Biased data labeling is hazardous to children (or is it?)

[mattbell]

I'm watching what looks like a lot of shitty parenting in the boat here. Kids are consciously and unconsciously trying to sort out how the world works, and simply being told “no” and getting yanked away when they do something wrong is an inefficient way of communicating to them. They kids are just running around behavior-space like little pinballs, bumping up against “no”s every few seconds,. Kids easily pick up and imitate what the grown-ups around them do, but it's hard to figure out what grown-ups don't do since they aren't doing it.  So why not, when a kid does something wrong, stop them and show them the correct alternative so they can imitate it?  Or you could help them understand what specifically they did that was bad by pantomiming the bad action, saying “no”, then pantomiming a correct alternative and saying “yes”. Or, (depending on what the bad action is) do it to them, ask them how they feel, and then tie that back to their actions.

I'm starting to think of allusions to my knowledge of machine learning. You can't effectively train a machine learning system with a partially labeled data set where only the negative examples are labeled and many negative examples are unlabeled. I should probably stop now. Kids are way more complex than that.

For example:

“Wow... this crying thing really gets me attention from the parents but everything else doesn't. I've got to remember that!”

Is this how drama queens are born?

Comments

[lynthia.livejournal.com]

Obviously, I've been reading a lot about how to teach children things lately. It's very interesting. One of the most useful things I've read in the last few days is the idea that you shouldn't use the word "no" unless the child is doing something dangerous, because otherwise they just get used to hearing it all the damn time. Instead, do something more like what you suggested - either distract them from the behavior you don't want them to engage in, or if they are old enough, tell them why they shouldn't do that and suggest what to do instead.

Reasoning with a 1 year old is dumb. They don't have the pathways to understand "you shouldn't do that because." It's frustrating for both adult and child to try. But everyone knows that young children are champions at mimicry.

Also, working on pictures. I'm usually the one with the camera, so it's hard to get them with both me *and* the baby. ;)

[lasofia.livejournal.com]

Yeah that. But I'll tell you why parents don't always do it - it takes a lot more time and thought and energy.

When Mateo was little but verbal, so a few months before two, we started working on the pattern:

Bad action happens (like shrieking when someone comes to visit)

Stop bad action
Establish what he wanted to happen with that action (like he wanted attention)
Establish what actually resulted from the action (upset looks, his removal)
Find an action that would more likely result in the desired outcome (like giving a greeting)
Teach that action to him

It was astonishingly effective at that age and through maybe age 4.

It didn't and doesn't work quite as well with Rio.

[nasu-dengaku.livejournal.com]

That procedure sounds like it makes a lot of sense, though as you say it's the sort of thing that does require patience on the parents' part, and if the parents never learned patience...

[lasofia.livejournal.com]

It's not just learning patience, it's maintaining patience that's the trick. There are tons of very effective parenting techniques taht work great so long as you are rested and focused enough to remember and apply them. And then there are the ineffective methods we use when we're tired and overwhelmed!

That said, two of the most annoying things to me in watching parenting are:
- rewarding behavior you don't want. Like giving in to whining because it's easier. Or to screaming, or temper tantrums.
- making kids not trust you by not following through (on promised threats or promised anything)

A side note on negative examples: It's amazingly more effective to tell a child what you want him to do as opposed to what you want him not to do. The trick is figuring out how to phrase the positive half of the instruction. As in, not "don't whack things with that push toy!" but, "keep the wheels of that toy on the ground."

[easwaran.livejournal.com]

It's a really fascinating question figuring out just what children and babies are able to pick up on, and how they manage to learn things from what sorts of stimulus. Because kids really do manage to learn so much even from such bad instruction. It really doesn't seem possible. (This is Chomsky's "poverty of the stimulus" argument in favor of the claim that we have specific language capacities that are innate, beyond our general cognitive abilities for learning things.)

Anyway, there's lots of positive examples that get labeled (basically everything they see an adult doing around them) - the interesting thing is really how few negative examples get labeled (just the things they try to do that someone sees and puts a stop to). (I think this is the site of another major argument in linguistics, between those who want to match syntax to actual verbal behavior in a corpus of data, and those who want to match it to the native speaker intuitions of the linguist. The corpus people say that intuitions are too easy to corrupt, but the intuitions people say that a corpus only gives you positive data, and you really need negative data as well in order to properly develop a theory.)

[nasu-dengaku.livejournal.com]

Interesting background, thanks.

[peristaltor.livejournal.com]

In my experience, yes.

What's worse, for the rest of the family, the whining becomes a trigger they don't recognize. "No," and "Shut up!" become automatic reactions on their parts. I used to see it in a friend's family. His younger brother grew up a bit screwed up as a result.