You are right that not all of our frameworks have picked up as well as we had hoped. The text input issue you outline works just fine for me though, in fact I don't remember switching language for spell-checking in ages, I simply stopped thinking about it. Nepomuk had a bit of a slow start, but is increasingly useful, although not used to its full potential yet. So there's surely some work left to do, but the foundations are there -- and let's face it, it's not something simple to do, it's actually quite natural that it takes a while until this kind of technology mature, which I think it has now with our 4.7 release.
From my experience as Plasma Active developer, I was surprised how well the recommendations already worked. A quick example: I was browsing the web, as I do quite often with Active on the tablet lately. When I wanted to go back to a page I had recently read, I thought that we'd really need to implement browser history. Turns out all I was looking for was actually in the recommendations panel -- I was pretty surprised that it already worked that well (sure, for this limited use case, but we're talking alpha here, so cut us some slack :)).
As to the software stack: It wasn't mentioned in the article, but Martin Grässlin, who develops and maintains Kwin, Plasma's window and compositing manager, presented our plans WRT to Wayland and how compositing and the related parts of our graphics stack look like in the future. Moreover, I've just got the graphics stack sorted, so that on my tablet, I'm running the compositing manager on top of OpenGL-ES now. This is part of our plan, of course.
The UI is not done yet, of course. I agree that it looks messy, and it has a fair amount of rough edges right now. This is something we're focusing on right now, basically trying to sort out as many papercuts as we can. The progress of the past week alone is really impressive, so pace-wise, I think we're doing quite fine. Again, it's not done yet, but work in progress. Another important aspect here is that in the design of Contour and Plasma Active, we've worked with professional UI and interaction designers from day one, and I think that also shows in many concepts, big and small. It's still an iterative process, since nobody has done anything like we plan before, but it looks pretty good already.
The interesting parts of Nokia's Swipe UI are all open source (again, after a period of a few months where Nokia had taken this part of the codebase behind closed doors). You can get them from Gitorious, what you are probably looking for is in the Qt components project there. QML makes it pretty easy to implement something like this, as you can see.