Like HowStuffWorks on Facebook!

How Electronic Language Translators Work

What Can an Electronic Language Translator Do for You?
The Franklin Electronics 14 Language Speaking Global Translator, talks to you, too, giving you an idea of just how awful/awesome your French accent is.
The Franklin Electronics 14 Language Speaking Global Translator, talks to you, too, giving you an idea of just how awful/awesome your French accent is.
Image courtesy

Before we dive into specs and features, it's important to understand what you can and can't do with electronic translators. You can't pick up a translator, start reading "To Kill a Mockingbird" and have Harper Lee's prose come out the other side, perfectly translated in the target language. Such machines, known as universal translators, don't yet exist on the consumer electronics aisle in 2012.

Instead, think of electronic translators as digital phrase books. They store lots of ready-made phrases in the device's memory, enabling users to query the database and return results based on the search parameters. To streamline the process, manufacturers usually organize words and phrases into 10 to 15 categories, like basics (hello, goodbye, thank you) and local transport.

Once you select a phrase category, you can drill down into subcategories and then into a specific list of phrases. At that level, the machine's translation accuracy increases dramatically. When you say or type, "I need an ATM" while in the bank category, the machine searches a much smaller list of possible translations, increasing the odds it will find the best match. It may respond with, "Where is the nearest ATM?" which isn't a direct translation but is an accurate representation of what you meant.

Not impressed yet? Consider the advances in speech recognition that have made speech-to-speech translators possible. In these devices, a sound card converts analog sound waves into a digital format, breaks down words into phonemes, the smallest units of spoken language, and then compares the phonemes to a dictionary of stored sounds to find a best match. On the flip side, when a translator speaks a word or phrase appearing on the screen, it's employing text-to-speech, or TTS, technology, a two-step process. In the first step, the computer analyzes a word, breaks it into phonemes, then decides how long and at what pitch to say each phoneme. Then it pulls matching sounds from a database of prerecorded human sounds and assembles it into an audio file.

Algorithms, however, could be the most important technology in electronic translators. Everything begins with a database of parallel texts in two different languages. The texts could be translated works of literature, United Nations speeches or Web documents. Next, a complex set of operations identifies short matching phrases across sources and measures how often and where words occur in a given phrase in both languages. Finally, the software uses this information to build statistical models that link phrases in one language to phrases in the second. An electronic translator uses similar calculations when a user drills down into a phrase category and speaks or types a phrase. The computer analyzes the input, finds a high-probability match and returns the results. It may not allow an American to discuss the causes and effects of the French Revolution with a Parisian, but it can help her get directions to the monument marking the location of the Bastille.