AI (artificial intelligence) technology in the machine translation industry has made a huge jump during the last decade. Just in 2016, Google released GNMT (Google Neural Machine Translation), one of the first widely used MT(machine translation) on AI, and already today, people are using ChatGPT and other similar bots that can greatly understand the context and partly replace humans for some tasks, including paraphrasing and translating. In turn, NMT translating systems also did considerable work (for example, in Google, the accuracy for English-Spanish translation reached 90%, which is an outstanding achievement). This fast development of technology has inspired us to explore the AI translation systems landscape more deeply to understand their current capabilities and future potential.

AI translation: overview, pros and cons

The AI translation market was valued at $2.17 billion in 2023, and the prediction for 2030 is $5.72 billion. The Google Trend reports also show how the search for AI services has increased over the last two years.

google trends AI

There are a few reasons that explain such forecasts and popularity:

  • The world is constantly moving toward blurring borders. Globalization has been a central component of success for countries and companies in recent centuries, and in the 21st century, the WEB became a new power. Businesses need tools to communicate and efficiently work with customers and partners worldwide.
  • There is an increased demand for accurate and fast translation. Traditional translating systems were more effective than people with dictionaries but needed help understanding context. That required more time for proofreading even simple sentences. Businesses need and are ready to pay for more accurate alternatives.
  • There is high competition between tech giants. AI is a vector of development for all the big technical corporations, which leads to the release of new features and increases user interest. For example, the AI software vendor Appen got 80% of its revenue from Microsoft, Meta, Apple, Amazon, and Google. These tech giants continuously buy promising AI startups or partner with them, giving users new abilities sooner or later. This race of large companies popularized the topic of artificial intelligence in general and in the field of translation.

Let's sum up the main pros of the AI translation, as we touched on many of them above.

  • fast translation;
  • better context understanding in comparison with traditional translating systems;
  • perfect scalability - one tool can be used for different purposes (languages, types of projects, etc.)
  • affordable cost in comparison with a live translator;

But nothing is perfect under the sun; otherwise, all translators would have lost their jobs by now. So, the cons of AI translation are:

  • Limited language support. Any AI language model requires a lot of content for "studying," so additional languages require more power and more training materials. One of the most popular AI ChatGPT today can "speak" 11 languages and know a few additional languages on the low level.
  • Quality of training material. The information for the AI translating model is taken from the web, and no one can check the quality of all these documents. So the system can study mistakes.
  • AI is training constantly. The AI model can learn incorrect information during communication with users.
  • Legal issues. Translating sensitive or confidential information with AI services is risky as data can be shared with external servers.
  • Cultural sensitivity. AI translation models can't understand all the cultural nuances, which can lead to offensive content.
  • Ethics. All AI models are trained on subjective data and will inherit biases. So, the quality and ethics of AI translation depend on the development team.
  • AI models can't work independently of humans. No matter how well the AI translator is trained, it should only be used with control. Every business bears responsibility for each word on its business pages. Thus, the proofreading stage will always exist.

Human translation: overview, pros and cons

Human-only translation is a less popular method. It requires more resources (time and cost) for an equal volume of data than machine translation. However, there are many situations where it is preferable to use only human efforts for translation. For example, the research made with 7 imperative sentences showed that the equation between Google Translate and human translation was only 29%.

Such a low figure explains the vast number of translators worldwide - more than 600K - and proves that the translator's job will be requested for many years. Other areas where human translation is always required are:

  • Marketing and advertising. Slogans, ad texts, and video scenarios cannot be translated for different audiences; they require adaptation with a deep audience understanding. Marketing and advertising are always focused first on attracting, informing, and retaining. For example, all of us will still remember some obsessive text from an ad song from childhood. Thus, the content for this industry is mainly transcreated rather than translated. AI machine translation models are useless in this field as they don't have a high level of creativity like humans.
  • Confidential information. Companies would be forced to implement a boxed solution with an AI translation model inside their company infrastructure to guarantee the data's safety, which is a very complex and expensive solution. Otherwise, there is always a risk that the data can be stored and sent to external servers. In this situation, businesses prefer human-based translation to minimize the chance of data breaches.
  • Specialized content: tech documentation, manuals, medical reports, guides, math or physics documents, etc. These types of content cannot be translated with AI as the terms and data are narrowly specialized. For example, content related to the mineral resource "lead" can be mistakenly translated by AI to mean "leader."
  • Literary works: сreative writing often contains wordplay, other literary elements, or deep cultural references. All of this requires an understanding of the context and the initial author's idea.

