NetNews Usenet Archive 1993 #3/NN_1993_3.iso/spool/sci/lang/japan/5194

Text File | 1993-01-26 | 8.6 KB | 167 lines

Newsgroups: sci.lang.japan
Path: sparky!uunet!cs.utexas.edu!usc!howland.reston.ans.net!bogus.sura.net!udel!gatech!swrinde!emory!sol.ctr.columbia.edu!The-Star.honeywell.com!umn.edu!i9!molenda
From: [e-mail] (Jason Molenda)
Subject: Re: ``Kanji Education'' paper available by anonymous-ftp
Message-ID: <[e-mail]>
Sender: [e-mail] (Usenet News Administration)
Nntp-Posting-Host: i9.msi.umn.edu
Organization: University of Minnesota
References: <[e-mail]> <[e-mail]> <[e-mail]>
Date: Tue, 26 Jan 1993 05:39:16 GMT
Lines: 154

[e-mail] (Lars Huttar) writes:

>        As you said, the kind of writing done on Usenet is an "unusual
>mix of spoken and written Japanese."  As such, it may not well
>represent the mainstream of written Japanese.  It may be
>representative of the kind of writing in manga, but not in newspapers.
>(I don't know -- I'm just raising the question.)  I think we need to
>address what reading level, or style of document, the student is
>aiming at being able to read.  Then we can ask, how much study does
>one need to be able to read 95% of a newspaper?  a comic?  a novel?

Unfortunately, (as you guess later in the message), usenet was the only
available on-line source of Japanese that I know of.  In all honesty, I
did this paper for fun originally and it just happened that I wanted to
graduate so I used it as my senior project.  If I had had time to stay
at the University I had a much more interesting project I wanted to
do.  Anyway, since I did it for fun I wasn't about to go through the
pain and agony of typing in any other sources of Japanese.  A book
would be great but I'd probably spend a zillion years typing it all
in.

As I mention in the paper, there was one study about the frequency of
kanji in newspapers back in the early seventies which found similar
overall numbers.  I do not have the kanji list they came up with,
though, so I'm unable to compare these two lists.  I'll be living in
Japan in about two weeks, so maybe I'll have a chance to contact the
National Language Research Inst (?) and find that paper.  (I think the
full paper is like 90 pages so it probably includes a list of the
kanji found)

>This leads to another question: how efficient are the various written
>media in terms of reinforcing common kanji, boosting confidence,
>learning new kanji, learning spoken idioms, etc.?  These criteria are
>somewhat at odds.

Empirically, you will see the same kanji and jukugo used over and over
if you stick to a particular domain.  Headline news will use `oil' and
`economy' many, many times.  `zebra' (or whatever) is not going to
occur nearly as often.

Although I didn't include this in the paper, I found that when I
restricted the scan to a particular sub-group (e.g. fj.bikes), the
graph had the same basic curve but the 95th percentile fell to around
700 or so kanji with a total of maybe 1300 unique kanji (this is from
memory, I can dig up/regenerate the data if you're interested).

>        To facilitate research on these questions, are there other
>electronic sources of Japanese text, which are more representative of
>the mainstream, or at least biased differently?

I think the newspaper study is the best I can cite.  Someone sent me a
mail message (I've been deluged recently, sorry I can't find it right
now!) which noted a similar results in Beginning Technical Japanese for
a physics text.  I think I have a copy of BTJ around here somewhere if
someone would like me to look that up.

>        Looking at the appendix where the kanji [3396 unique kanji
>found in a Usenet sample] are listed, I notice that the first two are
>$BBg(B (`dai', big) and $B3X(B (`gaku', school).
 [...]
>boosted in these samples by the many names of universities and
>commercial research institutions in the writers' addresses.  So again
>we have to ask how accurate these orderings are...

to be truthful, I found dai and gaku being the first two pretty
suspicious myself.  I'm thinking about trying to toss .signatures
before doing the analysis.  It would still get all those "konnichi ha,
matsumoto@wherever-daigaku desu" at the beginning of messages but
shikata ga nai desho^.

