It's not MT, and it's not TM, but it certainly helps accelerate translation work. What is it? A new generation of translation tools that builds on older principles of Example Based and Statistical MT and resolves deficiencies in classic TM. We call it ‘Advanced Leveraging'. It combines statistical analysis and linguistic intelligence tools to create a new category of fuzzy matches that can lead to significant increases in translation productivity.
Advanced Leveraging
Ever since the beginnings of MT in the 1960s and the development of multilingual publishing and information management in new European and international organizations, scientists and practitioners noticed that digital translation content could be recycled, as well as being translated (with great difficulty) directly by machines.
They realized that if "linkages" (we would call them alignments) between source text and target text could be maintained, translators would be more productive and terminology and style more consistent. But it was not until the late 1980s that linkage-driven translation memory (TM) tools were finally developed to meet the surge in localization and translation needs for the burgeoning computer industry.
TMs have been the principal translation productivity tool for two decades now. But today, large scale translation users are looking for even better results from their TM technology. This has led to research into new forms of "advanced leveraging" that overcome some of the limitations of classic TMs.
The problem with TMs
The key drawback of classic TM technology is the lack of a capability to leverage sub-sentence segments, either by searching for them as phrases in memories, or aligning them fuzzily or exactly as source and possible targets. Alignment will not operate on paragraphs, or from two target sentence to one source sentence and vice versa. Nor will it capture the similarity of content if two phrases are reversed in a new piece of text. In other words, classic TMs can only assemble new translations easily and effectively from old translations under very restricted conditions.
It is also impossible to use TMs to extract terminology automatically without employing separate tools. Finally classic TMs have proved to be most effective when used with restricted document collections with regular updates comprising a limited number of types of changes. This rules out potential productivity gains for translations of texts such as reports, news items, emails, websites, which may include plenty of reusable terms and phrases that "link" back to existing translations but which do not come into the "document update" category at sentence level where TMs do well.
To overcome these shortcomings and extend the power of the basic TM vision, Advanced Leveraging (AL) offers a mix of existing tools and techniques from natural language processing, statistical and example-based machine translation, search, and alignment. It is not a new paradigm of translation technology, nor is it an alternative name for MT. AL is best thought of as a new generation of enriched TMs that goes a long way towards bridging the gap with fully automated translation.
New generation alignment and search
Although no single system on the market will necessarily provide the same set of AL features, the key processes involved are:
- Alignment: where sentence alignment can pair one source to many target sentences, and many target sentences to one source; Legacy TMs can also be aligned at the sub-sentential level, providing added leverage to companies with very large TM repositories.
Sub-sentential alignment, using phrase matches, or morphological or statistical information or a combination of all these is the core strength of AL.
- Search: where cascading searches for fuzzy or exact matches on incoming (new) text can range down from paragraphs through sentences to sub-sentential segments - with the latter being the big productivity win.
In addition to adding measurable advantages to the actual process of translating, these AL features also correspond more adequately to the market environment. For many companies with heavy translation needs, the real challenge is no longer to publish and translate new releases with small quantities of new content to be added to the basic documentation. The agenda is increasingly about addressing user-driven needs for information on bugs, alerts and additions to knowledge bases, rather than pushing out massive static documents. In this new, customer focused market, the key requirement is to integrate the translation process into a semi-continuous stream of content production. This means very quick turn around times for the translation process, and maximum leverage of existing translated content.
Corporate content is increasingly created in XML as chunks of information, a standards shift that can be used to facilitate the translation process. Any source language change to a given chunk in one place (say a screen display) can be automatically updated right across the product document set, including for example the online user manual. When the same content change has to be translated, the same word, phrases or other sub-sentential item can be translated once and cascaded through the translation set for the same product.
In such a typical publishing model, AL provides a better solution to the kind of translation needs emerging in an XML-based and chunk-architected approach to content management. In a typical case, a classic TM approach might only find 25% recyclability from matches within a memory database, whereas AL, with its capacity to search down to sub-sentential segments and across broader memory bases could double or even triple that leverage.
Who's developing Advanced Leveraging tools?
Technology products currently on the market that provide various types of AL tools include Atril(Déjà Vu X), ESTeam (ESTeam Translator), KCSL(NoBabel Translator), Language Weaver (Statistical Machine Translation Software), Lingotek (Language Search Engine), Lingua et Machina (Similis), Multicorpora R&D (MultiTrans 4), and Oracle Worldwide Product Translation Group (Translation Factory - internal only).
More on Advanced Leveraging
TAUS has produced a report on Advanced Leveraging - The New Generation of TMs authored by Bob Kuhns that provides an in-depth examination of AL tools, including detailed profiles of each functionality, examples of translation results, numerical data on leveraging and a comparative chart of eight products. This report is available to TAUS members.