LiFR-Law is a corpus of Czech legal and administrative texts with measured reading comprehension and a subjective expert annotation of diverse textual properties based on the Hamburg Comprehensibility Concept (Langer, Schulz von Thun, Tausch, 1974). It has been built as a pilot data set to explore the Linguistic Factors of Readability (hence the LiFR acronym) in Czech administrative and legal texts, modeling their correlation with actually observed reading comprehension. The corpus is comprised of 18 documents in total; that is, six different texts from the legal/administration domain, each in three versions: the original and two paraphrases. Each such document triple shares one reading-comprehension test administered to at least thirty readers of random gender, educational background, and age. The data set also captures basic demographic information about each reader, their familiarity with the topic, and their subjective assessment of the stylistic properties of the given document, roughly corresponding to the key text properties identified by the Hamburg Comprehensibility Concept.
LiFR-Law is a corpus of Czech legal and administrative texts with measured reading comprehension and a subjective expert annotation of diverse textual properties based on the Hamburg Comprehensibility Concept (Langer, Schulz von Thun, Tausch, 1974). It has been built as a pilot data set to explore the Linguistic Factors of Readability (hence the LiFR acronym) in Czech administrative and legal texts, modeling their correlation with actually observed reading comprehension. The corpus is comprised of 18 documents in total; that is, six different texts from the legal/administration domain, each in three versions: the original and two paraphrases. Each such document triple shares one reading-comprehension test administered to at least thirty readers of random gender, educational background, and age. The data set also captures basic demographic information about each reader, their familiarity with the topic, and their subjective assessment of the stylistic properties of the given document, roughly corresponding to the key text properties identified by the Hamburg Comprehensibility Concept.
Changes to the previous version and helpful comments
• File names of the comprehension test results (self-explanatory)
• Corrected one erroneous automatic evaluation rule in the multiple-choice evaluation (zahradnici_3,
TRUE and FALSE had been swapped)
• Evaluation protocols for both question types added into Folder lifr_formr_study_design
• Data has been cleaned: empty responses to multiple-choice questions were re-inserted. Now, all surveys
are considered complete that have reader’s subjective text evaluation complete (these were placed at
the very end of each survey).
• Only complete surveys (all 7 content questions answered) are represented. We dropped the replies of
six users who did not complete their surveys.
• A few missing responses to open questions have been detected and re-inserted.
• The demographic data contain all respondents who filled in the informed consent and the demographic
details, with respondents who did not complete any test survey (but provided their demographic
details) in a separate file. All other data have been cleaned to contain only responses by the regular
respondents (at least one completed survey).
Corpus of Czech educational texts for readability studies, with paraphrases, measured reading comprehension, and a multi-annotator subjective rating of selected text features based on the Hamburg Comprehensibility Concept
Corpus of Czech educational texts for readability studies, with paraphrases, measured reading comprehension, and a multi-annotator subjective rating of selected text features based on the Hamburg Comprehensibility Concept
This data set contains four types of manual annotation of translation quality, focusing on the comparison of human and machine translation quality (aka human-parity). The machine translation system used is English-Czech CUNI Transformer (CUBBITT). The annotations distinguish adequacy, fluency and overall quality. One of the types is Translation Turing test - detecting whether the annotators can distinguish human from machine translation.
All the sentences are taken from the English-Czech test set newstest2018 (WMT2018 News translation shared task www.statmt.org/wmt18/translation-task.html), but only from the half with originally English sentences translated to Czech by a professional agency.
OAGK is a keyword extraction/generation dataset consisting of 2.2 million abstracts, titles and keyword strings from cientific articles. Texts were lowercased and tokenized with Stanford CoreNLP tokenizer. No other preprocessing steps were applied in this release version. Dataset records (samples) are stored as JSON lines in each text file.
This data is derived from OAG data collection (https://aminer.org/open-academic-graph) which was released under ODC-BY licence.
