Artificial intelligence (AI) has become a hot topic in recent months. From Google’s driverless cars to more subtle technologies such as Facebook suggesting people you may know, AI is already very much part and parcel of our daily lives, and increasingly so. At the same time, prominent scientist Stephen Hawking has warned that AI could spell the end of the human race, and technology entrepreneur Elon Musk has warned that AI is our biggest existential threat. Generally, scientists differentiate between narrow AI (limited to a machine performing a very specific task really well) and general AI (human-level intelligence). Hawking’s and Musk’s statements were referring to the threats of very high-level AI.
There is a lot of debate about ethics in AI and how legal systems would cope with truly intelligent machines. And quite understandably, speculation is rife about which jobs run the most risk of being replaced by machines in the long run. Take Japan, for example, where carebots (AI robots) are already taking care of patients, replacing human nurses. In an age where everything that can be automated will be automated in the name of progress, is the job of the translator next on the list? And if we really are, as some scientists and futurologists suggest, on the verge of “singularity” (the moment in time where artificial intelligence will surpass human intelligence), would it be conceivable to create a machine that can translate creatively?
We all know and have probably used Google Translate or other machine translation tools at one time or another in our lives. They are undoubtedly useful tools for very particular scenarios, but it’s not the kind of AI I want to look at in this article. Google Translate is essentially statistical machine translation (SMT), referring to a huge corpus of existing bilingual texts and then trying to find the best fit based on statistical analysis. Sometimes it works just fine; other times the translations are gobbledygook. SMT cannot, by its very design, be “creative” — it can only really reproduce what has been translated before, and so this particular type of machine translation is unlikely to pose an existential threat to the creative translation profession any time soon. But what about other machines that are not based on statistics but are trained with the help of machine learning? Could they be creative? To answer this, let’s start by taking a look at some of the skills creative translators, whether human or machine, need to bring to the table.
What’s involved in creative translation?
Here is a non-exhaustive list of the skills creative translators need — I am focusing on marketing translation as just one example of creative translation:
They need to be able to understand the product or service the customer is selling.
They need to understand why this product or service is helpful for potential buyers and what problems it will help to solve.
They need to understand in what context the text will appear (website, packaging, magazine) and what impact that will have on the text to be written (length, tone of voice, target audience and so on).
They need to understand and be able to write in a particular tone of voice and register.
They need to be able to recognize stylistic elements such as metaphors and alliteration in the source text and be able to find similar appropriate style elements that are culturally and emotionally appropriate in the target language.
They need to know differences in culture between the countries in which the source and target languages are respectively spoken.
They need to understand the overall message of the source text and not translate word for word or even sentence for sentence, but rather take a holistic approach to providing the best possible translation.
While some of these tasks (such as understanding a product, or understanding text length restrictions) are skills that can potentially be taught or programmed, other tasks are difficult for machines to accomplish.
Why is creative translation so hard for machines?
At the risk of being overly simplistic, the translation process can be roughly divided into two parts.
The first part is about understanding the meaning of the source text. That’s the first (and possibly an insurmountable) hurdle: beyond word meaning, how do you teach a machine to recognize sarcasm, for example, or euphemisms, or irony? Or even a simple joke? Could you, and how would you, teach a machine about implied meaning, the unsaid words that can speak volumes? How do you teach a machine to infer the right meaning from a word that has more than one possible meaning?
Creative texts in particular are packed with figurative language, metaphors, allegories, alliterations and snappy slogans that would be very hard for machines to process. Of course, you can teach a machine a set amount of existing metaphors, but in the business of creative marketing, we like to come up with new and novel ways of writing. After all, one of the first things we learn in a writing course is to avoid using well-trodden metaphors, to be original, to be innovative.
But let us assume for a second that a machine was able to truly understand the source text. The second part is the creative translation of this meaning into the target language.
