Machine Translation for Translators

Background Information: History of Machine Translation

To make it clear: Though the latest developments in the field of neural machine translation are quite remarkable, there is no reason to get hyped up. Examined on a broader basis, the development of machine translation has always taken place in the form of a "punctuated equilibrium". There have been several phases that many considered to be the final strike that would eliminate the need for translators.

For example, this was the case from 1948, when machine translation was first introduced. However, in the ALPAC (Automatic Language Processing Advisory Committee) report that was published in 1966, the experts of the National Academy of Sciences came to the conclusion that machine translation was slower, less precise, and twice as expensive as human translation. This sobering calculation doused the excitement.

The hype flared up once more when IBM introduced statistical machine translation in 1988. Thereafter, the statistical and rule-based machine translation methods were long considered as "mere" useful aids in the translation process. Again, the voices of the doomsday prophets who had foretold the demise of the translation profession faded out.

The buzz started again when the first scientific paper about neural machine translation was published in 2014. A year later, OpenNMT was presented as the first operational NMT engine, developed by Harvard in collaboration with SYSTRAN.

This article is not meant to be a history lesson, as we all know what has been going on since then. From Google to Amazon, Microsoft, Facebook and DeepL—all large and many smaller enterprises employ artificial intelligence and neural machine translation. The industry is changing, and with the right preparation, translators can benefit greatly from these winds of change.

Tip 1: Use Both Machine Translation and a CAT Tool

As long as machine translation was mainly offered via statistical systems, translators were unable to connect a high-quality system to their CAT tool. In view of the high cost and time input involved, configuring a customized system was out of question for most translators.

Today, neural machine translation enables you to connect your CAT tool to excellent, reliable systems for a reasonable fee. With these systems, you can process your translation tasks (subject to your customers' approval) and generate much more translation output in less time.

At the same time, you can prepare effectively for future post-editing tasks. Successful post-editing can be quite difficult initially and requires practice. To be productive, you need to be two to three times faster than in your translation work. According to KantanMT, the daily output would have to amount to about 7,000 words. To reach this speed, you should practice post-editing for some time before you offer the service as a specialist. By connecting a machine translation system to your Across Translator Edition or another CAT tool, you can learn and practice post-editing while handling your translation tasks.

Tip 2: Take an Open Approach to the Topic of Post-Editing

Enterprises use machine translation in order to save translation costs and speed up the process. Another aspect is the increasing amount of content, which would be impossible to handle otherwise.

Often, foreign-language texts merely need to be understood roughly and are therefore suitable for machine translation. For example, this could be the case when a German project manager receives an e-mail from a French colleague and merely wants to know what it is about. In this case, it might be of minor importance if the sentence structure is not fully correct or if the text contains grammatical errors. If, however, a machine-translated text is to be published, your expertise is required. Your customer will appreciate your advice concerning post-editing. Provide your customer with realistic information on the workload and honest feedback on the projects. To many customers, post-editing is uncharted territory. As post-editing is still a relatively new service, there are not yet too many experts on the market. Thus, the chances of gaining new customers and projects are good.

Open, honest communication is another key factor to ensure the success of your cooperation with your customer. For example, the final quality of a translation should be determined before the project even starts.

Remember that you have the control over your work and can decide which ones to accept. You are the specialist, and you have the needed expertise to decide whether or not a project is worthwhile to accept. Remember: If the output is poor, you will not be able to reach the required editing speed. Therefore, consider in advance which types of text are suitable for machine translation and which ones are not. For instance, short, standardized sentences can easily be translated by the systems. Accordingly, MT is especially suitable for technical texts. By contrast, MT systems have difficulties with creative content.

For this reason, machine translation is not recommend for texts characterized by a more demanding style, e.g. marketing or literary texts. Be sure to give your customer the needed feedback in case you decline a project due to its poor quality. Based on this feedback, the customer may be able to adapt his processes or introduce additional steps to improve the quality of the machine translation, allowing you to accept future projects. Good interaction between the customer and the translator can thus deliver a win-win situation.

