RE Microsoft Word: last I heard they use regular expressions or something similar, so I wouldn’t expect it to do great when there’s an interaction of number agreement with coordination. Their translation team has some grammar-based parsers but I don’t know if the Office team uses them, or if they even model agreement any better.
I also have little objection to Where is the key and the book? and even less with the contraction Where’s. Where are the key and the book? sounds almost pedantic in an informal register, but I also know better (when I notice it) when speaking/writing more formally. I would appreciate a model of standard English grammar to tell me when I’ve made such a mistake, and I think that’s part of the “policy” with the ERG: it should say when something is grammatical or not and not just provide a best guess when something isn’t. But it also reflects modern usage in some respects, such as This data shows us nothing parses but These data show us nothing does not. Sometimes it accommodates regional variation, such as The government is incompetent and The government are incompetent; both parse, but both also have
NUM: sg on
If you want to parse things that are a little ungrammatical, you can try using a version of the ERG with “mal-rules”. I’m not sure which configuration is best, but when I compile the ERG with it’s
ace/config-educ.tdl configuration, I’m able to get a parse for Where is the key and the lock?.