What´s new

November 17, 2005

• We uploaded the pictures taken at IWSLT 2005.
  Please check the Photo Gallery.

November 14, 2005

• The speech data for Chinese and Japanese was uploaded. Please check the downloads page. (Please use the username/password provided for the appropriate tracks).

November 11, 2005

• The English and Chinese references for the IWSLT2005 Testsets can be downloaded. (Please use the username/password provided for the Evaluation Server).

November 1, 2005

• The Proceedings are online.

August 22, 2005

• The Evaluation server for the Testsets will show your scores now. The submissions with the yellow/orange background color will be considered final submissions for the Evaluation campaign. If more than one or none submissions were marked as final submissions the last one was automatically chosen. The submission log will reflect this change.
  
  You can certainly continue to use the evaluation server to get improved numbers for your technical papers.
  

August 19, 2005

• The problem some groups encountered with the TESTSET submission for the English to Chinese tracks was fixed. It should from now on work for everyone.
  
  

August 18, 2005

• We removed the possibility to submit main submissions at 12:20 am EDT. If you had trouble submitting and submitted as a "non-final" submission after that please contact the organizers. We will absolutely not accept any submissions that were sent after 1:30 am.
  
  
• Some people are encountering problems with the TESTSET submission for the English to Chinese tracks. Please send your submissions via email to Chiori Hori at chiori@cs.cmu.edu and Matthias Eck at matteck@cs.cmu.edu.
  
  

August 16, 2005

• We sent the username and password for the testset evaluation server to each organization.
  If you don't get them, please contact us soon as possible.
  

• 12:13 am EDT: The Test Data is online. Please use the usernames/passwords provided for the development and training data.
  
  

August 15, 2005

• 11:49 pm EDT: The Evaluation server for the Testsets is online. Please use the username and password that will be provided for your submissions.
  
  
• The calculation of the WER and PER was changed in the evaluation servers. Up to now the WER and PER for the document was calculated as the arithmetic mean of the sentence WER/PER.
  
  We changed this WER/PER calculation to a calculation that selects the references based on the relative WER/PER and recalculates the WER/PER on the whole document using the errors and length of the closest references.
  
  WER/PER scores that were calculated before this change will not be changed in the Evaluation Server so please note this difference when comparing with earlier submission.

August 11, 2005

• We provided the English reference files for the development set 1 and development set 2 here. We also provided the exact script that is automatically applied to your hypothesis before scoring it.

August 1, 2005

• The Evaluation Server is now able to score the Japanese → English development sets.
  
  
• The Evaluation for the English → Chinese translations will be done based on the segmentation of the ASR output. Please submit your files in this segmentation. We will not re-segment your files prior to evaluating.
  To reduce the influence of the segmentation quality, we will also evaluate the hypotheses based on a character segmentation. (This character segmentation will be done automatically by our server)
  
  

July 29, 2005

• The Evaluation Server is now able to score the English → Chinese development sets.
  (The METEOR scoring seems to be deficient for Chinese translations; we are investigating this.)
  
  

July 28, 2005

• The Evaluation Server is now able to score the Arabic → English development sets and the Korean → English development set.
  
  

July 27, 2005

• Update: The Evaluation Server is online again.
  
  
• The Evaluation Server is offline for maintenance. We apologize for the inconvenience.
  
  
• The Chinese lattices for the Chinese → English ASR translations were updated. It now contains N-best lists. We provide the ASR output with 3 different accuracies. Check for the updated file at the Downloads page.
  
  

July 26, 2005

• The Arabic → English training data package was updated as 2 lines were incorrectly translated. Check for the updated file at the Downloads page.
  
  
• The paper submission deadline was extended until July 31, 2005. See Submission page.

• The online evaluation server is running here.
  

July 22, 2005

• Notice that A-E, C-E, J-E and K-E tracks have 16 English references. E-C track has only 1 Chinese reference. See Evaluation Specification page.

July 21, 2005

• MT results for the ASR output should include both source hypothesis and its translation. Check Evaluation Specification page for the required output format.


• In the Unrestricted and C-STAR tracks, the test set perplexity should be reported in the system description paper. See details here.
  


• Details of the Venue added.

July 20, 2005

• The Arabic → English training data package is now complete at 20,000 lines. Check for the updated file at the Downloads page.
  
  

July 19, 2005

• We added 11,000 lines of Arabic to the Arabic → English training data package making it 16,000 lines overall. The remaining 4000 lines will be added as soon as possible. Check for the updated file at the Downloads page.
  
  
• The missing 500th n-best list for the second development set is now included in the package Japanese → English..
  
  
• The segmented files were modified and include punctuations. Although the previous package included the segmented files with punctuations and the no-segmented files with punctuations, the current package includes only the segmented files with punctuations.
  
  
• The Japanese lattices and N-Best lists for the Japanese → English ASR translations are available for both development sets at the Downloads page. (The 500th n-best list for the second development set is currently empty. We will fix this as soon as possible.)
  
  

July 15, 2005

• The Chinese lattices for the Chinese → English ASR translations are available for both development sets at the Downloads page.
  
  
• The English side of the second development set (IWSLT2004) for the English → Chinese manual transcriptions was changed to a different paraphrasing to coincide with the ASR lattices and N-best lists. This does not make a serious difference for the data, and it is not necessary to update if you do not want to combine both data sets for manual transcriptions and lattices/nbest lists.

July 14, 2005

• The English lattices and N-Best lists for the English → Chinese ASR translations are available for both development sets at the Downloads page.

July 13, 2005

• The current Japanese training data contains a segmented corpus without punctuation marks and a non-segmented corpus with punctuation marks. We will provide a segmented corpus with punctuation marks as soon as possible.

• Please update your files.

  • The Japanese training data that was available on the previous downloads page contained Part-of-Speech Tags and other information. Please do not use this file.
  • There was in error in the Chinese CSTAR03 development sets. The error was fixed.
  • The Arabic development sets were reviewed and corrected.

• The subjective evaluation will be done for the Chinese → English track

  
  

July 12, 2005

• We introduced a new track: Supplied + Tools. More information can be found here

• The downloads page was reorganized and updated. Training data and development sets for the manual transcription track are available. Lattices and N-Best Lists for the development sets will follow.

  
  

July 7, 2005

• The Submission server for technical papers and MT System descriptions is online

  
  

July 1, 2005

• The deadline for technical papers was postponed until July 25th.

• The restrictions for each track are online.

• The evaluation specification is online.

May 20, 2005

• Registration form is online

April 5, 2005

• Data is online

March 22, 2005

• The Call for Paper posted

• Visa information