WIT3
Web Inventory of Transcribed and Translated Talks
| ▿ Latest version of XML files of the TED Talks (April 2016) |
| ▿ Releases |
| ▿ Note on Transcripts/Translations |
| ▿ Terms of Use |
| ▿ Acknowledgments |
| ▿ Related resources |
| ▿ Contact person |
| ▿ Author |
Since 2007, the TED Conference has been posting on its website all video recordings of its talks, English subtitles and their translations in more than one hundred languages. In order to make this collection of talks more effectively usable by the research community, the original textual contents are redistributed here, together with MT benchmarks and processing tools.
For a detailed description of this corpus, read:
M. Cettolo, C. Girardi, and M. Federico. 2012. WIT3: Web Inventory of Transcribed and Translated Talks.
In Proc. of EAMT, pp. 261-268, Trento, Italy. pdf, bib.
Please, cite the paper if you use this corpus in your work.
▸ Latest version of XML files of the TED Talks (April 2016):
- Talks in XML format (109 languages)
- Tools
▸ Releases
- 2020-01
- 2019-01
- 2018-01: special release for the IWSLT 2018 evaluation campaign
- Training and development sets for the Low Resource MT TED task
- additional archive with the original xml files of all the 2018 TED talks (excluding those in the tst2018 evaluation set)
- the list of talkid in the tst2018 evaluation set
- 2017-02: Special Release for the analyses carried out in the paper
Luisa Bentivogli, Mauro Cettolo, Marcello Federico, Christian Federmann. 2018.
"Machine Translation Human Evaluation: an investigation of evaluation based on Post-Editing and its relation with Direct Assessment".
In Proceedings of the 15th International Workshop on Spoken Language Translation (IWSLT 2018) , Bruges, Belgium.
- 2017-01: special release for the IWSLT 2017 evaluation campaign
- Training and development sets for the multilingual TED task
- Evaluation sets for the multilingual TED task
- Training and development sets for the TED task (optional)
- Evaluation sets for the TED task (optional)
- Normalizer for Arabic
- Three more pairs beyond the evaluation campaign: English-Indonesian, English-Portuguese and English-BrazilianPortuguese
-
Human evaluation data, including Direct Assessment (source-based and reference-based) and Post-Edits
- 2016-02: special release for human evaluation analysis of MT runs from IWSLT 2016 evaluation campaign
- 2016-01: special release for the IWSLT 2016 evaluation campaign
- Training and development sets for the MT track
- Evaluation sets for the MT track
- Normalizer for Arabic
- Four more pairs beyond the evaluation campaign: Arabic-Hebrew, Arabic-Czech/German and English-Spanish
- Human evaluation data, including additional references for evaluation sets
- 2015-01: special release for the IWSLT 2015 evaluation campaign
- Training and development sets for the MT track
- Evaluation sets for the MT track
- References of the progressive test set (for 1 language pair)
- Linguistic preprocessors for Vietnamese
- Human evaluation data, including additional references for evaluation sets
- 2014-01: special release for the IWSLT 2014 evaluation campaign
- Training and development sets for the MT track
- Evaluation sets for the MT track
- References of the progressive test sets (for 4 language pairs)
- MT baseline results
- Linguistic preprocessors for Arabic, Farsi and Hebrew
- Human evaluation data, including additional references for evaluation sets
- 2013-01: special release for the IWSLT 2013 evaluation campaign
- Training and development sets for the MT track
- Evaluation sets for the MT track
- References of the progressive test sets (English to German and English to French pairs)
- MT baseline results
- One more pair beyond the evaluation campaign: German-French
- Human evaluation data, including additional references for evaluation sets
- 2012-03: special release for the TED task of the IWSLT 2012 evaluation campaign
- Training and development sets for the MT track
- Evaluation sets for the MT track
- MT baseline results
- Human evaluation data
- 2012-02 (Update: May 2012)
- Plain texts for MT (23x22 language pairs)
- Talks in XML format (71 languages)
- MT baseline results
- Tools
- 2012-01 (Experiments of the above mentioned EAMT 2012 paper)
- Plain texts for MT (23x22 language pairs)
- Talks in XML format (79 languages)
- MT baseline results
- Tools
- 2011-01 (IWSLT 2011 evaluation campaign)
- Plain texts for MT (3 language pairs)
- MT baseline results
- Human evaluation data
▸ Note on Transcripts/Translations
TED transcripts and translations were generated following these guidelines: How to Tackle a Transcript and How to Tackle a Translation.WIT3 redistributes original TED texts in their original format, therefore infos included in TED guidelines are mostly valid also for WIT3 texts. An important difference regards texts of development and evaluation sets where metadata, for example those regarding sound information, are removed.
▸ Terms of Use
TED makes its collection of video recordings and transcripts of talks available under the Creative Commons BY-NC-ND license (look here). WIT3 acknowledges the authorship of TED talks (BY condition) and does not redistribute transcripts for commercial purposes (NC). As regards the integrity of the work (ND), WIT3 only changes the format of the container, while preserving the original contents. WIT3 aims to support research on human language processing as well as the diffusion of TED Talks!▸ Acknowledgments
The work was partially supported by the EU-BRIDGE and CRACKER projects, funded by the European Commission.
▸ Related resources
The NAIST-NTT TED Talk Treebank
▸ Contact person
Mauro Cettolo (cettolo
fbk.eu)