Types Of Machine Translation
With the world filled with people of different languages, there is a great need for people to overcome language barriers. In order to make each and everyone understand what the other person is saying, quick translations are required. Translation machines have been developed specifically to translate those different languages for easy communication.
Machine translation is achieved by computer software changing text or speech from one language to another. There are three different types of translation machines within the rule-based machine translation pattern: the direct, transfer and interlingual approach.
The most basic versions of these machines simply change single words from one language to another (red arrow). In this direct approach words are translated directly without passing through an intermediate language construct. The direct translators don’t take context, meaning and domain into account.
In transfer based machine translation (green arrow) the source language is changed into a summary, a less language specific version. This summary is based on the linguistic rules of both the source and target language and is specific for each language pair. The source language is first translated into an intermediate summary language version. And from there the sentences in the target language are generated.
The advanced machines are able to translate more difficult translations. They support the use of idioms, and they can easily recognize phrases as well. With the use of analytic software the machine interprets the meaning of the source text, constructs an intermediate summary language and then re-interprets the meaning in the required language. It will not only process words from the language of the speaker and convert those to the language of the listener. Therefore the translator must have enough knowledge of the source language to understand and examine all the features of the text. And the translator must be fluent in the target language to convey the proper message. On top of that, the translator must also understand the culture and domain of the people speaking the language.
The third approach is called Interlingual Machine Translation (blue arrow). In this method the text in the source language is changed into an Interlingua. This Interlingua is a summarized, abstract and mathemetical language edition. It is able to describe all of the characteristics of all supported languages, instead of simply translating between one source language and one target language.
There are also different approaches to the analytics used to change the source language into the intermediate language construct and consequently to the target language. There are grammatical methods and statistical methods.
Getting enough data of the right kind to support the particular method is sometimes difficult to find. The grammar method requires a person who is qualified in the required languages to set up the grammar for use by the machine. For the statistical methods an ample supply of reference documents in all languages is required to ‘learn’ the translation statistics.
Statistics are extremely important when words are used that have more than one meaning. When dealing with word-sense disambiguation the meaning of the word that is best suitable in that particular phrase should be used. In the past the machine had no capability of differentiating between the two meanings of a word, but this has greatly been overcome by the modern translation machines. There has been much progress in letting machines guess the best meaning based on statistics. This guesswork is called the shallow approach to translation.
In a deep approach, a human translator would for example place a couple of phonecalls to the writers of the text to make sure he understood exactly what was meant. In deep transfer machine translation, Artificial Intelligence is used to do this kind of research. Deep approaches are currently being developed as much is being done in the field of Artificial Intelligence.