After more than half a century, machine translation has evolved from science fiction to reality. Developers of machine translation systems no longer fret over being confronted by a problem without a solution, while translators are no longer worried lest the "rise of the machines" should deprive them of their livelihood. While a computer will hardly ever replace a human translator in all areas, machine translation is already being put to effective use, even if to accomplish only a strictly limited set of tasks. It is important to know the operating principles of machine translation, its limitations and fields of application so as to understand when machine translation should be used or avoided, which will save you the disappointment.
Kinds of Machine Translation Systems
Machine translation systems belong to one of the three categories: Rule-Based Machine Translation (RBMT) systems, Statistical Machine Translation (SMT) systems, and the most promising "hybrid systems" combining the benefits of RBMT and SMT. RBMT systems analyze the text and build the translation using built-in dictionaries and a set of rules applicable to the particular language pair. SMT systems rely on statistical analysis: large volumes of source text (amounting to millions of words) along with target translations performed by humans are loaded into the application. The application analyzes the statistics of interlingual matches, word usage, and syntactic structures, and relies on this analysis when choosing fitting translations. This process is known as self-training. The system can be also trained by human translators who edit the translations produced. The widely known Google Translate service uses this very principle. Owing to the self-training ability of statistical and hybrid machine translation systems, the translation quality improves as they accumulate linguistic data with each completed translation.
Dos and Don'ts of Machine Translation
The key merit of machine translation is that it can handle very large volumes of text quickly, which sometimes makes it more cost effective compared to human translation. That said, one should always remember that machine translation quality will always be inferior to that of human translation. It therefore makes sense to use it only in particular cases.
First, machine translation is suitable for materials intended for internal use. For example, you need to get the gist of a foreign-language website, article, or letter, or find news reports on a particular event in international publications. Second, machine translation can be used on technical and highly specialized texts that will then undergo post-editing by experts in the field. In this case the target text is used as interlinear translation based on which the technical expert will produce the final text relying on the knowledge of the subject area.
Many types of materials are not suitable for machine translation by definition. Machines cannot be trusted with texts where inaccurate translation could jeopardize human health, a sophisticated device, or a major contract. In this case the time savings do not justify the risks. Any documents associated with legal liability (contracts, warranty) must be processed by humans. Machine translation is not suitable for marketing materials where the text is actually recast in a new cultural context and recreated from scratch.
Generally adequate quality can be expected when translating strictly formalized technical texts, whereas literary translation and promotional texts are beyond the scope of machine translation.
Preliminary Text Preparation and Post-editing
Preliminary preparation of text can considerably simplify the job of the machine translation systems and editors tasked with refining the raw machine translation. Such preparation begins at the stage of composing source text. To this end, standards are developed for technical writers and authors, who abide by them to make the text easier to understand and translate, for machines and humans alike.
Following these three rules goes a long way in enhancing the quality of machine translation from English:
- Use infinitive verb forms instead of gerunds
- Use the active voice instead of passive voice
- Avoid compound sentences and homogeneous parts of sentences
Ideally, each sentence should convey a single logical thought. This rule in particular, which applies to all languages in equal measure, is the most effective of all three.
It has been proven in practice that following these simple rules and adapting the machine translation system correctly can considerably simplify the post-editing of the target text. This should give you an idea of the benefits that come with formalizing and standardizing texts ahead of machine translation, be that management of text authoring with the use of special applications, pre-editing of text, or simply requiring the author to follow a basic set of the most effective rules.
Post-editing involves refining raw machine translation by an editor, who normally has special training and is experienced in handling machine-translated texts. In most cases machine translation requires post-editing. However, it can be omitted occasionally, especially when texts are translated for internal use or to get the overall gist or locate and select specific materials. The amount of time and labor involved in post-editing is one of the key factors to be considered when evaluating the cost effectiveness of machine translation. Literary, promotional, and other texts that are not suitable for machine translation by definition are not subject to post-editing either: to improve text quality to a level achieved by human translators, the editors would have to rewrite the text from scratch, thus cancelling out any benefits of using machine translation.
Volume and Economy
Before resorting to machine translation, one should not only have a clear idea of the desired end result and realize the limitations of this method, but also bear in mind another factor. Machine translation systems require complex customization and improvements, including "training" in a particular subject area, without which they fall short of expectations. It therefore makes sense to use machine translation only on large volumes of similar texts. Only in this case will it be cost effective to spend a certain amount of time on system training before applying machine translation and obtaining text suitable for post-editing. Meanwhile, when a job involves several dozen pages, attempting machine translation would be futile and eventually costlier.
Conclusively, machine translation with post-editing can be worthwhile when you deal with large volumes of similar texts. Because large translation volumes pass through translation agencies that often specialize in particular subject areas, deploying fairly effective yet expensive machine translation solutions of the latest generation is economically justified at such companies in particular, while neither content providers, however large, nor freelance translators can make effective use of machine translation on their own.