[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: CONLANG: LogFlash and LESSY (!)



>1) Of the various mode, I could not understand almost anything! There have
>been a couple of messages on the list about that topic (one of them, quite
>detailed, from Logical Language Group <lojbab@ACCESS.DIGEX.NET>), but I cold
>not grasp that much from them either. Probably, I have to sit down with
>paper and pencil, draw some flow charts and hope for the best!


The user's manual, now somewhat outdated and not distributed because it is
so out of sync with the user interface, has extensive discussion on the
theory of the prgram, and its implementation, including flow charts.

One problem is that in the last version, we added so many options that one
is no longer compelled to follow the "official" flow chart, hence as you 
found by using it, that flowchart is not so obvious.

>3) Other are less interesting; for example, it is not easy to grasp from the
>screen which of the many things said about a word I'm supposed to learn.


Well, ideally you need to learn all of it %^).  But what is quizzed is
in recognition - the English keyword based on Lojban data, and in recall, 
the Lojban keyword based on a collection of English data.  This of course
is made clear in the user documentation ...

>4) something I consider a serious problem is the following:
>
>in the quiz mode (well, I cannot remember its formal name), I answered some
>questions as follows:
>
>gizmu   my answer       expected answer
>-----   ---------       ---------------
>bacru   "say"           "utter"
>gismu   "root"          "root word"
>tanru   "compound"      "phrase compound"
>tavla   "speak"         "talk"
>
>I am not saying that "say" is equivalent to "utter" or "speak" is equivalent
>to "talk"; however in the definition of the gismu's something like that
>appears:
>bacru   DEF: x1 utters/says ...
>tavla   DEF: x1 talks/speaks ...
>
>then my own answers were somehow considered also by the word list author;
>IMHO, the program should have accepted multiple translations for those
>words, otherwise the reult is exactly in the direction lamented by Edmund
>Grimley-Evans in a recent message: linking a particular question to a
>particular answer instead than a concept to another concept.


This is of course a limitation of the method.  It is possible for someone
using the program simpistically and without warning to establish simply
a one-to-one correspondence between English and Lojban keywords.  BUT
when one gets to REcall mode, the more important, the prompting information
is the full English definition as well as the short keyword.

That you have to type the single answer exactly is an implementation
restriction, one that we could probably design around if we wanted to put
the work into it.  In practice, we found it easiest to have there be a
UNIQUE word associated with each Lojban word, and that necessitated being 
picky.  If one focusses ONLY on the English keyword, for example, and not
the long definition, then if "root" were the prompt, how would you
know that the desired answer was "gismu"= root word as opposed to "genja"
the root of a plant.  You aren't SUPPOSED to work only with the keywords
as part of a broad effort to learn the language, but in actual "quiz mode"
when you are reading and typing fast, you want the cue to be as brief as
possible.

This also necessitates that the keyword not always be definitional.
WE have sfani (fly - the insect) and vofli (fly - the verb), and for 
the latter we chose the keyowrd "flight", making for minimum typing and
yet distinguishing the concepts.  Typing IS a significant problem in such
quizzing, long before you reach mastery, typos will be a major source of
errors (I think we even added in a typo correction option to allow people
to deal with this if they wanted to).

In practice, while at the beginning you link words only with other words 
and hence not with concepts, by the time you have mastered the whole
set of gismu this is not a problem.  You have a grasp of the whole semantic
space of the language at that point, and hence when trying to think of a
word for a concept, you run through the list of keywords to find those
most closely related to what you have in mind, and then (if necessary) check
the long form of the definition.  Lojban's structure, with multi-part
place structures pretty much forces you to do this anyway because often the
thing you are expressing is derivable in form from the long definition (the
"place structure", but NOT from the keyword alone, so you get used to 
reading and considering the long definition in actual usage.

Still, it would not be impossible if someone set up an appropriate file, to
modify the program to accept more than one answer for a given input.  We just chose not to try to do that in our data files.

>5) Another point to which I attache some importance is the archive language:
>I realized, while using LogFlash, that I had to do a double translation:
>from lojban to English (the language of the word archive) and from English
>to Italian (my own mother language). LogFlash is probably very useful for
>learning lojban, IF YOU ARE ANGLOPHONE; if not, well, you have to wait for a
>nationalized version, if it will ever come!


Clearly a problem with any language teaching program.  In the case of Lojban
we have a nearly ready data file with Spanish keywords, but it has not been
converted into a format compatible with the program, and may need some minor
modifications.  Someone could create an Italian/Lojban data file and it would
also work.  I suspectthat  pragmatically, since most people using the program can READ English, it is priomarily the keywords to be typed that have to be
translated into each language, but then this offers greater risk of the sort
E. G-E. warned of.  But we do have a keyowrd-only file for Esperanto, and 
hence then in theory could offer LogFlash for 3 languages, if there was a
market.

That is another limitation in all of this of course.  It takes an ENORMOUS 
amount of time to create the data files, and prepare them for uyse with the
program.  You have to be extremely motivated, and have lots of time to do 
it, and financial motivation is not likely to be enough for any conlang
effort.  LogFlash has perhaps generated as much as $1000 in income for LLG
over 9 years, and that hardly pays for the cost of distributing the program,
much less forthe development of program and lists.  When the work on the program
and list is easured in person-months or even person-years, you have problems
asking for a lot of functionality that is geared only for a small number of
users.

