[lojban] lojban teaching software suite (was: Re: lessons deadline)

[Jay F Kominek <lojban-out@lojban.org>]

On Thu, Sep 19, 2002 at 10:05:22AM -0700, Jim Carter wrote:
> Here's an idea for a suite of language learning software.  I'm not saying I
> have all of these programs, or even that I know how to write some of them.

I don't have any of them, but I have a fair idea how I'd go about writing
most of these. Probably no time to do so, though. Anyone interested, by
all means, inquire further.

> 1. Vocabulary practice, particularly structure words.  There would be
> an order, designed by the author, for learning the words, or
> importance grades on each one.  We have several of these programs
> already.

Yes, though something integrated with all the rest of the software you're
mentioning would be even more valuable.

> to show a picture of the referent of the word, when feasible: a
> picture of a cat is simple; a picture of "truth" is not, or perhaps
> a picture story: a comic book style illustration of how the word is
> used.  I understand that the graphic elements are a lot of work to
> make and would take a lot of disc space, but I quote: "Ask for the
> moon; they might give it to you!"

Clip art? People with Microsoft Office can access a huge amount of it,
and searching for something like "cat" will turn up hundreds of comic
book style cats. Not sure what the license on the resultant data would
be, or how it could be integrated into some new piece of software, though.

Speaking of a picture story, Allan Bailey has mentioned in the past an
Esperanto book which tries to teach Esperanto without using any other
language. I'd still like to see someone try to do that with Lojban.
Maybe just duplicate the way the Esperanto book does it, and then tweak
slowly as needed. (I'm too cheap to buy the book, myself, and see how
they do it.)

> 4. A writing analyser.  It would detect unrecognized words and
> ungrammatical or semantically bizarre constructions, and would point
> them out as possible or definite errors.  For the "Lojban as a
> second language" student, the ability to translate into low-quality
> natural language would be helpful.  Our existing software is a good
> beginning.  Besides working on whole files, it needs an interactive
> interface similar to the word finder.

Shew, this would/could be hard in that it would require error correcting
parsing. There is a fair amount of research out there on the topic,
though, so it would just be a matter of someone who can read it, applying
it. A collection of semantically bizarre constructs would be needed,
as well. Some would be easy, I suppose, like non-abstractions in places
that expect abstractions (or vice versa), or doubled tu'a, but many
would be quite a bit more subtle. And would probably require interaction
with students to work out the more common cases.

> 5. A "net buddy", an artificial intelligence that would conduct vapid
> but grammatical conversation.  I had in mind something along the
> lines of "Couch" or "Eliza", maybe with a little more work put into
> holding up its end of the conversation.  It would mostly stick to
> words and grammar already known, emphasizing recently learned
> material.  It should also push the student by using material from
> the next lesson, and for some students this will be a major means to
> advance in vocabulary.

http://www.megahal.net/

It would be very easy to make something similar to that, except which
spoke Lojban, and always did so grammatically. Instead of trying to
ascertain the probability of a transition between pairs of words as
Megahal (roughly) does, something using Lojban would be able to identify
the parts of speech, and slowly accumulate probabilities for expansions
of rules in the grammar. A lesson discussing tu'a could force the
sentence generator towards constructs using tu'a, and the selbri picker
towards things which need abstractions. Eliza's habit of mimicing things
could be emulated by having a list of all gismu (or lujvo) places which
are {super|equi(?)}sets of other places (and listing those places), so
that if the user says something like:

 da jibri mi

It would know that all se jibri's are also gunka's, and thus could ask:

 do gunka fi ma

Hrm. Whether or not Eliza's mimicry could be mixed in with Megahal-esque
random sentence generation, I'm not actually quite sure of, now that I
think about it.

-- 
Jay Kominek <jkominek@miranda.org>
If you can't do it in Perl,
you don't need to do it.