Multilingual communication plays a vital role in economic success and in the uninterrupted flow of information between businesses. Nevertheless, language remains a major obstacle for many corporations because the volume of information generated exceeds the scope of human translation in terms of deadline, cost and linguistic availability.

To manage this new demand, WhP offers a solution that is comprehensive, flexible, innovative, profitable, rapid and reliable; a solution that surpasses the inadequate results provided by free but non-customizable online Translation engines. It implements state of the art processes and technologies enabling organizations to attain their multilingual objectives while cutting costs and shortening deadlines significantly. WhP offers Machine Translation combined with linguistic post-review and can adapt the quality to meet all expectations.

WhP has teamed-up with the world leader in hybrid Machine Translation to propose a scalable translation model with the aim to address:

  • Budget constraints
  • Massive volumes of information
  • Quick access to strategic information required
  • Scarcity of specialized and experienced linguists
  • Real-time results expected by users of new technologies

This solution enables companies to manage translation as a profitable investment (rather than an expense) that can be fully integrated into their global business communication strategy and contribute directly to their growth.

Machine Translation Technology:

Machine Translation is actually old technology. Systran’s first system was marketed over 40 years ago.

Old systems were rule based. Using dictionaries and grammar rules, the machine tried to understand sentences and rephrase them in another language.

In the 90’s, researches came up with a new approach called statistical. The machine is fed with a large bilingual corpus. It finds fragments of sentences in this very large corpus and an efficient algorithm identifies the most probable translation for them. All such tools are based on the Moses algorithm which is available in public license.

While rule-based technology had proven efficient for controlled content (limited sentence structure) it was not able to learn from large corpora. On the other hand, performance of statistical engines was highly dependent on the size and focus of the corpus. These technologies made a significant step forward with the merge of both technologies into the first hybrid Machine Translation models in 2010.

A number of technology providers are active in the market but few are likely to survive. WhP has decided not to invest in designing its own system but rather create tied partnerships with leading suppliers and invest in training and best practices instead with one objective in mind, get the best out of these systems for the benefit of our customers.

Machine Translation in practice:

The main applications of Machine Translation are:

  • Translation of Blogs and Frequently Asked Questions (FAQ) , which would not otherwise be translated due to cost of the highly dynamic nature of the content. Ideal when users need to understand the meaning and are not that picky about quality.
  • Gisting by integrating it with mails or browsers. Translation will present the most important aspects of the information and facilitate exchanges.
  • Substantially reduce speed and cost of translation by using Machine Translation and post-editing the output by trained linguists. Analysis of the content and detailed feasibility study should be done to ensure quality and Return on Investment.
  • Translation of Business Portals and User Generated Content. In this case, Machine Translation systems have to overcome the challenges of processing mistakes, abbreviations or SMS style authors tend to use.
  • In the future, combined with voice recognition and text to speech synthesis Machine Translation will allow “real time automatic interpretation”.

As a Localization Service Provider, WhP uses this technology in full transparency to maximize services offered to its customers.

WhP achieved significant successes with the use of Machine Translation in multiple languages, on dynamic content not to be post-edited and large volume translation with high quality expectation, thus followed by post-editing.

Machine Translation implementation process:

The main steps in a successful implementation process, especially when used in conjunction with post-editing are:

  1. Define which translation flows will be processed by Machine Translation. A risk here is not to focus enough and thus never reach quality output
  2. Gather all available linguistic assets : bilingual corpus, terminology, relevant monolingual corpus (either source or target language)
  3. Clean linguistic assets to minimize inconsistencies
  4. Train the engine and check the output quality on a sample (many iterations may be needed)
  5. Integrate this engine in the translation workflow environment to be used during the localization cycle. It could be a Computer Aided Translation (CAT) tool for new content only to be translated by the engine while repetitions, existing content or fuzzy matches come from the Translation Memories.
  6.  Select and train linguists who will post-editors Machine Translation output. Not all good professional translator can become good post-editors
  7. Run pilot projects and tune the engine
  8.  Deploy it on all relevant projects
  9. Train the engine with the new bilingual corpus after each project
  10.  Whenever possible, improve the authoring process to gain further productivity increase

By applying best practices at each step of the process, WhP linguists and engineers have been able to demonstrate significant increase in efficiency to key customers. ROI tends to be less than six months.

WhP solved many challenges using Machine Translation, including those involved in processing structured content such as XML or XML DITA where segments include in line tags for variables or cross references.