Machine translation

Machine translation, or MT, is computer technology that takes text written in one language and converts it into another language without human intervention. The designers of the first computers in the 1950s hoped to accomplish machine translation. However, the task proved to be more complex than expected, and reliable and effective machine translation was not developed until the second decade of the twentieth century.

Four main formats of machine translation exist and efforts to improve the technology continue. These efforts include using artificial intelligence (AI) and other software that helps computers learn. However, computer translated language can lack the nuances that languages have and the linguistic quirks and idioms in each language can cause problems. As a result, machine translation is often subjected to post-editing by human translators, especially when mistranslation could result in medical or legal issues.

Background

The word “translation” comes from the Latin word transferre, which means “to bring or carry across.” The word was first used in France in the mid-fourteenth century to refer to transferring the bones or other relics of a saint from one place to another.

Translation refers to the process of taking something that is written in one type of symbols or language and converting it into another type of symbols or language while retaining the original meaning. Since the earliest days of human communication, being able to translate the spoken and written language of one group of people to that of another has been necessary for many reasons. These include basic interactions such as greetings and exchanges of information, as well as trade and negotiations.

The earliest recorded translations date back to the Mesopotamian era, from around 4000 BCE to around 2000 BCE. Archeological records contain the Sumerian poem “Gilgamesh,” translated from its original language into an ancient Asian language. Many other early translations from the first centuries CE were of the Hebrew Bible and the books that became the Christian Bible, as well as other religious texts. This work was generally unappreciated and sometimes even dangerous. The English priest William Tyndale (1494–1536), who was fluent in seven languages and understood the biblical languages of ancient Greek and Hebrew, was executed by strangulation and burning because he translated the Bible into English against the will of the British crown.

Despite challenges like those faced by Tyndale and others translating religious texts, the printing press—invented in 1436—increased the demand for texts translated from one language to another. However, translation in those times was an inexact profession and often a very difficult task. Translating the exact words was not enough; the translator also had to understand the culture and habits of those speaking each language to ensure an accurate translation.

A variety of factors can affect translation. Languages use different sentence structures. For example, common English sentences are constructed using the “subject, verb, object” pattern, as in “He pets the dog.” This is not the case in many languages, so translators cannot merely substitute words in the same order. Compound words—those created by combining two or more words—can be difficult to translate, especially when the words that comprise them are not used literally. For example, “deadline” refers to the time when something is due and has nothing to do with literal death. Idioms and figures of speech, such as “she’s under the weather,” and other situations where words are not used literally, are also difficult to translate. As a result, multiple human translators working on the same original text could produce translations with many variations, some of them significant.

For many years, translation was done by anyone who knew both languages and sometimes involved multiple translations. For instance, an eighteenth-century English-speaking American who needed to communicate with a Native American might use the services of a French trapper who understood the Native American language and another French-speaker, who also spoke English, translating each message twice as it went back and forth.

Early computer programmers thought computers could alleviate these difficulties. In 1956, a computer was created that could translate Russian into English. However, its translation was limited to about 250 words. Any additional words—and additional languages—required more labor-intensive programming done by hand by human programmers. Programmers also discovered that the translation process required far more storage and processing powers than early computers possessed. A reasonably proficient computerized system of machine translation was not developed until the early 2000s.

Overview

Machine translation is part of a broader field of technology known as computational linguistics. This field also includes related technology such as machine-aided human translation and computer-aided translation, where a human does most of the translating with support from databases or other programs on a computer. Another area of computational linguistics is automated translation. This uses triggers built into portions of a document that appear frequently—such as a legal disclaimer—to add content from a prepared database.

Truly effective machine translation devices came into use in the 2010s. They generally required the use of artificial intelligence and deep learning technology that enabled them to recognize and translate the intricacies of language that was not possible with earlier devices. Deep learning technology uses an artificial neural network modeled after those in the human brain to help computers “learn” as they are exposed to new material. Translation is generally considered the most complicated function for an artificial intelligence device to complete.

Four forms of machine translation exist:

• Rule-based machine translation (RBMT): This was the first form available. It uses the rules of grammar and spelling from online databases to perform translations. RBMT is generally the least accurate form and requires the most proofreading and post-editing from human translators.
• Statistical machine translation (SMT): Instead of relying on rules, this method analyzes the relationships between words, phrases, and sentences in a language and builds a statistical model. It then applies that model to the second language and uses this as a basis for translation. SMT is somewhat more accurate than RBMT but still requires a great deal of proofreading and correction.
• Hybrid machine translation (HMT): This form combines RBMT and SMT and adds additional memory to the process. This improves the quality of the translation but still requires significant post-editing by human translators.
• Neural machine translation (NMT): This form applies artificial intelligence and deep learning technology to the translation process. It has several benefits. In addition to producing more accurate translations, it is easier to add information for new languages and easier to use. This is the most effective form of machine translation that has been developed in the first quarter of the twenty-first century.

