Machine translation facts for kids

A phone app translating Spanish words into English.

Machine translation is when computers translate text or speech from one language to another. It uses special computer programs to understand and change words, sentences, and even the way people express ideas (like jokes or common sayings) from one language to another. Think of it like a super-smart dictionary that can also understand grammar and meaning!

History of Machine Translation

Early Ideas

The idea of machine translation is quite old! Back in the 800s, a smart Arabic cryptographer named Al-Kindi worked on ways to translate languages using math and patterns. Later, in the 1600s, René Descartes thought about a universal language where different languages would share the same symbols for ideas.

The idea of using computers for translation came up in 1947. People like Warren Weaver thought computers could help. In 1954, there was a first small test where a computer translated some English into French. This showed that it was possible!

Growing Interest

In the 1950s, more people started working on machine translation. The first researcher, Yehoshua Bar-Hillel, began his work in 1951. A team at Georgetown University even showed off their translation system in 1954. Soon, other countries like Japan and Russia also started their own research.

However, progress was slow. In 1966, a report said that after ten years, machine translation hadn't met expectations, so funding was cut. But some systems, like Logos MT, still helped translate military manuals during the Vietnam War.

Computers Get Smarter

By the late 1980s, computers became much more powerful and cheaper. This made people more interested in machine translation again. Companies like SYSTRAN, which had been working on this since the 1960s, started offering their services. In 1988, the French Postal Service used SYSTRAN for its online service.

In 1996, SYSTRAN offered free translation of small texts on the internet. Then, in 1997, AltaVista Babelfish used SYSTRAN to offer free translations, getting half a million requests a day! This made machine translation popular for everyone.

Later, in 2003, Franz Josef Och, who would later lead Google's translation team, won a competition for fast machine translation. More cool things happened, like open-source translation tools (MOSES in 2007) and mobile phone translation services. In 2012, Google Translate was translating so much text that it was like translating 1 million books every single day!

How Machine Translation Works

Before, machine translation used lots of rules. Now, it often uses smart computer learning methods.

Rule-Based Translation

This method uses dictionaries and grammar rules. Imagine teaching a computer every single grammar rule and every word meaning. The problem was that you had to tell the computer everything, even how to handle mistakes or words with many meanings.

Transfer-Based Translation

This method was a bit like creating a middle language that showed the meaning of the original sentence. Then, it would turn that middle language into the new language. It was partly based on the two languages being translated.

Interlingual Translation

This was another rule-based method. It turned the original text into a "language neutral" form, like a universal code that didn't belong to any specific language. Then, it would create the new language from this universal code. One system, KANT, was used to translate technical English for Caterpillar.

Dictionary-Based Translation

This is the simplest method. It translates words just like a dictionary would, one by one.

Statistical Translation

This method uses statistics. It learns by looking at huge amounts of text that have already been translated by humans. For example, it might look at records of Canadian parliament debates translated from English to French. By seeing how words and phrases are translated many times, it learns to guess the best translation.

IBM created the first statistical machine translation software called CANDIDE. In 2005, Google improved its translation by using billions of words from United Nations documents to train its system.

The main problem with this method is that it needs a lot of already translated texts. It also struggles with languages that have many word forms (like different endings for verbs).

Neural Machine Translation

This is the newest and most advanced method. It uses deep learning, which is a type of artificial intelligence that works a bit like the human brain. Neural machine translation has made huge progress recently.

Tools like DeepL Translator are very good, but even they usually need a human to check and fix the translation afterward. Sometimes, very large AI models like GPT can also translate, but they use a lot of computer power.

Challenges in Machine Translation

Even with all the progress, machine translation still faces some big challenges.

Machine translation can sometimes create funny or confusing phrases, like "wikipedia" for a type of mushroom.

A sign in Bali with a confusing Chinese translation.

Understanding Meaning

One big problem is when a word has more than one meaning. For example, the word "bank" can mean the side of a river or a place where you keep money. A human knows which meaning to use based on the rest of the sentence, but a machine can get confused.

Claude Piron, a translator for the United Nations, explained that the easy part of a translator's job is translating words. The hard part is figuring out what the author *really* meant, especially when things are unclear. For example, if a text mentions a "Japanese prisoners of war camp," does it mean a camp in Japan with prisoners from other countries, or a camp in another country holding Japanese prisoners? A human translator would research this, maybe even call someone, but a machine can't do that.

