| CPC G06F 16/90332 (2019.01) [G06N 5/022 (2013.01); G06N 7/01 (2023.01)] | 16 Claims |
|
16. A system comprising:
a pretrained language agnostic transformers model configured to generate first embeddings of a user query received in a first language;
a plurality of embedding layers configured to generate second embeddings of the user query in a second language different from the first language;
a first classification model configured to predict a first intent probability distribution for the user query based on the first embeddings;
a second classification model configured to predict a second intent probability distribution for the user query based on the second embeddings, the second intent probability distribution corresponding to the first intent probability distribution; and
a weighted average calculator configured to calculate a weighted average between the first intent probability distribution and the second intent probability distribution,
wherein the weighted average is used to generate a response to the user query,
wherein the first classification model and the second classification model are trained based on training examples in a combination of different languages,
wherein the first classification model is trained based on an original language of the training examples, and
wherein the second classification model is trained based on a machine translation of the training examples from the original language to the second language, when the original language is different from the second language.
|