Uses

Machine translation is used for many reasons. The type of machine translation system and level of complexity needed depends on the purpose, required speed, and budget. NMT systems are the most expensive and the fastest; however, other systems may produce adequate results in some instances at a lower cost.

Governments use machine translation in several ways that generally require a high degree of accuracy. Machine translation may be needed when negotiating a treaty or trade agreement or for intelligence-gathering purposes. Law firms and other enterprises and companies often use machine translation as part of the information gathering and data mining processes when the required sources are in a variety of languages. Companies also use machine translation technology when developing package information and instruction manuals for customers in different countries. Language service providers, also known as professional translation industry services, use machine translation to increase the speed and efficiency of human translators. Machine translation can also be used for other functions, such as closed captioning and subtitles in movies and video productions, as well as speech translation. These uses all require sophisticated systems that can deliver fast, accurate translations such as NMT systems.

On the other hand, travelers who want to order dinner in a foreign country usually need to translate single sentences or phrases. Less sophisticated machine translation programming can easily handle this function. SMT is especially functional when the languages have similar roots, such as the Romance or Latin languages of French, Spanish, Italian, and Portuguese.

Future of machine translation

Some people believe the development of machine translation will put human translators out of work. While some sources estimate that 40 percent of the overall workforce worldwide will be replaced by AI technology, experts in the machine technology field expect that changeover to take considerably more time. This is due to several factors. One is that the existing technology is relatively new and far from perfect. It is still not as accurate as a human translator. While machine translation technology improves at a rate of 3 to 7 percent a year, this technology mostly addresses the linguistic aspects of translation. It is much more difficult to incorporate the cultural aspects that are essential to accurate and appropriate translation, such as not violating any language taboos with the translation. There is also a need to improve the technology used to translate spoken language, as opposed to written text.

Machine translating technology does provide an additional future benefit beyond speeding up the translation process, however. Experts say that coding languages into machine translation systems helps to preserve the language. This is important because some sources estimate that one language becomes extinct every fourteen days, and it is possible to lose as many as seven thousand languages over the next century. Many of these can be preserved through machine translation.

Bibliography

“Brief History of Translation: Everything You Need to Know.” Language Networks, 30 Sept. 2021, www.languagenetworkusa.com/blog/brief-history-of-translation-everything-you-need-to-know. Accessed 20 Aug. 2022.

“Common Challenges of Translation.” Smartling, www.smartling.com/resources/101/common-challenges-of-translation/. Accessed 20 Aug. 2022.

Kirchhoff, Petra. "Machine Translation in English Language Teaching." ELT Journal, vol. 78, no. 4, Oct. 2024, pp. 393-400, doi.org/10.1093/elt/ccae034. Accessed 15 Nov. 2024.

“Machine Translation.” American Translators Association, www.atanet.org/client-assistance/machine-translation/. Accessed 20 Aug. 2022.

Mamane, William. “The Future of Machine Translation and AI.” Marketing Institute, 28 Jan. 2022, www.marketingaiinstitute.com/blog/the-future-of-machine-translation-and-artificial-intelligence-in-the-evolution-of-languages. Accessed 20 Aug. 2022.

Quach, Ann. “What Types of Machine Translation Should You Use?” GTE Localize, 15 Dec. 2021, gtelocalize.com/what-types-of-machine-translation-should-you-use/. Accessed 20 Aug. 2022.

“What Is Machine Translation?” Memsource, www.memsource.com/machine-translation/. Accessed 20 Aug. 2022.

“What Is Machine Translation? Rule Based Machine Translation vs. Statistical Machine Translation.” Systran, www.systransoft.com/systran/translation-technology/what-is-machine-translation/. Accessed 20 Aug. 2022.

“William Tyndale.” Christianity Today, www.christianitytoday.com/history/people/scholarsandscientists/william-tyndale.html. Accessed 20 Aug. 2022.

“Who Uses Machine Translation?” Omnicien, omniscien.com/faq/who-uses-machine-translation/. Accessed 20 Aug. 2022.