Slang and Informal Language

Machine translation struggles with slang, informal speech, or unusual ways of speaking. It's trained on standard language, so if you use common everyday phrases or text messages, the translation might not be as accurate. This is why translating casual conversations on mobile devices can be tricky.

Proper Nouns

Proper nouns are names of specific people, places, or organizations, like "George Washington" or "Chicago." Machines sometimes struggle to translate these correctly. They might try to translate a name like a normal word, or even leave it out.

For example, if you say "Southern California," the machine should translate "Southern" but just write "California" as it is. Sometimes, machines try to translate both parts, which can make the translation sound strange.

Using Multiple Sources

Some research looks at using texts that have been translated into three or more languages. By comparing these different translations, the machine can sometimes make a more accurate translation into a new language than if it only used one source.

How Machines Learn About the World

Computers can use something called an "ontology" to understand things better. An ontology is like a huge knowledge base that stores information about objects, ideas, and how they are connected.

For example, if you say "I saw a man with a telescope," a human knows the telescope is used for seeing, and it's with the man. A machine might not understand this easily. But with a big enough "ontology," the machine can learn that telescopes are tools for seeing and that they are often used by people. This helps the machine pick the right meaning for words.

Where Machine Translation is Used

While no machine can translate perfectly every time, many systems do a good job, especially if the text is about a specific topic. Machine translation can help speed up translations or provide quick, rough translations when you need them fast.

Travel

Many apps on phones and other devices offer machine translation. This is super helpful for travelers! For example, the Google Translate app can use your phone's camera to translate signs in real-time, showing the translated text right on your screen. It can also listen to someone speak and translate it for you.

Public Services

Governments and big organizations use machine translation too. The European Commission, for example, uses it to translate many documents between different European languages.

Wikipedia

Machine translation is also used to help translate Wikipedia articles. There's even a "content translation tool" that helps editors translate articles more easily. This helps make more information available in different languages, as the English Wikipedia often has more articles than other languages.

Security and Military

After events like 9/11, countries like the U.S. became very interested in translating languages like Arabic, Pashto, and Dari. They use machine translation to help military members communicate quickly with people and understand important information.

Medicine

Machine translation is being explored in medicine, especially when human translators aren't available. However, it's very important to be careful! A wrong translation in a medical diagnosis could be dangerous. Experts say that machine translations in medicine should always be checked by a human translator to make sure they are accurate.

Law

Legal language is very precise and can be tricky for machines. Special programs have been made for legal texts. But just like in medicine, there's a risk of mistakes. Lawyers are also warned that using free online translation tools might accidentally share private client information. Some courts even have rules about using machine translation in official legal meetings.

Ancient Languages

Thanks to new computer learning methods, machine translation can even help with ancient languages like Akkadian (an old language from Mesopotamia). This is amazing because there's very little information available for these languages.

Checking Translations

How do we know if a machine translation is good?

Human Judges: The oldest way is to have people read the translation and say how good it is. This takes time but is the most reliable.
Computer Scores: There are also computer programs that give a score to a translation, like BLEU or NIST.

Even with these tools, it's important to remember that human language is complex. A machine can't always understand the full meaning or context. So, for important translations, a human should always review and edit what the machine produces.

Sometimes, machine translation can even be funny! People have made videos showing how Google Translate can turn simple Japanese words into silly, nonsensical phrases like "DECEARING EGG."

Machine Translation and Sign Languages

In the early 2000s, it was hard to translate between spoken and sign languages. Sign languages, like American Sign Language (ASL), use hand movements, body language, and facial expressions in different ways than spoken languages use voice tone.

Researchers created a program called TEAM (translation from English to ASL by machine). This program would analyze English text, then use a "sign synthesizer" (like a dictionary for ASL signs) to create the signs. Finally, a computer-generated person would appear on screen and sign the English text in ASL for the user.

Copyright and Translations

Copyright protects original creative works. Some people wonder if machine translations should have copyright protection, because a machine isn't "creative." However, the person who wrote the original text still owns the copyright. If someone wants to publish a translation, they usually need permission from the original author.