公开(公告)号：US20240346246A1

公开(公告)日：2024-10-17

申请号：US18300930

申请日：2023-04-14

Applicant: ServiceNow, Inc.

Inventor： Torsten Scholak ,  Dzmitry Bahdanau

IPC: G06F40/279

CPC classification number: G06F40/279

Abstract: A trained natural language model is provided that uses an input session history to generate outputs to an interpreter. Outputs to the interpreter, and inputs responsively received therefrom, are added to the history to generate additional model outputs as the history is updated. The model is trained to engage in goal-oriented dialog with the interpreter and with the user (optionally through interpreter function calls) to identify the user's goals, to learn information about modules, functions, and methods available in the interpreter that are relevant to the user's goals, and to execute function calls and/or commands, based on the learned information, that accomplish the user's goals. The use of a history that may be completely blank at the beginning of the session reduces the computational requirements of running the model, as well as allowing the model to ‘update’ itself as the available modules are update, added, or removed.

公开(公告)号：US12243518B2

公开(公告)日：2025-03-04

申请号：US17978690

申请日：2022-11-01

Applicant: ServiceNow, Inc.

Inventor： Dzmitry Bahdanau

IPC: G10L15/18 ,  G10L15/06

Abstract: The present disclosure relates to a data augmentation system and method that uses a large pre-trained encoder language model to generate new, useful intent samples from existing intent samples without fine-tuning. In certain embodiments, for a given class (intent), a limited number of sample utterances of a seed intent classification dataset may be concatenated and provided as input to the encoder language model, which may generate new sample utterances for the given class (intent). Additionally, when the augmented dataset is used to fine-tune an encoder language model of an intent classifier, this technique improves the performance of the intent classifier.

公开(公告)号：US11768831B2

公开(公告)日：2023-09-26

申请号：US17397117

申请日：2021-08-09

Applicant: ServiceNow, Inc.

Inventor： Torsten Gerald Scholak ,  Dzmitry Bahdanau

IPC: G06F16/2452 ,  G06F16/242

CPC classification number: G06F16/24522 ,  G06F16/243

Abstract: A natural language query to domain-specific language query (NLQ-to-DSLQ) translation system includes a language model and a domain-specific language (DSL) parser that constrains the output of the language model to a DSL, such as structured query language (SQL). At each decoding step, the language model generates a predicted next token for each of a set of potential translations of a NLQ. The DSL parser evaluates each of the potential translations at each decoding step based on a set of stored DSL rules, which define valid terminology, syntax, grammar, and/or other constraints of the DSL. The DSL parser may reject and remove from consideration partial potential translations that are invalid or receive a low parsing score, such that the language model only continues to generate new tokens at the next decoding step for partial potential translations that are determined to be valid and/or sufficiently high scoring.

公开(公告)号：US20220358125A1

公开(公告)日：2022-11-10

申请号：US17397117

申请日：2021-08-09

Applicant: ServiceNow, Inc.

Inventor： Torsten Gerald Scholak ,  Dzmitry Bahdanau

IPC: G06F16/2452 ,  G06F16/242

Abstract: A natural language query to domain-specific language query (NLQ-to-DSLQ) translation system includes a language model and a domain-specific language (DSL) parser that constrains the output of the language model to a DSL, such as structured query language (SQL). At each decoding step, the language model generates a predicted next token for each of a set of potential translations of a NLQ. The DSL parser evaluates each of the potential translations at each decoding step based on a set of stored DSL rules, which define valid terminology, syntax, grammar, and/or other constraints of the DSL. The DSL parser may reject and remove from consideration partial potential translations that are invalid or receive a low parsing score, such that the language model only continues to generate new tokens at the next decoding step for partial potential translations that are determined to be valid and/or sufficiently high scoring.

公开(公告)号：US20230141398A1

公开(公告)日：2023-05-11

申请号：US17978690

申请日：2022-11-01

Applicant: ServiceNow, Inc.

Inventor： Dzmitry Bahdanau

IPC: G10L15/18 ,  G10L15/06

CPC classification number: G10L15/1815 ,  G10L15/063 ,  G10L2015/0631

Abstract: The present disclosure relates to a data augmentation system and method that uses a large pre-trained encoder language model to generate new, useful intent samples from existing intent samples without fine-tuning. In certain embodiments, for a given class (intent), a limited number of sample utterances of a seed intent classification dataset may be concatenated and provided as input to the encoder language model, which may generate new sample utterances for the given class (intent). Additionally, when the augmented dataset is used to fine-tune an encoder language model of an intent classifier, this technique improves the performance of the intent classifier.