##################################################
                    IWSLT 2009 Corpus : BTEC Chinese-English
               ##################################################

+------------------------------------------------------------------------------+
| Copyright (C) 2004-2009, National Laboratory of Pattern Recognition (NLPR),  |
|                          Beijing, China.                                     |
|                          [Chinese text data sets (train)]                    |
|                                                                              |
| Copyright (C) 2004-2009, Advanced Telecommunications Research Institute      |
|                          International (ATR-I), Kyoto, Japan                 |
|                          [English data sets (train,devset[123456])           |
|                          [Chinese text data sets (devset[123456])]           |
|                          [Chinese ASR output data sets]                      |
|                                                                              |
| Copyright (C) 2008-2009, National Institute of Information and Communications|
|                          Technology (NICT), Kyoto, Japan                     |
|                          [English data sets (devset7,testset)]               |
+------------------------------------------------------------------------------+

================================================================================
=== File Encoding

Chinese:  UTF-8
English:  UTF-8

================================================================================
=== Data Sets

devset1_CSTAR03 - develop data set of IWSLT 2004 (BTEC Task)
devset2_IWSLT04 - evaluation data set of IWSLT 2004 (BTEC Task)
devset3_IWSLT05 - evaluation data set of IWSLT 2005 (BTEC Task)
devset6_IWSLT07 - evaluation data set of IWSLT 2007 (BTEC Task)
devset7_IWSLT08 - evaluation data set of IWSLT 2008 (BTEC Task)

IWSLT09_BTEC.testset - evaluation data set of IWSLT 2009 (BTEC Task)

================================================================================
=== Data Files

IWSLT/2009/Corpus/BTEC/Chinese-English
   |
   |-- train
   |   +-- TXT
   |       |-- IWSLT09_BTEC.train.zh.txt
   |       +-- IWSLT09_BTEC.train.en.txt
   |-- dev
   |   |-- TXT
   |   |   |-- IWSLT09.devset[12367]*.zh.txt
   |   |   +-- IWSLT09.devset[12367]*.mref.en.txt
   |   +-- SGM
   |       |-- IWSLT09.devset[12367]*.case+punc.src.zh.sgm
   |       |-- IWSLT09.devset[12367]*.case+punc.mref.en.sgm
   |       |-- IWSLT09.devset[12367]*.no_case+no_punc.src.zh.sgm
   |       +-- IWSLT09.devset[12367]*.no_case+no_punc.mref.en.sgm
   |-- test
   |   +-- TXT
   |       +-- IWSLT09_BTEC.testset.zh.txt
   |
   |-- tools
   |
   +-- README.BTEC_CE.txt

================================================================================
=== Data Format

[TXT data files]

    + same sentence IDs for corresponding translation examples
    + same sentence order for corresponding data files

    + single reference files:

      (a) file name: IWSLT09.<TASK>.<LANG>.txt

          > IWSLT09.train.en.txt
          > IWSLT09.devset3_IWSLT05.ar.txt
          > IWSLT09.testset.zh.txt

      (b) file format: <SENTENCE_ID>\01\<TEXT>

          > TRAIN_00001\This is the first sentence.
          > TRAIN_00002\This is the second sentence.
          > ...

    + multiple reference files:

      (a) file name: IWSLT09.<TASK>.mref.<LANG>.txt

          > IWSLT09.devset3_IWSLT05.mref.en.txt

      (b) file format: <SENTENCE_ID>\<PARAPHRASE_ID>\<TEXT>

          > DEV3_IWSLT05_001\01\This is the first reference translation of the first sentence ID
          > DEV3_IWSLT05_001\02\This is the second reference translation of the first sentence ID
          > DEV3_IWSLT05_001\03\...
          > ...
          > DEV3_IWSLT05_002\01\This is the first reference translation of the second sentence ID
          > DEV3_IWSLT05_002\02\This is the second reference translation of the second sentence ID
          > DEV3_IWSLT05_002\03\...
          > ...

[SGM data files]

    + two types of SGM files are provided for each data set

      (a) case-sensitive, with tokenized punctuations (official evaluation specifiactions)

          - IWSLT09.<TASK>.case+punc.src.<SRC_LANG>.sgm
          - IWSLT09.<TASK>.case+punc.mref.<TRG_LANG>.sgm

       (a) case-insensitive, without punctuations (additional evaluation specifiactions)

          - IWSLT09.<TASK>.no_case+no_punc.src.<SRC_LANG>.sgm
          - IWSLT09.<TASK>.no_case+no_punc.mref.<TRG_LANG>.sgm

    + format: SGML format required by BLEU/NIST scoring script

      (a) SRC files:
          > <srcset setid="set" srclang="source">
          > <DOC docid="document">
          > <seg> ...</seg>
          > ...
          > </DOC>
          > </srcset>

      (b) REF files:
          > <refset setid="set" srclang="source" trglang="target">
          > <DOC docid="document" sysid="r01">
          > <seg> ...</seg>
          > ...
          > </DOC>
          > </refset>
          > ...
          > <refset setid="set" srclang="source" trglang="target">
          > <DOC docid="document" sysid="r16">
          > <seg> ...</seg>
          > ...
          > </DOC>
          > </refset>

      (c) TST files:
          > <tstset setid="set" srclang="source" trglang="target">
          > <DOC docid="document" sysid="system">
          > <seg> ...</seg>
          > ...
          > </DOC>
          > </tstset>

================================================================================