Natural language to API conversion

Patent metadata

Field	Value
Publication number	US-11842205-B2
Application number	US-202217898701-A
Country	US
Kind code	B2
Filing date	Aug 30, 2022
Priority date	Mar 13, 2018
Publication date	Dec 12, 2023
Grant date	Dec 12, 2023

Abstract

Representative embodiments disclose mechanisms to map natural language input to an application programming interface (API) call. The natural language input is first mapped to an API frame, which is a representation of the API call without any API call formatting. The mapping from natural language input to API frame is performed using a trained sequence to sequence neural model. The sequence to sequence neural model is decomposed into small prediction units called modules. Each module is highly specialized at predicting a pre-defined kind of sequence output. The output of the modules can be displayed in an interactive user interface that allows the user to add, remove, and/or modify the output of the individual modules. The user input can be used as further training data. The API frame is mapped to an API call using a deterministic mapping.

First claim

What is claimed is:

1. A computer implemented method, comprising:
receiving a natural language utterance;
submitting the natural language utterance to a trained machine learning model comprising an encoder and a plurality of decoder modules, each of the plurality of decoder modules coupled to the encoder and being trained to recognize one or more tokens output from the encoder and to map the one or more tokens to one or more corresponding application programming interface (API) parameters;
receiving, from the trained machine learning model, a plurality of API parameters provided by two or more decoder modules of the plurality of decoder modules, the two or more decoder modules being selectively activated on an individual basis based at least on the one or more tokens output from the encoder;
mapping the plurality of API parameters into an API call; and
outputting the API call.

2. The method of claim 1 , wherein the two or more decoder modules include a first decoder module configured to output a first API parameter of the API call and a second decoder module configured to output a second API parameter of the API call.

3. The method of claim 1 , wherein the trained machine learning model comprises a controller configured to selectively activate individual decoder modules based at least on an encoded utterance output by the encoder.

4. The method of claim 1 , wherein individual decoder modules are configured to receive an encoded utterance output by the encoder and selectively self-activate based at least on the encoded utterance.

5. The method of claim 1 , further comprising:
using a set of deterministic rules to map the plurality of API parameters into the API call.

6. The method of claim 1 , further comprising:
outputting a first API parameter provided by a first decoder module;
receiving user input indicating whether the first API parameter is correct; and
training the first decoder module based at least on the user input.

7. The method of claim 6 , wherein the outputting the first API parameter comprises displaying a graphical user interface having an edit control configured to edit the first API parameter to a corrected first API parameter in response to the user input.

8. The method of claim 7 , further comprising including the corrected first API parameter in the API call and outputting the API call to a service.

9. A system comprising:
a processor; and
a device-storage media storing executable instructions which, when executed by the processor, cause the processor to:
process a natural language utterance using an encoder to obtain an encoded utterance;
selectively activate individual decoders of a plurality of decoders based at least on the encoded utterance;
using the activated individual decoders, map tokens of the encoded utterance into corresponding application programming interface (API) parameters; and
generate an API call having the corresponding API parameters.

10. The system of claim 9 , wherein the executable instructions, when executed by the processor, cause the processor to:
receive the natural language utterance over a network from a user device.

11. The system of claim 10 , wherein the executable instructions, when executed by the processor, cause the processor to:
send the API call over the network to the user device.

12. The system of claim 9 , wherein the executable instructions, when executed by the processor, cause the processor to:
receive the natural language utterance locally from a user; and
invoke a local service on the system using the API call.

13. The system of claim 9 , wherein the executable instructions, when executed by the processor, cause the processor to:
receive the natural language utterance locally from a user; and
invoke a remote service by sending the API call over a network to the remote service.

14. The system of claim 9 , wherein the executable instructions, when executed by the processor, cause the processor to:
map the corresponding API parameters into a corresponding intermediate representation of the API call; and
generate the API call by applying one or more conversion rules to the intermediate representation.

15. The system of claim 14 , the intermediate representation comprising an API frame.

16. The system of claim 9 , the encoder comprising a word embedding layer configured to map the tokens into vector representations that are provided to the activated individual decoders.

17. A computer-readable storage medium storing comprising executable instructions which, when executed by a processor of a machine, cause the machine to perform acts comprising:
obtaining an encoded utterance;
based at least on the encoded utterance, selecting two or more decoder modules from a plurality of decoder modules to decode the encoded utterance;
providing individual token representations of the encoded utterance to the selected two or more decoder modules;
receiving a plurality of application programming interface (API) parameters provided by the selected two or more decoder modules;
mapping the plurality of API parameters into an API call; and
outputting the API call.

18. The computer-readable storage medium of claim 17 , the acts further comprising:
selecting different sets of decoder modules to process different encoded utterances based at least on different token representations in the different encoded utterances.

19. The computer-readable storage medium of claim 18 , the different token representations comprising word embeddings.

20. The computer-readable storage medium of claim 17 , wherein outputting the API call comprises invoking a service by submitting the API call to the service.

Assignees

• Microsoft Technology Licensing Llc

Inventors

• Awadallah Ahmed Hassan
• Wang Miaosen
• White Ryen
• Su Yu

Classifications

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06F9/448 Primary

  Execution paradigms, e.g. implementations of programming paradigms · CPC title