NIST 2008 Machine Translation Evaluation - (Open MT-08)

Official Evaluation Results

Date of release: Fri Jun 06, 2008

Version: mt08_official_release_v0

The NIST 2008 Machine Translation Evaluation (MT-08) is part of an ongoing series of evaluations of human language translation technology. NIST conducts these evaluations in order to support machine translation (MT) research and help advance the state-of-the-art in machine translation technology. These evaluations provide an important contribution to the direction of research efforts and the calibration of technical capabilities. The evaluation was administered as outlined in the official MT-08 evaluation plan.

Disclaimer

These results are not to be construed, or represented as endorsements of any participant's system or commercial product, or as official findings on the part of NIST or the U.S. Government. Note that the results submitted by developers of commercial MT products were generally from research systems, not commercially available products. Since MT-08 was an evaluation of research algorithms, the MT-08 test design required local implementation by each participant. As such, participants were only required to submit their translation system output to NIST for uniform scoring and analysis. The systems themselves were not independently evaluated by NIST.

Certain commercial equipment, instruments, software, or materials are identified in this paper in order to specify the experimental procedure adequately. Such identification is not intended to imply recommendation or endorsement by the (NIST), nor is it intended to imply that the equipment, instruments, software or materials are necessarily the best available for the purpose.

There is ongoing discussion within the MT research community regarding the most informative metrics for machine translation. The design and implementation of these metrics are themselves very much part of the research. At the present time, there is no single metric that has been deemed to be completely indicative of all aspects of system performance.

The data, protocols, and metrics employed in this evaluation were chosen to support MT research and should not be construed as indicating how well these systems would perform in applications. While changes in the data domain, or changes in the amount of data used to build a system, can greatly influence system performance, changing the task protocols could indicate different performance strengths and weaknesses for these same systems.

Because of the above reasons, this should not be interpreted as a product testing exercise and the results should not be used to make conclusions regarding which commercial products are best for a particular application.

Evaluation Tasks

The MT-08 evaluation consisted of four tasks. Each task required a system to perform translation from a given source language into the target language. The source and target language pairs that made up the four MT-08 tasks were:

• Translate Arabic text into English text
• Translate Chinese text into English text
• Translate Urdu text into English text
• Translate English text into Chinese text

Evaluation Conditions

MT research and development requires language data resources. System performance is strongly affected by the type and amount of resources used. Therefore, two different resource categories were defined as conditions of evaluation. The categories differed solely by the amount of data that was available for use in the training and development of the core MT engine. The evaluation conditions were called "Constrained Data Track" and "Un-Constrained Data Track".

• Constrained Data Track - limited the training data to data in the LDC public catalogue existing before July 1st, 2007.
• Note: For the URDU task, the constrained data condition required that the core system development use only the provided resource DVD. No other data was allowed for the primary condition of interest.
• Un-Constrained Data Track - extended the training data to any publicly available data existing before July 1st, 2007.

Other submissions not in categories described above will not be reported in the final release.

Evaluation Data

Source Data

MT-08 evaluation data sets contained documents drawn from newswire text documents and web-based newsgroup documents. The source documents were encoded in UTF-8.

The test data was selected from a pool of data collected by the LDC during July 2007. The careful selection process sought to have a variety of sources (see below) and publication dates while hitting the target test set size.

Reference Data

MT-08 reference data consists of four independently generated high quality translations that were produced by professional translation companies. Each translation agency was required to have native speaker(s) of the source and target languages working on the translations.

Current versus Progress Data Division

For those willing to abide by the strict processing rules, a "PROGRESS" test set was distributed to use as a BLIND benchmark for several evaluations. Teams that processed this data submitted their translations to NIST and deleted all related files (source, translations, and any other derivitive file). The scores of the progress test sets were reported to the participants but were not reported here. Future Open MT evaluations will report PROGRESS test set scores from year to year.

Performance Measurement

Machine translation quality was measured automatically using an N-gram co-occurrence statistic metric developed by IBM and referred to as BLEU. BLEU measures translation accuracy according to N-grams or sequence of N-words that it shares with one or more high quality reference translations. Thus, the more co-occurrences, the better the score. BLEU is an accuracy metric, ranging from "0" to "1" with "1" being the best possible score. A detailed description of BLEU can be found in the paper Papineni, Roukos, Ward, Zhu (2001). "Bleu: a Method for Automatic Evaluation of Machine Translation" (keyword = RC22176).

Although BLEU was the official metric for MT-08, measuring translation quality is an ongoing research topic in the MT community. At the present time, there is no single metric that has been deemed to be completely indicative of all aspects of system performance.

Automatic metrics reported:

• BLEU-4 (MTeval-v11b: official metric)
• IBM BLEU (IBM's BLEU with original brevity penalty)
• NIST (NIST's refinement of BLEU, commonly referred to as NIST)
• TER
• METEOR

Other metrics (to be) reported:

• Human Assessments of Adequacy (judged by participants and others)
• Human judgments of Preference (judged by participants and others)
• MT Comprehension Test (implemented by MIT-LL)

Evaluation Participants

The table below lists the organizations entered as participants and the evaluation tasks they are registered for in MT-08.

Evaluation Systems

Each site/team could submit up to four systems for evaluation with one system marked as its primary system. The primary system indicated the site/team's best effort. This official public version of the results report the results only for the primary systems. Note that these charts show an absolute ranking according to the primary metric.

Systems that fail to meet the requirements for either track will not be reported here.

"significance groups*" shows areas where the wilcoxon signed rank test was not able to differenciate system performance at the 95% confidence level. That is, if two systems belong to the same significance group (by sharing the same number), then they are determined to be comparble, based n BLEU-4 scoring.

Results Section

Contains Valid On-Time Submissions

Overall System Results

Arabic to English (primary system) Results

Chinese to English (primary system) Results

Urdu to English (primary system) Results

English to Chinese (primary system) Results

Results by Genre

All reported scores are limited to the entire "CURRENT" data sets. All primary submissions are shown here.

Site Results ( alphabetic order )