official cmavo form

250 views
Skip to first unread message

guskant

unread,
Oct 19, 2014, 8:19:33 PM10/19/14
to bpfk...@googlegroups.com
coi ro me byfy

peg morphology allows cmavo beginning with "consonant glide":

http://www.lojban.org/tiki/BPFK+Section%3A+PEG+Morphology+Algorithm
cmavo-form <- !h !cluster onset (nucleus h)* (!stressed nucleus / nucleus !cluster) / y+ / digit
onset <-  h / consonant? glide / initial

while CLL3.4 disallows it:
"the ten following ones [diphtongs with on-glide, i.e. beginning with i or u] are used only as stand-alone words and in Lojbanized names and borrowings."

Which is official now? La jbovlaste allows it based on peg morphology, and la gleki and I don't agree to the new rule. Concretely, an experimental cmavo "jie'e'e" is now discussed:
http://jbovlaste.lojban.org/comments.html?valsi=27527;natlangword=0;commentid=0;definition=0

If this new rule is official, CLL3.4 must be modified.

mi'e la guskant mu'o

Jorge Llambías

unread,
Oct 19, 2014, 9:26:28 PM10/19/14
to bpfk...@googlegroups.com
On Sun, Oct 19, 2014 at 9:19 PM, guskant <gusni...@gmail.com> wrote:
coi ro me byfy

peg morphology allows cmavo beginning with "consonant glide":

http://www.lojban.org/tiki/BPFK+Section%3A+PEG+Morphology+Algorithm
cmavo-form <- !h !cluster onset (nucleus h)* (!stressed nucleus / nucleus !cluster) / y+ / digit
onset <-  h / consonant? glide / initial

while CLL3.4 disallows it:
"the ten following ones [diphtongs with on-glide, i.e. beginning with i or u] are used only as stand-alone words and in Lojbanized names and borrowings."

It's not very clear that it disallows them, it just says that they are not used. I don't think for example that the form "ie'a" should be disallowed as a cmavo form, even though it uses "ie" not as a stand-alone word.  I wouldn't mind disallowing "consonant glide" in cmavo if they were also disallowed in zi'evla. If they were disallowed in cmavo but not in zi'evla, would that make "jiebroda" a valid zi'evla?
 
Which is official now?

CLL, even though it leaves some points of the morphology unclear.
 
La jbovlaste allows it based on peg morphology, and la gleki and I don't agree to the new rule. Concretely, an experimental cmavo "jie'e'e" is now discussed:
http://jbovlaste.lojban.org/comments.html?valsi=27527;natlangword=0;commentid=0;definition=0
 
If this new rule is official, CLL3.4 must be modified.

The PEG morphology is not yet official. I would have been happy to disallow "consonant glide" as a valid onset altogether, but other people were opposed to that at the time. Some would have even allowed "CC glide" as a syllable onset, so allowing a single consonant was something of a compromise. I would be happy to go with the more strict "onset <-  h / glide / initial" if that's the new consensus, although in that case there are probably several words in jbovlaste that would have to be revised.

mu'o mi'e xorxes

mukti

unread,
Oct 19, 2014, 11:25:22 PM10/19/14
to bpfk...@googlegroups.com
On Sunday, October 19, 2014 9:19:33 PM UTC-3, guskant wrote:
La jbovlaste allows it based on peg morphology, and la gleki and I don't agree to the new rule. Concretely, an experimental cmavo "jie'e'e" is now discussed:
http://jbovlaste.lojban.org/comments.html?valsi=27527;natlangword=0;commentid=0;definition=0

A note to clarify why, when and how jbovlaste came to use the camxes BPFK morphology.

In March a bug was filed against jbovlaste, citing problems with words that were being rejected on the basis of not being approved by "vlatai" -- a tool which is part of "jbofihe", and which jbovlaste previously used to validate words. Since jbofihe/vlatai was not maintained, in April I posted a message to the lojban list proposing to replace it with camxes, and asking for feedback. I ran all of the words in the database through camxes, and only a small number of cmevla and fu'ivla were classified differently by camxes than they had been by jbofihe/vlatai. I published a list of those words for confirmation.

