Sakhr's Arabic Language Buddy
Automated machine translation (MT) has always fascinated me. I grew up with people who spoke languages other than English. I lived in France for two years, and I now correspond with a friend from that time who speaks very little English. As a consultant, I travel to countries whose languages I barely speak or read. So I’ve experienced firsthand the challenges of speaking to people in a language whose subtleties (or even nonsubtleties) I do not control.
So when Sakhr Software asked me to evaluate Arabic Language Buddy, an English-to-Arabic MT product, I readily agreed. As a cloud-based iPhone application, Arabic Language Buddy is a product of its time. It required sending my iPhone’s unique device identifier (UDID) to Sakhr. Fortunately, Apple’s Ad Hoc Helper application, which automatically sends the phone’s UDID to any application that needs it, made this process easy. Once that was done, I downloaded and installed Arabic Language Buddy without a hitch.
I did an informal assessment at two Chicago locations: a commercial district on the Northside (which is a hotbed of cell phone activity) and the Egyptian Consulate in downtown Chicago. My subjects were three native speakers of Arabic: a Tunisian, a Syrian, and an Egyptian. The utterances I used included greetings (What is your name?), directions (Where is the post office?), medical queries (Where does it hurt?), shopping questions (How much does that cost?), and dining requests (What are today’s specials?).
Arabic Buddy employs a four-step process:
Step 1: Say a free-form English utterance. Arabic Language Buddy exploits the iPhone’s touchscreen capabilities for push-to-talk input of English. The speaker presses a virtual button on the screen and holds it until the utterance is complete. The button is at the top of the screen, so your hand does not block the microphone.
Step 2: Convert the speech to text using speech recognition (SR). The SR was reasonably good, although it still made errors. For example, it insisted on transcribing “glasses” as “glances” in “There are two glasses on the table.” Also, despite the press-to-talk approach, background noise still affected SR accuracy.
As with any push-to-talk system, sometimes users start speaking before pressing the button or release the button before they stop speaking. This causes clipping.
The biggest challenge was Internet connectivity. A service like Arabic Language Buddy relies on the Internet, but that is a double-edged sword: It brings the power of the cloud, but users can be stranded when connectivity fails. At the Egyptian consulate, for example, the system complained about connectivity problems, and I was not able to run the program. At the other location, the speech-to-text conversion slowed to a crawl.
Step 3: Translation. Arabic Language Buddy displays the English text for the speaker, which ensures the correct utterance is translated. Translation is initiated when the speaker presses the “translate” button. When the translation is complete, Arabic Language Buddy shows the result in Arabic script.
Step 4: Text-to-speech output. Because the screen is small, my three subjects preferred to hear an audio translation. That is easily accomplished by pressing the “play audio” button.
For this product, Sakhr used the Modern Standard Arabic dialect rather than a multitude of local dialects. The rationale for using that dialect is it is widely understood, if not used.
Both my Syrian and Tunisian subjects reported the translations were accurate. The Tunisian objected to the use of Modern Standard Arabic, characterizing it as a dialect for religious or academic contexts rather than for everyday conversations among average speakers. He would have preferred to hear a dialect he could more easily understand. The Syrian also identified the dialect as a formal style of speech, but, unlike the Tunisian, he understood that the use of the standard dialect makes Arabic Language Buddy more universal. He assured me people from many countries and localities would understand what is being said, even if it were stilted.
Arabic Language Buddy is a nicely designed iPhone application that, according to my subjects, produces accurate translations. My biggest concern is its reliance on accessing the Internet. True, the connectivity challenge is not a function of Arabic Language Buddy (or any comparable application), but, as I saw in my evaluation, it can render it unusable.
Judith Markowitz, Ph.D., is president of J. Markowitz Consultants and a leading independent analyst in the speech and voice biometrics field. She can be reached at judith@jmarkowitz,com.