Autocorrect candidate selection

A data processing system comprising:

• a processor;
• and
• a machine-readable medium storing executable instructions that, when executed, cause the processor to perform operations comprising:
• receiving textual content from a first application on a first client device associated with a first user for analysis by a writing assistant configured to provide suggestions for improving the textual content;
• analyzing the textual content using one or more natural language processing models to identify one or more candidate suggestions for improving the textual content;
• causing the first application of the first client device to display the one or more candidate suggestions;
• receiving feedback from the first client device comprising user acceptance information associated with the one or more candidate suggestions;
• aggregating the user acceptance information included in the feedback with historical user acceptance information to obtain aggregated user acceptance information;
• obtaining model performance information associated with the one or more natural language processing models;
• analyzing the aggregated user acceptance information and the model performance information to make a determination whether aggregate user acceptance information and the model performance information satisfies autocorrect threshold criteria;
• and
• training an autocorrect model based on the determination whether the aggregate user acceptance information and the model performance information satisfies the autocorrect threshold criteria, the autocorrect model being trained to receive as an input a suggestion category associated with a suggestion for improving textual content generated by the writing assistant and to output an indication whether the writing assistant can autocorrect the textual content based on the suggestion for improving the textual content.

A method implemented in a data processing system for providing suggestions for improving textual content, the method comprising:

• receiving user acceptance data at a model training pipeline, the user acceptance data indicating whether users have accepted autocorrections to textual content suggested by a writing assistant configured to provide suggestions for improving the textual content, each autocorrection being associated with a suggestion category;
• aggregating the user acceptance data by suggestion category using the model training pipeline to generate aggregated user acceptance data;
• training an autocorrect model using the model training pipeline based on the user acceptance data to receive a suggestion category and to output an indication whether the writing assistant can automatically apply autocorrections for the suggestion category, the model training pipeline training the autocorrect model to output an indication that the writing assistant can autocorrect for a respective suggestion category responsive to the aggregated user acceptance data indicating that the respective suggestion category satisfies an autocorrection threshold;
• receiving textual content from a first application on a first client device associated with a first user for analysis by the writing assistant;
• analyzing the textual content using one or more natural language processing (NLP) models to identify one or more candidate suggestions for improving the textual content;
• determining that a first candidate suggestion of the one or more candidate suggestions is associated with a first suggestion category;
• providing the first suggestion category as an input to the autocorrect model to obtain an indication whether the writing assistant can autocorrect the textual content by automatically applying the first candidate suggestion;
• generating autocorrected textual content by updating the textual content based on the first candidate suggestion responsive to the autocorrect model indicating that the writing assistant can autocorrect the textual content;
• and
• providing the autocorrected textual content to the first client device.