SpeechCycle Awarded Patent for Spoken Language Understanding
SpeechCycle, a provider of customer interaction management solutions, has been awarded a U.S. patent for spoken language understanding (SLU).
The patent (U.S. Patent No. 7,962,339), covers a way to deal with different levels of specificity with a caller's input to a natural language understanding system. It is the latest patent in SpeechCycle's IP portfolio covering key areas of customer interaction management and speech technology solutions.
This advancement in spoken language understanding enables speech applications to discriminate among different levels of vague and specific caller requests and act on them, resulting in higher degrees of automation, reduced handling times, and a better overall experience. It deals with how calls to an automated system are mapped and routed to the correct path.
When caller utterances are vague, the technology covered in the patent allows the system to ask additional questions to determine his true intent, according to Roberto Pieraccini, chief technology officer at SpeechCycle.
By creating a hierarchy of specificity related to the call mapping, the technology allows a speech recognizer to understand a larger variety of meaning subtleties from natural language expressions. It outlines a method of resolving errors in an automated spoken language recognition system, identifying whether the error is a high-cost-error or low-cost-error.
A high-cost-error occurs between two categories that are not on the same path or between categories that are in the same path but the true category is at a higher level in the hierarchy. Most low-cost errors, on the other hand, can be corrected with a follow-up dialogue by asking the user to select a category from all the categories that have a higher level of specificity than the recognized one.
"This newest patent is further proof of our commitment to create innovative customer care solutions that provide superior automation and customer experience improvement," Pieraccini says. "Our customers are now able to use more accurate natural language understanding for all applications, especially for a call router, which is now able to discern the requests that can be routed immediately from those that require further disambiguation with the highest degree of precision.
"Rather than make a mistake, the system asks more questions," he explains.
SpeechCycle originally applied for the patent in March 2008.
Watson and Siri are only the beginning.