguistic categorization [PhD Thesis, LMU Munich]. https://doi.org/10.5282/edoc.29853

The ArchiMob corpus represents German varieties spoken on the territory of Switzerland. It is the first electronic resource containing long samples of transcribed text in Swiss German, intended to be used for studying spatial distribution of morphosyntactic features and for natural language processing.

This first release of the corpus, initially published in 2016, contains 34 transcribed interviews with a total of 528 381 tokens.

The ArchiMob corpus represents German varieties spoken on the territory of Switzerland. It is the first electronic resource containing long samples of transcribed text in Swiss German, intended to be used for studying spatial distribution of morphosyntactic features and for natural language processing.

This second release of the corpus, initially published in 2019, contains 43 transcribed interviews (i.e., 9 more than release 1) with a total of 581 976 tokens. We also make the audio files available with release 2.

The ArchiMob corpus represents German varieties spoken on the territory of Switzerland. It is the first electronic resource containing long samples of transcribed text in Swiss German, intended to be used for studying spatial distribution of morphosyntactic features and for natural language processing.

This dataset contains a sample of audio files for evaluation purposes. The samples represent small segments of speech from 5 different participants. The complete set of audio files is available on request.

The ArchiMob corpus represents German varieties spoken on the territory of Switzerland. It is the first electronic resource containing long samples of transcribed text in Swiss German, intended to be used for studying spatial distribution of morphosyntactic features and for natural language processing.

This dataset contains the complete audio files of the transcriptions available in release 2.

This dataset includes only behavioral results of stress processing tasks performed in Spanish by French- and German-speaking participants.

The collection of manuscript photographs presented here was provided by Arlo Griffiths, who assembled them during a series of field trips to various locations in Odisha (eastern India) in the years 1998 to 2001 and later added to.
The data were prepared for publication and supplemented by reference tables by Paul Widmer and Thomas Zehnder.

This resource was created in an exploratory study in which we compare how university students aged 19-25 (Experiment 1) evaluate the acceptability of objective and subjective relations expressed with French "car" and "parce que".

This dataset was created in an exploratory study in which we compare how native speakers, aged 20-70 and recruited via a crowdsourcing platform (Experiment 2), evaluate the acceptability of objective and subjective relations expressed with car and parce que.

The folder contains everything that is needed to reproduce findings, figures and tables presented in the following publication:
Krasselt, J., & Dreesen, Ph. (in press). Topic models indicate textual aboutness and pragmatics: Valuation practices in Islamophobic discourse. Journal of Cultual Analytics.

in detail:

==================================================
1. Script
==================================================

- R Script to reproduce figures and tables: supplementary_material_cultural_analytics.rmd
- the script is also provided as a commented html markdown version: supplementary_material_cultural_analytics.html
- to run the script, open the file supplementary_material_cultural_analytics.rmd in Rstudio, install the necessary packages and run each chunk

==================================================
2. LDA topic model
==================================================

- document-topic-distribution: doc_topics_df.csv
- topic list (top 20 words): top_words_df.csv
- word-topic-assignment: topic_word_assignment.csv (columns with actual tokens and lemmata were deleted due to copyright)


==================================================
3. Sura citations
==================================================

- a file containing 3grams counted for sura citations only: citations_suras_3grams.txt

A detailed readme-file describes all the data, documents, and publications that are available on SWISSUbase.
The data contain results (as csv-tables), descriptions of the methodology, but not audio-files for reasons of privacy.
Data are thus reusable and interoperable.

The dataset contains the Supplementary Information (two data tables and the R script) used for the analysis presented in the article "Daten und Zahlen zur Struktur rigvedischer Metren" published in the "Zeitschrift der deutschen morgenländischen Gesellschaft" 174.1 (2024), p. 193-219 by Salvatore Scarlata and Paul Widmer.

The first book of the Paippalāda Recension of the Atharvaveda comprises of 112 sūktas (hymns) with a total of 483 stanzas. Each stanza is provided with an English translation, detailed linguistic and Indological comments, critical apparatus, parallel passages in the Sanskrit literature and a full morpho-lexical analysis. This information is encoded in TEI-XML format.

This database contains more than 2200 clippings, their respective source words, and annotations for a wide range of variables that pertain to language structure and language use, as represented by corpus data.

The What's New, Switzerland? Corpus is a dataset of 72 authentic WhatsApp chats between 118 French-speaking users in Switzerland, collected in the framework of the "Evolving Language" NCCR. Chats were donated by users between August and October 2022. The data have been de-identified using a partly automated and partly manual workflow. Each chat is provided in two versions: an XML-TEI version (which includes extensive metadata about chats, users, and messages) and a plain text version. The dataset is available on demand for research purposes, under a restricted license contract.

The tenth book of the Paippalāda Recension of the Atharvaveda comprises 10 sūktas (hymns) with a total of 165 stanzas. Each stanza is provided with an English translation, detailed linguistic and Indological comments, critical apparatus, parallel passages in the Sanskrit literature and a full morpho-lexical analysis. This information is encoded in TEI-XML format.

The fourth book of the Paippalāda Recension of the Atharvaveda comprises 40 sūktas (hymns) with a total of 302 stanzas. Each stanza is provided with an English translation, detailed linguistic and Indological comments, critical apparatus, parallel passages in the Sanskrit literature and a full morpho-lexical analysis. This information is encoded in TEI-XML format

The twelfth book of the Paippalāda Recension of the Atharvaveda comprises ten sūktas (hymns) with a total of 198 stanzas. Each stanza is provided with an English translation, detailed linguistic and Indological comments, critical apparatus, parallel passages in the Sanskrit literature and a full morpho-lexical analysis. This information is encoded in TEI-XML format.

Der Datensatz enthält die originalgetreuen Schreibungen von 39 Schüler:innen von Anfang erster bis Anfang zweiter Klasse (Alter Anfang erste Klasse: 5 Jahre, 7 Monate - 7 Jahre, 1 Monat).