To answer the question of whether AI would be capable of creative translation, we first need to look at what creativity actually is. There is a plethora of definitions of creativity. According to the website creativityatwork.com, “Creativity is the act of turning new and imaginative ideas into reality,” and is characterized “by the ability to perceive the world in new ways, to find hidden patterns, to make connections between seemingly unrelated phenomena, and to generate solutions.”
AI could probably fulfill some of these criteria. Machine learning is a big part of AI, and machines are really good at sifting through large amounts of data and finding hidden patterns, so they may well be able to find connections between seemingly unrelated phenomena. Whether the outcome would be meaningful or valuable to humans, however, is another matter. One of the main issues is that currently, AI lacks emotional intelligence. It lacks intuition. And most crucially, it lacks imagination, the first step of the creativity process.
Does creativity require consciousness?
But there is perhaps an even more fundamental question we need to ask: would a machine need to have consciousness of itself and the world around it in order to be creative? Or in other words, can creativity exist without consciousness?
To quote David Gelernter, professor of computer science at Yale: “Suppose you could build a conscious software mind. Some cognitivists believe that such a mind, all by itself, is AI’s goal. Indeed, this is the message of the Turing test. […] But such a mind could communicate with human beings only in a drastically superficial way. It would be capable of feeling emotion in principle. But we feel emotions with our whole bodies, not just our minds; and it has no body. (Of course, we could say, then build it a humanlike body! But that is a large assignment and poses bioengineering problems far beyond and outside AI. Or we could build our new mind a body unlike a human one. But in that case we couldn’t expect its emotions to be like ours, or to establish a common ground for communication.) […] No such mind could even grasp the word ‘itch.’”
Even if a machine could have artificial consciousness, it’s safe to say that it would have a very different kind of consciousness from that of humans. The way we experience the world through our senses, the way we feel with our bodies, the mental and physiological mechanisms that make up our emotions — that is what makes us distinctly human. And our empirical view of the world is entrenched in the language we speak. From simple prepositions (being “in” the rain in English, but sous la pluie or “under” the rain in French) to metaphors such as “feeling like a million dollars” to euphemisms like “feeling under the weather” to similes such as “as strong as an ox.” All these examples are based on a shared view of the world and our experiences in it.
And when we write or translate creative texts, emotions, feelings, shared experiences and shared cultural values all play a huge role in creating engaging content that resonates with the reader in their particular culture.
So, what’s the verdict?
The jury is out at the moment as to whether AI can achieve consciousness and if so, if it could be truly creative in a way that’s meaningful for humans. We can only really speculate. I think it’s safe to assume that in comparison to the whole spectrum of translations (general, legal, medical, technical), creative translations (such as literary translation and marketing translation) will be the hardest nut to crack for AI and would most likely be the last to be automated, if they were at all. There seems to be a general consensus that jobs involving creativity and empathy — the very qualities that make us human — are the hardest to replace by machines. The example of the Japanese carebots would perhaps be a counterexample to this argument. Right now, at this moment in time, I would be inclined to say that it will never be possible to create a machine that can translate creatively in a meaningful way, given the many complexities of language — we haven’t even touched on language change or intercultural differences. But at the same time, I — we — cannot afford to be complacent. Translation and technology companies are heavily investing in machine learning. Not just statistical machine translation, but using machine learning with the help of real translators to feed data into machines that might well eventually replace them.
But even aside from current developments within the translation industry, AI in general is getting better all the time with the help of deep learning. AI is developing at an exponential rate, so there is a real chance that human-level AI could happen much sooner than we think. It will develop at such a pace that, right now, we can’t even think up all the things that AI could achieve. Hordes of scientists are using big data to help machines learn about emotions, for example. Machines are being taught to recognize facial expressions such as laughing and frowning, and they are being taught to recognize the sentiment of written text. This could essentially spell the beginning of emotional machine intelligence. And while these developments or new technologies may not seem directly connected to creative translation right now, in their entirety they will contribute to machines getting increasingly better at handling emotions, at language, and eventually at translation.