Current machine translation software often allows for customisation by domain or profession (such as weather reports) — improving output by limiting the scope of allowable substitutions. This technique is particularly effective in domains where formal or formulaic language is used. It follows then that machine translation of government and legal documents more readily produces usable output than conversation or less standardised text.
Improved output quality can also be achieved by human intervention: for example, some systems are able to translate more accurately if the user has unambiguously identified which words in the text are names. With the assistance of these techniques, MT has proven useful as a tool to assist human translators, and in some cases can even produce output that can be used "as is". However, current systems are unable to produce output of the same quality as a human translator, particularly where the text to be translated uses casual language.
Real progress was much slower, however, and after the ALPAC report (1966), which found that the ten-year-long research had failed to fulfill expectations, funding was greatly reduced. Beginning in the late 1980s, as computational power increased and became less expensive, more interest was shown in statistical models for machine translation.
The idea of using digital computers for translation of natural languages was proposed as early as 1946 by A.D.Booth and possibly others. The Georgetown experiment was by no means the first such application, and a demonstration was made in 1954 on the APEXC machine at Birkbeck College (London Univ.) of a rudimentary translation of English into French. Several papers on the topic were published at the time, and even articles in popular journals (see for example Wireless World, Sept. 1955, Cleave and Zacharov). A similar application, also pioneered at Birkbeck College at the time, was reading and composing Braille texts by computer.
Recently, the Internet has emerged as a global information infrastructure, revolutionizing access to any information, as well as fast information transfer and exchange. Using Internet and e-mail technology, people need to communicate rapidly over long distances across continent boundaries. Not all of these Internet users, however, can use their own language for global communication to different people with different languages. Therefore, using machine translation software, people can possibly communicate and contact one another around the world in their own mother tongue, in the near future.
Behind this ostensibly simple procedure lies a complex cognitive operation. To decode the meaning of the source text in its entirety, the translator must interpret and analyse all the features of the text, a process that requires in-depth knowledge of the grammar, semantics, syntax, idioms, etc., of the source language, as well as the culture of its speakers. The translator needs the same in-depth knowledge to re-encode the meaning in the target language.
Therein lies the challenge in machine translation: how to program a computer that will "understand" a text as a person does, and that will "create" a new text in the target language that "sounds" as if it has been written by a person.
This problem may be approached in a number of ways.
Machine translation can use a method based on linguistic rules, which means that words will be translated in a linguistic way — the most suitable (orally speaking) words of the target language will replace the ones in the source language.
It is often argued that the success of machine translation requires the problem of natural language understanding to be solved first.
Generally, rule-based methods parse a text, usually creating an intermediary, symbolic representation, from which the text in the target language is generated. According to the nature of the intermediary representation, an approach is described as interlingual machine translation or transfer-based machine translation. These methods require extensive lexicons with morphological, syntactic, and semantic information, and large sets of rules.
Given enough data, machine translation programs often work well enough for a native speaker of one language to get the approximate meaning of what is written by the other native speaker. The difficulty is getting enough data of the right kind to support the particular method. For example, the large multilingual corpus of data needed for statistical methods to work is not necessary for the grammar-based methods. But then, the grammar methods need a skilled linguist to carefully design the grammar that they use.
To translate between closely related languages, a technique referred to as shallow-transfer machine translation may be used.
Transfer-based machine translation
Interlingual machine translation is one instance of rule-based machine-translation approaches. In this approach, the source language, i.e. the text to be translated, is transformed into an interlingual, i.e. source-/target-language-independent representation. The target language is then generated out of the interlingua.
Machine translation can use a method based on dictionary entries, which means that the words will be translated as they are by a dictionary.
Statistical machine translation tries to generate translations using statistical methods based on bilingual text corpora, such as the Canadian Hansard corpus, the English-French record of the Canadian parliament and EUROPARL, the record of the European Parliament. Where such corpora are available, impressive results can be achieved translating texts of a similar kind, but such corpora are still very rare. The first statistical machine translation software was CANDIDE from IBM. Google used SYSTRAN for several years, but has switched to a statistical translation method in October 2007. Recently, they improved their translation capabilities by inputting approximately 200 billion words from United Nations materials to train their system. Accuracy of the translation has improved.
Example-based machine translation (EBMT) approach is often characterised by its use of a bilingual corpus as its main knowledge base, at run-time. It is essentially a translation by analogy and can be viewed as an implementation of case-based reasoning approach of machine learning.
Shallow approaches assume no knowledge of the text. They simply apply statistical methods to the words surrounding the ambiguous word. Deep approaches presume a comprehensive knowledge of the word. So far, shallow approaches have been more successful.
The late Claude Piron, a long-time translator for the United Nations and the World Health Organization, wrote that machine translation, at its best, automates the easier part of a translator's job; the harder and more time-consuming part usually involves doing extensive research to resolve ambiguities in the source text, which the grammatical and lexical exigencies of the target language require to be resolved:
The ideal deep approach would require the translation software to do all the research necessary for this kind of disambiguation on its own; but this would require a higher degree of AI than has yet been attained. A shallow approach which simply guessed at the sense of the ambiguous English phrase that Piron mentions (based, perhaps, on which kind of prisoner-of-war camp is more often mentioned in a given corpus) would have a reasonable chance of guessing wrong fairly often. A shallow approach that involves "ask the user about each ambiguity" would, by Piron's estimate, only automate about 25% of a professional translator's job, leaving the harder 75% still to be done by a human.
Although no system provides the holy grail of "fully automatic high quality machine translation" (FAHQMT), many systems produce reasonable output.
Despite their inherent limitations, MT programs are used around the world. Probably the largest institutional user is the European Commission.
Google has claimed that promising results were obtained using a proprietary statistical machine translation engine. The statistical translation engine used in the Google language tools for Arabic <-> English and Chinese <-> English has an overall score of 0.4281 over the runner-up IBM's BLEU-4 score of 0.3954 (Summer 2006) in tests conducted by the National Institute for Standards and Technology. Uwe Muegge has implemented a demo website that uses a controlled language in combination with the Google tool to produce fully automatic, high-quality machine translations of his English, German, and French web sites.
With the recent focus on terrorism, the military sources in the United States have been investing significant amounts of money in natural language engineering. In-Q-Tel (a venture capital fund, largely funded by the US Intelligence Community, to stimulate new technologies through private sector entrepreneurs) brought up companies like Language Weaver. Currently the military community is interested in translation and processing of languages like Arabic, Pashto, and Dari. Information Processing Technology Office in DARPA hosts programs like TIDES and Babylon Translator. US Air Force has awarded a $1 million contract to develop a language translation technology.
Relying exclusively on unedited machine translation ignores the fact that communication in human language is context-embedded, and that it takes a human to adequately comprehend the context of the original text. Even purely human-generated translations are prone to error. Therefore, to ensure that a machine-generated translation will be of publishable quality and useful to a human, it must be reviewed and edited by a human.
It has, however, been asserted that in certain applications, e.g. product descriptions written in a controlled language, a dictionary-based machine-translation system has produced satisfactory translations that require no human intervention.