Common Language Resources and Technology Infrastructure - Slovenia
Not a member yet
840 research outputs found
Sort by
GLOBAL French-Chinese Dictionary - MLDS (ELEXIS)
No full textA general language French to Chinese dictionary
Middle High German Dictionary by Benecke, Müller, Zarncke - BMZ (ELEXIS)
No full textMittelhochdeutsches Wörterbuch von Benecke, Müller, Zarncke. The Middle High German Dictionary by Benecke, Müller and Zarncke captures the Middle High German language and was published in the middle oft he 19th century (1854-1861)
Middle High German Dictionary + Supplements to the Middle High German Dictionary - Lexer (ELEXIS)
No full textMittelhochdeutsches Handwörterbuch von Matthias Lexer + Nachträge zum Mittelhochdeutsches Handwörterbuch von Matthias Lexer. The Middle High German Dictionary of Matthias Lexer published 1872-1878 captures the Middle High German language. It was created as an alphabetical index to the Middle High German Dictionary by Benecke, Müller and Zarncke (BMZ) and can therefore be seen as a supplement to it.
Please note that the Lexer itself consists of two parts: the main dictionary and the Lexer supplements
The CLASSLA-StanfordNLP model for morphosyntactic annotation of standard Croatian 1.2
No full textThe model for morphosyntactic annotation of standard Croatian was built with the CLASSLA-StanfordNLP tool (https://github.com/clarinsi/classla-stanfordnlp) by training on the hr500k training corpus (http://hdl.handle.net/11356/1183) and using the CLARIN.SI-embed.hr word embeddings (http://hdl.handle.net/11356/1205). The model produces simultaneously UPOS, FEATS and XPOS (MULTEXT-East) labels. The estimated F1 of the XPOS annotations is ~94.1.
The difference to the previous version of the model is that the pre-trained embeddings are limited to 250 thousand entries and adapted to the new code base
Spoken corpus Gos VideoLectures 4.2 (transcription)
No full textGos VideoLectures is an add-on to the Gos reference corpus of spoken Slovene (http://hdl.handle.net/11356/1040), and covers public academic speech. It can be used for training continuous speech recognition for Slovene language, for phonetic research or any other research of Slovene academic speech. The corpus contains a selection of public lectures available through the web portal Videolectures.net provided by the Jožef Stefan Institute, and covers 55 lectures and 22 hours of speech. This resource contains only annotated transcriptions of the corpus, while the audio recordings are available at http://hdl.handle.net/11356/1222.
The transcriptions for Gos VideoLectures were done manually and carefully checked. The main guidelines for transcription were those of the Gos corpus (http://www.korpus-gos.net/Support/About). The transcription tool Transcriber 1.5.1 (http://trans.sourceforge.net/en/presentation.php) was used for making transcriptions. It can be also used for reading or exporting transcriptions (.trs files) to different formats.
The transcriptions comprise the TRS files with tabular metadata, their conversion to TEI and to vertical file format (as used e.g. by Sketch Engine). Each recording has two TRS files, one with pronunciation-based and the other with the standardised/normalised transcription. The TRS zip also contains files with automatically produced word and phone-level alignment with the speech signal, as well as the annotation guidelines (in Slovenian).
The TEI and vertical encodings join the two transcriptions at the token level, with the normalised words also linguistically annotated. The annotations comprise the word lemma, the MULTEXT-East MSDs and the Universal dependencies morphological features.
As opposed to version 4.1, this version corrects some errors and slightly changes the TEI and vertical encodings
Italian YouTube Hate Speech Corpus
No full textWe present an Italian YouTube dataset manually annotated for hate speech types and targets. The comments to be annotated were sampled from the Italian YouTube comments on videos about the Covid-19 pandemic in the period from January 2020 to May 2020. Two sets were annotated: a training set with 59,870 comments (IMSyPP_IT_YouTube_comments_train.csv) and an evaluation set with 10,536 comments (IMSyPP_IT_YouTube_comments_evaluation.csv). The dataset was annotated by 8 annotators with each comment being annotated by two annotators. It was used to train a classification model for hate speech types detection that is publicly available at the following URL: https://huggingface.co/IMSyPP/hate_speech_it.