Die Schreibungen wurden zu folgenden Zeitpunkten erhoben:
Arbeitsblatt 1: 15./16. Schulwoche (1. Klasse)
Arbeitsblatt 2: 29./30. Schulwoche (1. Klasse)
Arbeitsblatt 3: 38./39. Schulwoche (1. Klasse)
Arbeitsblatt 4: 5./6. Schulwoche (2. Klasse)

This dataset is part of our research paper at LREC-COLING 2024:

Jiang, Zifan, Göhring, Anne, Moryossef, Amit, Sennrich, Rico, Ebling, Sarah. 2024. SwissSLi: The Multi-parallel Sign Language Corpus for Switzerland. In: Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). Online: https://aclanthology.org/2024.lrec-main.1342

Our study contains the sign language version of two sets of television programs in three spoken/sign languages released between 2020 and 2023 fall. They were created by the Swiss Broadcasting Corporation (SRG) in spoken language and then translated into sign language.

This dataset contains mitenand (German/DSGS), Ensemble (French/LSF-CH), and Insieme (Italian/LIS-CH) which report the work of charity organizations, are parallel and produced weekly. Most episodes originate from German and are then translated into other languages.

mitenand in Gebärdensprache:

https://www.srf.ch/play/tv/sendung/mitenand-in-gebaerdensprache?id=1d478b33-be72-4258-a330-c8c15bc1f6aa

Ensemble - signé:

https://www.rts.ch/play/tv/emission/ensemble-signe?id=12546856

Insieme LIS:

https://www.rsi.ch/play/tv/programma/insieme-lis?id=703974

This dataset is part of our research paper at LREC-COLING 2024:

Jiang, Zifan, Göhring, Anne, Moryossef, Amit, Sennrich, Rico, Ebling, Sarah. 2024. SwissSLi: The Multi-parallel Sign Language Corpus for Switzerland. In: Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). Online: https://aclanthology.org/2024.lrec-main.1342

Our study contains the sign language version of two sets of television programs in three spoken/sign languages released between 2020 and 2023 fall. They were created by the Swiss Broadcasting Corporation (SRG) in spoken language and then translated into sign language.

This dataset contains Helveticus in three languages which covers historical content with animations and is produced irregularly. Most episodes originate from German and are then translated into other languages.

Helveticus in Gebärdensprache:

https://www.srf.ch/play/tv/sendung/helveticus-in-gebaerdensprache?id=46863500-56d8-4414-a3d9-e84a6a843287

Helveticus signé:

https://www.rts.ch/play/tv/emission/helveticus-signe?id=10953398

Helveticus LIS:

https://www.rsi.ch/play/tv/programma/helveticus-lis?id=703913

This dataset resulted from an online self-paced experiment in which we tested how readers process sentences in which the French Passé Simple is employed in a perspectival manner within two types of subjective contexts: those with Free Indirect Discourse (FID) and those without FID markers. In addition, we tested whether the processing is influenced by the type of emotion expressed in these subjective contexts, whether it is positive or negative. The experiment was designed with E-prime. Each experimental item was composed of three sentences: the context, a sentence biasing the interpretation towards subjective or objective readings, and the target sentence. The dataset consists of the reading times for each of the three sentences composing each experimental item in each experimental condition.

This dataset resulted from a self-pacedreading experiment, in which we sought to compare the cost of inferring the presence of causal vs. temporal relations in the absence vs. presence of a connective indicating a given relation in French. For the explicit marking, two types of connectives were tested: one specialized for each relation ('donc' for causality and 'puis' for temporality) and one underspecified ('et' in its temporal and causal readings).

Includes test item list and cognate guessing test results, 163 research subjects at 100 answers per subject; R scripts used to analyse the data; description of the variables used in the survey data included in CSV format.

Zu diesem Datensatz gehören 9 Aufnahmen (6 aus dem Kanton Jura, 2 aus dem Kanton Neuenburg, 1 aus dem Kanton Bern).
Sprachen: Frankoprovenzalisch und Französisch.

Zu diesem Datensatz gehören 166 Aufnahmen (32 aus dem Kanton Bern, 23 aus dem Kanton Zürich, 22 aus dem Kanton Graubünden, 11 aus dem Kanton Freiburg, 9 aus dem Kanton Glarus, 8 dem Kanton Schwyz, 9 aus dem Kanton St. Gallen, 6 aus dem Kanton Thurgau, 6 aus dem Kanton Luzern, 5 aus dem Kanton Zug, 4 aus dem Kanton Aargau, 4 aus dem Kanton Appenzell Ausserrhoden, 4 aus dem Kanton Basel-Landschaft, 4 aus dem Kanton Uri, 3 aus dem Kanton Basel Stadt, 3 aus dem Kanton Solothurn, 2 aus dem Kanton Appenzell Innerrhoden, 2 aus dem Kanton Schaffhausen, 2 aus dem Kanton Tessin, 1 aus dem Kanton Wallis; 3 aus Italien, 2 aus Liechtenstein, 1 aus Österreich).
Sprache: Schweizerdeutsch, Deutsch.

Zu diesem Datensatz gehören 15 Aufnahmen aus dem Kanton Graubünden (CH).
Sprache: Rätoromanisch.

Zu diesem Datensatz gehören 10 Aufnahmen (8 aus dem Kanton Tessin; 2 aus Italien).
Sprache: lombardische Mundart.

