The IWSLT 2007 Evaluation Campaign will feature two challenge tasks, namely, the translation of spontaneous conversations in the travel domain:
as well as two classical tasks, that is, the translation of read speech in the travel domain:
- from Italian to English and
- from Chinese to English,
For all tasks, participants will be provided with training and development data sets, some links to relevant and linguistic resources, and to open software tools for developing state-of-the-art statistical machine translation systems.
- from Arabic to English, and
- from Japanese into English.
In contrast with other MT evaluation campaigns, translation will occur not on written text but on transcriptions of speech. There will be two types of transcripts: manually created and the output of automatic speech recognition systems. In the latter case, this year, ASR engine output will be provided as n-best lists and word-graphs.
To take part in IWSLT07, prospective participants will need to fill out the usage agreement and return it to FBK, and register for the tasks. Upon receiving the usage agreement, we will send access information for downloading data sets to the contact email given during task registration. We will also use this email address to contact participants when there are news or changes.