The history of machine translation began in the middle of the last century. The quality of the translation has grown many times, various techniques and approaches have been developed. However, at the moment, the best technological minds of humankind have not been able to create a system that is capable of providing an ideal, or at least, comparable to human, result.
In this article, we decided to figure out what the near future of machine translation is, and do translators need to start looking for a new job?
One of the main achievements of recent years is that machine translation has become available to all Internet users. Google has long accustomed users to the fact that translating texts of varying complexity can be completely free. Ordinary users are less likely to acquire specialized translation applications and electronic dictionaries. And using the usual paper dictionary is already perceived as an anachronism, although, of course, this old school approach still makes sense in some cases.
However, until now there is no unambiguous criterion for assessing the quality of machine translation. The main thing that is required is that the user can eventually understand what is written in the source text. And even if the result contains a lot of syntax errors, the relative correctness of the result is more decisive.
This makes it possible to say that machine translation technologies are still under development and there are some problems that we will discuss in the following paragraph.
The current stage in the development of neural network technology for translation is characterized by three main difficulties that so far scientists have not been able to overcome. All these problems are directly interconnected, so there is a chance that if scientists manage to overcome at least one of the problems, the remaining two will also be easily resolved.
The program translates the text mechanically, completely without remembering the context. If in one sentence, following its data patterns, the machine was able to use the correct word or pronoun, this does not mean that in the next sentence it will do exactly the same. For example, a text may be written about a woman, but at some point, following its own data patterns, the program decides that in this particular sentence we are already talking about a man. As a result, the integrity of the text is violated.
When a translation is done by a specialist from The WordPoint translation service, for example, the translator may notice errors or inaccuracies in the source document. The program does not conduct such an analysis, but simply translates the words according to its algorithms, and arranges them into sentences, according to the rules that it knows. Therefore, if the original text contains an error, the same mistake will be made in the translation, and most likely, the correctness of the translation will also begin to limp, because the program will simply write something the most appropriate according to its views, and go on.
And of course, at the moment, AI-based translation programs cannot perceive and evaluate context. They know nothing about artistic symbols, humor, allegories, established expressions and situations in which they are used. Therefore, any translation at this stage will be performed simply according to the template, and the ability to analyze the content remains an advantage for the person.
Currently, it is known that many companies are working on programming in machine translation of language equivalents in different languages, which allows these programs to translate from one language to another. However, scientists still do not talk about programming background knowledge in machine translation, which will allow electronic translators to translate taking into account the peculiarities of style and situation. And at the moment this is task number one.
Therefore, now it is difficult to imagine that, by means of machine translation, it is possible to carry out, for example, translation in business or political negotiations, where it is very important to translate, smoothing sharp corners, because often the future of two or even more countries depends on such negotiations.
According to Microsoft, speaking about machine translation, mathematical technology has been created that can work on the principle of the human brain. More precisely, deep neural networks recognize senons (consecutive combinations of phonemes) and convert them into speech, which makes it possible to make three times fewer errors compared to traditional systems.
But speech recognition is only the first stage of translation. The main component here is the conversion of sentences in a foreign language, the application of which is often complicated by specific grammar rules. For this reason, in spite of the fact that all words in the sentence will be familiar, machine translation may produce a completely incorrect translation, which is excluded when translated by a person.
In addition, machine translation is not yet able to cope with literary texts and poems. The reason is that these texts are characterized by the individual style of the author, and the machine only follows the given technical algorithms, and nothing more.
Of course, as we said above, if you ask the program to translate any passage of a literary text, then most likely we can catch the minimum essence. But we will not get the effect of the work that we either read in the original or in high-quality human translation when the style of the author and the style of the translator merge into a single talent. AI doesn’t have talent yet, it just has skills. Therefore, by analogy, if you try to translate any poem, then the rhyme, symbolism, images, and hints will be lost. The program does not know how to think figuratively and artistically but is able to simply determine data patterns.
So, the question of teaching AI to translate correctly is a bit like teaching its ethics. In this context, some scientists simply offered to let the machine read all the information that a person knows about ethics. But in fact, the result will be as if we let the program read all the possible texts in all languages of the world. Yes, it read and remembered, but so far we still could not teach the machine the correct behavior in a given situation, nor would we teach the most accurate translation, taking into account the characteristics of each language, context, and artistic images.
Therefore, in the coming years, translators may not be afraid to be expelled from the labor market. Human translation still remains more qualitative, meaningful and accurate, although it takes more time and its cost is higher. Here we are faced with the main task on the path to the development of AI – we taught it how to collect and analyze information, but so far it can only think by patterns. Flexibility in decision-making is still a human prerogative.
Terrie E. Key is a responsible and prompt freelance writer and marketer, who is currently cooperating with the company The Word Point translation service. She has over 10 years of experience in marketing and 3 years in writing articles. She feels great delight in having turned her hobby into a new career. Her objective is to provide people with 100% original, interesting, easy-to-read content.