This set contains the following data:
- Full Task Lab data set
- Task Lab component tests
- Task Lab questionnaire answers
- Task Lab reading test results
- Task Lab component tests problems and exercises
- Task Lab study questionnaires
- Task Lab reading test tasks
- Task Lab data treatment & student IDs
- Task Lab component tests items IDs & scoring instructions
- Task Lab questionnaire data analysis treatment
- Task Lab reading test information

The database is a collection of sentences aimed at investigating the use of the particle fii ‘in/at’ across Arabic varieties. It consists of 11 parallel datasets of Arabic sentences, their phonological transcription, morphological glosses, and their translation in English.
Sentences have a unique identifier encoding the Arabic variety that they represent, the phenomenon, and a progressive number, and this labeling system was devised to allow the comparison of the same sentence, or group of them, across varieties.
Each sentence is associated to a list of 3 numerical values expressed as either 0 or 1, where the two are respectively the transposition of the Boolean values false and true. The values have been collected with the help of language participants in one-to-one interviews conducted either in person or on Zoom. The Boolean values represent the answers given by a participant to questions about the interpretation of a given sentence.
For instance, given a sentence referred here for the purpose of this explanation as S1 and a potential translation of that same sentence, call it T1, a participant was presented with the stimuli and asked the question:
“Sentence T1 is a suitable translation for sentence S1, is that true or false?”
Each sentence of the dataset aims at investigating a linguistic variable assumed to be potentially relevant for the expression of one of the syntactic functions of the particle fii across Arabic varieties. Every variable of each phenomenon is targeted by two sentences. The variables targeted by the sentences is indicated in the databases too.
Finally, the list of stimuli, the targeted syntactic phenomena and the choice of variable depends on the findings described in the literature. See section: "bibliographical references".

Two Excel spreadsheets for each corpus containing 132 examples of disease curses with 'che ti venga NP'. The spreadsheets are also provided in a CSV format with UTF-8 coding.

This dataset features pictures of the grapholinguistic landscape of Novi Sad. It includes a graffiti subcorpus consisting of more than 90 pictures documenting script systems and ideological membership expression in graffiti.

This dataset consists in a questionnaire on heritage languages in Switzerland. The questionnaire comprises questions and innovative rating tasks about language capital and language prestige.
The questionnaire was developed between January 2024 and January 2025 and released in February 2025 by Dr. Cristiana Lucchetti at the University of Zurich.

Étude principale :
• Données orales (récits narratifs séquencé, avec structure informationnelle non prototypiques permettant d'elliciter différents types de relations anaphoriques, Finite Story, Dimroth 2006)
• Résultats d’un questionnaire sur la dominance langagière (Bilingual Language Profile, Birdsong, Gertken, et Amengual, 2012)
• Tests de compétences lexicales par des locuteurs francophones et/ou germanophones en Suisse (LexTaLE, Lemhöfer et Broersma, 2012)

(Les données issues de l'étude secondaire sur la perception de la structure informationnelle et l’organisation du discours par des locutrices et locuteurs francophones en Suisse ne sont pas archivées sur SWISSUBase car elles ne sont pas anonymes. Les détails sont disponibles dans la thèse de doctorat d'Audrey Bonvin)

- un dossier contentant des données orales (données liées à l'étude principal, donc qui concernent le dataset 1) qui comprend à son tour les sous-dossiers suivants :
• un dossier avec les productions orales en italien des locutrices/locuteurs bilingues italien/suisse-allemand
• un dossier avec les productions orales en suisse allemand des locutrices/locuteurs bilingues italien/suisse-allemand
• un dossier avec les productions orales des locuteurs monolingues italophones
• un dossier avec les productions orales des monolingues germanophones

- un dossier avec des données écrites, qui comporte à son tour les sous-dossiers suivants :
• un dossier avec les données concernant l’étude principale sur la production de la structure informationnelle et qui comprend des autres sous-dossiers avec les résultats du questionnaire Bilingual Language Profile (mesure utilisée comme proxy de la dominance langagière), les résultats des tests de vocabulaire (mesure utilisée comme proxy de la compétence linguistique) et un dossier avec les annotations des données orales
• un dossier avec les données concernant l’étude secondaire sur la perception de la structure informationnelle

PolarIs4CAUS is a dataset of causal statements derived from climate-related Twitter conversations. It includes manually annotated and curated expressions. The data set is aimed at enabling qualitative analysis and computational modeling of causality in climate debates. Ideal for researchers in argumentation, discourse studies, and NLP.

The dataset consists of data and experimental items for the three experiments reported in Blochowiak J., Grisot C. & Sander E. (2025) How does language shape formation of concepts? Empirical investigation of generics and conditionals in French. Lingua. Volume 324. https://doi.org/10.1016/j.lingua.2025.103959.

This dataset contains entries referring to a survey on language use, language attitudes and identities in the Russian-speaking diaspora of Israel and Germany. The survey was carried out in march 2021 and reached more than 700 participants.

The corpus of the dissertation focuses on the border speech of the Artigas Department (Uruguay), considering both an urban setting (the city of Artigas itself) and a rural one (La Guayubira/La Estiba). The fieldwork was conducted solely by Bárbara Garrido Sánchez-Andrade. The corpus consists of audio material along with the corresponding transcriptions, totaling more than 230,000 words and approximately 24 hours of audio. All material has been anonymized. A specific methodology was applied: semi-structured interviews as well as free conversations. This means that each interviewed speaker also participated in a free conversation with another interviewee.

The final sample is distributed as follows: for the urban setting, the third generation includes individuals aged sixty and above, while the second generation comprises those aged thirty-four to fifty-nine. The total number of speakers in the urban sample is thirteen, consisting of five women and eight men. For the rural setting, the total is seventeen speakers, with nine women and eight men.

In the rural context, a later adjustment was necessary due to the lower number of younger individuals. As a result, the third generation includes speakers aged fifty-five and above, and the second generation includes those aged thirty-four to fifty-four.

The age distribution is not uniform across both settings because, in the urban area, the population density of speakers who use the Artigas variety and reside in the city of Artigas is lower than in the rural setting. Therefore, although the initial plan was to apply the same age distribution for both the urban and rural areas (i.e., using the urban criteria for both), this adjustment proved necessary. This modification also allowed for a more balanced distribution of speakers from the third and second generations in the rural sample. Had the same age ranges from the urban context been maintained for the rural setting, the result would have been six speakers (instead of nine) for the third generation and eleven speakers (instead of eight) for the second generation in the rural area. Given the difficulty in finding suitable speakers for the purposes of this research, this adjustment was considered important.

