Machine Translation (MT) has become a vital part of the language industry, but that does not mean that human translation will become obsolete. On the contrary, there is a strong need for human translators to work in conjunction with machine translators to achieve first-rate results.
MT (Machine Translation): An Overview
With roots dating back to the seventeenth century, MT was launched in the 1950s, when research funded by the U.S. government first garnered global interest in the concept. During the machine translation process, content is automatically transferred between languages through the following methods:
- Rule-based MT, which generates translated text by developing grammar rule sets for the source and target languages.
- Statistical MT, which creates translations from statistical models based on the source and target languages.
- Neural Machine Translation, which uses a large artificial neural network to compute the likelihood of a words sequence, resulting in contextually accurate translations.
Since computers are able to rapidly process a translation, the primary benefits of machine translations are a quick project turnaround time and cost savings for the client.
Limitations of Machine Translation
Although MT can produce fast outputs, it comes with many limitations, including:
- MT engines do not have the capability to consider the context of the text, leading to translated phrases that are incomprehensible.
- Languages are an intricate blend of cultural nuances, and aren’t always compatible with machine processing in terms of native phraseology.
- The same language in two separate countries can be different, with varying idioms and expressions in each location. This is another area in which machine translation can fall short, simply because it might not be able to capture even slight language differences from one region to another one.
- MT (Machine Translation) engines often make mistakes in specialized areas that require technical writing, because the meaning of the word is different than what the machine recognizes. In addition, the program might simply not be able to decipher a word that is unique to a specialized technical area.
Because of these limitations, raw machine translations are often best served for providing the gist of what the translated text means, rather than a complete translation. With this in mind, machines can never replace human translators; instead, machine translating tools can complement humans by making our work more efficient.
Human Post-Editing: A Complementary Approach
Perhaps one of the best ways to integrate the work of humans and machines to achieve optimum translation quality is by adding the step of human post-editing. This step involves a human translator reviewing machine translated content to make it more readable or better suited for regions, cultures, or specialties. Based on the quality of the original text and the machine translation, there are three primary levels of human post-editing available:
- Light post-editing, meaning that the translator focuses primarily on ensuring that the content is readable. Instead of concentrating on nuances or advanced grammar, the translator confirms that the main gist is acceptable.
- Full post-editing, which includes making certain that the text flows well, is understandable, and is highly accurate.
- Human-level post editing, which ultimately results in an output that reads like it was written in the target language. With this level of post-editing, all figurative languages, cultural nuances, idioms, and unique parts of the target language are translated so that the content is fully understood by the reader.
Since “Human Post-Editing” can be costly and requires a certain amount of time, several steps can be taken to make the process move faster:
- Provide clean source text. By following this step, you can save a large amount of time in the post-editing phrase. Clean text includes consistent language, specific terminology, and culturally sensitive word choices, as well a format that leaves room for the translated output.
- Update memory databases. Ensuring that translation memory databases are brought up-to-date for the target language before translation begins will help make certain that the output is accurate. This step is especially helpful during the translation of technical documents that include specialized terminology or long documents with repetitive words and phrases.
- Start with a translation style guide. This important step can save quite a bit of time, because it means that all translators will be working from the same template.
Guidelines to Achieving Optimum Quality with Machine Translation
As the use of machine translation continues to expand, there are a number of useful tips that can be followed to achieve superior quality:
- Use short sentences and standard terminology.
- Make sure that primary terminology is accurately translated, and create a list of terms that should not be translated.
- Carefully review the content to be certain that no information has been deleted or added.
- Bee on the lookout for anything in the content that is culturally unacceptable or inappropriate.
- Ensure that the formatting is acceptable and has the correct space needed for the text.
Google Translate’s daily translation count of 100 billion tells you that there’s clearly a market for machine translation. And it saves you a lot of money, making it ideal for companies on a tight budget; it’s pretty much a given that MT’s here to stay. However, consider this study conducted by Sejong Cyber University in Korea, wherein a team of human translators competed against three machine translation applications. Although the machines were quicker, they made more errors in the final output. In addition, 90% of the text that was translated by the machines was considered to be grammatically awkward. This tells us that however handy, MT is nowhere near the finished article. In the long run, collaboration between machines and humans is what will provide the best balance between cost savings and achieving the level of precision and language understanding that only humans can provide.
In fact, combining complementary technologies and harnessing their power with human expertise will keep paying richer dividends for companies. As alluded to in our recent blog A New Frontier in Translation Automation – Leveraging AI for Greater Efficiency and Customization, AI-led machine translation used for “linguistic mapping” based on content type and subject matter to build customized “linguistic teams” and as part of automating workflows, can have a powerful impact on efficiencies achieved, while optimizing the quality delivered. The results are favorably comparable even to MT backed up by human post-editing.
