Skillet,
Are the scoring algorithms generic or are they developed based on the topic?
Well, both, actually. There are a lot of programs out there, and some are generic and others are based on the topic. The more advanced ones actually learn (in a matter of speaking) by being fed sample responses that are marked.
To get a very rough idea of this, turn the grammar checker on in a document you've written. You can see that the computer can do a decent job of catching noun/verb disagreements, number disagreements, run-on sentences, etc. So it can generate a rough idea of how well constructed the essay is. Then, it can shift to words and word phrases that are appropriate.
If you look up natural language processing (wikipedia will do), you'll get a better idea of what is happening there. For example, there is a cool thing they use also called "colloquies." A colloquy is a word pair or short phrase that appears more often than any two or three word phrase might be expected to appear. "Scoring algorithms" is an example of a colloquy. Consider: we like strong tea and powerful computers, but we don't like powerful tea and strong computers. Two colloquies versus two phrases that aren't really a natural part of English.
If you are clever, you could still write nonsense that would get a good score from a computer, but it's gotten to the point where you really have to be clever to do so. As a measurement person, the thing I like about computer-based essay scoring is that it doesn't matter whether the marker agrees with your position or not, or likes your handwriting (if the test is initially hand-written -- we know that quality of handwriting affects scoring) or whether the marker has just had it on a Friday afternoon and is giving you grade X no matter what you've written. That's why a balance between a neutral computer, and a human scorer (who may rightly see brilliance in something the computer could miss "To be or not to be") would be how I would want my work scored. And if the essay is written on computer, saves a lot of time and effort.
Also, I might mention that computer-assisted feedback is highly effective in helping kids improve their work in a course. This research was carried out in an unnamed, large, East Coast public university that recently just missed on two highly ranked QBs but hopefully will get a commitment out a third QB, who while not as highly ranked, looks damn good to me on tape.