US 12,488,184 B1
Alternative input representations
Zhongkai Sun, Bellevue, WA (US); Sixing Lu, Bellevue, WA (US); Chengyuan Ma, Bellevue, WA (US); Xiaohu Liu, Bellevue, WA (US); and Chenlei Guo, Redmond, WA (US)
Assigned to Amazon Technologies, Inc., Seattle, WA (US)
Filed by Amazon Technologies, Inc., Seattle, WA (US)
Filed on Mar. 30, 2022, as Appl. No. 17/708,639.
Claims priority of provisional application 63/309,738, filed on Feb. 14, 2022.
Int. Cl. G06F 40/295 (2020.01); G06F 40/284 (2020.01); G06N 5/022 (2023.01); G10L 15/26 (2006.01)
CPC G06F 40/295 (2020.01) [G06F 40/284 (2020.01); G06N 5/022 (2013.01); G10L 15/26 (2013.01)] 18 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
receiving audio data corresponding to a spoken input;
determining automatic speech recognition (ASR) data using the audio data;
determining natural language understanding (NLU) data using the ASR data, the NLU data including first entity data representing a first entity;
determining, using a knowledge base, a plurality of entities using the first entity data, wherein the knowledge base includes connections between at least the first entity and a second entity based on inclusion in at least a first pair including a first past user input and a first past system response associated with the first past user input;
determining, using a classifier, that the first entity corresponds to a first level, the classifier being configured using at least a second pair including a first user input and a first alternative input representation, wherein the classifier is configured to categorize the first entity to one of:
the first level representing that the first entity is included in the spoken input,
a second level representing that the first entity is included in a second alternative input representation corresponding to the spoken input, and
a third level representing that the first entity is included in both the spoken input and the second alternative input representation;
determining, using the classifier, that a second entity of the plurality of entities corresponds to the second level;
determining, using the classifier, that a third entity of the plurality of entities corresponds to the third level;
determining a ranked list of entities based on classification of entities by the classifier, wherein the ranked list of entities includes the third entity, corresponding to the third level, ranked higher than the second entity based on the second entity corresponding to the second level, and wherein the ranked list of entities excludes the first entity based on the first entity corresponding to the first level;
determining, using the third entity, a third alternative input representation corresponding to the spoken input; and
determining, using the third alternative input representation, output data responsive to the spoken input.