[possibly boring to most] Bob, what you noticed with your grandchild is the norm, or at least that's what I picked up from studies I read while I was fiddling around with computational linguistics way back when I was naive enough to think that teaching a computer to talk could be done independently of teaching a computer to think ... still, a dissertation completed and accepted is a dissertation finished.
I no longer have the papers I read - they got dumped several house moves ago - and so I can't quote, but nearly all children seem to learn language in much the same way, the detail of their approach depending on the language being learned.
At the same time as I was scrivening my way towards my post-grad diploma, my niece was rising two and a half years old and talking fit to bust. She was quite bright and learned new vocab at an impressive rate. But she had a grammatical/linguistic quirk which I found fascinating. She would say:
"I talk" for the present
"I talked" for the imperfect past, and
"I have talkted" for the perfect past.
The "talkted", as far as I could make out, was an attempt to add an extra "ed" to "talked", creating "talkeded", so to speak, the beginner's approach to parsing perfect past participles. It seemed to show that she had absorbed the idea that the perfect tense sometimes took a different form of the verb than the imperfect, and it was better to be "safe" than sorry.
Examples of this of course are fairly common, but for example:
"I grow"
"I grew"
"I have grown"
Why she, or any child, would be struck by this relatively small difference in verb formation to the extent that she would form a word-formation rule based on it, I have no idea.
And the rule of thumb she used kind of stuck, because I would correct her when she did it to one particular verb, e.g. "talked", but she would merrily continue to apply it to the next verb along.
At the time, I put it down to her being just her, but I've heard similar constructs out of the mouths of other near-babes and ex-sucklings since, so the papers I was reading would appear to be empirically corroborated even if it is based on a rather small sample.
Didn't help with the computational linguistics in the slightest, unfortunately ...
FWIW
