Machine Translation (30%) (10%) In a couple of paragraphs, Explain the ideas, is

Question

Machine Translation (30%)

(10%) In a couple of paragraphs, Explain the ideas, issues, problems, (or maybe the fun) of machine translation between two closely related languages such as English and German (or maybe German and Dutch, or maybe American English and British etc.), or may be between two Chinese dialects, two Asian Indian languages. Pick your own pair of two close but different languages (or two close but different dialects).

(10%) Repeat the previous question here but pick a pair of two very unrelated languages such as English and Chinese (you pick your own pair; these two languages have to be very unrelated).

(10%) In a couple of paragraphs, Discuss or explain how machine translation is done programmatically. Pick your pair of natural languages and also your computer language (and example of machine translation using computer programming).

Explanation / Answer

(1)

Machine translation translated text from one language to another using simple word substitution of the given language and converting them to another language. The main problem with machine translation that the grammar of most languages is ambiguous meaning most of the time there is no one correct translation for a given text. Hence, each time the translator adds some noise to the process and after a few translations, the original meaning of the text is completely lost.

For example,  

Translating "I love her" from English to French on Google Translate gives "je l AIME". However, translating it back to English gives "I love it". The grammatical mistakes in similar languages are subtle and a human can easily understand the meaning of the context and also rectify the mistake.

(2)

In non-similar languages, the errors go a long way. In some instances of using machine translation the meaning of the text is entirely lost and there is no way to possibly know what the original text meant.

For example,

When "Incredible India!" is translated from English to Tamil it gives "Nambamudyatha India". However, the actual meaning of "Nambamudyatha India" is "India cannot be trusted". In these cases, even one-way translation loses its meaning.

(3)

There are two modern approaches to perform machine translation

(a) Rule-based MT

Rule-based MT relies on linguistic and grammatical rules and words in dictionaries to create a translational representation from which the text is generated. It requires extensive lexicons with syntactic and semantic information and complex grammatical structure. Translations are built on synonyms from data processed from gigantic dictionaries.

(b) Statistical MT

Statistical MT uses statistical techniques based on complex machine learning algorithms to create text translation. The models are trained on data encompassing billions of words and then the models learn the expected outputs which are then provided to the user. On advantage of this technique, is that models keep on taking user feedback and are constantly improving.

Related Questions

Navigate

• Browse (All)
• Browse M
• Subjects

Integrity-first tutoring: explanations and feedback only — we do not complete graded work. Learn more.