Increasing semantic coverage with semantically irrelevant insertions

• US 9,129,598 B1
• Filed: 03/27/2015
• Issued: 09/08/2015
• Est. Priority Date: 02/28/2013
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Mengibar, Pedro J. Moreno, Epstein, Mark Edward, Biadsy, Fadi
• Primary Examiner(s): Pham, Thierry L
• Application Number:
  US14/671,353
• Time in Patent Office: 165 Days
• Field of Search: 704/1, 704/2, 704/8, 704/231, 704/246, 704/257
• US Class Current: 1/1
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules
• CPC Class Codes: G10L 2015/0631 Creating reference templates; Clustering

Abstract

A method includes accessing data specifying a set of actions, each action defining a user device operation and for each action: accessing a corresponding set of command sentences for the action, determining first n-grams in the set of command sentences that are semantically relevant for the action, determining second n-grams in the set of command sentences that are semantically irrelevant for the action, generating a training set of command sentences from the corresponding set of command sentences, the generating the training set of command sentences including removing each second n-gram from each sentence in the corresponding set of command sentences for the action, and generating a command model from the training set of command sentences configured to generate an action score for the action for an input sentence based on: first n-grams for the action, and second n-grams for the action that are also second n-grams for all other actions.

Claims

1. A computer-implemented method performed by a data processing apparatus, comprising:

for each action of a set of actions, generating a training set of command sentences from a corresponding set of command sentences for the action, the generating the training set of command sentences including retaining first n-grams determined to be semantically relevant for the action and removing second n-grams that are determined to be semantically irrelevant for the action from each sentence in the corresponding set of command sentences for the action; and

for each action, generating a command model from the training set of command sentences and persisting the command model in a computer memory device, the command model facilitating the generation of an action score for the action for an input sentence based on;

the first n-grams determined to be semantically relevant for the action; and

second n-grams for the action that are also second n-grams for all other actions.

2. The computer-implemented method of claim 1, wherein for each action, generating a training set of command sentences from the corresponding set of command sentences includes removing each n-gram from each sentence in the corresponding set of command sentences for the action that is not a first n-gram for the action.

3. The computer-implemented method of claim 1, wherein determining second n-grams in the set of command sentences that are semantically irrelevant for the action comprises:

determining, for each action, corresponding candidate second n-grams that are semantically irrelevant for the action; and

selecting the candidate second n-grams that are not first n-grams for any of the actions as the second n-grams.

4. The computer-implemented method of claim 1, wherein for each action, generating or updating a command model comprises generating a grammar rule model based on the first n-grams for the action.

5. The computer-implemented method of claim 1, wherein for each action, generation a command model comprises generating a statistical based model based on the first n-grams for the action.

6. The computer-implemented method of claim 1, wherein the command sentences are voice command sentences.

7. A system, comprising:

a data processing apparatus and a computer storage apparatus in data communication with the data processing apparatus and storing instructions executable by the data processing apparatus and that upon such execution cause the data processing apparatus to perform operations comprising;

for each action of a set of actions, generating a training set of command sentences from a corresponding set of command sentences for the action, the generating the training set of command sentences including retaining first n-grams determined to be semantically relevant for the action and removing second n-grams that are determined to be semantically irrelevant for the action from each sentence in the corresponding set of command sentences for the action; and

for each action, generating a command model from the training set of command sentences and persisting the command model in a computer memory device, the command model facilitating the generation of an action score for the action for an input sentence based on;

the first n-grams determined to be semantically relevant for the action; and

second n-grams for the action that are also second n-grams for all other actions.

8. The system of claim 7, wherein for each action, generating a training set of command sentences from the corresponding set of command sentences includes removing each n-gram from each sentence in the corresponding set of command sentences for the action that is not a first n-gram for the action.

9. The system of claim 7, wherein determining second n-grams in the set of command sentences that are semantically irrelevant for the action comprises:

determining, for each action, corresponding candidate second n-grams that are semantically irrelevant for the action; and

selecting the candidate second n-grams that are not first n-grams for any of the actions as the second n-grams.

10. The system of claim 7, wherein for each action, generating or updating a command model comprises generating a grammar rule model based on the first n-grams for the action.

11. The system of claim 7, wherein for each action, generation a command model comprises generating a statistical based model based on the first n-grams for the action.

12. The system of claim 7, wherein the command sentences are voice command sentences.

13. The system of the claim 12, wherein for each action, generating a training set of command sentences from the corresponding set of command sentences includes removing each n-gram from each sentence in the corresponding set of command sentences for the action that is also not a first n-gram for the action.

14. The system of claim 12, wherein determining second n-grams in the set of command sentences that are semantically irrelevant for the action comprises:

determining, for each action, corresponding candidate second n-grams that are semantically irrelevant for the action; and

selecting the candidate second n-grams that are not first n-grams for any of the actions as the second n-grams.

15. The system of claim 9, wherein the command sentences are voice command sentences.

16. The system of claim 9, further comprising:

a user device and a computer storage apparatus in data communication with the data processing apparatus and storing instructions executable by the data processing apparatus and that upon such execution cause the data processing apparatus to perform operations comprising;

receiving from the data processing apparatus the command models for the action;

receiving an input sentence of n-grams;

applying the command models to the input sentence of n-grams to generate respective action scores;

determining, from the command models, an action invoked by the input sentence based on the action scores, the action being one of the actions in the set of actions; and

performing the action.

17. The system of claim 13, wherein:

each command model is a grammar rule model; and

applying the command models to the input sentence of n-grams comprises, for each grammar rule model, parsing the input sentence of n-grams.

18. The system of claim 13, wherein applying the command models to the input sentence of n-grams comprises, for each command model scoring the input sentence based, in part, on first n-grams that have been determined to be semantically relevant to the action and second n-grams that have been identified as being semantically irrelevant for all actions.

19. A non-transitory computer readable storage medium in data communication with the data processing apparatus and storing instructions executable by the data processing apparatus and upon such execution cause the data processing to perform operations comprising:

accessing data specifying a set of actions, each action defining a user device operation that is different from user device operations for each other action;

for each action of a set of actions, generating a training set of command sentences from a corresponding set of command sentences for the action, the generating the training set of command sentences including retaining first n-grams determined to be semantically relevant for the action and removing second n-grams that are determined to be semantically irrelevant for the action from each sentence in the corresponding set of command sentences for the action; and

for each action, generating a command model from the training set of command sentences and persisting the command model in a computer memory device, the command model facilitating the generation of an action score for the action for an input sentence based on;

the first n-grams determined to be semantically relevant for the action; and

second n-grams for the action that are also second n-grams for all other actions.