Partially filling mixed-initiative forms from utterances having sub-threshold confidence scores based upon word-level confidence data

1. A speech processing method, implemented at least in part by at least one computer comprising at least one hardware processor, the method comprising:

prompting, via the at least one computer, for a spoken response that provides input for multiple elements;

receiving at the at least one computer, a single spoken utterance comprising content for multiple elements, each of which is mapped to a data field;

speech-to-text converting, using the at least one computer, the spoken utterance to derive values for each of the multiple elements;

determining, using the at least one computer, that an utterance-level confidence score for the spoken utterance falls below an associated certainty threshold;

ascertaining, using the at least one computer, element-level confidence scores for each of the derived elements;

determining, using the at least one computer, that a first set of the multiple elements each has an element-level confidence score above an associated certainty threshold and that a second set of the multiple elements each has an element-level confidence score below an associated certainty threshold;

storing, on the at least one computer, values for data fields mapped to elements in the first set; and

prompting, via the at least one computer, for a new spoken response that provides input for elements of the second set.

2. The method of claim 1 , wherein the ascertaining step is based on word-level confidence scores.

3. The method of claim 2 , further comprising:

establishing a configurable percolation algorithm, said percolation algorithm defining a manner in which confidence scores associated with child nodes of a parse-tree are applied to parent nodes of the parse-tree, wherein leaf nodes of the parse-tree are words, each associated with one of the word-level confidence scores.

4. The method of claim 2 , said method further comprising:

for an element node having multiple component nodes, determining component-level confidence scores for each of the component scores for each of the component nodes; and

setting the element-level confidence score to a lowest one of the determined component-level confidence scores.

5. The method of claim 1 , wherein the multiple elements are defined within a mixed-initiative form.

6. The method of claim 5 , wherein the mixed-initiative form is written in a standardized language that includes language constructs for specifically handling voice input.

7. The method of claim 5 , wherein the mixed-initiative form is written in VoiceXML.

8. The method of claim 5 , wherein the mixed-initiative form is associated with a grammar document which defines a recognition grammar used in the speech-to-text converting step.

9. The method of claim 8 , wherein the grammar document is written in an augmented Backus-Naur form (ABNF) based language.

10. The method of claim 8 , wherein the grammar document is written in an Extensible Markup Language (XML) Speech Recognition Grammar Specification (SRGS) based language.

11. The method of claim 1 , further comprising:

repeating the steps of claim 1 in a recursive fashion until an utterance-level confidence score for a received utterance falls above an associated certainty threshold at which point data fields mapped to the received utterance are all completed.

12. The method of claim 1 , wherein said steps of claim 1 are performed by at least one machine in accordance with at least one computer program stored in a computer readable media, said computer programming having a plurality of code sections that are executable by the at least one machine.