In June I posted a follow-up, announcing that jbovlaste had been updated to use a python implementation of camxes, and detailing the attendant reclassifications: Out of 21,940 words, only about 100 were affected by the change in the morphological verifier. Most of the effected words were non-conforming fu'ivla.

All of the implementations of the implementations of camxes distributed at that time, including the the one added to jbovlaste, were using an older version of the camxes BPFK morphology: Probably version 108, from November 2005. The most recent version of the morphology had been completed in June 2008. 

The fact that an older morphology was being used came up in discussions of {relmast}, which had been permitted by jbofihe/vlatai, but was forbidden by camxes. After it became apparent that camxes was not using the latest BPFK morphology, I updated camxes/vlatai to do so. 20 cmevla that were previously disallowed were reinstated.

Since that time, the various implementations of camxes have been updated to use the 2008 camxes BPFK morphology, including Ilmen's camxes.js implementation, which powers the camxes bot in IRC, and python-camxes, which is actually a wrapper around Robin's original Java implementation, and which is used by camxes.lojban.org.

mi'e la mukti mu'o

Gleki Arxokuna

unread,
Oct 20, 2014, 1:54:25 AM10/20/14
to bpfk...@googlegroups.com
I suggest disallowing at least /^C[iu]V$/ cmavo like {kia} for some time until we stabilize the language on the new level which for the most part is 
1. a new printed CLL with errata applied
2. a printed dictionary of cmavo, upper ontology + most useful in real life words. The last two would probably consist mostly of gismu.


--
You received this message because you are subscribed to the Google Groups "BPFK" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bpfk-list+...@googlegroups.com.
To post to this group, send email to bpfk...@googlegroups.com.
Visit this group at http://groups.google.com/group/bpfk-list.
For more options, visit https://groups.google.com/d/optout.

John Cowan

unread,
Oct 20, 2014, 2:50:23 AM10/20/14
to bpfk...@googlegroups.com
Gleki Arxokuna scripsit:

> I suggest disallowing at least /^C[iu]V$/ cmavo like {kia} for some time

I suggest disallowing them forever.

--
John Cowan http://www.ccil.org/~cowan co...@ccil.org
Historians aren't constantly confronted with people who carry on
self-confidently about the rule against adultery in the sixth amendment to
the Declamation of Independence, as written by Benjamin Hamilton. Computer
scientists aren't always having to correct people who make bold assertions
about the value of Objectivist Programming, as examplified in the HCNL
entities stored in Relaxational Databases. --Mark Liberman

guskant

unread,
Oct 20, 2014, 3:05:43 AM10/20/14
to bpfk...@googlegroups.com


Le lundi 20 octobre 2014 15:50:23 UTC+9, John Cowan a écrit :
Gleki Arxokuna scripsit:

> I suggest disallowing at least /^C[iu]V$/ cmavo like {kia} for some time

I suggest disallowing them forever.


I agree.

I should point out that a fu'ivla creator should think of "jiebroda test" for fu'ivla morphology if "consonant glide" is allowed for cmavo. If both cmavo and fu'ivla may include "consonant glide", "jiebroda" is not a fu'ivla (or zi'evla) but two words "jie broda". In order to make it fu'ivla form, we should modify it, for example, to "djiebroda" or "ji'ezbroda".

Actually, vlatai of jbofi'e says "jie" is not a lojban word, "jiebroda", "djiebroda" and "ji'ezbroda" are fu'ivla (stage-4), while camxes says "jie" is cmavo, "jiebroda" is two words, "djiebroda" is not a lojban word, and "ji'ezbroda" is fu'ivla.

Whether cmavo with "consonant glide" is allowed or not is very important for defining Lojban phonology. I, who have been believed that "consonant glide" is basically a kind of non-lojban sound, I have not been much careful in listening and speaking "je". If "jie" is allowed as cmavo, I should change my attitude, and be careful whether the pronunciation is "je" or "jie". Please, BPFK-members, make clear the decision on this matter.

