US 12,266,355 B2
Shared encoder for natural language understanding processing
Jonathan Jakob Hueser, Aachen (DE); Fabian Triefenbach, Meschede (DE); Chandana Satya Prakash, Revere, MA (US); Jin Cao, Forest Hills, NY (US); Wael Hamza, Yorktown Heights, NY (US); and Mariusz Momotko, Tczew (PL)
Assigned to Amazon Technologies, Inc., Seattle, WA (US)
Filed by Amazon Technologies, Inc., Seattle, WA (US)
Filed on Mar. 9, 2022, as Appl. No. 17/690,609.
Prior Publication US 2023/0317066 A1, Oct. 5, 2023
Int. Cl. G06F 40/30 (2020.01); G06F 40/295 (2020.01); G10L 15/06 (2013.01); G10L 15/18 (2013.01); G10L 15/22 (2006.01); G06F 40/279 (2020.01); G10L 15/08 (2006.01); G10L 15/30 (2013.01)
CPC G10L 15/1815 (2013.01) [G06F 40/295 (2020.01); G06F 40/30 (2020.01); G10L 15/063 (2013.01); G10L 15/22 (2013.01); G06F 40/279 (2020.01); G10L 2015/088 (2013.01); G10L 15/1822 (2013.01); G10L 2015/223 (2013.01); G10L 15/30 (2013.01)] 20 Claims
OG exemplary drawing
 
5. A computer-implemented method comprising:
receiving first input data representing a first natural language input;
determining, using an encoder and the first input data, first encoded representation data, the encoder comprising a plurality of sequential layers connected in series, the plurality of sequential layers including at least a first layer and a second layer different from the first layer, the first encoded representation data including a first value output by the first layer and a second value output by the second layer;
processing, using a first attention component associated with a first decoder, the first encoded representation data to determine the first value;
processing, using the first decoder, the first value to determine first natural language understanding (NLU) data corresponding to the first natural language input;
processing, using a second attention component associated with a second decoder, the first encoded representation data to determine the second value;
processing, using the second decoder, the second value to determine second NLU data corresponding to the first natural language input; and
determining, based at least in part on the first NLU data and the second NLU data, first output data responsive to the first natural language input.