Human Translation vs. Machine Translation

Human Translation vs. Machine Translation

In the past few decades, technology has evolved at an incredible rate. Our professional and personal lives have been impacted by wonderful technological advancements, marvels beyond our imagination. Technological progress has made our current degree of globalization possible. The world has certainly become “smaller” and more interconnected. 

The commercial and cultural exchange made possible by technology has immensely increased the need for language services. And the translation industry itself is no stranger to technological disruption. While the possibility of human translation vs. machine translation has been around since the 1950s, in recent years it has become a very palpable reality.

Professional translation services providers rely on CAT tools and other translation software to facilitate their process. The development of translation technology, as well as the development of AI (artificial intelligence) in the form of artificial neural networks capable of processing natural language naturally, results in a question: Will machine translation services ever be capable of replacing human-powered solutions? 
Human Translation vs. Machine Translation, a website compiling the findings of a 2013 study by Carl Benedikt Frey and Michael A. Osborne suggests that there’s a 38% chance of “robots” (AI) taking linguists’ jobs. The website also features polls, where the humans in question can vote on the likeliness of their being replaced. Linguists are of the opinion that there’s a 58% chance of automation in the next two decades.

In this post, we’ll compare human translation vs. machine translation. We’ll be taking a look at:

  • The importance of machine translation, and how it evolved in the last few decades
  • The benefits and downsides of machine translation
  • The benefits and downsides of human translation
  • The very common practice of machine translation post-editing
  • How the human translation vs. machine translation dichotomy will evolve in the near future.

What Is Machine Translation?

“Machine translation” is an umbrella term. And it can obscure the incredible technological diversity we find in this field. Machine translation is the translation process, carried out in a fully automated way, by software designed (or “trained”) for that purpose. Currently, when referring to machine translation technology, we’re most often referring to the work of artificial neural networks. But, what does it consist of? Let’s have a quick look at the history of machine translation.

Machine translation has always had a bad reputation, as it was deemed to have poor, inaccurate outputs and be inefficient in terms of costs. But this has changed in the last two decades: scientific development (especially in the field of AI) has made MT more competitive. And the numbers are on its side: the machine translation market is growing at such a fast pace that it’s expected to reach a cumulative value of 1.5 billion dollars by 2026.

Machine translation evolved through different protocols, from its beginnings in the ‘50s. The early Rule-Based MT models were succeeded by Example-Based MT in the ‘80s, and by Statistical Machine Translation in the ‘90s. Artificial neural networks are the latest standard in the field. 
Human Translation vs. Machine Translation

Rule-based machine translation tools compare the key grammar and syntax rules between the languages at hand. As their name indicates, these tools work by understanding the rules of each language. Its outputs are limited to literal translations. And, the more complex the language, the more rules are needed to translate. And rules need to be developed manually by linguistic experts, making this method time-consuming and costly, as well as potentially unreliable. At the end of the day, language rules are often broken. A language is always diverse, multifaceted, and changing. 

Example-based machine translation is essentially based on searching analog sentence pairs for the source and target languages. This requires building large collections of sentences in both languages, which makes the method quite accurate, but costly and not very flexible. Polysemic words or phrases also bring about the possibility of ambiguities.

Statistical machine translation (or SMT) is powered by statistical models. The parameters in use are derived from the analysis of the corpus of bilingual texts, meaning, from the parallelism of two texts that are the same but in different languages, and generating statistical models of language usage. This protocol struggles with highly inflected languages. This happens because a large number of inflected words generate a lot of data sparsity. Before the introduction of neural machine translation, SMT was by far the most widely studied machine translation method.

In an article for MITNews, Larry Hardesty defines neural networks as “a means of doing machine learning, in which a computer learns to perform some task by analyzing training examples. Usually, the examples have been hand-labeled in advance. An object recognition system, for instance, might be fed thousands of labeled images of cars, houses, coffee cups, and so on, and it would find visual patterns in the images that consistently correlate with particular labels.”

Unlike the systems before them, neural networks are integrally trained to maximize performance. In simpler terms, neural networks are capable of “understanding” the differences between the source language and the target language.

The Benefits of Machine Translation

Machine translation can result in low turnaround time, short time-to-market, and reduced translation costs. But, at the same time, machine translation falls short. Machines are yet to understand languages with the same nuance as humans. This may result in inconsistent or unintelligible translations. Especially when the text at hand is long, complex, or has several layers of meaning, the result of a machine translation will probably be subpar. 

And that subpar translation will carry hidden costs with it. For starters, it’ll need to be revised by a team of human translators. 

Machine Translation Post-Editing

Machine translation and human translation are often approached as opposites. Especially by people outside of the language industry. What if it didn’t have to be human translation vs. machine translation? 

In recent years, machine translation technology has entered translation agencies, not to replace human translators’ work, but as an aid. Machine translation is often used to produce an initial, “raw” version of the translation, which will then be edited by human linguists. This process is called “machine translation post-editing” (MTPE).