>With a general-purpose tool, whose archive is an editable text file, a
>nationalized version could be done in, I think, a week end.

Not for LogFlash.  YOu could perhaps do the keywords in a very intense weekend
if you did not read the definitions carefully.  But translating the list into
Spanish took Jorge Llambias spare time over many months, and he had a lot of
time to work on the language - more than most.  The raw data file for
LogFlash, BTW is around 350K bytes, about the same as a 125 page book.  This
is NOT a weekend of translation, I am sure you agree.

The MacIntosh version of the porgram, as described by Richard Kennaway,
has somewhat more limited functionality, biut it does take a raw text file.
The MS-DOS version  also takes a raw text file, but it has to be run through
a preprocessor that builds indexes and our internal instrumenttion files
(as I have described, we also have built LogFlash to allow us to do research
on how individuals learn language vocabulary through such data gathering).
That preprocessor is trivial though entirely undocumented, therefore we do
not have it posted.  We can convert anyones data file that is in the form
of the Lojban gismu list (gismu.lst or logdata.raw on our ftp site) into
a data file for the program.  needless to say, if you need to include
the diacritic charcaters, you need our undocumented codes for them to create
the data files.

>Someone said that LogFlash IS a general-purpose tools; this is not said in
>the version I have (downloaded from the official lojban www site,
>xiron.pc.helsinki.fi), nor there is any direction about how to customize its
>word archive.


As I said, we don't have current user documentation.  The program as
distributed IS a general purpose program, but we aren't marketing it as such.
It just has the capability built in if someone wants to build the files.
Since either documenting the general purpose nature or supporting people who
want to build versions for specific languages is not part of the Lojban
project, and we aren't making money from this, supporting such usages is
going to be on a negotiated basis between me (and my wife who wrote the
program) and whoever wants to use the version.  You can call this part of
the difference between unreggistered and registered Shareware %^).  But don't
expect general purpose instructions for LogFlash to be put up on the net
very soon.

>Seriously: I really want to stress it again: I DO NOT THINK OF LESSY AS AN
>EXCLUSIVE TOOL FOR LEARNING LANGUAGES FROM SCRATCH.
>
>Instead, it is tool for acquiring familiarity with the memory-dependent part
>of a language. If all you have of a language is some phonetics, some
>morphology and a list of 100 words, well, presumably, you do not need any
>tool for practising it; you probably do not need practicing it at all!
>
>For more developped languages, I stick to may idea: of course, a finished
>product, "good for teaching", takes long to create; but if you have a
>simple, maintainable, editable archive, you can achieve it step by step,
>while your requirements grow and your familiarity with that particular
>language increases.


This is basically how I haveuse LogFlash to learn Russian vocabulary.  But
I quickly found that it was more work to create the vocabulary lists than
to simply memorize them, so I stopped at around 200 words - not enough to
do anything with in the language.  If you build the word list as you go,
it simply proves not to be that useful to YOU the person who is creating
the list - for one thing, you then need a special program  (which we had for
an earlier version of LogFlash, but it might not work for the current
one) that can process additions to the list without resetting the status of
all the words that you have studied.  Otherwise youhave to start over as
soon as you ad a few dozen more words.  Simpler less sophsiticated algorithms
can deal with this problem better than LogFlash can - JCB's first version
of what is now LOgFlash - a program he calls MacTeach - has a simpler algorithm
and may allow people to add words on the fly.

>You may begin by importing some material from books, by traslating from
>reference works in other languages, by typing rough word lists, your lesson
>notes, etc. etc., each time adding something and your archive will grow with
>you.

Well, I said what the implemtation problem is with this.  But in addition,
with things like keyowrds and prompts, you find that later on you start
getting duplicate prompts and answers, and it gets rather difficult to
resolve.  The more you want to get the program away from simple word-word
memorizing to concept-concept mapping, the more detailed your data files 
have to be, AND the more they have to be designed and looked at as a whole.

For example, when you start developing your list, you might have the concept
"run", fairly straightforward in most languages.  But as anyone who has explored an English dictionary knows, "run" has  more than 100 subsidiary definitions,
and these might not go over into your new language using the same newlang
word.  So at the beginning you might just have put in "run", but later you
need to revise and discriminate.  But that entails relearning of whatever
you have memorized.  And if you are supporting a community of learners as we
are, you end up with configuaration management - getting new versions of the
list out to old users etc. so that everyone is talking the same language.
Because when you suddenly decide that the prompt for vofli will be "flight"
instead of "fly", nd the latter will instead prompt "sfani", then the two
word lists are incompatible.  (It isn;t hard to relearn this kind of thing
by the way, since you DO learn the words as concepts, but your fingers
acquire a "typing memory" when you are working at high speed, and you 
type the old word before you even realize that this was the word that you
changed lastr week.)  Again, all this can be dealt with by careful file
design, but the more you want to make it a concept-concept and not a word-word
learning thing, the more work has to go into that data file.  And a weekend
will simply not be enough.

The long term vision of LogFlash is not that unlike what you describe for
LESSY, and as I say, it already has the functionality and even the
multilingula support built in now.  But the porblem of designing the data
files 9as well as the economics of developing a set of data files without
knowing you will have an audience to justify the amount of work it requires)
is truly enormous.

lojbab