Deciding on machine vs. human translation has been a trending topic in the translation industry in the past few years.
Machine translation is a fast-moving technology that has become more accurate over the years, shaping a new way of approaching translation quality, and businesses are reaping their ever-growing benefits.
But, despite the great advances in artificial intelligence (AI), there is still a long way to go before a machine can mimic human creativity, soundness, style, and tone in translation.
This has led translation clients to decide on machine vs. human translation for some specific translation tasks.
But what about style, soundness, and creativity?
To a large extent, the aim of MT has been achieved with many systems of gisting quality enabling a conversation between speakers of different languages.
But before a machine can create amazing texts, we need to understand our brain processes so we can mimic this puzzling mechanism by using software code.
We cannot understand what art is and, with this in mind, we still have to deepen our research to know how our brain works, what processes we are using to come up with thoughts, ideas, and concepts.
It’s currently unknown how these processes interact and translate into emotions and feelings. That’s why art touches our hearts!
Understanding our emotions is a tricky task at the moment and chances are there still is a long way to go before we can achieve it, especially if we remember we don’t have to understand emotions but experience them instead.
But first, let’s get down to the basics. Machine translation is the automated and instant transformation of text from a language to another using artificial intelligence. But what is AI?
Well, according to technopedia AI is “an area of computer science that emphasizes the creation of intelligent machines that work and react like humans.”
In other words, AI is the simulation of human intelligence processes by machines, especially computer systems.
These processes may include speech recognition, learning, planning, problem-solving, expert systems, self-correction, and the list goes on.
Some of the intriguing types of AI include:
- Reactive machines: This is the most basic type of AI, as it can only react to current situations and is not able to learn from past experiences to influence future decisions. Examples are the IBM chess program known as Deep Blue, and Google´s AlphaGO, Chinese strategy board game, specialized to play against human competitors.
- Limited memory: Unlike the previous one, LM can use past experiences to inform future decisions, of which a well-known example is self-driven cars.
- Theory of Mind: This term derives from psychology and refers to a person’s belief, intentions and desires and the way it influences our decisions. However, this type of AI is only a theoretical construct and does not exist yet in the real world.
- Self-awareness: This is probably the most amazing and disturbing type of AI, as it comprises systems that have a sense of self and consciousness. These systems can infer what other´s feelings are based on the information they receive. This type of AI does not yet exist, neither.
However, according to some researchers and industry experts, the term “artificial intelligence” may not be the most convenient or appropriate, as many people see it as a threat to human beings, causing unrealistic fears and expectations on how it will impact our lives.
Most experts recommend using the term “augmented intelligence”, in the hope it will help people understand that AI will simply contribute to improving products and services and that it will not impact the role of humans in the marketplace and in society overall.
The question is how all these technologies influence machine translation? Well, MT may be included in the limited memory category and, specifically, in statistical machine translation (SMT), which is the dominant paradigm in this field.
This way, statistical models generate parameters learned from bilingual data, developing automated tools for translating from one human language to another.
So, we have automated translation tools that depend on the human input to generate large databases known as translation memories (TM), which are the core of translation technology that help translators and translation agencies to speed up the translation process.
More specifically, according to Wikipedia, TM is “a database that stores segments, which can be a sentence, paragraph or sentence-like units (headings, titles or elements in a list) that have previously been translated, to aid human translators.”
Using “translation units”, a type of translation pairs, translation memories can store a source text and its translation into the target language.
These translation units can then be used in new translations with the corresponding translation tool known as computed aided translation (CAT) tool.
But what are the pros and cons of machine translation?
- MT can be inexpensive.
- It can speed up the translation process.
- It´s simple to use.
- It´s unlikely for machine translation to encompass cultural traits. That´s true. We first would have to fully understand all the nuances of cultures in its social, economic, political, philosophical, moral and countless more aspects.
- Language is a human output, and it´s unfeasible, if not impossible, for a non-human entity to produce a human creation. Not to mention we don’t even realize what the human brain limits and processes are at the moment.
- Machines cannot replicate artistic skills, such as style and tone of poetic and persuasive texts and cannot put in the human finishing touches.
- The likelihood of producing a literal translation, often not the best choice, is higher with MT.
- MT fails in selecting the best word when several options are available for a term.
- Machines cannot solve specific linguistic problems such as ambiguity or polysemy (several meanings for a word or phrase).
When considering the machine vs. human translation dilemma, we find there are some advantages of MT, such as cost and speed, the disadvantages outweigh the advantages, meaning it would only be a good choice if the texts are relatively simple, for a non-specific domain and it’s always reviewed by a linguist with a solid background in the field.