Clippy, User Experiences, and Microsoft Search

Chris Pratley blogs of Clippy: ...A big part of the Assistant plan was to use it as the gateway to help. You could click on the Assistant and ask it questions in normal English (or other language depending on your version of Office) did not do this as it turned out - people still overwhelmingly type a single word in the hopes that will get them the answer...the Assistant was actually a wash in terms of user acceptance. Many users told us that they really liked it and found it useful, something which technical people have a hard time believing, since they were the ones who pretty much uniformly didn't like the assistant.... via [Chris_Pratley's WebLog]

One of the things that we're hearing about Microsoft's forthcoming search engine is that it will focus on "natural language queries," as did, say, Ask Jeeves. The thing about natural language query is that you need a user-model. For instance, at iMind we could analyze the educational objectives of lesson plans because we knew "this lesson plan for 9th grade American History is from this teacher in this school district" and we had already spent gazillions of cycles preprocessing all of that context into a model that had "just" a couple of hundred dimensions. I don't think you can do relevant English-language query without a user model: "What's the best movie of the year?" "What's the most important news of the day?" "Why is my computer crashing?"

Is general-purpose English-language query achievable or is it destined to be the next Microsoft Bob or, if not Bob, the next Clippy (hated by some, appreciated by some)? If Web search could be dramatically improved by a server-side user model (essentially, "We store every query you ever enter and induce that you're a programming geek") would that be acceptable?