| |
 |
International Workshop on Spoken Language Translation
|
|
|
|
|
|
Evaluation campaign >> Evaluation specification
Evaluation Scores
- Subjective Evaluation (for Chinese → English translations)
- Automatic Evaluation (BLEU, NIST, WER, PER, METEOR and GTM)
Arabic → English, Chinese → English, Japanese → English and Korean → English tracks have 16 English references.
English → Chinese track has only 1 Chinese reference.
Specification: Case and Punctuation marks
The evaluation will be done by submitting your translations to an online evaluation server.
We will automatically detect and possibly change (lower casing and removal of punctuation marks) your submission format.
-
Word compounds will always be split (hyphen '-' will be replaced with space, "second-hand" → "second hand").
- Apostrophes ("I'll", "He's") will not be changed.
Lower Case & Punctuation removed
All evaluations will be done under this condition.
Every sentence is evaluated lower-cased, punctuation-removed
Optional: Mixed Case & Punctuation marks
Mixed case evaluation is not obligatory but an option of participants. Every sentence is evaluated mixed-cased with punctuation marks. Punctuations will be separated.
Specification: Segmentation
For English
- The word segmentation by space in the original training data will be used for text and ASR ouput translation.
For Chinese
- The data will be segmented using the ASR segmentation.
Specification: Output Format
MT results for ASR output
- When submitting MT results for the ASR output, please submit both source hypothesis and its tranlation. The output file format should be as follows:
Source Sentence 1
Target Sentence 1
Source Sentence 2
Target Sentence 2
...
MT results for Manual transcriptions
- One target sentence per line, in the same order of the source sentences given as test set.
Examples
- Example submitted sentence:
"They didn't buy any duty-free items, did they?"
- Automatically changed for lower case/punctuation removed:
"they didn't buy any duty free items did they"
- Automatically changed for mixed case/with punctuations:
"They didn't buy any duty-free items , did they ?"
|
|
|
|
Copyright(c) 2005 interACT All rights reserved.
|
|
|