In order to guarantee the highest quality of translation, WhP uses permanent, dedicated translators who are not only experienced in the client’s given industry domain, but also well versed with WhP’s tools and processes. Furthermore, to ensure high productivity and sustain consistency throughout the translation, WhP uses CAT tools, including translation memories.

Translation Memories (TM) deliver higher productivity, quality and consistency

Every single translation is broken down into segments which are stored in a database together with their equivalents in other languages. When a new text needs to be translated, the tool breaks it down into segments, checks whether similar segments already exist in the TMs, and proposes the corresponding previous translation to the translator, thus ensuring high productivity and consistency.

TMs deliver the best results for documentation translations and software localizations where repetitions are the most frequent. Obtaining an 80% re-use rate on a technical document's second or third revision is not uncommon.

However, in order to be really efficient, TMs must be properly structured from the onset. It would be tempting to have one TM per client to try to re-use the maximum amount of content, however the risk is that a segment may have been translated in a different context, resulting in a bad translation. Furthermore, the management costs of using different TMs for each document category with a view to optimizing re-use of common text can be prohibitive.

WhP will thus suggest how best to organize the translation memories and their application domains based on both its internal translation processes and on the client’s own documentation processes. WhP will also give the client full and exclusive Intellectual Property Rights (IPR) to the translation.

Translation memories are now used by most Localization Service Providers, and have been highly instrumental in reducing costs while improving the quality of translations. They deliver well defined, re-usable terminology that can be associated with the client’s Terminology Management System, if one exists. In addition, TMs optimize content translation and ease the sharing of glossaries and terms, both internally and externally. It is also important to note that a Translation Memory standard, TMX, has been defined in order to ensure database portability.

Translation memories add value to your partnerships and can enhance your company brand

Sharing TMs and glossaries with partners can be a real strategic advantage for a company, which can then enforce the proper usage of its terminology, and progressively ensure that its own or preferred terms are adopted by the industry.

Combining Translation Memory with Machine Translation may improve speed and cost efficiency

While Machine Translation has great potential, it is still incapable of delivering a satisfactory level of quality. Nevertheless, WhP believes this technology could be applied to dynamic web sites in the coming years, allowing for a significant reduction in translation costs.

WhP anticipates that within a few years, Statistical Machine Translation will give acceptable results for some language combinations (mainly from English to the main European languages), on condition that:

• there is a focus on specific domain
• up-to-date translation memories and terminology lists have been developed
• a large corpus of multilingual content is available (500,000 to 1 million words)
• the result is reviewed by a specialized linguist

When quality requirements are high, cost savings alone do not justify the use of MT.

At present, there are two main MT technologies: linguistic processing and statistical processing.

Linguistic engines have now reached a certain maturity, and are unlikely to improve much further in coming years. They require extensive lexicons with morphological, syntactic and semantic information, and large sets of rules.

Statistical Machine Translation generates translations using statistical methods based on a bilingual corpus fed into it. The results can be quite impressive, but it is rare to have a corpus large enough for the system to give effective results.

Statistical Machine Translation combined with Translation Memory can improve the output quality by interjecting human approved translation. It helps "teach" statistical MT engines, and automatically improve the quality of subsequent translations. However, once again, the quality of the results correlates to the size of the application-specific bilingual corpus, which needs to be sufficiently large.

For more information, request our White Papers on:
- Machine Translation: Dream or Reality?
- The translation memories