Machine translation: a wonderful tool, but not for every occasion

Machine Translation (MT) is no doubt a useful tool in many environments, and represent a major step as a communication technology, as it fosters exchanges between people and companies that, until now, would have never dreamed of usefully communicating with counterparts all over the world.

MT bring something valuable to human relationships, and open enormous possibilities to commercial and personal exchanges all over the planet.

But MT cannot be properly considered per se a translation, as shown by the fact that research departments of prestigious companies and universities may work on its possible uses, but do not publish their findings using MT, and tribunals do not use MT for their elements of proof or the issuance of judgments.  

The obvious conclusion is that MT is very useful in many cases, but not in all cases.

In times of increased financial difficulties, many organizations consider cutting in what they consider “non-essential” activities (cleaning, postal expenses, redundant staff, auxiliary services and so on). But no potential investor would expect them to cut expenses in their core activities, and no one would think that Microsoft or Mercedes Benz should reduce their spending in their engineering labs or by lowering the level of qualification of candidates for the recruitment of developers. They will lower the level of spending, for instance by reducing the number of projects or delaying  some of them, but they will do everything possible to retain their top level staff and keep the high standards required for production. They have proven many times that they can survive major economic crises by keeping the focus on their added-value chain and adapting to the market.

Until the 1990’s, the IGOs used to spend a considerable portion of their budgets in translation services, and saw good reasons for it: equality of treatment of member states, by ensuring to all the possibility to participate with their best quality reflection (i.e. using their own mother tongue) in order to succesfully contribute in solving serious problems affecting great numbers of people; almost instant dissemination of the outcome of their reflections; and offering to Member States the means for a democratic dialogue among nations.

But in the 1990’s, technology came in and created great expectations in terms of the cost effectiveness of operations. Railways and airlines started dreaming of the almost unbelievable savings that would bring unmanned trains and planes. Armies immediately saw  the potential of drones and “surgical hits”, and so on. And they started diverting the resources provided in their budgets towards that objective. And then reality came, and driving a train or piloting a plane appeared to be much more resistant than originally expected to automation. Wars fought using state-of-the-art technology were either lost or, at least, not won. What these unexpected results had in common? That, when it comes to resolve complex issues requiring more than the mere technical training in repetitive actions and the need to adapt to unforeseen circumstances, the human brain is the one factor that makes the difference.

So, being the human factor the key for the activity itself, does that mean that nothing has changed, that translators have to do their work today as they used to do it 20 years ago? Progress in communication and information technologies has necessarily had an impact on the scenario, and advances in these two domains have benefited translators to a certain extent.

Translation is one of the most specialized activities of the human being, because translating is not only connected with linguistic competence, but also with intertextual, multicultural, psychological and narrative competence. As any good reader knows, translation can never be reduced to the rule-based or statistical substitution of one set of lexical and grammatical forms (or segment) by another set in the other language. Translation is not the mere similarity in meaning. While carrying out translation work, the translator has constantly to make decisions in order to reach what Mason calls a "functional equivalence", which means that the translation must generate in the target language the same effect aimed at by the original.

Is there any statistical or rule-based program able to provide the functional equivalence of any articulate work? The practical answer is somewhat easy: no publishing house, for instance, will publish a MT of the latest best-seller for which it bought the rights. It might underpay a translator to do the job, but it will use a translator. Would any of us submit to a tribunal in a foreign country a statement processed by MT?   Would any company sign a contract on the basis of a MT of it?

Any Statistical MT system provides an output that resembles more and more to human translation, but it must be clearly understood that no MT system would pass the translation tests to which candidates are subject in any competitive examination for entering a professional translation organization, be it public or private. 

So, the use of MT in such organizations appear clearly as a limited option (repetitive documents or, in cases where end-users would be ready to accept "gisted" texts, textual approximations offering, in the best of cases, “similarity of meaning” through post-editing) and as a political choice aimed at divert translation’s budgetary resources to other fields of activity.

No doubt that there is room for MT use in public and private organizations, either for gisting or, if resources so require, machine translation, provided that final users of such documents are made aware of the limitations of such methods of work;  but when it comes to documents of major importance (treaties, legal and financial content, technical processes, political statements, etc.), no matter what the level of available resources is, translation proper, i.e. finding functional equivalents from a source language into a target language, can only be done by human translators who, beside having an extended professional experience, must be able to quickly get familiar with the issues and domains relevant for the organization for which they work.