Gleki Arxokuna

unread,
Oct 20, 2014, 3:17:58 AM10/20/14
to bpfk...@googlegroups.com
Actually I just don't want them because in future I want to be able to pronounce {tu'a} as {tua}.

Alex Burka

unread,
Oct 20, 2014, 4:01:50 AM10/20/14
to bpfk...@googlegroups.com
I like {tu'a} just how it is of course, but I would agree with disabling C[iu]V. I agree with guskant that it's too similar to CV.

Gleki Arxokuna

unread,
Oct 20, 2014, 4:22:25 AM10/20/14
to bpfk...@googlegroups.com
2014-10-20 12:01 GMT+04:00 Alex Burka <dur...@gmail.com>:
I like {tu'a} just how it is of course, but I would agree with disabling C[iu]V. I agree with guskant that it's too similar to CV.
I didn't say I wanted {tu'a} forbidden. I just want a new alternative accent for a limited number of dipthongs

selpa'i

unread,
Oct 20, 2014, 8:23:27 AM10/20/14
to bpfk...@googlegroups.com
la .xorxes. cu cusku di'e
> The PEG morphology is not yet official. I would have been happy to
> disallow "consonant glide" as a valid onset altogether, but other people
> were opposed to that at the time. Some would have even allowed "CC
> glide" as a syllable onset, so allowing a single consonant was something
> of a compromise. I would be happy to go with the more strict "onset <-
> h / glide / initial" if that's the new consensus, although in that case
> there are probably several words in jbovlaste that would have to be revised.

My personal position is that I definitely would want to disallow any CiV
where C is a sibilant (e.g. ?{sia}) or a dental (e.g. ?{tia}) [1],
because they tend to degenerate into simpler forms over time (e.g. {ca}
and {tca} respectively) and are hard to distiniguish for many people. I
would ban them in any word, not just in cmavo.

Regarding forms like {kia} (which must be pronounced [kja]), they bother
me slightly less, but I wouldn't miss them if they got removed.

So one could either ban just the ones that are likely to cause problems,
or ban all for simplicity's sake.

I see no problems with {ie'o} as a cmavo form.

mi'e la selpa'i mu'o

--
[1] The are some additional phonotactic constraints I would install, but
the details seem irrelevant here.

And Rosta

unread,
Oct 20, 2014, 9:02:05 AM10/20/14
to bpfk...@googlegroups.com
selpa'i, On 20/10/2014 13:23:
> la .xorxes. cu cusku di'e
>> The PEG morphology is not yet official. I would have been happy to
>> disallow "consonant glide" as a valid onset altogether, but other people
>> were opposed to that at the time. Some would have even allowed "CC
>> glide" as a syllable onset, so allowing a single consonant was something
>> of a compromise. I would be happy to go with the more strict "onset <-
>> h / glide / initial" if that's the new consensus, although in that case
>> there are probably several words in jbovlaste that would have to be revised.
>
> My personal position is that I definitely would want to disallow any CiV where C is a sibilant (e.g. ?{sia}) or a dental (e.g. ?{tia}) [1], because they tend to degenerate into simpler forms over time (e.g. {ca} and {tca} respectively) and are hard to distiniguish for many people. I would ban them in any word, not just in cmavo.
>
> Regarding forms like {kia} (which must be pronounced [kja]), they bother me slightly less, but I wouldn't miss them if they got removed.

Why must Lojban have onsets and syllabification? /Cia/ is problematic only if /Ci-/ must constitute an onset.

> So one could either ban just the ones that are likely to cause
> problems, or ban all for simplicity's sake.

I'd ban none.

> I see no problems with {ie'o} as a cmavo form.

How about {a'ua}?

> [1] The are some additional phonotactic constraints I would install, but the details seem irrelevant here.

I think Lojban already has way more phonotactic constraints than is necessary...

--And.


John Cowan

unread,
Oct 20, 2014, 9:59:58 AM10/20/14
to bpfk...@googlegroups.com
selpa'i scripsit:

> My personal position is that I definitely would want to disallow any
> CiV where C is a sibilant (e.g. ?{sia}) or a dental (e.g. ?{tia})
> [1], because they tend to degenerate into simpler forms over time
> (e.g. {ca} and {tca} respectively) and are hard to distiniguish for
> many people. I would ban them in any word, not just in cmavo.