The dataset consists of the following fields:
ID_Commento - YouTube ID of the comment
ID_Video - YouTube ID of the video under which the comment was posted
Testo - text of the comment
Tipo - type of hate speech
Target - the target of hate speech
Additionally, we have included the Italian YouTube data (SR_YT_comments.csv) which was collected in the same period as the training data and was annotated using the aforementioned model. The automatically labeled data was used to analyze the relationship between hate speech and misinformation on Italian YouTube. The results of this analysis are presented in the associated paper.
The analyzed data are represented with the following fields:
ID_Commento - YouTube ID of the comment
Label - automatically assigned label by the model
is_questionable - the type of channel where the comment was collected from; the channels could either be categorized as spreading reliable or questionable information
The corpus of older Slovenian narrative prose PriLit 1.0
No full textThe PriLit corpus contains 37 texts of older Slovenian narrative prose by 12 authors. One text, Sreča v nesreči (Fortune in Misfortune) by Janez Cigler (first published in 1836), is present in 7 editions, leading to 43 texts in the corpus. The texts were published 1643 - 1991, or, excluding reprints, 1643 - 1866. The corpus is encoded in two variants: one contains the corpus encoded in TEI, while the other adds automatic linguistic annotations comprising word modernisation, lemmatisation, and syntactic annotation following the Universal Dependencies formalism for Slovenian. Besides the two TEI encoded version, the corpus is also available in derived plain text and in vertical format
The CLASSLA-StanfordNLP model for morphosyntactic annotation of standard Serbian 1.2
No full textThe model for morphosyntactic annotation of standard Serbian was built with the CLASSLA-StanfordNLP tool (https://github.com/clarinsi/classla-stanfordnlp) by training on the SETimes.SR training corpus (http://hdl.handle.net/11356/1200) and using the CLARIN.SI-embed.sr word embeddings (http://hdl.handle.net/11356/1206). The model produces simultaneously UPOS, FEATS and XPOS (MULTEXT-East) labels. The estimated F1 of the XPOS annotations is ~95.2.
The difference to the previous version of the model is that the pre-trained embeddings are limited to 250 thousand entries and adapted to the new code base
Choice of plausible alternatives dataset in Croatian COPA-HR
No full textThe COPA-HR dataset (Choice of plausible alternatives in Croatian) is a translation of the English COPA dataset (https://people.ict.usc.edu/~gordon/copa.html) by following the XCOPA dataset translation methodology (https://arxiv.org/abs/2005.00333). The dataset consists of 1000 premises (My body cast a shadow over the grass), each given a question (What is the cause?), and two choices (The sun was rising; The grass was cut), with a label encoding which of the choices is more plausible given the annotator or translator (The sun was rising).
The observed agreement of the English annotator and the Croatian translator is perfect on the training and the validation dataset, with one different label (agreement of 99.8%) on the test dataset.
The current state-of-the-art on this dataset is held by the BERTić model (https://huggingface.co/CLASSLA/bcms-bertic), achieving an accuracy of 66% (50% is random)
Frequency lists of collocations from the Gigafida 2.1 corpus
No full textFrequency lists of collocations were extracted from the Gigafida 2.1 Corpus of Written Standard Slovene (https://www.clarin.si/noske/run.cgi/corp_info?corpname=gfida21) using specialised scripts for extraction of data from syntactically parsed corpora.
The lists contain collocations with absolute frequency 10 and above, split into files corresponding to 81 predefined syntactic structures. The formal description of syntactic structures with information on restrictions and representations applied to POS and dependency parsing annotations is included in the dataset.
The lists are sorted according to absolute frequency of collocations and include frequency information on individual lemmas, together with the most frequent representative forms of combined lemmas. The lists also include calculation of logDice score for collocations, and the number of distinct forms of lemmas appearing in corpus hits for a particular collocation