Regarding the urban speakers, it is worth noting that not all of them come from the city center of Artigas. In fact, six are from the suburban area. The decision to group both the urban and suburban areas together under the label “urban” was based on the following reasons: first, the number of speakers exhibiting the border variety within the city is relatively low; second, the geographic distance between the urban and suburban areas is minimal (in the case of the Ayuí and Cerro Ejido neighborhoods, the distance does not exceed four kilometers), and residents frequently travel between these areas and the city.

The dataset contains:
• the main data table, RV_data.csv, with morphosyntactic, syntactic and metrical information on each Rigvedic word form, and
• a script, disticha.rmd, for the analysis of disticha in the main types of Rigvedic stanzas which were studied as an example for the application of the data table, resulting in the published article:
Salvatore Scarlata and Paul Widmer, Syntactic evidence for metrical structure in Rigvedic stanzas, Indo-European Linguistics 13 (2025), 1-21, doi:10.1163/22125892-bja10041, issn: 2212-5892.

In addition the dataset contains:
• a further data table, RV-polylex.csv, wherein all compounded word forms are analyzed, and
• some ancillary basic scripts for linking the two tables respectively for simplified representations: join.r resp. pivot01–03.r.

Finally, the dataset contains:
• a data table, RV-polylexREJECTS.csv, containing words for which it was not possible to assess them as compounded

HF4ATS (Human Feedback For Automatic Text Simplification) LLM checkpoints following supervised fine-tuning (SFT) and direct preference optimization (DPO) for German-language automatic news sentence simplification.

The pretrained models we performed SFT on were DiscoLeo-Llama-3-8B-Instruct, Llama-3.1-8B-Instruct, and LeoLM-Mistral-7B-Chat.

The SFT checkpoints resulted from fine-tuning on up to 2,800 data instances from HF4ATS-SFT (the companion dataset also available on SWISSUbase).

The DPO checkpoints resulted from preference alignment on up to 2,280 data instances from HF4ATS-DPO (the companion dataset also available on SWISSUbase).

The number of training instances seen by a given model checkpoint during either SFT or DPO is shown in its filename.

HF4ATS-SFT (Human Feedback For Automatic Text Simplification - Supervised Fine-Tuning) is a German-language collection of automatic news sentence simplifications suitable for supervised fine-tuning. It is composed of 5,200 complex-simple sentence pairs (3,600 train, 800 development, 800 test) drawn from DEplain-APA data (https://zenodo.org/records/8304430) and paired with randomly assigned automatic text simplification prompts.

HF4ATS-DPO (Human Feedback For Automatic Text Simplification - Direct Preference Optimization) is a German-language collection of news sentence simplification preferences suitable for preference alignment.

HF4ATS-DPO is composed of 6,018 preference annotations expressed on 3,009 ATS pairs (preferences_all.jsonl) as well as several subsets of these preferences. Each ATS pair was annotated at least once by an individual with an intellectual disability (indicated by "ta" in the dataset) and at least once by a text simplification professional (indicated by "ea" in the dataset). The checkpoints used to generate these preference pairs were 7-8B parameter LLMs tuned using HF4ATS-SFT data (these checkpoints as well as the SFT data are also available on SWISSUbase).

The preference pair subsets are as follows:

1. preferences_ie.jsonl: all preferences on ATS pairs with information equality according to pair creators.
2. preferences_iter.jsonl: all preferences from the four target annotators or two expert annotators with the highest inter-annotation agreement.
3. preferences_itra.jsonl: all preferences from the four target annotators or two expert annotators with the highest intra-annotation agreement.
4. preferences_<model_name>.jsonl: all preferences on ATS pairs generated by the <model_name> checkpoint (these checkpoints result from supervised fine-tuning on HF4ATS-SFT).

Additionally, we include all raw annotation files collected during the course of this project. These files include duplicated ATS pairs and pairs that had not been annotated at time of submission. They are aggregated and de-duplicated at the annotator-day level inside preferences_raw.jsonl. preferences_raw.jsonl also contains a field indicating a given annotation's reason for exclusion from preferences_all.jsonl (as well as all subsequent subsets).

CSV-file containing the data used for statistical analysis of correlations between questions types and discourse participants. This file is the result of the questions classification.

This dataset includes R-Code used to conduct and visualise the logistic regression analyses.

This dataset includes python scripts and test data for the QClass question classifier. It was developed by Gunilla Kaibel (Universität Zürich) and Seraina Betschart (Universität Zürich).

This dataset contains the complete 135 transcripts of the narrative task ("Frog Story") that were collected in Equatorial Guinea as part of the EG-Spanish project. This narrative task was elicited using visual input in the form of a short story consisting of 12 pictures. The story represents a shortened and adapted version of the story “Frog, where are you?” by Mercer Mayer (1969) (see documentation document).
The total of 135 Frog Stories has a duration of 6:33:30 (43,925 tokens, wherof 39,736 by the participants). The transcriptions have been prepared manually with EXMARaLDA.
The dataset presents the transcripts in the following extensions: .xml (ISOTEI), .rtf (PartiturView), .html (ReaderView), .exs (segmented), .exb. All files corresponding to one and the same participant have the same file name, which contains the code of the participant. The codes used to identify each participant provide the following information: project-internal number of participant, place of data collection (town/village), age, sex (M = male, F = female) (e.g., 001_Malabo33M = participant number 1, data collected in Malabo, 33 years old, male).