That is exactly why Lojban Central banned them by construction back in
1988 (they existed and still exist in Loglan).
Nobody expects the RESTifarian Inquisition! Our chief weapon is
surprise ... surprise and tedium ... tedium and surprise ....
Our two weapons are tedium and surprise ... and ruthless disregard
for unpleasant facts.... Our three weapons are tedium, surprise, and
ruthless disregard ... and an almost fanatical devotion to Roy Fielding....

John Cowan

unread,
Oct 20, 2014, 10:13:18 AM10/20/14
to bpfk...@googlegroups.com
And Rosta scripsit:

> Why must Lojban have onsets and syllabification? /Cia/ is problematic
> only if /Ci-/ must constitute an onset.

It's settled that "ia" is /ja/.

> >I see no problems with {ie'o} as a cmavo form.

I agree.

> How about {a'ua}?

By avoiding things like this, we keep Lojban /h/ safely ambisyllabic.
Allowing /ahwa/ means we have to choose between coda /h/ (hard for
anglophones and many others) and onset /hw/ (hard for most anglophones,
who no longer have a /w/ ~ /W/ distinction).

We have no shortage of possible cmavo. Let's not go toward hC or Ch
clusters (where C = any consonant including /j/ and /w/).

> I think Lojban already has way more phonotactic constraints than is
> necessary...

I wish we had more, but our lujvo-making machinery prevents some that
would be really useful, like not allowing both "denbro" and "dembro".
What asininity could I have uttered that they applaud me thus?
--Phocion, Greek orator

And Rosta

unread,
Oct 20, 2014, 11:16:48 AM10/20/14
to bpfk...@googlegroups.com
John Cowan, On 20/10/2014 15:13:
> And Rosta scripsit:
>
>> Why must Lojban have onsets and syllabification? /Cia/ is problematic
>> only if /Ci-/ must constitute an onset.
>
> It's settled that "ia" is /ja/.

Does /j/ (corresponding to orthographic <i> rather than <j>) actually exist? Are minimal pairs possible with /i/:/j/?

I'd propose that "ia" should be /ia/.

>> How about {a'ua}?
>
> By avoiding things like this, we keep Lojban /h/ safely ambisyllabic.
> Allowing /ahwa/ means we have to choose between coda /h/ (hard for
> anglophones and many others) and onset /hw/ (hard for most anglophones,
> who no longer have a /w/ ~ /W/ distinction).

These problems arise when {a'ua} is analysed as something other than /a'ua/. If Lojban has no /u/:/w/ contrast -- as the impossibility of minimal pairs would show -- then /ahwa/ and /awha/ are not possible analyses. I think syllabification is likely an unnecessary complication, but I don't see why /'/ in /a'ua/ couldn't be ambisyllabic.

> We have no shortage of possible cmavo. Let's not go toward hC or Ch
> clusters (where C = any consonant including /j/ and /w/).

The simplest phonological analysis of Lojban is one in which there are no clusters at all. The only phonotactic rules necessary are that a C many be adjacent only to a V, a V may be adjacent only to a C or a glide V of a type other than its own, and /%/ (or however we symbolize the buffer vowel) may be adjacent only to a C.

I realize the current rules are much more complex. I'm just noting that with no detriment they could be simplified to what I've set out. (I would actually advocate slightly more restrictions on VV sequences, tho.)

>> I think Lojban already has way more phonotactic constraints than is
>> necessary...
>
> I wish we had more, but our lujvo-making machinery prevents some that
> would be really useful, like not allowing both "denbro" and "dembro".

Those are problematic only when there has been excessive syncope of /%/.

--And.

Gleki Arxokuna

unread,
Oct 20, 2014, 11:53:46 AM10/20/14
to bpfk...@googlegroups.com
2014-10-20 18:13 GMT+04:00 John Cowan <co...@mercury.ccil.org>:
And Rosta scripsit:

> Why must Lojban have onsets and syllabification? /Cia/ is problematic
> only if /Ci-/ must constitute an onset.