Besides the high quality of human translation and its versatility, there is also a list of cons:

  • Losing time-to-market. Human translation speed is significantly inferior to AI or traditional machine translation. In today's business world, competing means having a quality product and being the first to launch it to reach the target audience.
  • Difficult scalability. Finding professional translators is always a challenge, especially for rare language pairs. At the same time, even if you build the translators team, they can only cover a certain amount of work, and you can't scale their efforts 1.5 or 2 times.
  • Human factor. People can bring their interpretations into translation, which leads to different styles and interpretations. Also, the human factor assumes that translators can quit or get sick, shifting the schedule.
  • Increased translation cost. Because of all the points above, translation costs will steadily increase. The final price depends on factors like language pairs, content complexity, translation speed (urgent translations will always cost more), translators' experience, and expertise.

The information above makes it easy to understand that the best method for content translation is somewhere in the middle. Optimal indicators for businesses can be achieved by combining AI machine translation speed with human context understanding. How to make that in the best way? We know the answer.

How does Lingohub combine machine translation and human translation?

As a translation management service, we see the cases of different businesses and hear their needs in the first stages. Different feedback allowed us to build a system that perfectly combines the machine translation power and harmoniously includes translators to get all the best from both types of translation.

First, Lingohub combined Google Translate, Microsoft Translate, and DeepL to produce the internal machine translation result. This decision was made because each of these engines has its own strong and weak sides, which we balanced by this solution.

The Lingohub machine translation feature is provided in two ways: first, the machine translation tool, where the user can type the query or click on the segment to get the ready translation, and second, automated machine translation for selected segments via prefill. We will overview the prefill feature in more detail, as it allows the pre-translation of a vast amount of data in a few clicks.

perform prefill

Note: Lingohub can use another language or translation memory besides machine translation to fill the empty text segments.

So, how does the schema work? The localization manager can calprefill on its own or set up automatic prefill rules. Thus, translators will get machine translation results instead of working with blank spaces. Such an approach saves a vast amount of time because:

  • Translators don't manually type translations for simple content, such as text segments with buttons like "Start," "Send," "Submit," etc. All they need to do - proofread and approve the text's correctness.
  • Translators have a starting point for complex texts and can lean on the pre-translated content.

The next tool is designed to set a fair and transparent price for translations based on MT results. The thing is that some text segments can be fully translated via machine translation, and the translator's efforts are light in this case, but another segment will require much more time and effort. To solve this dilemma, Lingohub offers EES (Edit Effort Score). We provided a detailed description in the blog article about machine translation post-editing; below is a brief description.

EES quantifies the level of editing by ranging from 0 to 100, where 0 is proofreading (no edits), and 100 is fully edited text. So, there are four effort levels, and customers can set the range and price for each. For instance:

  • Proofreading - 0.02$ per word
  • Low effort - 0.04$ per word
  • Middle effort - 0.06$ per word
  • High effort - 0.08$ per word
manage efforts rates

Thus, Lingohub combines machine translation and human efforts and provides all the necessary tools to manage this collaboration conveniently.

The AI market is constantly growing, and we believe that one day, AI language models will cover a more significant amount of translating work. However, humans will always keep everything in the correct direction and be part of the translation process. With this understanding in mind, our team does our best to provide the optimal balance by using the strengths of each approach.

If you would like to get more information about Lingohub and our features, schedule a quick demo call with our team or sign up for a 14-day free trial to try all the benefits of Lingohub TMS (translation management system).

Stay awesome, and let's localize together!

Try lingohub 14 days for free. No credit card. No catch. Cancel anytime