System and method for customized text macros
AssigneeNuance Communications, Inc.
InventorsPaul Joseph Vozila, Joel Praveen Pinto, Frank Diehl
Life of this patent
Estimating the term…
Abstract
A method, computer program product, and computer system for encoding, by a computing device, a transcript and text macros into vector representations. A word by word report may be predicted based upon, at least in part, the encoding. An attention mechanism may be queried based upon, at least in part, a decoder state. An attention distribution may be produced over an encoder output. An interpolation of the encoder output may be produced based upon, at least in part, the attention distribution. The interpolation of the encoder output may be input into a decoder for report modeling that includes text macro location and content.
Claims20
A computer-implemented method comprising:
- encoding, by a computing device, a transcript and text macros into vector representations;
- predicting a word by word report based upon, at least in part, the encoding;
- querying an attention mechanism based upon, at least in part, a decoder state;
- producing an attention distribution over an encoder output of the encoding based upon, at least in part, the querying of the attention mechanism;
- producing an interpolation of the encoder output as a context vector based upon, at least in part, the attention distribution;
- inputting both the context vector and a predicted word of the predicted word by word report as inputs into a decoder for report modeling, wherein an output of the decoder includes a location and content of text of a first text macro of the text macros associated with the context vector;
- and
- outputting the text of the first text macro associated with the context vector in the location of a report.
A computer program product residing on a non-transitory computer readable storage medium having a plurality of instructions stored thereon which, when executed across one or more processors, causes at least a portion of the one or more processors to perform operations comprising:
- encoding a transcript and text macros into vector representations;
- predicting a word by word report based upon, at least in part, the encoding;
- querying an attention mechanism based upon, at least in part, a decoder state;
- producing an attention distribution over an encoder output of the encoding based upon, at least in part, the querying of the attention mechanism;
- producing an interpolation of the encoder output as a context vector based upon, at least in part, the attention distribution;
- inputting both the context vector and a predicted word of the predicted word by word report as inputs into a decoder for report modeling, wherein an output of the decoder includes a location and content of text of a first text macro of the text macros associated with the context vector;
- and
- outputting the text of the first text macro associated with the context vector in the location of a report.
A computing system including one or more processors and one or more memories configured to perform operations comprising:
- encoding a transcript and text macros into vector representations;
- predicting a word by word report based upon, at least in part, the encoding;
- querying an attention mechanism based upon, at least in part, a decoder state;
- producing an attention distribution over an encoder output of the encoding based upon, at least in part, the querying of the attention mechanism;
- producing an interpolation of the encoder output as a context vector based upon, at least in part, the attention distribution;
- inputting both the context vector and a predicted word of the predicted word by word report as inputs into a decoder for report modeling, wherein an output of the decoder includes a location and content of text of a first text macro of the text macros associated with the context vector;
- and
- outputting the text of the first text macro associated with the context vector in the location of a report.
Dependencies are inferred from the claim text.