She's also starting to sing and, while she knows loads of songs, she doesn't know loads of lyrics so she just makes the closest sound she can to something like the tune. But, unlike her talking, the context gets her through and it's mostly easy enough to pick up the song and then you can sing along and over time she gets closer and closer to the words.
When I'm trying to implement something, if I do it bit by bit, designing each component to completion before the next, it's hard to get right. I have to repeat things and correct things and when I try to demo it to people, they can't work out what it is I'm trying to do.
If I try to do it end-to-end, not worrying too much about completeness but just trying to get one path from start to finish before polishing it, then the context of the process gets me through and it's mostly easier for others to pick up the gist of what I'm attempting, and then as I refine it with feedback or more spec or new test cases I get closer and closer to what I want.
Out of the mouths of babes...