The Conversation Engine is a patented development environment that Voxify uses to develop speech applications. There is a Conversation Engine Runtime that is included in the Automated Agents enabling powerful conversational capabilities. The Conversation Engine takes a giant leap in speech technology moving beyond basic utterances and grammars to conversational intelligence that understands intent across multiple topics.
Speech Technology Evolution
Speech technology first started attempting to automate basic yes/no questions which we call "Basic I/O". We call the first generation "Exchange" where pre-built and flexible grammars were used to find the best match for a single utterance. The second generation we call "Interaction" where behaviors such as confirmation are used to understand the intent of a single topic. The Conversation Engine moves speech technology into the third generation which we call "Conversation". Conversation uses more sophisticated behaviors to understand intent across multiple topics. The diagram below shows an example of how a caller gives multiple pieces of information that was not requested.
What's your departure city?
- Austin
OK Boston. And when are you flying?
-No I meant Austin, and I'm flying on June 2nd
This appears to be a simple dialogue because humans know how to process this. But up until now, speech technology did not. Today airline fare shopping and reservations, entertainment ticketing reservations, and product up-sell calls are being handled successfully by Voxify Automated Agents. This unprecedented usability is made possible because of the Conversation Engine.
Conversational Behavior Models
The Conversation Engine is designed with Conversational Behavior Models which are typical human behaviors that happen between callers and Customer Service Representatives (CSRs). While these human behaviors come naturally when we speak to CSRs, they must be programmed into the speech application.
The Conversation Engine applies patented Conversational Behavior Models at specific points in the call flow, enabling a caller to speak in normal conversation and the Automated Agent to understand the caller's intent. A basic example is the behavior of affirmation. Callers express affirmation in different ways with different words and do this throughout normal conversation without being prompted. Ordinary speech applications would stumble when a caller gives an un-prompted "Uh-Huh" in the middle of the Automated Agent's sentence.
There are many Conversational Behavior Models such as affirmation, rejection, correction, conditional confirmation, disambiguation, and multi-slotting that are part of the Conversation Engine. These behaviors are pervasive throughout the Automated Agent and can be applied at any step or dialogue within the application. The dialogue below demonstrates how Conversational Behavior Models are applied throughout the conversation.
Dynamically Generates Call Flows
Unlike ordinary speech applications, Automated Agents do not have hard-coded call flows for all the potential routes that a customer service conversation might take. Instead, the Conversation Engine dynamically generates the call flows as needed, applying Conversational Behavior Models throughout the conversation. So rather than a caller being forced down an inflexible directed dialogue, the caller is gracefully guided through to collect needed information to complete the transaction and the call. The Conversation Engine is groundbreaking technology enabling a whole new world of speech application usability.
|
