Spelling and grammar correction

[arvindpdmn]

Interactive spell checking is handled by #259.

The proposed task will check for both spelling and grammar. It will also suggest corrections, similar to what Gmail does. However, within the scope of this task the NLP model is expected to run offline. Suggestions will not be given interactively to the user. This will be scheduled later as a separate task.

Spelling checks should be sensitive to context. For example, "I showed him the write way to format the text" has a spelling error but most spellcheckers will not catch it. Word "write" should be "right". Gmail catches this error.

During inference, input to the NLP model will be an entire article. Thus, model has sufficient context. Output should be problem locations and suggestions. BERT-based model is expected to be suitable for this task but other approaches can be researched.

[arvindpdmn]

Resources:

• https://dev.languagetool.org/ (not sure if this is relevant)