|
|
Effectiveness of Machine Translation
Kvapil, Lukáš ; Reich, Pavel (oponent) ; Kotásek, Miroslav (vedoucí práce)
The thesis considers machine translation(MT) in terms of difficulties it deals with, describes the most common methods and, with practical examples of MT, evaluates its quality and possible applications. In the first place, the MT has to deal with differences between languages, which can have different inflection, grammatical categories and syntax. Methods to deal with morphological, grammatical and syntactical differences are therefore required. Another problem is on the level of semantics; the MT systems must successfully identify meaning of words and choose appropriate translation. However, the computers have only limited capability in understanding of the meaning and considering context, as well as in making greater decisions about the whole text. To successfully deal with all problems of translation, a complete artificial inteligence would be required, which is not yet available. The most advanced in terms of AI seems to be the neural machine translation, which is the most modern method already used by online translators. The practical example of translation of several types of texts from English to Czech (and from CS to EN) with Google Translate shows that NMT can cope with many language differences and it can often successfully translate terminology and longer phrases, but it still produces a large number of mistakes, reason for which cannot be observed directly, and its behavior is inconsistent and sensitive to any change. To this day, there is still no universal system that would be able to produce Fully Automatic High-Quality Translation. MT application is restricted either by reduced quality of the output or by designing MT system only for specific field or purpose. MT can overall improve translation efficiency, while human involvement is required, but it will not replace human translators in the near future.
|
|
|
Effectiveness of Machine Translation
Kvapil, Lukáš ; Reich, Pavel (oponent) ; Kotásek, Miroslav (vedoucí práce)
The thesis considers machine translation(MT) in terms of difficulties it deals with, describes the most common methods and, with practical examples of MT, evaluates its quality and possible applications. In the first place, the MT has to deal with differences between languages, which can have different inflection, grammatical categories and syntax. Methods to deal with morphological, grammatical and syntactical differences are therefore required. Another problem is on the level of semantics; the MT systems must successfully identify meaning of words and choose appropriate translation. However, the computers have only limited capability in understanding of the meaning and considering context, as well as in making greater decisions about the whole text. To successfully deal with all problems of translation, a complete artificial inteligence would be required, which is not yet available. The most advanced in terms of AI seems to be the neural machine translation, which is the most modern method already used by online translators. The practical example of translation of several types of texts from English to Czech (and from CS to EN) with Google Translate shows that NMT can cope with many language differences and it can often successfully translate terminology and longer phrases, but it still produces a large number of mistakes, reason for which cannot be observed directly, and its behavior is inconsistent and sensitive to any change. To this day, there is still no universal system that would be able to produce Fully Automatic High-Quality Translation. MT application is restricted either by reduced quality of the output or by designing MT system only for specific field or purpose. MT can overall improve translation efficiency, while human involvement is required, but it will not replace human translators in the near future.
|
host ::
RSS.