I get the idea that it might be more powerful to not use wrong word in context, but a rule template that does the same, but will allow for antipatterns, which might be a great help.
One could make the template in a way that word 1 is always warned for, unless in its correct context; same for word 2.
This will cause more warnings. These could be reduced using antipatterns.
Each word pair will be one rulegroup, containing 2 rules.
Are there any objections to this approach? (Apart for having to create examples for all of those confusions)