>        To get some indication of the precision of the samples, I
>would like to compare the kanji-frequency orders gained from the two
>samples taken 6 months apart.

pretty easy to do.  I did that, in fact.  I think I noted that in my
paper but I might have edited it out.

In fact, after I did my last scan (late November), I changed my news
configuration and now I have about 100 days of fj news instead of 50
(as I used to have).  If you'd be interested in seeing the output of
another run, I'd be happy to re-run it (although I think most of
November would still be on disk) and send you the output.  It's really
rather painless to run and takes maybe 30-40 minutes to process the
~125MB of fj I have around.

>                               Do any kanji vary widely as to their
>place in the list?  (I suppose a statistician could come up with a
>nice metric on this.)  What if a student learned the first 1000
>characters in sample 1, and tried to read sample 2 -- what percentage
>would be readable?

This is an interesting idea.  I'm not sure about how I would go
computing this.  Maybe add up the absolute value of number of places
each character moved and divide it by the total number of characters.
After I get to Japan I'll have to find time to try this.  (anyone
have a better idea?  this is just off the top of my head).

the idea is that the really frequent characters will be pretty static
in their location.  As you get farther down the frequency list, the
probability (I would think) of a character moving is higher.  All
bets are off for the 1,000 (?) or so that occur only once.

>lists with ones already published.  The only one I know of off-hand is
>the one of the Joyo kanji in the back of Halpern's Shin-kanei-jiten.
>You probably know of others.  These lists are probably based on more

Hm.  I hadn't seen anything like this.  Please don't assume I know any
great deal about this -- I really only did it for fun.  I have never
used Halpern's kanei-jiten.  I didn't know it had any frequency stuff
in there.

I really think -- for the individual studying japanese -- more
important than focusing on any one particular list (as it is easy and
presuasive to do), it is important to realize the underlying theme:  In
each study I've seen, whether it's an unrestricted domain analysis of
fj, restricted to just fj.bikes (or any one particular newsgroup), or
newspaper stuff or even a physics book, all kanji lists showed the same
characteristics (which most of our teachers have been harping on all
along :-):  Pick something of interest and focus on it.  There may be
1,945 characters and a few thousand others that will turn up in
literature, but you won't need more than 700 or so to read a motorcycle
magazine.  For me, at least, that is enough.

For non-individuals, I think the most important thing, more important
any silly frequency lists, is to maintain some type of consistency in
whatever list is picked over a multi-year program.  This was lacking
at the University of Minnesota (although there will be some pretty
major changes here as of next year).  It takes more than a raw list
of kanji to teach students; a teacher needs some kind of comprehensive
program with sample texts, characters that will build into easily
taught jukugo, etc.  I am not an educator and know nothing about this.

>                                  Initially, however, I think it's
>important to learn to write some kanji well, and to write a lot of
>kanji at least once, so that the student can learn kanji elements, and
>thus be able to recognize them, and look them up, even when they don't
>appear in the neatest book font.

I agree.  I think there is an important base of maybe 200-400, but
after that maybe it isn't so important to remember that in `tsukau' (to
use), on the right side the vertical line starts above the top
horizontal line but in BENRI no BEN (nelson #451), it starts exactly on
the horizontal line.  (yes, I was marked off a point for this in
first year Japanese and it always pissed me off. :)

I should state a caveat:  I'm just another student of Japanese who
happens to also be a computer nerd.  I have no experience as an
educator and have only been studying Japanese for a little over three
years.  Although the numbers I have in my paper are pretty factual,
please take the discussion and recommendations and such with a grain of
salt.
--
Jason Molenda, University of Minnesota, Supercomputer Inst., Technical Support
SGI Iris Admin        [e-mail]       DoD #1867     '77 CB750F