Der Datensatz besteht aus Anhängen der Dissertation mit dem Titel "Zeigen im pandemieskeptischen Diskurs. Bedingung, Praxis und Ordnung":
1) Inputdatei und Projektdatei des Zeigenetzwerks; 2) Kodebuch; 3) URL-Liste, die die Datengrundlage des PANSKEP-Korpus bildet

Existing speech corpora for phonetic analysis are typically recorded in controlled environments, have restricted sample sizes (in terms of voices and materials recorded), and frequently focus on read speech, making them less suitable for studying naturalistic speech and voice phenomena. The CANDOR corpus (Reece et al., 2023) was recently introduced as a large multimodal dataset of naturalistic conversations containing audio and video from 1656 dyadic conversations in English. This dataset presents CANDORspeech - a transcribed, force aligned and manually screened audio subset of the original CANDOR corpus. We outline the processing pipeline and the resulting CANDORspeech corpus with around 30 utterances from more than 900 speakers (more than 26,000 utterances in total). The corpus is suitable for large-scale phonetic analysis, voice feature extraction, and perceptual testing of naturalistic within-speaker variability.

This dataset contains one readme file, as well as all the open access research data from the project, generated out of intercomprehension tasks with think-aloud-protocols. The data has been generated during two different phases, and in the framework of two pilot studies. The research data is organised in 3 subfolders. A short description of all the items in each subfolder is to be found in the readme file.

This dataset contains a readme file with all of the necessary metadata, plus seven audio files (MP3) of the "Alltagsdialoge" ("Everyday Dialogues"). A list with the title of every audio file is to be found in the readme file. These are supplementary materials for beginner lessons in the area of German as a Second Language (DaZ).

Datensätze, Codebooks und Fragebögen aus dem Projekt "Beurteilung der Sprachkompetenz im DaF-Unterricht"

Dieser Datensatz umfasst alle ohne weitere Restriktionen öffentlich freigegebenen Daten zu dem Projekt "Fremdsprachen lehren und lernen im Zeichen der Mehrsprachigkeit", d.h. insbesondere das im Rahmen des Projekts erstellte merhsprachigkeitsdidaktische Material. Der Datensatz richtet sich an Forschende und Lehrende in diesem Bereich.

This dataset contains a series of documents produced by the Federal Statistical Office (FSO) that are currently publicly accessible. These documents are administrative letters and presentations on the collected data. A readme file is also included. The rest of the data from this project are not available in open access because they are sensitive, private, and/or not anonymized. An exhaustive list of these files is to be found in the readme file.

Enthält alle öffentlich verfügbaren Daten zum Projekt "Literatursynthese Sprachaustausch in der obligatorischen Schulzeit und der Sekundarstufe II (LiSa)". Sie umfassen die Dokumentenvariablen und das Codebuch, welche der Studie zugrunde lagen.

Dieser Datensatz umfasst eine Präsentation mit Videos, die in der Studie als Stimuli verwendet wurde, um Beschreibungen von Bewegungsereignissen zu erheben. Eine ausführliche Beschreibung und Anleitung liegt dem Datensatz bei: Die Datei Readme.md erklärt, wie die Präsentation konfiguriert werden kann, während die Datei description_stimuli.pdf den Aufbau des Experiments erläutert.

This dataset consist of a PDF/A document with the metadata of the project's datasets, as well as a readme file with general information on the project. The research data of this project is stored in paper format in the archives of the Institute of Multilingualism and it is not accessible, since it is sensitive/private and it was not possible to anonymise it.

The FreCaDiS corpus consists of a total of 1081 excerpts (52527 words): 200 from blogs, 220 from contemporary literature, 221 from web discussions, 205 from SMS and 215 from written press. The excerpts were randomly collected through search using “car” and “parce que” as keywords in Sketch Engine (blogs and web discussions), the Corpus of novels from CENTAL (Université Catholique de Louvain) (contemporary literature), the Belgian SMS Corpus (Cougnon 2012) and the Le Monde Corpus (year 2012). The shared dataset includes only annotations of subjectivity and discourse relations (explanation vs. justification) for sentences containing the causal connectives "parce que" and "car" extracted from the FreCaDiS corpus. Excerpts cannot be shared with open access due to copyright restrictions.

Ethnographische Daten in Deutsch: Feldnotizen aus den teilnehmenden Beobachtungen, Interviewtranskripte.

Dieser Datensatz enthält eine Übersicht der Stichprobe, den pseudonymisierten Gesamtdatensatz für die Befragungen t0 und t1 der Schülerinnen und Schüler sowie der Lehrpersonen im SPSS- und CSV-Format und die SPSS-Syntax der Datenanalyse für den Schlussbericht. Er enthält weiterhin die Fragebögen zur Studie und weitere Daten, die eine Kontextualisierung des Projekts ermöglichen.

The corpus comprises texts from high-reach and regional Swiss daily and weekly newspapers in German, obtained through the Swiss Media Database (swissdox@LiRi).

Digital translation tools and dictionaries have become an indispensable part of language use. DeepL, Leo and Co. are used extensively and for various purposes, however, their benefits for foreign language teaching and learning remain contested. Some consider these tools to be of educational value, while others express their doubts on the sustainability of their contribution to language learning. This project examines the advantages and limitations of digital translation aids with regards to the following questions:
- To what extent do digital translation tools contribute to language learning?
- How do they influence the development of language skills, in particular writing and vocabulary learning?
- What learner groups are likely to benefit particularly from these tools?

This dataset includes data from two field experiments conducted at vocational schools for business, management, and services, where students in the federal vocational baccalaureate programme are required to learn a national language and English. The experiments examined the effects of different digital vocabulary training methods including online dictionaries and online flashcard systems.