This data (OAGK Keyword Generation Dataset) is released under CC-BY licence (https://creativecommons.org/licenses/by/4.0/).
If using it, please cite the following paper:
Çano, Erion and Bojar, Ondřej, 2019, Keyphrase Generation: A Text Summarization Struggle, 2019 Annual Conference of the North American Chapter of the Association for Computational Linguistics, June 2019, Minneapolis, USA
OAGKX is a keyword extraction/generation dataset consisting of 22674436 abstracts, titles and keyword strings from scientific articles. The texts were lowercased and tokenized with Stanford CoreNLP tokenizer. No other preprocessing steps were applied in this release version. Dataset records (samples) are stored as JSON lines in each text file.
The data is derived from OAG data collection (https://aminer.org/open-academic-graph) which was released under ODC-BY license.
This data (OAGKX Keyword Generation Dataset) is released under CC-BY license (https://creativecommons.org/licenses/by/4.0/).
If using it, please cite the following paper:
Çano Erion, Bojar Ondřej. Keyphrase Generation: A Multi-Aspect Survey. FRUCT 2019, Proceedings of the 25th Conference of the Open Innovations Association FRUCT, Helsinki, Finland, Nov. 2019
To reproduce the experiments in the above paper, you can use the first 100000 lines of part_0_0.txt file.
OAGL is a paper metadata dataset consisting of 17528680 records which comprise various scientific publication attributes like abstracts, titles, keywords, publication years, venues, etc. The last field of each record is the page length of the corresponding publication. Dataset records (samples) are stored as JSON lines in each text file. The data is derived from OAG data collection (https://aminer.org/open-academic-graph) which was released under ODC-BY license. This data (OAGL Paper Metadata Dataset) is released under CC-BY license (https://creativecommons.org/licenses/by/4.0/).
If using it, please cite the following paper:
Çano Erion, Bojar Ondřej: How Many Pages? Paper Length Prediction from the Metadata.
NLPIR 2020, Proceedings of the the 4th International Conference on Natural Language
Processing and Information Retrieval, Seoul, Korea, December 2020.
OAGS is a title generation dataset consisting of 34993700 abstracts and titles from scientific articles. Texts were lowercased and tokenized with Stanford CoreNLP tokenizer. No other preprocessing steps were applied in this release version. Dataset records (samples) are stored as JSON lines in each text file. The data is derived from OAG data collection (https://aminer.org/open-academic-graph) which was released under ODC-BY licence. This data (OAGS Title Generation Dataset) is released under CC-BY licence (https://creativecommons.org/licenses/by/4.0/). If using it, please cite the following paper: Çano, Erion and Bojar, Ondřej, 2019, "Efficiency Metrics for Data-Driven Models: A Text Summarization Case Study", INLG 2019, The 12th International Conference on Natural Language Generation, November 2019, Tokyo, Japan. To reproduce the experiments in the above paper, you can use oags_train1.txt, oags_train2.txt, oags_train3.txt, oags_test.txt and oags_val.txt files. If you need more data samples you can get them from oags_train_backup.txt and oags_val-test_backup.txt.
OAGSX is a title generation dataset consisting of 34408509 abstracts and titles from scientific articles. The texts were lowercased and tokenized with Stanford CoreNLP tokenizer. No other preprocessing steps were applied in this release version. Dataset records (samples) are stored as JSON lines in each text file.
The data is derived from OAG data collection (https://aminer.org/open-academic-graph) which was released under ODC-BY license.
This data (OAGSX Title Generation Dataset) is released under CC-BY license (https://creativecommons.org/licenses/by/4.0/).
If using it, please consider citing also the following paper:
Çano Erion, Bojar Ondřej. Two Huge Title and Keyword Generation Corpora of Research Articles.
LREC 2020, Proceedings of the the 12th International Conference on Language Resources and Evaluation,
Marseille, France, May 2020.