Would it be possible to train model to offer multiple suggestions?
Sometimes it is unclear what is the right fix.
For example, in this sentence:
The books is on the table.
There are 2 equally correct versions:
The book is on the table
The books are on the table
Would be incredibly useful in such a situation to have the model providing both options.
That sounds very useful, but imagine you have a dozen such ambiguities, due to combinatorial explosion you get thousands of valid interpretations. How do you get any form of usable output out of something like that?
One solution I can see is to have inline variants. e.g.,
The [book is|books are] on the table.
This would not be useful for the use case of "just clean up the text for me", but would be helpful for a GUI type editor to offer the alternative in a dropdown, and the default being the first option. Maybe a secondary prompt format could trigger such a formatting.