> From: "WALLYWORLD" <ran...@[EMAIL PROTECTED]
>
> I want to point out that the use of internet involves a cache of
> the data.
That's one of those aspects that is usually transparent, so it
shouldn't be mentionned unless you have some really good reason.
> Although there may be two ends of the connection, there may not
> be a connection.
You mean there isn't a real-time end-to-end connection? Of course
there *must* be an end-to-end virtual connection when using HTTP.
(End-to-end means client-to-server. There needn't be a
user-to-database virtual connection in real time. User-to-client,
client-to-server, and server-to-database, can be relatively
asynchronous to some degree, while each of the three is internally
synchronous.)
> I know a web database course with a working model of a test
> taking system, Each test has many questions, and there are several
> tests to take. Each question is multiple choice.
That's garbage! With manually taken test on mark-sense paper, of
course multiple-choice is the only feasible way. But with
InterNet-based client/server tests, short-answer fill-in with
automatic guidance toward correct spelling of correct answer with
partial points-off for such corrections needed, is much much better
than multiple choice. There's really no excuse for pretending we
have a batch system of submitting decks of mark-sense cards, when
we actually have an interactive Web client/server system.
Guessing multiple choice by elimination is nowhere as good as
short-answer fill-in in determiming whether somebody really knows
the correct answer to a question. Multiple choice ****fts the
thought process away from actually knowing the answer, to learning
heuristics for eliminating obviously false answers the guessing
randomly from the rest, which distracts from actual learning.
Fill-in restores the emphasis in learning the correct answer and
testing whether the student really knows the correct answer even if
slightly misspelled.
English: "perchance"
Spanish: "*** casualidad"
Don't you think it rather defeats the lesson if five multiple
choices are given instead of the user just typing the correct
3-letter word?
English: "The word of the day"
Spanish: "La palabra *** d{i'}a"
Ditto?
English: "my parents love each other"
Spanish: "mis padres ** aman")
Ditto (2 letters)?
English: "forever"
Spanish: "**** siempre"
Ditto (4 letters)?
I actually think a person who knows the correct answer is more
likely to make a mistake matching the letter A,B,C,D,E with correct
answer than just typing the short answer in directly, and typing
directly is faster than scanning multiple choices too.
> There are of course right answers, and wrong answers associated
> with each question.
With short-answer fill-in there is no need to set up strawmen wrong
answers. Just store the secret correct answer and be done with it!
Question goes database->server->client->user
Student's answer goes user->client->server
Correct answer goes database->server
Comparison occurs on server, which is secure against student's
eavesdropping.
> He uses the Access database to store the questions, and answers.
Yes. (And for computational questions, such as solving algebraic
equations, which are parameterized in real time, the database would
store an algorithm for computing the correct answer, or a one-way
algorithm for comparing correct/student answer without first
computing the correct answer.)
> But in your model, you could store the connections to the web
> pages which contain sets of answers, and indexed questions (by test).
Note that for a final exam we'd need some way to prevent the
student from cheating by doing a Google search to find the Web
pages containing the Q/A pairs and then simply copying and pasting.
Google Images keyword-tagging game provides a start to a solution:
Suppose we have native English speakers learning Spanish, and
native Spanish speakers learning English. One fact about foreign
languages is that it's easier to translate from foreign to native
language than vice versa. Another fact is that supplying one
missing word is easier than supplying the entire translation. So
here's my idea for a "final exam" which can't be cheated: Random
text in language A is picked from the Web. It's passed to native
speakers of language A to proofread. If a vast majority accept it
as valid, it goes to the next step, else it's rejected. Natives of
language B are asked to translate from language A. If a vast
majority agree on the translation, then it goes to the next step,
else it's rejected. Natives of either language are asked to fill in
just one missing word of their foreign language, given full text of
their native language. Grading "on the curve" per Shannon entropy
is performed. I.e. if 100% of the students get it correct, in both
directions A->B and B->A then it's regarded as "too easy". If just
one person gets it wrong, that person gets a big bad mark while
everyone else gets a small good mark. If just a few get it right,
they get big good credit while everyone else is penalized only a
little. If there's a spread from exactly correct answer through
slightly misspelled correct ansers to completely wrong answers,
partial credit is given. Since all translations are on-the-fly via
the proofreading game and the full-translation test, it's very
unlikely that the translation pair already exists online, and on
the average it wouldn't be profitable for any student to try to
cheat by using Google to look for the translation pair. As a extra
protection, the test-maker system might explicitly do a Google
search for each translation pair, and also ask BabelFish for
translation, just to make sure neither affords a way to cheat. Of
course for the *final*exam*, students would be locked in a room
with no access to cellphone or any other way to cheat. Then they
could be given 100% questions they already studied online, and
nearly 100% correct answers could be expected to match what the
student claimed to demonstrate online.
|
