LISTSERV mailing list manager LISTSERV 16.5

Help for CONLANG Archives


CONLANG Archives

CONLANG Archives


CONLANG@LISTSERV.BROWN.EDU


View:

Message:

[

First

|

Previous

|

Next

|

Last

]

By Topic:

[

First

|

Previous

|

Next

|

Last

]

By Author:

[

First

|

Previous

|

Next

|

Last

]

Font:

Proportional Font

LISTSERV Archives

LISTSERV Archives

CONLANG Home

CONLANG Home

CONLANG  August 2009, Week 2

CONLANG August 2009, Week 2

Subject:

Re: Engelang with unnatural statistical properties?

From:

Jörg Rhiemeier <[log in to unmask]>

Reply-To:

Constructed Languages List <[log in to unmask]>

Date:

Thu, 13 Aug 2009 22:47:41 +0200

Content-Type:

text/plain

Parts/Attachments:

Parts/Attachments

text/plain (105 lines)

Hallo!

On Thu, 13 Aug 2009 12:20:54 -0400, Jim Henry wrote:

> On Thu, Aug 13, 2009 at 12:07 PM, Jörg Rhiemeier<[log in to unmask]> 
wrote:
> > On Thu, 13 Aug 2009 03:58:52 -0400, Jim Henry wrote:
> 
> > The most random imaginable language would be a "free monoid", i.e. a
> > language where every possible string of symbols would be a grammatical
> > utterance, but I doubt that something like that is possible.  As soon
> > as you introduce grammar, you get a set of grammatical as opposed to
> > ungrammatical strings, and thus some degree of redundancy.
> 
> Don't some grammars allow a higher proportion of possible strings to
> be grammatical than others, though?  I'm fairly sure that's so.

You are of course right.  The more redundant the language is, the
smaller is the ratio between grammatical strings of length N and
ungrammatical strings of the same length.  And surely, languages with
lower redundancy than natlangs could be designed, but I don't know
how much can actually be gained, and how.

>       It 
> seems that a language with relatively free word order, especially if
> case-marking and adpositions are sometimes in free variation, would
> have more grammatical strings of N characters than a language with
> similar phonology and lexicon that had stricter word order.  But maybe
> that's not the case if you look at the phoneme or morpheme level
> rather than the word level...?

Maybe not.  Those grammars which allow for more freedom in word order
usually introduce other recurrent elements such as case markers which
more rigidly ordering languages get along without.  It is a trade-off
of some sort.
 
> >> Probably you could approximate that property by making all the
> >> phonemes that are rarest in the lexicon occur mostly in very common
> >> words, so that the weighted frequency of phonemes cancels out and all
> >> phonemes are of roughly equal frequency in running text.
> >
> > Now that is an interesting idea that would indeed make the "code"
> > harder to "crack".
> 
> I suppose you could apply this at higher levels too, making sure that
> all digraphs or even all trigraphs are about equally frequent in
> running text.  But at whatever level you don't take precautions, order
> will sneak back in.

Indeed.  This looks like a *very* hairy endeavour.

>       The thing is to make the language look random at 
> enough levels that possible decipherers give up and decide the text is
> meaningless before they get to the level where some strings are
> significantly more common than others.

I don't know whether that can be attained in any practical way.
As I said before, if you have grammar, you also have redundancy.
It's like those bumps in the carpet that won't ever disappear,
only move into another part of the room when try to smoothen them
out.  But I am not sure about this reservation; it would be useful
to try it out.

> > only a short step to Plan B).  Plan B also uses a kind of Huffman
> > coding (the bit sequence at the beginning of each morpheme encodes
> > the morpheme length), which could be helpful.
> 
> It's helpful for Plan B and yours and Ray's engelangs inspired by Plan
> B as a means of self-segregation.

Yes.

>       I'm not sure it would help with
> this project's main goal, though; at least I can't see how.  The
> phonemes that indicate the start of a short word would tend to occur
> more often at the start of a sentence or paragraph than the phonemes
> that indicate the start of a long word, and that's the kind of thing
> we don't want here.

Yes, that's a problem with it.  You need more possible beginnings for
the shorter morphemes than for the longer ones in order to even it
out.

> > Human beings are indeed abysmally poor random number generators.
> > The allomorph approach would thus probably fail; you'd end up with
> > people prefering particular allomorphs, which is not at all what was
> > intended.
> 
> Free variation of allomorphs at the speaker's whim wouldn't have as
> much effect in flattening the statistical properties of the language
> as free variation using a random-number generator