SAIC Introduces Spoken MT App for the iPhone

Article Details

Global Watchtower

Common Sense Advisory Blogs

Posted by Donald A. DePalma on May 22, 2012 in the following blogs: Technology, Translation and Localization
Pages \| \|

SAIC today made Arabic and Spanish versions of its Omnifluent Travel machine translation (MT) app available at the iTunes store for free, joining a growing category of handheld MT and mobile interpreting offerings. While online MT software has long offered computer users translation at their fingertips, this technology on mobile phones brings language technology into real-life situations, such as traveling to distant lands.

The Omnifluent Travel app accepts either keyboard input on the iPhone or spoken requests using SAIC’s automated speech recognition (ASR) technology. Both apps use the company’s MT engine in the cloud for bidirectional translation to or from English, and produce both a textual and spoken version of the translation. Translation requires a connection to the web.

When the phone doesn’t have web access, the app allows one-way communication of emergency, health, dining, and transportation issues through a set of more than 100 phrases that should prove useful for travelers. For example, the set expressions run the gamut from “call a doctor!” to “where is the nearest all-night drug store?” When the iPhone is connected to the cloud, users can record, save, and even post-edit translations of expected interactions, such as “please take me to the Jeddah Hilton” to play for the cab driver at the airport.

We spoke with SAIC Senior Vice President Jonathan Litchman about the product and his company’s plans for the technology. He told us that the major reason for making the app available for free was to showcase the company’s ASR and MT technology to a broader community than the usual professional and government workers who use SAIC’s products. Longer term, the Travel App is a test bed for future versions tailored to the needs of a specific vertical. SAIC has already produced a version for the New Jersey Association of County and City Health Officials to support the group’s influenza outreach program to Spanish-speaking residents of the state.

Litchman acknowledged that Omnifluent Travel is a first-version product in explaining an input-output issue that we noticed with the Arabic version of the application. Because it was trained with speakers in Egypt and the Gulf States, the speech recognition component stumbled with our Algerian-born test speaker. Also, the MT outputs classic Arabic, not the dialect used in North Africa. Litchman said that the broader community that will be reached through the iTunes store – and through future Android and Windows Mobile versions – should increase the app’s ability to deal with other dialects and regional accents. However, classical Arabic will be the preferred dialect until there’s demand for other variants.

Finally, we discussed why MT apps haven’t reached the all-time popularity of Angry Birds or Fruit Ninja, although one voice-driven app, iTranslate Voice, has broken in the top 10 paid apps on iTunes. A major problem with any travel-related app is the travel part – roaming fees for wireless data tend toward the extortionary end of the pricing spectrum, causing many international travelers to turn off data roaming while outside their home countries. Another issue is that the MT development community has oversold the capability of these tools, leading some users to expect these tools to allow them to engage in an active conversation with a native speaker. The technology is far from that goal.

Why do we find yet another smartphone MT app interesting? With Google’s grip on free web-based machine translation strengthening, other suppliers are looking for ways to get a hearing for their products in venues that Google doesn’t control. Replicating that web traffic is impossible for sites other than Facebook and Yahoo, so many MT developers are looking to handheld, omnipresent devices like mobile phones. Others, like Asia Online and SYSTRAN have chosen high-visibility projects such as translating Wikipedia into Thai or creating a corpus for all European Union legislation, respectively (SYSTRAN has hedged its bet with an iPhone app).

Developers offering handheld or these customized translation projects hope to draw the attention of prospective enterprise and information publishing buyers to the feature sets that distinguish them from Google – things such as technology customized to their domain or company for higher quality, data security through multi-tenant or on-premises deployment, and a number to call when there’s a problem. Given the enormous gap between translation that could happen and that which actually does happen, there’s room enough for them to make their case (see “Translation Future Shock,” Apr12).

Categories