WikiDiscuss

Posted by arj on Sat 12 of Feb, 2005 21:46 GMT posts: 953

On Sat, 12 Feb 2005, Robert LeChevalier wrote:

>> Based almost entirely on pacna.
>>
>> I don't know how to generate potential rafsi and evaluate them; I think
>> this word is of high utility, though. Could someone else take a look at
>> that issue?
>>
> No opinion at this point on the proposal. If a rafsi is desired, then that
> should be factored into the gismu algorithm word selection. I don't know how
> and when Arnt and Cowan generated words, but if the original program was
> used, there would be a list of multiple candidates. What are the sources
> from the 6 languages?

John implemented re-implemented the gismu scoring algorithm in Perl, from
the description in the Red Book.

My part in this is that I re-ran his program and got a different result.
It turned out that his script for generating candidate gismu forms didn't
generate all possibilities, while my is vastly overgenerating — but that
doesn't matter, so long as the winner has a legal form.

The sources from the 6 languages were:
Chinese: uan
Hindi: rakan
English: ekspekt
Spanish: esper
Russian: predpologa
Arabic: mulud (no effect on scoring)

> I'll try to run the program on it, and see what comes up (note that the
> current weightings are slightly different from those of 1986, and running
> with different weightings also gives different answers).

Please do. John used the most recent weighthings he could get, viz.:
zh: 0.347
hi: 0.196
en: 0.160
es: 0.123
ru: 0.089
ar: 0.085

--
Arnt Richard Johansen http://arj.nvg.org/
Att handla en kvarting p Systemet kan vara oerhrt sofistikerad p ett
digitalt plan. — Z mag@zine