It's settled that "ia" is /ja/.

> >I see no problems with {ie'o} as a cmavo form.

I agree.

> How about {a'ua}?

By avoiding things like this, we keep Lojban /h/ safely ambisyllabic.
Allowing /ahwa/ means we have to choose between coda /h/ (hard for
anglophones and many others) and onset /hw/ (hard for most anglophones,
who no longer have a /w/ ~ /W/ distinction).

We have no shortage of possible cmavo.  Let's not go toward hC or Ch
clusters (where C = any consonant including /j/ and /w/).

wut? no way, {'} is a vowel separator.
If you want new phonemes so much add click sounds, add [D] and [G].


> I think Lojban already has way more phonotactic constraints than is
> necessary...

I wish we had more, but our lujvo-making machinery prevents some that
would be really useful, like not allowing both "denbro" and "dembro".

--
John Cowan          http://www.ccil.org/~cowan        co...@ccil.org
What asininity could I have uttered that they applaud me thus?
        --Phocion, Greek orator

John Cowan

unread,
Oct 20, 2014, 3:42:48 PM10/20/14
to bpfk...@googlegroups.com
And Rosta scripsit:

> >It's settled that "ia" is /ja/.
>
> Does /j/ (corresponding to orthographic <i> rather than <j>) actually
> exist? Are minimal pairs possible with /i/:/j/?

I should have written [ja]. And no, there is no separate /j/ phoneme,
and I hope there will be none. An alternate analysis is possible by
which there is /j/ and /w/ but not /h/; that is, "kai" is /kaj/ and
"ka'i" is /kai/ with epenthetic [h]. But I see no particular merit in
this analysis.

> These problems arise when {a'ua} is analysed as something other than
> /a'ua/. If Lojban has no /u/:/w/ contrast -- as the impossibility of
> minimal pairs would show -- then /ahwa/ and /awha/ are not possible
> analyses. I think syllabification is likely an unnecessary complication,
> but I don't see why /'/ in /a'ua/ couldn't be ambisyllabic.

Again I should have written [ahwa] with square brackets. My concern
is that this form would decay to [aWa] and then be merged with [awa].

> The simplest phonological analysis of Lojban is one in which there are
> no clusters at all. The only phonotactic rules necessary are that a
> C many be adjacent only to a V, a V may be adjacent only to a C or a
> glide V of a type other than its own, and /%/ (or however we symbolize
> the buffer vowel) may be adjacent only to a C.

That would still exclude "a'ua", because there is no valid place to insert a %.
You let them out again, Old Man Willow!
What you be a-thinking of? You should not be waking!
Eat earth! Dig deep! Drink water! Go to sleep!
Bombadil is talking.

Jorge Llambías

unread,
Oct 20, 2014, 5:43:08 PM10/20/14
to bpfk...@googlegroups.com
On Mon, Oct 20, 2014 at 10:02 AM, And Rosta <and....@gmail.com> wrote:

Why must Lojban have onsets and syllabification? /Cia/ is problematic only if /Ci-/ must constitute an onset.

There's no special need to realize /Cia/ as one syllable, but it must be counted as one syllable for penultimate stress rule purposes. Formulating the rule without mentioning syllables would be more complicated. It also has to be distinguished from "ciia"

I see no problems with {ie'o} as a cmavo form.

How about {a'ua}?

"a'ua" is currently not allowed by camxes, but "a'uua" is (it's two words)

Jorge Llambías

unread,
Oct 20, 2014, 6:24:43 PM10/20/14
to bpfk...@googlegroups.com
On Mon, Oct 20, 2014 at 4:05 AM, guskant <gusni...@gmail.com> wrote:

Whether cmavo with "consonant glide" is allowed or not is very important for defining Lojban phonology. I, who have been believed that "consonant glide" is basically a kind of non-lojban sound, I have not been much careful in listening and speaking "je". If "jie" is allowed as cmavo, I should change my attitude, and be careful whether the pronunciation is "je" or "jie". Please, BPFK-members, make clear the decision on this matter.

