Artificial intelligence is transforming machine translation (MT). Deep learning and neural networks have evolved into a new model – neural machine translation (NMT) – that has made MT faster and more sophisticated. But the future of translation lies in combining this new technology with human translators. Machine translation with post-editing is a rapidly developing service, and it heralds a swath of new opportunities for companies with a global footprint.
What Is Machine Learning with Post-Editing?
In 2014, researchers introduced the concept of neural networks to machine translation. Rather than memorize “windows” of words in sentences as previous models had done, neural networks look for the whole context of a particular translation. As a result, NMT produces a higher-quality translation in a shorter amount of time. What’s more, it continues to “learn” and improve.
Still, the technology has limitations. Though it’s suitable on its own for some use cases – such as getting the “gist” of an email message – it falls short when greater precision is required. That’s why machine translation with post-editing is one of the fastest growing services in the translation industry. The process is simple. Content is initially translated using NMT. Then, a skilled human translator reviews the output and edits for clarity and precision. The result is a better-quality translation – in less time.
This process is also highly efficient, an important consideration for businesses that need to translate larger volumes of content. For instance, as the internet has expanded access to a more global customer base, marketing teams must make social media and digital advertising content available in local languages. This is not an insubstantial task, especially as six in 10 international marketers don’t have support in all the local markets in which they operate, according to data from eMarketer.
However, machine translation with post-editing streamlines and simplifies translation projects. This process allows organizations to:
- Handle large volumes of content. Many translation projects involve hundreds or even thousands of documents or pages. Examples include website pages, product support documents, human resource materials, and litigation documents used in the discovery process.
- Accelerate turnaround time. Using NMT for the initial translation means human translators can focus on smaller sections of content, enabling them to deliver a high-quality translation in much less time than if they’d started the entire project from scratch.
- Achieve greater precision. Though it is fast, NMT has its limitations. While it produces significantly better translations than statistical machine translation, it isn’t yet adequate for specialized text. However, when human translators conduct post-editing on machine-translated output, they can clean it up faster while ensuring creative and nuanced text is precisely translated.
- Improve efficiency. By providing a foundational translation from which to start, NMT helps human translators be more productive. For instance, one longtime translator estimated that a good quality machine translation improves his productivity by 30% to 40%.
Working with Machine Translation with Post-Editing.
When evaluating a project for machine translation with post-editing, be sure to consider the following.
- Languages. Because of the large amount of training data needed for NMT, the technology serves some languages better than others. For example, English, Spanish, and Chinese are more widely spoken and thus will require less post-editing and time to produce. Other languages will require a native speaker to produce a higher quality translation, extending the post-editing time.
- Specialized text. Machine translation with post-editing will become more common for standard translations, but specialized content will still need the involvement of human experts. One of the limitations of neural machine translation is domain adaptation. The technology performs best in general use cases, but it’s not quite ready for specialized use cases, such as patents or healthcare.
- Sector expertise. In order to ensure the subject matter is accurately represented and is compliant with government regulations, you need sector expertise. Certain disciplines such as life sciences and manufacturing have their own shorthand and terminology. Precise translation of this type of content is more difficult for artificial intelligence to achieve, making post-editing a critical step in the process. The American Translators Association recommends working with a translator who knows your subject inside and out.
- High-stakes communications. What happens when more is at stake? Legal advisers can’t afford to misunderstand text as they determine how to proceed in a significant lawsuit. Products may not sell if the translation misses the cultural mark. Precision is essential in high-stakes communication, and the cost of errors can be high. For example, literal translations can lead to serious misunderstandings, especially in domains where words may have different meanings. A skilled linguist can ensure cultural accuracy.
When to Use Machine Translation with Post-Editing.
Machine translation services such as Google Translate are valuable for personal use or when “a sense of the content” is an adequate outcome. For example, Google Translate is ideal for tourists who want a quick translation of foreign-language signage at places they visit. Businesspeople conducting first-line research may value speed more, if they only need to get the gist of the news covered in an online article.
But for large scale projects that need to be completed quickly and accurately, machine learning combined with post-editing produces the best results. Artificial intelligence offers an efficient, cost-effective way to handle the large volume of content that businesses need to translate. However, human translators are better at capturing style, cultural nuances, and context than their machine counterparts.