Machine Translation of Languages in Artificial Intelligence
The automatic translation of text from one natural language (the source) to another is known as machine translation (the target). It was one of the first applications for computers that were imagined (Weaver, 1949).
There have been three primary uses of machine translation in the past:
- Rough translation, such as that given by free internet services, conveys the “gist” of a foreign statement or document but is riddled with inaccuracies. Companies utilize pre-edited translation to publish documentation and sales materials in several languages.
- The original source content is written in a limited language that makes machine translation easier, and the outputs are often edited by a person to rectify any flaws.
- Restricted-source translation is totally automated, but only for highly stereotyped language like a weather report.
The translation is tough since it necessitates a thorough understanding of the text in the most general scenario. This is true even for very basic messages, such as one-word “texts.” Consider the word “Open” on a store’s front door. It conveys the sense that the shop is currently welcoming clients. Consider the same phrase “Open” emblazoned on a giant banner outside a recently built store. It indicates that the business is now open on a daily basis, yet viewers of this sign would not be misled if the store closed at night and the banner was not removed. The two signs have the same term but have distinct meanings. “Offen” would be the sign on the entrance in German, and “Neu Eroffnet” would be the flag.
The issue is that different languages classify the world in various ways. The French word “doux,” for example, has a wide range of connotations that are roughly equivalent to the English terms “soft,” “sweet,” and “gentle.” Similarly, the English term “hard” encompasses nearly all meanings of the German word “hart” (physically resistant, harsh) as well as certain meanings of the word “schwierig” (physically recalcitrant, cruel) (difficult). As a result, translating the meaning of a statement is more complex than understanding it in a single language. Predicates like Open(x) might be used in an English parsing system, but the representation language would need to make further distinctions for translation, possibly with Open1(x) expressing the “Offen” sense and Open2(x) reflecting the “Neu Ero ffnet” sense. An interlingua is a representation language that makes all of the differences required for a group of languages.
A translator, whether human or computer, must frequently comprehend the entire scenario described in the source, not simply the individual words. To translate the English word “he” into Korean, for example, one must choose between the humble and honorific forms, which is determined by the speaker’s social connection with the referent of “him.” Because honorifics in Japanese are relative, the decision is based on the speaker’s, referent’s, and listener’s social ties. It can be challenging for translators (both machine and human) to make this decision. “The baseball struck the window,” for example, maybe translated as “The baseball hit the window.” In order to translate “it broke” into French, we must pick between the feminine “elle” and the masculine “il,” indicating whether “it” refers to the baseball or the window. To get the translation properly, you’ll need to know both physics and language.
There are situations when there is no other option that will result in a truly satisfactory translation. An Italian love poetry, for example, that utilizes the masculine “il sole” (sun) and feminine “la luna” (moon) to signify two loves must be adjusted when translated into German, where the genders are reversed, and further changed when translated to a language having same genders.
Although all translation systems must model the source and destination languages, the models used by each system differs. Some systems try to convert the source language text into an interlingua knowledge representation, which is subsequently used to construct sentences in the target language. This is challenging because it entails three unresolved issues: building a comprehensive knowledge representation of everything, parsing into that representation, and producing words from that representation.
A transfer model is used in other systems. They store a database of translation rules (or instances), and they translate straight when the rule (or example) matches. At the lexical, syntactic, or semantic levels, the transfer can happen. A stringent syntactic rule, for example, transfers English [Adjective Noun] to French [Noun Adjective]. French [S1 “et puis” S2] is translated into English [S1 “and then” S2] using a hybrid syntactic and lexical rule.