Dieser Datensatz enthält folgende Dateien:
- ReadMe
- Vollständiges analoges Testheft (Deutsch)
- Testheft für die schriftlichen Aufgaben (Arabisch, Deutsch, Tigrinya)
- Ergebnisse der drei Subtests (Deutsch)
- Ausgewählte kontrastive Ergebnisse (Arabisch, Deutsch, Tigrinya)
- Beschreibung der Ergebnissdateien im ReadMe
- Skripte für die verwendete Software (PsychoPy und SpeechRecorder) sowie visuelle Stimuli
- Einverständniserklärung in Leichter Sprache (Deutsch, Englisch)

Nicht enthalten sind die auditive und audiovisuelle Stimuli sowie die Sprachaufnahmen der Teilnehmenden. Ausserdem sind die soziodemographischen Informationen der Teilnehmenden ebenfalls nicht im Datensatz enthalten. Diese Daten sind auf Anfrage erhältlich.

Data from the studies conducted within the project "Language aptitude: why and how to assess it?". Includes questionnaires, a codebook, audio files used as testing instruments and various files related to the organization of the tests.

This dataset contains videos of DSGS L2 learners who did not give consent for their data to be shared under a CC licence. The participants completed the same exercises as the native signers included in the L1 subcorpus. The dataset includes the original video recordings, corresponding MediaPipe videos, and annotations for two exercises. All data were prepared following appropriate data management and ethical guidelines, and relevant preprocessing steps were applied to ensure consistency and usability for research purposes as described in the data statements.

Usage Information: This dataset contains .xz compressed files. While their content is safe, the .xz format may be blocklisted by your institution's IT infrastructure, and handling these files could result in an account suspension. Consider recompressing them locally before use.

This dataset includes data related to word stress detection task performed in Spanish by French-speaking and Korean ab initio and experienced learners of Spanish.

The dataset consists of two CSV files containing all the corpora dedicated to French as a foreign language (FLE) and German as a foreign language (DaF), respectively. The dataset also includes a readme file with metadata on the project, as well as information about the CSV files.

We present a multimodal speech-production dataset combining simultaneous electromagnetic articulography (EMA; 1,250 Hz), electroencephalography (EEG; 2,048 Hz), and 48 kHz audio from 29 German-speaking adults. All participants have external craniofacial anthropometry; an anatomical subset (N = 18) adds acoustic pharyngometry, rhinometry, and 3D head meshes. Speech materials include alternating and sequential motion-rate syllable sequences at habitual and maximally fast rates with high trial counts; EMA + audio also cover passage reading, sustained vowels, palate tracing, and non-speech oromotor actions. EEG was recorded for syllable blocks. Cross-modal synchronisation is achieved via hardware triggers, yielding sub-millisecond alignment. All streams are released as raw and minimally processed files with a stable event-code map, machine-readable metadata, and open-source Python scripts (with a reproducible container) for data loading, synchronisation, and minimal preprocessing. This resource supports analyses of neural activity time-locked to articulatory onsets, gesture sequencing and rate effects, benchmarking of overt-speech EEG artefact handling, and studies linking vocal-tract anatomy to articulatory dynamics.

This dataset contains videos of DSGS productions by native signers and may serve as a control group for studies. The participants completed the same exercises as the learners included in the L2 subcorpus. The dataset includes the original video recordings, corresponding MediaPipe videos, and annotations for two exercises. All data were prepared following appropriate data management and ethical guidelines, and relevant preprocessing steps were applied to ensure consistency and usability for research purposes as described in the data statements.

Usage Information: This dataset contains .xz compressed files. While their content is safe, the .xz format may be blocklisted by your institution's IT infrastructure, and handling these files could result in an account suspension. Consider recompressing them locally before use.

A readme in this dataset contains general information on the project, on the methodology, on the data included and on the data not included.

This dataset contains videos of Phase 2/4 of DSGS L2 learner productions. The participants completed the same exercises as the native signers included in the L1 subcorpus. The dataset includes the original video recordings, corresponding MediaPipe videos, and annotations for two exercises. All data were prepared following appropriate data management and ethical guidelines, and relevant preprocessing steps were applied to ensure consistency and usability for research purposes as described in the data statements.

Usage Information: This dataset contains .xz compressed files. While their content is safe, the .xz format may be blocklisted by your institution's IT infrastructure, and handling these files could result in an account suspension. Consider recompressing them locally before use.

This dataset contains videos of Phase 3/4 of DSGS L2 learner productions. The participants completed the same exercises as the native signers included in the L1 subcorpus. The dataset includes the original video recordings, corresponding MediaPipe videos, and annotations for two exercises. All data were prepared following appropriate data management and ethical guidelines, and relevant preprocessing steps were applied to ensure consistency and usability for research purposes as described in the data statements.

Usage Information: This dataset contains .xz compressed files. While their content is safe, the .xz format may be blocklisted by your institution's IT infrastructure, and handling these files could result in an account suspension. Consider recompressing them locally before use.

Dieses Release enthält die SAC-Bände der Jahre 1864 bis 2015, die EdA-Bände 1872 bis 1924, sowie die Alpine Journal-Bände von 1969 bis 2008. Es handelt sich um:
- Das Jahrbuch des SAC (1864-1923, gemischt-sprachlich)
- Das Echo des Alpes (1872-1924, französisch)
- Die ALPEN (1925-1956, gemischt-sprachlich)
- Die ALPEN (1957-2011, parallel deutsch und französisch)

Seit 1957 gibt es Die ALPEN übersetzt in parallelen deutschen und französischen Versionen. Dieses Release enthält sowohl die deutschen wie auch die französischen Bände seit 1957. Die parallelen Bände enthalten grösstenteils übersetzte Artikel, aber (von 1957 bis 1982) auch identische Artikel. Wir haben die Duplikate aus den französischen Bänden entfernt und durch einen Verweis auf den jeweils entsprechenden Artikel im deutschen Band ersetzt. Ausserdem enthalten die übersetzten Artikel je einen Verweis auf den anderssprachigen Artikel im Parallelband, also eine Alignierung der übersetzten Artikel. Wenn die Information verfügbar ist, haben wir die Originalsprache vermerkt (bis 1982).
- Die ALPEN (2012-2015, parallel deutsch und französisch und italienisch).
- Die Jahrbücher des britischen Alpine Club "The Alpine Journal" (1969 bis 2008, englisch).
In den Jahren 1870, 1915 und 1924 sind keine SAC-Bände erschienen. Deshalb enthält das aktuelle Release insgesamt 302 Jahres-Bände:
- 90 Bände von 1864 bis 1956 in je einer Version,
- plus 55 Bände von 1957 bis 2011 in sowohl deutscher und französischer Version,
- plus 4 Bände von 2012 bis 2015 in deutscher, französischer und italienischer Version,
- die 53 Bände Echo des Alpes von 1872 bis 1924,
- die 40 Bände des Alpine Journal von 1969 bis 2008.
Insgesamt enthält das Release nach der Tokenisierung 53,3 Millionen Wörter.

