In the new age of cloud computing, AI and Deep Learning, the way we used to translate is changing rapidly. Harvest the power of Neural Machine Translation by utilizing your language resources and your domain know-how for your benefit. Let us help you with your MT needs in areas such as:
Collection and selection of appropriate publicly available and private bilingual corpus toward the creation of domain specific MT Systems. Clean-up and optimization of training data through removal of unsuitable text and formatting, deduplication, special symbol harmonization, data anonymization etc. Creation of specialized dictionaries with standard phrases and non-translatable text for training specific MT Systems.
Pre and Post MΤ Text Processing
Custom MT System specific software development for Pre MT translation text selection (what should go to MT) and text preparation (symbol harmonization, non-translatable text replacement, anonymizing etc.) for optimal MT output. Post MT System text treatment for adhering to client-based specific formatting and stylistic rules, de-anonymizing, terminology and other known errors fixes etc.
MT System Training and Evaluation
Find out if you can benefit from MT in 66”
Rating a Machine Translation Engine
- 36 Custom Machine Translation Systems
- 27 Language Pairs
- 3 Different Content Domains
An exciting MT project was creating 36 Custom Machine Translation Engines, for one of our clients, spread across 3 domains and 27 Language Pairs. To train the systems, we used both public and private data that was cleaned and appropriately prepared. For selecting each final MT System, at least three data sets were created with different degrees of quality vs volume, and the resulting MT Systems were evaluated by language specialists to select the best.
The project was a great success with:
- BLEU Scores around 0.6 even for the most demanding of languages;
- Client reporting overall savings of more than 25%.
MT System Evaluation and Feedback
• Pre MT document processing
• Evaluation of post edited documents
• Identification of human changes
• Feedback collection for MT improvement
This is a very interesting project that can continuously rate a Machine Translation Engine and pinpoint its weak points. First, documents are prepared for MT and then the relevant parts are being MT translated. Then the MT-translated documents go through the usual, translation/post-editing and review phase and they are finalized. The program keeps track of the MT segments and when they are completed, it analyses the original MT text against the corrected translations. It calculates hLEPOR and WER Scores and collects all changes made to the text. This data is used to improve the MT System and the pre and post translation process. Since the program took off, the client now relies on it to:
• Improve pre and post MT translation processes;
• Improve MT training data and glossaries;
• Evaluate the MT System performance against different inputs;
• Improve MT System quality and reduce human effort.
This is How We have Helped
SDL Language Weaver Inc./Datamundi BVBA
“TMServe has been helping us for 5 years at SDL LanguageWeaver and 10 years at Datamundi. They create power tools and advise us on best strategies for complex projects. They open up translation memories and various data formats like no other. Top service!”
Gert Van Assche
Data Acquisition Manager
SDL Language Weaver Inc.
“The result of TMServe’s work is excellent. Our collaboration in terms of both Machine Translation and the creation of unique and innovative applications for the automation of repetitive processes as well as the quality control of day-to-day operations is impeccable. Even the smallest issues are quickly resolved.
The time required to perform important corporate processes, as well as the cost for the quality control before the delivery to the customer has now been minimized.
Communication with TMServe, which is of primary importance for any cooperation, is excellent and they reply promptly, even on holidays!”
Machine Translation and Localization Engineering Manager
“TMServe has guided through the whole process of building several MT systems; I have thus managed to leverage my linguistic data and also achieve a significant increase in productivity. As a result, they have helped us work with tight deadlines and maintain high quality standards even in large-scale and challenging projects.”
16 Thermopylon Street
175 63 Paleo Faliro