{"id":49987,"date":"2021-08-05T13:38:57","date_gmt":"2021-08-05T11:38:57","guid":{"rendered":"https:\/\/39896150.servicio-online.net\/types-of-machine-translation\/"},"modified":"2024-11-22T11:30:26","modified_gmt":"2024-11-22T10:30:26","slug":"types-of-machine-translation","status":"publish","type":"post","link":"https:\/\/montero-ls.com\/en\/types-of-machine-translation\/","title":{"rendered":"TYPES OF MACHINE TRANSLATION"},"content":{"rendered":"

The translation industry has experienced a genuine paradigm shift as the implementation of machine translation has become increasingly common among language service providers. Using computers and software for interlingual reproduction, however, has not always been the most trusted solution among translation agencies and sector professionals. In fact, since its beginnings in the mid-20th century thanks to an experiment by Georgetown University, its application has been strongly questioned in different fields and for various reasons. In general, it can be said that programming a system capable of establishing equivalences between a language pair that adapt to communication needs is no easy task. How could a software program successfully incorporate the syntax, pragmatics, semantics and linguistic variations (among many other parameters) of one or more languages, as well as offer dynamic translation options between them? Specialists finally realized that computers perform much better when searching for and finding elements that have already been translated, instead of acting as independent translators and making their \u201cown\u201d decisions. And so, this marks the beginning of the history of machine translation, a tool with great potential that was extensively researched and which has led to different approaches that will be discussed in this article.<\/p>\n

Rule-based machine translation<\/em> was the first tool to be developed. This approach teaches the computer program to emulate the internal workings of a language. It includes the grammar rules of the different languages and their lexicon, and the transfer rules are programmed so that exact equivalences between the syntax of one language and that of another can be established. This aspect makes the tool incredible useful when translating languages that are similar, such as Spanish and Catalan, Galician, French or Italian. Nevertheless, there are some drawbacks. Developing this approach essentially requires a major investment in time and resources, and moreover, it is important to note that this type of MT cannot provide a translation if the computer program has not learned the grammar\/syntax of the language. As a result, it implies the need for constant updates and the incorporation of lexicon on a regular basis. That is also why using this method to translate expressions and idioms is also impossible, since translating these elements departs from more literal translation procedures. Despite said limitations and the fact that this type of MT is rarely used today, it is still considered useful for feeding language combinations into machine translation engines, combinations which, due to their infrequency, cannot be provided with another more current type of MT. One rule-based machine translation resource is Apertium<\/a>. We encourage you to check out the translations that are generated when trying out grammar from different languages.<\/p>\n

After this approach, a new type of MT was developed which consisted of gathering a corpus of translations and offering translation options based on the previously recorded language matches.<\/p>\n

La que vino antes fue la denominada traducci\u00f3n autom\u00e1tica estad\u00edstica (in. statistical machine translation<\/em>). Esta necesita un conjunto de textos monoling\u00fce y su equivalente en una lengua de destino. It has three main components. First, the language model, which calculates the probability that a sentence in the target language is correct and that the translation is fluent. Next, a translation model, which establishes matches between the source and the target by using a corpus aligned sentence by sentence. And finally, a decoder, which performs a search among all the possible translations and selects the most likely one. In short, all the possible translations are created and the most likely one is proposed. Naturally, this variability often leads to negative results. In fact, a few years ago, Google translator<\/a>, which previously used this type of machine translation engine, became famous for delivering poor, literal results that were taken out of context. Luckily, this approach has become increasingly popular as more corpora between different language pairs have been fed into the translation engine and alignment and training processes have been refined.<\/p>\n

Later, a much more refined and cutting-edge version arrived, known as neural machine translation<\/em>. This is the most recent corpus-based MT engine approach, and it requires a much larger corpus volume. If statistical machine translation required millions of words, this model requires millions of sentences. Its aim is to imitate the human brain\u2019s neural network, such that the proposed translations are similar to the decisions that a human translator would have made. Although this approach is a very complex mechanism to explain, these engines essentially not only analyze how a word is translated, but also how this word relates to the other words around it (which ones it combines with, which ones it usually appears more often with, etc.). Google translate uses this approach for its current service. That being said, it may so happen that a word is not included in the corpus, and therefore these engines try to replicate it according to the rules and words that it has learned. This leads to the risk of creating words that do not exist, incorrectly conjugating verbs, using terms with different meanings, incorrectly ordering sentences, etc.<\/p>\n

Todo ello apunta a que las herramientas de traducci\u00f3n autom\u00e1tica presentan grandes ventajas, pero que no pueden encargarse por s\u00ed solas de hacer una traducci\u00f3n. Ya que van asociadas a una serie de errores t\u00edpicos que deben ser subsanados por un profesional de la traducci\u00f3n y con arreglo a unas normas determinadas orientadas a garantizar la m\u00e1s alta calidad. Given this need, machine translation post-editing arose, a language service that offers many benefits and one that we will discuss in future articles. MT is constantly evolving and it opens up a world of possibilities for the translation industry, all with the goal of improving its work.<\/p>\n

This article is based on the information found at the following links:<\/p>\n