AT&T’s John Donovan provides more detail on the future offering here. In essence, when the APIs are released in June, developers can look forward to making it easier for the people who use their apps to use their voice (and probably their mobile phones) to carry out general web searches, find local businesses, engage in question and answer discussions and originate messages, including voice mail and text messages. Anticipating tighter coupling with Web-enabled TVs, there will also be a pre-built interface to AT&T’s U-verse® electronic programming guide.

Here’s AT&T’s YouTube video providing background on Watson and “how it will work” both for application developers and users:

In short, AT&T will be encouraging the developer community to take advantage of the market conditioning that Apple has undertaken by promoting the Siri Virtual Assistant through mass media advertising. But Apple, with Siri, is promoting the “closed garden” approach that has made the iPhone so successful. AT&T is making it clear that it’s engine, and the one million person hours of R&D will be there for the taking (though the mention of 600 patents) looks like a veiled warning that some metering of the use is inevitable.

Watson is already in use for voice search on iPhones and Android-based devices. An app called Speak4It is a mashup that includes AT&T Watson, and has been available since mid-2010. In addition, the AT&T Translator App, described here, has been a showcase for Watson technologies’ ability to support 7 different languages.

While many in the business and tech press think AT&T’s new offer portends a “smack-down with Siri,” that is not necessarily true. What we are destined to see is geometric growth in the energy put forth by developers to bring take advantage of voice as the “natural interface” for phones and other mobile/personal devices. Apple will continue to use its experience with Siri to refine the service offerings are Siri-enabled. Brisk sales of the iPhone 4S (which is almost identical to the iPhone 4, save for the inclusion of Siri), attest to its economic value. Apple’s implementation of Siri as a “beta” feature of the phone, give it sole access to the “Home” button as the mechanism to summon Siri (as well as the native accelerometer as a mechanism to support the “Raise to Speak” function).

Meanwhile, developers can look forward to more tools for using natural language in specific domains. AT&T has deemed general search, Q&A, business search and control of TVs to be worthy of early attention. Our belief is that positive experience in these domains will lead to the development of more “speechable moments,” as I started calling them back in mid-2010. Today there are more than a dozen firms – coming out of a diverse development community that, in addition to speech processing, spans customer care analytics, automotive electronics, telematics and pure academic research – that have formidable software platforms optimized for understanding the semantic structure, logical content and practical context of spoken phrases.

At this stage in the markets maturity, we’re happy to see AT&T entering the market with tools and resources to support the developer community.

The company, which has 1,500 customers, is now rolling out the service to organizations of all sizes and in all industry sectors. VPI’s value proposition is to “do all the work” of setting up an agent. The start-up process takes 4-6 weeks and accounts for about 3-4 “person days” all told. That includes feeding it the live recordings, suggested scripts and other tuning information. Unlike a typical speech IVR system, VirtualSource does not use or require companies to maintain logic trees. Instead, VPI uses about 10-20 recorded conversations (preferably with successful outcomes) and augments it with information, rules and logic from training manuals, call-flow diagrams or interviews with contact center managers or supervisors. Performance should improve with time, as the system carries out more and more conversations.

VPI customers can run the virtual agents side-by-side with live customer service reps. Comparisons of the two can be used to tune the virtual agent or to determine which tasks should are most eligible for automated handling. Visit here (scroll down) to listen to sample conversations. At this point, the company uses a slightly robotic aspect to the voice rendering, which allows the virtual agent to adapt more rapidly to dynamic conversations.

The “brain” or “intelligence engine” integrates important cognitive functions including “short- and long-term memory, natural language understanding, reasoning, learning mechanisms, goal-directed behavior and meta-cognition.” It’s service cloud is PCI-compliant and Botz believes that companies will see value in using virtual agents to receive payment instructions securely (less prone to human error or theft.)

VPI bills on a per use basis at $0.25 per minute (including tuning) with a monthly minimum and a one-year term agreement.

The program serves as an antidote to the reticence that IT and customer care professionals have shown (especially during the economic downturn) toward launching the lengthy design, deployment and debugging processes associated with natural language understanding (NLU) and enhanced call processing. The pitch is quite simple. Nuance first shows that callers benefit from shorter wait times and fewer call misdirections when a well-designed Call Steering app is deployed. Then they demonstrate how companies can benefit from shorter wait times and greater cost certainty when they adhere to a fixed scope deployment approach.

The term “fixed scope” is well chosen. Based on years of experience with Call Steering implementations among large global companies across multiple vertical industries, Nuance understands the application characteristics and communications architectures that will benefit from Fast Start. It has distilled those learnings into a “configuration file” which provides a foundation and design framework for getting started. Customers and clients are not being offered a “crippled app” or proof-of-concept, instead they are deploying the same code that has worked in full scale deployments.

The constraints have to do with the demographics and architecture of targeted companies. Based on experience in the field, Nuance recognizes that Fast Start will work most effectively for firms limited to 80-100 call destinations and 5-10 instances that require serious interpretation of “disambiguation” of a caller’s intent. In those cases, implementations can be completed in as few as 90 days and the total start up costs are a significant discount from the “full service” approach.

The offer seems very attractive to healthcare providers and utilities where companies have high standards for customer satisfaction but often lack the call volumes that justify investment in NLU. As Foley explained, Nuance had observed a threshold in the 10-12 million calls per year range (with some of the largest systems handling over 100 million calls each year). Fast Start can make Call Steering attractive when annual call volumes are less than 4 million. The clear message coming from Nuance is that “Natural Language Call Steering” is not too expensive for mid-sized businesses.

One of the early fruits of the partnership is an application for the Apple iPhone and RIM Blackberry that performs real time translation between spoken English and Arabic. This YouTube video provides a demonstration. Mobile translation of this sort is something akin to the Holy Grail of natural language voice processing. Offering the function as a Smartphone app is a real coup. It benefits from nearly twenty years invested by Sakhr in building a knowledge base of Arabic phrases to support optical character recognition (OCR), machine translation (MT), data mining and search.

Sakhr’s long-standing expertise in the field attracted the attention of BBN Technologies, which incorporated the technology into its work on the U.S. Defense Advanced Research Projects Agency’s (DARPA’s) GALE Project (referring to “Global Autonomous Language Exploitation” efforts to capture and recognize huge amounts of spoken words and text in a variety of languages. IBM and SRI have been vying with BBN over the past few years to make breakthrough advances in “natural language understanding” across a multitude of languages, dialects and modalities.

The mobile application is not related to GALE, but it is an impressive demonstration of Sakhr’s underlying technology. It is the product of a partnership between Dial Directions and Sakhr that dates back to 2008 when the two companies formed a developmental partnership to address “language application technology for mobile, cloud-computing environments.” The result was a “first of its kind open speech-to-speech mobile translation application for the U.S. government and business customers.”

The video demonstration is, indeed, impressive. The challenges surrounding transcription (even in a single language) are formidable. As the company brings the application to the marketplace it will need to manage user expectations surrounding “accuracy” – which is a very tough nut to crack in the transcription world where mis-recognizing a single word can distort the meaning of an entire utterance. In the commercial marketplace, that is the proverbial, technological elephant in the room.