18,221 lexical items (1000 are reserved for test) in Swiss German Sign Language, French Sign Language of Switzerland, and Italian Sign Language of Switzerland represented as videos and glosses along with signed example sentences (videos) as well as corresponding spoken language translations.

https://signsuisse.sgb-fss.ch/

Test set for WMT-SLT 23 SRF:

These are Standard German daily news (Tagesschau) and Swiss German weather forecast (Meteo) episodes broadcast and interpreted into Swiss German Sign Language by hearing interpreters (among them, children of Deaf adults, CODA) via Swiss National TV (Schweizerisches Radio und Fernsehen, SRF) (https://www.srf.ch/play/tv/sendung/tagesschau-in-gebaerdensprache?id=c40bed81-b150-0001-2b5a-1e90e100c1c0). For a more extended description of the data, visit https://www.wmt-slt.com/data.

This dataset contains videos of Phase 1/4 of the DSGS L2 learner productions. The participants completed the same exercises as the native signers included in the L1 subcorpus. The dataset includes the original video recordings, corresponding MediaPipe videos, and annotations for two exercises. All data were prepared following appropriate data management and ethical guidelines, and relevant preprocessing steps were applied to ensure consistency and usability for research purposes as described in the data statements.

Usage Information: This dataset contains .xz compressed files. While their content is safe, the .xz format may be blocklisted by your institution's IT infrastructure, and handling these files could result in an account suspension. Consider recompressing them locally before use.

This dataset contains videos of Phase 4/4 of DSGS L2 learner productions. The participants completed the same exercises as the native signers included in the L1 subcorpus. The dataset includes the original video recordings, corresponding MediaPipe videos, and annotations for two exercises. All data were prepared following appropriate data management and ethical guidelines, and relevant preprocessing steps were applied to ensure consistency and usability for research purposes as described in the data statements.

Usage Information: This dataset contains .xz compressed files. While their content is safe, the .xz format may be blocklisted by your institution's IT infrastructure, and handling these files could result in an account suspension. Consider recompressing them locally before use.

This data set is the first annotated corpus of spoken (Swiss) German of bilingual Albanian-German speakers in Zurich collected in 2019 as part of an SNSF-funded project on migrant and heritage Albanian in Zurich and Munich (SNSF grant number 182126). Communicative context is kept constant by including only one type of linguistic practice, i.e., renarrations of Wallace Chafe’s video Pear Story (http://pearstories.org). In total, the corpus encompasses 56 transcribed and grammatically annotated renarrations in (Swiss) German from Zurich.

The dataset provides transcribed German Pear Story renarrations of migrant and heritage speakers of Albanian living in Munich (recorded in 2019). Transcripts are automatically annotated with part-of-speech (POS) tags, syntactic dependencies, and morphological features, similar to the Universal Dependencies, and stored in CoNLL-U files.

These are the finished .csv language files of the PosiCat database. They are tables containing hand-collected information from language descriptions that are generally available. These language files may be updated whenever necessary.

The Sparcling corpus is built on top of a cleaned version of the Europarl corpus. (https://pub.cl.uzh.ch/wiki/public/PaCoCo/Sparcling)

617 WhatsApp chats gathered in the summer 2014, with permission to be used, in all four national languages of Switzerland: Swiss German dialect, non-dialectal German, French, Italian and varieties of Romansh.
763644 messages, 944 informants, 5155476 tokens, 382116 emojis.

Sub-corpus of WUS where Romansch provides the most messages and including demographic information from all communication partners.

Large-scale and multilingual corpus of WhatsApp messages for linguistic research. In this sub-corpus, argument drop has been manually annotated according to http://www.unige.ch/lettres/linge/syntaxe/journal/Volume11/11_Stuntebeck_2018.pdf

BlogSpot blogs.
NOAH’s Corpus of Swiss German Dialects consists of various text genres, manually annotated with Part-of-Speech tags.

Sub-corpus of WUS_ARGDROP where French provides the most messages.

Le corpus Oral de Français de Suisse Romande.

Sub-corpus of WUS_ARGDROP where Italian provides the most messages.

Sub-corpus of WUS. Chats that are either smaller than 100 messages or where the majority of messages are not in a national language.

The Swissdox@LiRI database comprises 23+ million published media articles in German, French and Italian from a wide range of Swiss media sources (both print and digital) covering many decades.

Sub-corpus of WUS_ARGDROP where Romansch provides the most messages.

Sub-corpus of WUS. Chats that are either smaller than 100 messages or where the majority of messages are not in a national language, and where we have demographic information from all communication partners.

Sub-corpus of WUS where Romansch provides the most messages

Sub-corpus of WUS where non-dialectal German provides the most messages

The "Sammlung Schweizerischer Rechtsquellen" (SSRQ) corpus is a comprehensive historical collection of legal documents originating from the territory of modern-day Switzerland, spanning from the Early Middle Ages to 1798. Its contents include a vast array of city laws, rural statutes, court records, and administrative acts that document the evolution of Swiss legal and social life. The corpus provides a rich, multilingual dataset featuring sources in Latin as well as various historical stages and dialects of German, French, and Italian. This diversity makes it a primary resource for studying the development of regional languages and the intersection of legal terminology across different linguistic traditions.

