Using KantanBuildAnalytics
This video will show you how to check the automatic evaluation scores relating to an engine.
1. Click the relevant engine from the Dashboard to make it active. Then click the ‘BuildAnalytics’ button on the right of the engine row.
2. You will be redirected to the BuildAnalytics page, where you can find the summary, rejects report, automatic evaluation scores, gap analysis and timeline.
3. You can check the results in each tab, download the reports in Excel format and receive tips to improve the scores.
These results will help you improve the quality of your builds and define which actions you can take to refine your translation output. To do so, as well as the reports for F-Measure, Bleu and TER scores, pay particular attention to the gap analysis and the rejects report.
• The Gap Analysis highlights word gaps or unknown words within your KantanMT engine. These can be downloaded, translated and uploaded back as training data. This step will improve phrase tables, fluency and vocabulary of your engine.
• The Rejects Report displays the training data segments which were removed during the data cleansing step as well as the reasons why they were removed. It’s important to review the rejects report to improve future data selection. Working on specific issues, rejections can be reduced by developing a pre-processor file using regular expressions (source.ppx and target.ppx).
If you have any problems or need help, please contact us at [email protected]