Tip 3: Understand Machine Translation

As mentioned previously, you can become a post-editing expert by knowing as much as possible about machine translation. For example, you should know how the various systems work and how enterprises can prepare their texts for machine translation.

Though neural machine translation is increasingly becoming the system of choice, it is good to know the basics of rule-based and statistical systems.

Rule-based Machine Translation

The rule-based approach is the conventional MT method. The development of a rule-based system is very costly and time-consuming, as every linguistic peculiarity needs to be entered manually. For this reason, this approach is gradually being abandoned. Nevertheless, it is to be noted that rule-based machine translation delivers good terminology proposals, as the system is systematically trained with corporate terminology. Moreover, the translations are always complete, and the results are predictable. The main disadvantage is that the translations sound very mechanical, and the sentence structure is not presented very effectively.

Statistical Machine Translation

Statistical machine translation is based on the approach of creating translations on the basis of probability calculations. The information required for this is extracted from bilingual corpora. As the sentence structures and the terminology are different in every corpus, the output might be marred by a lack of consistency, which can impair the legibility. Moreover, these systems can produce incomplete translations, wrongly add information, or make capitalization and spelling mistakes.

Neural Machine Translation

Neural machine translation is based on an artificial neural network (ANN) that mimics the neural connections in the brain. This approach, too, involves the analysis of parallel corpora for the translation. The difference is that in the ANN, the grammatical context of the sentences is implicitly taken into consideration. This means that the texts are not translated at phrase level, but at sentence level, which greatly contributes to the legibility. To date, the greatest challenge of neural machine translation is the still limited vocabulary that the models are able to process (currently 50,000 to 80,000 words). Therefore, the post-editing needs to focus more on the lexicon than on the grammar. The disadvantages of this approach are the same as those of statistical machine translation. As translations created with NMT are very pleasant to read, there is a risk of overlooking errors. Nevertheless, neural machine translation is the approach that currently delivers the best results. Therefore, a more careful review at the lexical level is usually worth the effort.

Tip 4: Make Distinction between Light and Full Post-Editing

It is not possible to formulate globally valid post-editing guidelines, as the quality requirements for light post-editing and full post-editing are different. Light post-editing merely corrects the most conspicuous errors, while full post-editing conditions the text to an extent that it can no longer be distinguished from a human translation. Prior to the start of the project, the customer needs to tell the translator what kind of post-editing is required. In combination with the output quality, the translator can thus assess his or her workload. You can find further information about post-editing and the different typey in our article "post-editing - higher quality for machine translation".

Tip 5: Include Post-Editing in the Service Portfolio

Many translators wonder whether it is possible to make as much money with post-editing as with conventional translation. As in the case of translation jobs, you can calculate your rate per word on the basis of your output and your target income. Usually, some 50 to 60 percent of the translation rate per word is charged for post-editing. Of course, this greatly depends on the type of post-editing and the text quality to be achieved. You will gradually become faster in your post-editing work and thus achieve higher rates per hour.

For example, Sara Grizzo offers useful tips to become faster. She is a freelance translator and post-editor, and she offers workshops and lectures and the subject of post-editing. If you know German, you might want to read her article in the "technische kommunikation" magazine, issue 03/2018.

• First, go over the text superficially in order to spot any repeated errors.
• Correct repeated errors by using the search and replace function of your CAT tool.
• Use the automatic quality management functions of your CAT tool in order to find common errors, such as capitalization errors, spelling errors, punctuation errors, etc.
• Use your keyboard instead of the mouse. Using the arrow keys and various keyboard shortcuts, you will be able to work faster.
• In the post-editing phase, the decision process if very short. As the machine output is to be used as often as possible, the post-editor must restrain him or herself from correcting each and every grammar or style error when engaging in light post-editing.

Tip 6: Become a Post-Editing Expert

Participation in training measures is recommended in order to prepare intensively for post-editing. For example, TAUS offers a six-hour online course, and the Federal Association of Interpreters and Translators (BDÜ) regularly holds seminars on the subjects of machine translation and post-editing.

Other Articles That You May Find Interesting:

• Machine Translation for Enterprises
• 6 Reasons Why You Should Update to ATE v7
• Time Management: 7 Tips for Freelance Translators