Translations.Health uses translation memory on the cloud to provide healthcare organizations and medical companies with professional life sciences localization services that are faster, better, and cheaper. Translation memory (TM) is a term that was first used in the early 90’s within the localization industry to describe the technique for translation leverage and reuse. The process involves saving previously translated content such as sentences and paragraphs called ‘segments’ in source-target language pairs within a bilingual database so they can be reused for future projects. The goal of translation memory is threefold: 1) to reduce translation costs by eliminating the need to translate the same sentences twice, 2) to achieve faster project turnaround, and 3) to maintain better linguistic consistency between versions.

As healthcare companies spend more on translation and localization services over time, translation memory becomes an important enterprise asset. This is because the TM database contains not only corporate IP and multilingual knowledge, it also represents significant cost reductions for a healthcare organization’s international business development. This is why translation memory transparency or TM ownership has become an important topic within the translation industry. Typically, there are two ways to manage translation memory: a) the vendor manages translation memory files on their own and away from the client, and b) the translation memory is stored online so both the client and vendor have access to it at all times. The first approach has been around since TM technologies were first introduced. In fact, many translation companies think they own the translation memory because it’s produced through internal localization process. Some vendors even refuse to turn over the TM files when the client decides to switch vendors. In this case, the client may end up with significant economic losses. Regardless of ownership, by keeping the TM content away from the client, the translation vendor has complete control over the level of translation memory leverage.

At Translations.Health, we believe the client owns the right to all their translation memory assets because they have paid for all the translations, and TM is just a byproduct of that effort. Our cloud-based translation memory management system allows our clients to get up-to-the-minute TM updates as well as access to their TM content at all times. Our highly transparent translation memory process gives our clients the peace of mind, knowing their localization investment is in good hands. Better yet, our clients can also download the entire TM content at any time, without having to get into an embarrassing situation to request for translation memories should they decide to engage a different translation service provider.

Believe it or not, the translation industry is one of the earliest sectors to experiment with big data and machine learning, long before those terminologies became popular. Translation memory management uses machine learning algorithm to store previously translated sentences in a big data repository, along with other attributes such as text formatting and subject field information. When a document is submitted to Translations.Health for translation, our intelligent translation management system first parses the document into sentences. These sentences are then systemically searched and compared against existing content within the translation memory database. Depending on a healthcare company’s translation volume, translation memory databases may contain hundreds and thousands of sentences or segments. Our TM algorithms are able to efficiently detect and leverage different levels of similarities, from 100% to various fuzzy matches.

In order to help our life sciences customers to most effectively leverage existing translation assets, our translation memory management system includes different levels of content matches such as 100% matches, in-context exact (ICE) matches, and fuzzy matches. The different levels of TM matches are designed to allow health care organizations to maximize translation leverage for a given document. For example, a medical device IFU (information for use) manual may contain some updated content of new features, but also share some exact same content pertaining to legal disclaimers, as well as common content with the previous product model but with minor updates. In this case, ICE matches may be great for the legal disclaimers that haven’t changed at all so the entire section can be leveraged at no charge. 100% matches are those sentences (segments) that haven’t changed, but sentences around them have been updated within the same paragraphs. Once translated, these sentences should be reviewed to ensure they are still correct (in relation to language gender, single/plural forms etc.) Fuzzy matches are sentences that have been modified, but most of the words remain the same. The following is a chart on how Translations.Health processes the various translation memory matches.