Meta Platforms, Inc.

Facebook, Inc.

Facebook Technologies, LLC

Ahmed Aly

Shrey Desai

Akshat Shrivastava

Yue Liu

Xiaohu Liu

Baiyang Liu

Vivek Natarajan

Pierce I-Jen Chuang

Armen Aghajanyan

Arun Babu

Denis Savenkov

Pooja Sethi

Abhinav Arora

In one embodiment, a method includes receiving training utterances associated with a domain, receiving ontology labels for the domain, wherein the ontology labels comprise one or more of an intent or a slot, generating an inventory for the domain, wherein the inventory comprises at least a respective index and respective span for each intent or slot, wherein the respective span comprises a respective descriptive label associated with the intent or slot, and wherein the respective descriptive label comprises a natural-language description of the intent or slot, generating frames for training utterances based on the training utterances and the inventory by a natural-language understanding (NLU) model, wherein each frame comprises a structural representation of the respective training utterance, wherein the structural representation is generated based on a comparison between the corresponding training utterance and the inventory, and updating the NLU model based on the frames.

Low-resource task-oriented semantic parsing via intrinsic modeling for assistant systems

In one embodiment, a method includes receiving a user input comprising input tokens from a client system, parsing the user input to determine ontology tokens and utterance tokens corresponding to the input tokens, decoding the ontology tokens and the utterance tokens to generate a span-based frame representation comprising intents, slots, and a span, wherein the ontology tokens are decoded into the intents and slots, and wherein the utterance tokens are decoded to determine the span comprising one or more tokens of the input tokens, wherein the span comprises a first index endpoint associated with a first token of the one or more tokens and a second index endpoint associated with a second token of the one or more tokens, and executing, responsive to the user input, one or more tasks based on the span-based frame representation.

Span pointer networks for non-autoregressive task-oriented semantic parsing for assistant systems

In one embodiment, a method includes receiving a user request to automatically debug a natural-language understanding (NLU) model, accessing a plurality of predicted semantic representations generated by the NLU model, wherein the plurality of predicted semantic representations are associated with a plurality of dialog sessions, respectively, wherein each dialog session is between a user and an assistant xbot associated with the NLU model, generating a plurality of expected semantic representations associated with the plurality of dialog sessions based on an auto-correction model, wherein the auto-correction model is learned from dialog training samples generated based on active learning, identifying incorrect semantic representations of the predicted semantic representations based on a comparison between the predicted semantic representations and the expected semantic representations, and automatically correcting the incorrect semantic representations by replacing them with respective expected semantic representations generated by the auto-correction model.

Continuous learning for natural-language understanding models for assistant systems

In one embodiment, a method includes receiving a user input comprising one or more words at a client system, wherein each word comprises one or more characters, inputting the words to a convolutional neural network (CNN) model stored on the client system, accessing a plurality of character-embeddings for a plurality of characters, respectively, from a data store of the client system, generating one or more word-embeddings for the one or more words, respectively, based on the accessed character-embeddings by processing the accessed character-embeddings with one or more convolutional layers and one or more gated linear units of the CNN model, determining one or more tasks corresponding to the user input for execution based on an analysis of the one or more word-embeddings by the CNN model, and providing an output responsive to the user input based on the execution of the one or more tasks at the client system.

On-device convolutional neural network models for assistant systems

In one embodiment, a method includes receiving a user input by the first user from a client system associated with a first user, parsing the user input to identify one or more n-grams associated with the user input, accessing a user profile associated with the first user, wherein the user profile is stored in a first data store, accessing ontology data based on the one or more identified n-grams from one or more information graphs, wherein the one or more information graphs are stored in one or more second data stores, respectively, determining contextual information associated with the user input, generating semantic information by aggregating the user profile, ontology data, and contextual information, generating a feature representation for the identified one or more n-grams based on the semantic information, and resolving one or more entities associated with the one or more n-grams based on the feature representation.

Aggregating semantic information for improved understanding of users

Growing community of inventors

Kenmore, WA, United States of America

Ahmed Aly