Heh, I have been asked this question a thousand times! ;-) Basically, my goal was to learn to C++, and I did that by programming a text adventure game. I figured a text adventure was a significant enough challenge for me at this stage. I quickly found out, though, that while it's relatively easy to program a "simple" text parser, doing the job right (i.e., up to Infocom standards or beyond) would really require a massive amount of work. It's exciting stuff, though, and I really think the next generation IF should take advantage of data mining and expert systems techniques to really up the ante. It seems like the hardest part to get right is not so much guessing what the player wants the avatar to do, but rather generating believable responses to wild inputs. Language is infinite, so there is no way to ever make a system that could simply spit out a pre-programmed response to every possible input. Rather, the system would have to use comparative analysis and determine which of a set of pre-programmed responses would be most appropriate for that given situation. If we ever want to succeed in this, we have to stop thinking in terms of the "correct response" and think instead of many acceptable responses.