That's what happens when you don't disambiguate.
Although I thought iStock's DA was based on Getty's; weren't these files properly DAd?

I have this weird theory that the disambiguate process that began at IS a few years ago consigned a huge number of images to some bottomless pit. I know my sales at IS dropped 40-60% immediately thereafter and never recovered. And, yes, I did go through the torturous process of updating my keywords although it took a few months. It didn't seem to make any difference at all in my sales level that remained far below past comparable months.

My memory sucks.  Wasn't Istock disambiguation before the sale?

I don't think so. I joined in late 2006, when Getty had already bought iStock, and everyone was whining about DAing their files - especially the long-established members with already several hundreds or thousands of files.
According to Wikipedia, "On February 9, 2006 iStockphoto was acquired by Getty Images for $50 million USD. ... ... On September 18, 2006 iStockphoto displayed the first results of the Getty Images ownership. A new keywording taxonomy called a �Controlled vocabulary� was borrowed from Getty Images and implemented on iStockphoto to control and manage keywords and searches, and to provide multilingual searching abilities."

My memory sucks.  Wasn't Istock disambiguation before the sale?

No. DA'ing was a result of adopting Getty's CV. We had a prior iStock-invented conversion from old to new categories (which I think was how they were going to try and solve the problem Gettty thought the CV addressed), so those of us who had been around a while had already been through one "new and improved" conversion already.

There were also bulk editing tools promised to contributors to help them with the conversion, and they were never released. They never helped in any way to educate contributors about the structure of the CV (which some of us did by walking the tree in the edit image dialog to learn how it was put together). There were massive amounts of contributor hours put into that effort. Hence the complaining.

This is a huge area, perhaps worthy of its own topic if you're going to try and make a smart search out of "regular" keywords, but...

Taking your example of a search for girl, I picked Little Girl from the meanings and a series of images of this woman were included. Girl is not a keyword on the image, but with little black dress and young woman I suppose your search assumed little girl. Incorrectly.

If I try to search for little girl blond I get this image and this one., neither of whom are little girls. You could have something simple that excludes any image with nude or sexy keywords from the results for Little Girl (SS won't allow those combinations to be entered on images).

There have been a number of innovative approaches to search, including using visuals (more like this one or this group) and making assumptions based on the sets of keywords. I'm no fan of Getty's CV (a big problem is that it's so inflexible and good keywords on less popular topics or places are hard to do), but if you're going to try a CV on the fly, it'll have to deal with a lot of complex issues to make it work well.