How to Talk Like an Iraqi
Overcoming language barriers can be a matter of life or death in Iraq. Soldiers, medical personnel, and Iraqi citizens struggle to convey crucial information on a daily basis. While human translators are used in many situations, there simply aren’t enough who are willing to assist in every important conversation.
Last month, Menlo Park-based SRI International announced that it had deployed 32 Windows XP laptops loaded with advanced translation software for military evaluation in Iraq. The software, called IraqComm, facilitates an English-Arabic conversation by recording a person’s spoken words, translating them, and playing the translation in a matter of seconds.
IraqComm’s predecessor, Phraselator, is a handheld device loaded with phrases. The user selects an English phrase, and the device plays its translation in Iraqi Arabic, for instance. While useful, that system has drawbacks. The number of possible phrases is limited, and it can’t translate phrases into English, resulting in a conversation that relies heavily on gestures.
IraqComm, in contrast, can facilitate an impromptu, two-way conversation. A person speaks into a microphone and the words are collected and analyzed by speech-recognition software, called DynaSpeak, a system developed at SRI. The laptop screen then shows the phrase as the computer heard it. With a tap (of the “T” key), the phrase is spoken in the Iraqi Arabic. If the software has misheard some words, the speaker can choose from a list of other likely phrases, explains Kristin Precoda, director of the Speech Technology and Research (STAR) Laboratory at SRI and lead developer on the project.
Text can also be typed directly into the machine and translated; but the researchers believe the ability to translate speech will be particularly useful for soldiers, who need to keep their eyes on their surroundings. Translating spoken phrases also makes the system useful for telephone conversations.
After DynaSpeak converts the spoken words into text, software performs the translation. The software consists of two components, developed with the assistance of the Information Sciences Institute (ISI) at the University of Southern California (USC) in Los Angeles. The first module uses rule-based algorithms, explains Precoda, written to recognize specific rules of grammar and usage. They’re somewhat limited, however, because people use so many different words to convey ideas.
Thus, for more complicated sentences, the translation software turns to a type of algorithm that performs a kind of statistical analysis on the language. It works by assigning the likelihood that a word or phrase will follow another word or phrase, like a complicated version of predictive text messaging used on some mobile phones.
The algorithm works by chopping up input sentences into “phrasal units,” says Kevin Knight, senior research scientist at ISI. For instance, the phrase “Show me your permit to carry that weapon” could be broken into “show me,” “your permit,” “to carry,” and “that weapon.” Then those units get translated and re-ordered, explains Knight, according to their likelihood of being positioned next to each other in an Iraqi Arabic sentence. The statistical algorithm quickly goes through “millions of possible translations and scores each one of them,” he says.
In order to quickly access all possible sentence structures and words, the algorithm has numerous choices that it can assess at once, explains Knight. “It’s similar to a chess program,” he says, because it searches through many different paths to find the best one. The translation algorithm searches for the path that leads to the right sequence of words, and plays the most likely phrase on the laptop’s speakers within seconds. Currently, IraqComm can draw from a vocabulary of 40,000 words in English and 50,000 in Iraqi Arabic – not surprisingly, with a heavy emphasis on military and medical terminology.
IraqComm was developed under the Spoken Language Communication and Translation System for Tactical Use, a Defense Advanced Research Projects Agency (DARPA) program. SRI is one of six organizations developing English-Iraqi Arabic translation systems that are mobile and can translate military- and medical-oriented conversations in the presence of ambient noise. Similar research projects are going on at USC, Carnegie Mellon University, IBM, BBN Technologies, and Sehda, Inc.
Whether or not IraqComm will be the most accurate system to come out of the DARPA initiative remains to be seen, says Craig Schlenoff, an electronics engineer at the National Institute of Standards and Technology in Boulder. He and a team of researchers will begin testing all the translation systems within the next year, so they can be measured against each other. SRI’s system has an early start, though, and has garnered positive feedback from military officials who’ve already used it. Last week, according to SRI, the military ordered 21 more of its translation systems.
The goal of IraqComm is not to put human translators out of business, emphasizes SRI’s Precoda. Language experts will still be needed to judge body language and colloquial subtleties that could reveal information not obvious to a computer, which is “not as smart as a human translator,” she says.
Unlike human translators, however, IraqComm can be deployed anywhere and everywhere. Ultimately, then, says Precoda, it can give the military more translation options and help to mitigate the wartime hazard for Iraqi translators.
The inside story of how ChatGPT was built from the people who made it
Exclusive conversations that take us behind the scenes of a cultural phenomenon.
Sam Altman invested $180 million into a company trying to delay death
Can anti-aging breakthroughs add 10 healthy years to the human life span? The CEO of OpenAI is paying to find out.
ChatGPT is about to revolutionize the economy. We need to decide what that looks like.
New large language models will transform many jobs. Whether they will lead to widespread prosperity or not is up to us.
GPT-4 is bigger and better than ChatGPT—but OpenAI won’t say why
We got a first look at the much-anticipated big new language model from OpenAI. But this time how it works is even more deeply under wraps.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.