Corpus of texts in 12 languages. For each language, we provide one training, one development and one testing set acquired from Wikipedia articles. Moreover, each language dataset contains (substantially larger) training set collected from (general) Web texts. All sets, except for Wikipedia and Web training sets that can contain similar sentences, are disjoint. Data are segmented into sentences which are further word tokenized.
All data in the corpus contain diacritics. To strip diacritics from them, use Python script diacritization_stripping.py contained within attached stripping_diacritics.zip. This script has two modes. We generally recommend using method called uninames, which for some languages behaves better.
The code for training recurrent neural-network based model for diacritics restoration is located at https://github.com/arahusky/diacritics_restoration.
In NLP Centre, dividing text into sentences is currently done with
a tool which uses rule-based system. In order to make enough training
data for machine learning, annotators manually split the corpus of contemporary text
CBB.blog (1 million tokens) into sentences.
Each file contains one hundredth of the whole corpus and all data were
processed in parallel by two annotators.
The corpus was created from ten contemporary blogs:
hintzu.otaku.cz
modnipeklo.cz
bloc.cz
aleneprokopova.blogspot.com
blog.aktualne.cz
fuchsova.blog.onaidnes.cz
havlik.blog.idnes.cz
blog.aktualne.centrum.cz
klusak.blogspot.cz
myego.cz/welldone
The corpus contains speech data of 2 Czech native speakers, male and female. The speech is very precisely articulated up to hyper-articulated, and the speech rate is low. The speech data with a highlighted articulation is suitable for teaching foreigners the Czech language, and it can also be used for people with hearing or speech impairment. The recorded sentences can be used either directly, e.g., as a part of educational material, or as source data for building complex educational systems incorporating speech synthesis technology. All recorded sentences were precisely orthographically annotated and phonetically segmented, i.e., split into phones, using modern neural network-based methods.
COSTRA 1.0 is a dataset of Czech complex sentence transformations. The dataset is intended for the study of sentence-level embeddings beyond simple word alternations or standard paraphrasing.
The dataset consist of 4,262 unique sentences with average length of 10 words, illustrating 15 types of modifications such as simplification, generalization, or formal and informal language variation.
The hope is that with this dataset, we should be able to test semantic properties of sentence embeddings and perhaps even to find some topologically interesting “skeleton” in the sentence embedding space.
Costra 1.1 is a new dataset for testing geometric properties of sentence embeddings spaces. In particular, it concentrates on examining how well sentence embeddings capture complex phenomena such paraphrases, tense or generalization. The dataset is a direct expansion of Costra 1.0, which was extended with more sentences and sentence comparisons.
Segment from Český zvukový týdeník Aktualita (Czech Aktualita Sound Newsreel) issue no. 9B from 1943 captures an ice sports course for mandatory youth service instructors, which was organised by the Board of Trustees for the Education of Youth as part of the Ice Sports Week event held at Štvanice Ice Arena in Prague from 1 to 6 February. Training in speed skating and ice hockey was led by hockey players Josef Maleček, Vladimír Zábrodský and Jiří Tožička. The event was attended by General Secretary of the Board František Teuner.
CsEnVi Pairwise Parallel Corpora consist of Vietnamese-Czech parallel corpus and Vietnamese-English parallel corpus. The corpora were assembled from the following sources:
- OPUS, the open parallel corpus is a growing multilingual corpus of translated open source documents.
The majority of Vi-En and Vi-Cs bitexts are subtitles from movies and television series.
The nature of the bitexts are paraphrasing of each other's meaning, rather than translations.
- TED talks, a collection of short talks on various topics, given primarily in English, transcribed and with transcripts translated to other languages. In our corpus, we use 1198 talks which had English and Vietnamese transcripts available and 784 talks which had Czech and Vietnamese transcripts available in January 2015.
The size of the original corpora collected from OPUS and TED talks is as follows:
CS/VI EN/VI
Sentence 1337199/1337199 2035624/2035624
Word 9128897/12073975 16638364/17565580
Unique word 224416/68237 91905/78333
We improve the quality of the corpora in two steps: normalizing and filtering.
In the normalizing step, the corpora are cleaned based on the general format of subtitles and transcripts. For instance, sequences of dots indicate explicit continuation of subtitles across multiple time frames. The sequences of dots are distributed differently in the source and the target side. Removing the sequence of dots, along with a number of other normalization rules, improves the quality of the alignment significantly.
In the filtering step, we adapt the CzEng filtering tool [1] to filter out bad sentence pairs.
The size of cleaned corpora as published is as follows:
CS/VI EN/VI
Sentence 1091058/1091058 1113177/1091058
Word 6718184/7646701 8518711/8140876
Unique word 195446/59737 69513/58286
The corpora are used as training data in [2].
References:
[1] Ondřej Bojar, Zdeněk Žabokrtský, et al. 2012. The Joy of Parallelism with CzEng 1.0. Proceedings of LREC2012. ELRA. Istanbul, Turkey.
[2] Duc Tam Hoang and Ondřej Bojar, The Prague Bulletin of Mathematical Linguistics. Volume 104, Issue 1, Pages 75–86, ISSN 1804-0462. 9/2015
CUBBITT En-Cs translation models, exported via TensorFlow Serving, available in the Lindat translation service (https://lindat.mff.cuni.cz/services/translation/).
Models are compatible with Tensor2tensor version 1.6.6.
For details about the model training (data, model hyper-parameters), please contact the archive maintainer.
Evaluation on newstest2014 (BLEU):
en->cs: 27.6
cs->en: 34.4
(Evaluated using multeval: https://github.com/jhclark/multeval)
Web corpus of Czech, created in 2011. Contains newspapers+magazines, discussions, blogs. See http://www.lrec-conf.org/proceedings/lrec2012/summaries/120.html for details. and GA405/09/0278
The Prague family of annotated corpora has a new member, the Czech Academic Corpus version 2.0 (CAC 2.0). CAC 2.0 consists of 650,000 words from various 1970s and 1980s newspapers, magazines and radio and television broadcast transcripts manually annotated for morphology and syntax.