The Swiss SMS corpus consists of 25'947 SMS (~650'000 tokens), which were sent in by the Swiss public in 2009/2010. Of all SMS, 41% are in Swiss German (dialect), 28% in non-dialectal German, 18% in French, 6% in Italian, and 4% in Romansh. Click the URL for more information.

Sub-corpus of WUS where non-dialectal German provides the most messages and including demographic information from all communication partners.

Sub-corpus of WUS where French provides the most messages

This is a collection of all speeches by Donald Trump from June 2018 to April 14, 2026. Downloaded from https://rollcall.com/ Factba.se. In speeches where other speakers are involved, only the parts uttered by Trump are integrated.

SWATCH annual report.
NOAH’s Corpus of Swiss German Dialects consists of various text genres, manually annotated with Part-of-Speech tags.

Sub-corpus of WUS where French provides the most messages and including demographic information from all communication partners.

The ArchiMob corpus represents German linguistic varieties spoken within the territory of Switzerland. This corpus is the first electronic resource containing long samples of transcribed text in Swiss German, intended for studying the spatial distribution of morphosyntactic features and for natural language processing.

Newspaper articles from “Blick am Abig”, Version Zurich, Nr. 97, 28. Mai 2013.
NOAH’s Corpus of Swiss German Dialects consists of various text genres, manually annotated with Part-of-Speech tags.

The British National Corpus (BNC) is a 100-million-word text corpus of samples of written and spoken English from a wide range of sources.

Articles from the Alemannic Wikipedia.
NOAH’s Corpus of Swiss German Dialects consists of various text genres, manually annotated with Part-of-Speech tags.

Sub-corpus of WUS where dialectal German provides the most messages

Sample of the TIGR corpus curated for test purposes

Digitale Edition des Die Alpen, Les Alpes, Le Alpi 1925-2014, The Alpine Journal 1969-2008

Digitale Edition des Echo des Alpes 1872-1924

Sub-corpus of WUS where Italian provides the most messages

Digitale Edition des Jahrbuch des SAC 1864-1923.
This corpus includes articles in High German, Swiss German, French, Italian and Romansch

Extracts of criminal novels by Viktor Schobinger.
NOAH’s Corpus of Swiss German Dialects consists of various text genres, manually annotated with Part-of-Speech tags.

Sub-corpus of WUS where Italian provides the most messages and including demographic information from all communication partners.

A preprocessing and encoding tool designed for the Corpus Workbench (CWB) and the LiRI Corpus Platform (LCP).

Pretrained CANINE model on Swiss German using a masked language modeling (MLM) objective. It was introduced in the paper CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation by Google.

ParZu is a dependency parser for German. This means that it analyzes the linguistic structure of sentences and, among other things, identifies the subject and object(s) of a verb.

The xlm-roberta-base model trained on Swiss German text data via continued pre-training.

Zmorge is a morphology tool that combines a lexicon automatically extracted from Wiktionary with a modified version of the finite-state morphological grammar SMOR. The extraction script is open source, enabling new lexicon versions to be extracted from future, expanded Wiktionary versions.

SignCLIP re-purposes CLIP (Contrastive Language-Image Pretraining) to project spoken language text and sign language videos, two classes of natural languages of distinct modalities, into the same space. SignCLIP is an efficient method of learning useful visual representations for sign language processing from large-scale, multilingual video-text pairs.

Attention-based encoder-decoder model for neural machine translation built in Tensorflow.

Bleualign is a tool to align parallel texts (i.e. a text and its translation) on a sentence level.

A Python application for transforming keystroke logs into linguistic units. It extracts all intermediate text versions produced during writing (text history) and tracks the evolution of sentences within the text (sentence histories).

Mobile (Android/iOS) app enabling participation in cloze tests, lexical decision tasks, multiple-choice reading comprehension, n-back working memory tasks, picture naming, and reaction time tests in English, German, French, and Italian.

A part-of-speech tagging model for Swiss German, trained on Universal POS tags (upos), supporting dependency parsing and lemmatisation for NLP tasks.

SwissBERT is a masked language model for processing Switzerland-related text. It has been trained on more than 21 million Swiss news articles retrieved from Swissdox@LiRI.

NMTScore is a library of translation-based text similarity measures, providing reference-free evaluation by scoring translations based on neural machine translation models.

A part-of-speech tagging model for Swiss German trained on STTS POS tags, designed for dependency parsing and lemmatisation.

Python-based commandline-tool to extract collocations from VERT formatted corpora.

SwissADT is the first audio description translation system implemented for three main Swiss languages and English. By collecting well-crafted AD data augmented with video clips in German, French, Italian, and English, and leveraging the power of Large Language Models (LLMs), it aims to enhance information accessibility for diverse language populations in Switzerland by automatically translating AD scripts to the desired Swiss language.

A Jupyter notebook demonstrating a comprehensive pipeline for Sign Language Processing (SLP). It converts sign language videos into compact, machine-readable representations through pose estimation, sign segmentation, and gloss recognition. Provides a practical guide for researchers and developers building SLP systems using current tools and datasets.

Unsupervised Word Segmentation for Neural Machine Translation and Text Generation

A Python package that provides a basic dictionary-based lemmatizer and part-of-speech tagger for the Romansh language. It supports the five standard Romansh idioms (Sursilvan, Sutsilvan, Surmiran, Puter, and Vallader) as well as Rumantsch Grischun by looking up words in the Pledari Grond dictionaries.

clevertagger is a German part-of-speech tagger based on a CRF tool and SMOR. Its main component is a module that extracts features from SMOR's morphological analysis.

A web application and model collection enabling state-of-the-art machine translation between different Romansh idioms, as well as between Romansh and German.