A 2013 report noted that approximately 30% of LSPs are using machine translation in this fashion. We could say that MTPE is the current state of machine translation in the language industry.

Some translation companies handle both machine translation and machine translation post-editing. Others work with AI services providers, to examine and edit the machine-translation’s output.

Edits can be either very light (focused on stylistic changes and improving naturality), or heavy, when the MT has rendered some passages incomprehensible. The latter may be needed when:

  • The MT software is outdated or low-cost
  • The text is particularly complex due to its subject matter
  • The text is particularly complex due to its prose style
  • Not enough context has been provided to the MT

    Human Translation vs. Machine Translation

The Benefits of Human Translation

Human translation can be defined as the translation process, as handled by professional human translators. 

Google Translate is an absolutely invaluable tool for language learners. For instance, it allows them to find the word that was just at the tip of their tongue. It’s also helpful for international travelers who need to quickly understand their surroundings. In these contexts, human translation vs. machine translation is out of the question.  But machine translation tools like Google Translate are incredibly unreliable and shouldn’t be used in professional settings. Google Translate may seem like a biased example of machine translation tools. But, in fact, studies have shown that it’s pretty advanced. 

Regardless of how advanced machine translation solutions currently are, professional human translation is still unbeatable when it comes to nuance and precision. As we saw when covering machine translation post-editing, human translation makes it possible to compensate for errors that may arise with automatic translation. While machine translation tools can speed up the human translation process, they can’t completely replace it. 

The human factor remains indispensable for achieving a quality translation. But why? What makes human translation so important? For starters, human translators can fill the gap left by those words and expressions that do not exist in the target language. In these cases, human translators make the careful choice to replace that word with another that allows the text to be understood as accurately as possible. The ability to creatively solve problems and see the source material through the lens of culture is vital to attain an accurate, high-quality translation. Machines are yet to have the capacity to solve problems creatively or understand culture with the same intuition as a human. 

One of the most significant steps of the translation process is that in which the translator reads and analyzes the original text before beginning to translate. This understanding of the text from a reader’s perspective is invaluable to understand how the text should flow and how the text should work. This is a great advantage that human translation has over machine translation: The possibility of empathizing with both the author and reader, and accommodating their intentions and needs. 

Being able to experience a text as a reader, the translator can notice details that a machine is not prepared to address. When comparing human translation vs. machine translation, we need to notice that, unlike machines, humans can:

  • Recognize and maintain the style and tone of a translation
  • Ensure consistent terminology usage
  • Properly capture puns and metaphors

Eventually, scientists may be able to teach machines how to solve these challenges. But, for the time being, there’s a whole dimension of the translation process that only a human can handle. The experience of the translator and the ability to adapt to the client’s request make the process much more reliable and efficient than machine translation.  When used as an initial step, before human translation, machine translation greatly reduces the cost of the process and accelerates it. But machines are yet to have the skills that make quality human translation possible. 

It’s also worth noting that translation is a core process within localization. Localization is the cultural, regulatory, and technical adaptation of a piece of media or software, so it can be fully enjoyed by people from another culture. Localization has been vital to the success of numberless companies. From Starbucks to Netflix. We could say that localization is the process that allows companies to do business in the very competitive but very promising international landscape.

Empathy, understanding customer needs, and how mere translation may fail to meet them, is key to localization. While machines may be able to translate word-by-word, only a human can have a deep understanding of how language actually works.

Sign up to our newsletter to receive the latest blogs and news

Machine translation post editing MTPE

Human Translation vs. Machine Translation, Going Forward

In this post, we compared human translation vs. machine translation. In order to do so, we took a brief look at both methodologies. But, what are the latest high-profile developments in this area? 

Earlier this year, Elon Musk-backed OpenAI published a limited version of their latest artificial neural network model, GPT-3. Those who accessed OpenAI’s API and tested the model’s possibilities discovered an unprecedentedly powerful tool. For instance, some designers and developers taught the neural network to design and code.

But, as a Slator article notes, GPT-3 may not open up a world of endless possibilities for LSPs. For starters, the results displayed on social media may be exceptional. OpenAI’s CEO himself noted that the model still has some serious weaknesses and makes “silly mistakes”. Twitter users who had access to it also note that they found it a lot clunkier than some have reported.

On the other hand, OpenAI’s decision to make the model widely available prevents it from being a competitive advantage, reducing (if not nullifying) its usage’s commercial appeal. Perhaps the future of machine translation will rely on companies developing their own language processing models.

But, for the time being, the question of human translation vs. machine translation seems to have a single, simple answer: Machines for the heavy lifting, humans for the human touch.

Consequently, if you’re looking for high-quality translation solutions, there’s nothing like a translation company with an experienced team of professional human translators

 Create Translations that Maximize Your Global Impact

Click on a star to rate this post!

5 / 5. 1



Let's Stay In Touch

Sign up to our newsletter and receive the latest industry news, insights, and trends straight to your inbox.

You might also like