a {text-decoration: none; } En la punta de la lengua: Machines vs. Humans: Is Machine Translation ready to replace us?

7 sept. 2014

Machines vs. Humans: Is Machine Translation ready to replace us?




Life was given to us a billion years ago. During that time we have gradually developed a great set of abilities, knowledge and skills that allow us to identify ourselves as human beings. We can feel, think, create, but overall we can express what is in our minds thanks to language, which is without a doubt the watershed that placed humanity on a distinct level from the rest of the animal kingdom, since as far as we know, we are the only ones capable of sharing our life experience by means of graphic and phonetic linguistic codes. 

Sometimes I lose myself thinking on all the events and circumstances that took place in order for humans to develop a more or less fixed system that has been transmitted from one generation to another, and acquired since a very early age as if it did not represent any difficulty whatsoever. We arrive to this world and become speakers of a language only through biological predisposition, even if we have not had any kind of formal instruction. At least for me, that is the most evident proof of evolution.
However, that evolutionary process has not been fast one. Many years went by before we came up with the ability of producing sounds, then understanding each other through a specific language, then creating a system to represent it graphically, then coining a set of rules to govern it, and finally putting together a technological platform that allowed us to be connected with the rest of the world whenever and wherever we were using written and spoken language as our major tools.
More recently the focus has been even more ambitious. Since the mid-20th century, many linguists and computer scientist have worked together to create the major wrecking ball that would bring down all language barriers. The dream of a software capable of translating and even interpreting any form of human language instantaneously, is an unprecedented project that aims to take over one of the world’s most ancient professions, with the promise of creating an intelligible global society that might no longer need a lingua franca.
Nonetheless, the journey has just started. Despite the fact that we count with technological enhancements capable of literally moving mountains, it is still very soon to assure that a computer could be capable of processing language and information just like a bilingual person would. I have to say that I personally agree with the stated by Crystal (2003) when saying that machines are still far from reaching a human-like condition that could carry out the sometimes underestimated task of translation, as good as we do. From a simplistic perspective, machine translation only deals with converting a word in one language to another; thus, most people would probably say: what’s the big problem then?  You grab a bilingual dictionary and write all the words in it on a computer and voilà, there you have your machine translator. Wrong. Although during the last decades we have already overcame this obstacle and nowadays we count with several online dictionaries that are rapidly replacing paper dictionaries; the unachieved purpose of the still non-existent Fully Automatic High Quality Machine Translation (FAHQMT) is not only to translate from one language to another, but to translate complete context-dependent discourses.
            I have to admit that online dictionaries have been a great help for computer-aided translation, and for those who find themselves in the need of a fast and free translation, regardless of its lack of accuracy and feasibility.  However, the authentic machine translator capable of translating long texts without any human help, must be able to overcome the following obstacles before singing our praises.

Disambiguation and Lexicon
There are countless words in all languages that have more than one meaning (polysemy). The problem here is how to explain a computer that ball could be both a round object to play with and a huge dance. Relying only on statistical word-use without an Interlingua system or some kind of universal encyclopedia is not enough to process those differences (Bar-Hillel, 1960). Also, other issues regarding lexicon are the dialectal use and meaning of words, (e.g. how shitty = que cagado {Northern Mexican Spanish} that’s funny (inf.) = que cagado {Central Mexican Spanish} as well as cultural concepts in one language that need a longer description to be fully explained in a foreign language (e.g. Fremdscham {GER} = Embarrassment felt on behalf of someone else {ENG} = Pena Ajena {SPA}.
  Register
Translation software is normally not so good when it comes to translate informal or non-standard language uses. Therefore, all those texts that are written or said in slang or uncommon dialects, are almost impossible to be translated accurately.
  Ontology and Syntax
Ontologies are formal representation of knowledge which includes concepts in a domain and the relations between them (Vossen, 2003). Most of time we as humans are aware of this relations because we can rely on the context or situation in which the utterance occurs. For instance, I saw a friend with my glasses, could mean that you saw your friend wearing your sunglasses or that you saw him because you were wearing your glasses.
Morphology
Depending on the language you are translating into, it is very likely that there will be problems with inflection, derivation, number, compounds, gender, etc.

Based on these unsolved problems with machine translation, the notion of having a completely independent translation software capable of translating whatever input you give it, starts to seem a little distant. Therefore, it might be more feasible to continue spreading the English as language as an international vehicle for communication, instead of sitting around waiting for a still utopic software that would make our lives a lot easier.
Now, in a hypothetic future in which FAHQMT is available, many scenarios are possible. Honestly, I would like to live long enough to experience a software that could translate with the same accuracy that a bilingual/bicultural professional translator would. If this happens any time soon, there are many factors that will have to be addressed.

Cost
I have heard many people who think that translation is an expensive service. Most translators charge either by page or by word, which might seem expensive when we are talking about a book or a scientific paper. However, most clients do not see that each translation each different is complexity and therefore on the time the translator spends working on it. If we had FAHQMT would it be cheaper to buy the complete software to translate some texts sporadically, or would it still be better to pay the old-fashioned human translator to do the work?
Time
The proto-translators to which we have free access online do the work mostly wrong, but instantaneously. It is hard to hypothesize if the idyllic error-free Machine Translator, would be able to carry out the same work a human does in less time. Do you know how long does it take for a 3D Printer to create a small scale model building? Not less than 8 hours…     Quality
Here we go again. Before we can say that the quality of both human and machine translation is alike, many years and different kinds of evaluations will go by. Linguists overall are the ones who need to work the most in order to assure future clients that weather they choose a machine or human translator the results will be an accurate, feasible, grammatical, coherent, meaningful and so on.
Domains
There are some proto-machine translator software today that is mainly bought by big companies to translate from one language to another in only one specific domain. The same is true for human translators. There are many specialized translators who have a specific training in one fields such as law, medicine, literature, etc. The question here is if a machine translator could translate whatever you give it regardless of the domain.

   Working languages
There are at least 6,000 language in the world. However, only around 30 are the ones that are normally required for translation. How possible could it be that a single translation software could have all those codes in it?

     Need of a Bridge Language 
Crystal (2003) also suggests that the presence of a global language might eliminate the demand for world translation services, or that the economics of automatic translation will weaken the cost of global language learning that the latter will become useless. I personally think that if we had a function translation software, the need of a bridge language would not fade away. English is today’s lingua franca as French and Latin once were; however, due to the social inequality when it comes to education, it is very unlikely that English could be one day written, read, spoken and understood by the whole world’s population. The way I see it language learning will never be outdated. After the United States ceases to be most powerful nation, another country will take over along with its language. And even if we could go around the world using our iPhone as an interpreter to and from any language, it will just not be as practical, as actually speaking it.


References

Bar-Hillil, Y. (1960). The present status of automatic translation of languages. Advances in Computers, 1(1), 91-163. Retrieved form http://www.mt-archive.info/Bar-Hillel-1960.pdf

Vossen, Piek: Ontologies. (2003). In Mitkov, Ruslan (ed.) Handbook of Computational Linguistics. Oxford: Oxford University Press.

Crystal, D. (2003). English as a global language. Cambridge: Cambridge University Press

No hay comentarios :

Publicar un comentario