The question shouldn't be about cmavo. If CiV/CuV are disallowed it should be for all words, since distinguishing prije from prijie is as hard or as easy as distinguishing je from jie. I currently find only three experimental cmavo in jbovlaste containing CiV/CuV, but there are lots of fu'ivla, so the proposal to disallow them should contemplate what is to be done with all those fu'ivla (change them to something else? leave them in as archaic words?) It would also be useful to know of how many words we are talking about.

And Rosta

unread,
Oct 20, 2014, 7:32:21 PM10/20/14
to bpfk...@googlegroups.com


On 20 Oct 2014 22:43, "Jorge Llambías" <jjlla...@gmail.com> wrote:
>
>
>
> On Mon, Oct 20, 2014 at 10:02 AM, And Rosta <and....@gmail.com> wrote:
>>
>>
>> Why must Lojban have onsets and syllabification? /Cia/ is problematic only if /Ci-/ must constitute an onset.
>
>
> There's no special need to realize /Cia/ as one syllable, but it must be counted as one syllable for penultimate stress rule purposes.

How come? Is it simply that because word-segmentation is sensitive to stress, and stress is sensitive to syllabicity, varisyllabicity risks undermining the consistency of the word-segmentation rules? If so, the stress or word-segmentation rules can be reformulated so that VV counts as a single metrical unit rather than a sequence of two.

> Formulating the rule without mentioning syllables would be more complicated.

It's not yet clear to me that that is so, but that could be because I don't know the rule.

> It also has to be distinguished from "ciia"

How come? Is "ciia" licit? Is it licit even if "cia" isn't?

>
>>> I see no problems with {ie'o} as a cmavo form.
>>
>>
>> How about {a'ua}?
>
>
> "a'ua" is currently not allowed by camxes, but "a'uua" is (it's two words)

Ah. So the two-word version needn't be "a'u.ua"? I had been thinking that all words must begin with a consonant.

I would allow "a'ua" (assuming I was not allowed to kill /'/), disallow "uu", and have all words begin with a consonant, but if "a'uua" must be a variant of "a'u.ua" then I see why "a'ua" must be forbidden.

--And.

And Rosta

unread,
Oct 20, 2014, 7:56:15 PM10/20/14
to bpfk...@googlegroups.com


On 20 Oct 2014 20:42, "John Cowan" <co...@mercury.ccil.org> wrote:
>
> And Rosta scripsit:
>
> > >It's settled that "ia" is /ja/.
> >
> > Does /j/ (corresponding to orthographic <i> rather than <j>) actually
> > exist? Are minimal pairs possible with /i/:/j/?
>
> I should have written [ja]. 

But must Lojban so specify the duration of the /i/ in /ia/? That seems unnecessarily pernickety, given the quite proper laxity of all other realization rules in Lojban. If, rather, what is settled is that /i/ in /ia/ is an onset, then I am asking why that must be so.

> And no, there is no separate /j/ phoneme,
> and I hope there will be none.  An alternate analysis is possible by
> which there is /j/ and /w/ but not /h/; that is, "kai" is /kaj/ and
> "ka'i" is /kai/ with epenthetic [h].  But I see no particular merit in
> this analysis.

A merit would be getting rid of /'/ as a phoneme -- a good outcome given its anomalousness (not counting as a consonant for morphological purposes). The illicitness of /w%w/ could be accounted for by a rule that /%/ can't be adjacent to a vowel and that /w/ is both a consonant and a vowel.

>
> > These problems arise when {a'ua} is analysed as something other than
> > /a'ua/. If Lojban has no /u/:/w/ contrast -- as the impossibility of
> > minimal pairs would show -- then /ahwa/ and /awha/ are not possible
> > analyses. I think syllabification is likely an unnecessary complication,
> > but I don't see why /'/ in /a'ua/ couldn't be ambisyllabic.
>
> Again I should have written [ahwa] with square brackets.  My concern
> is that this form would decay to [aWa] and then be merged with [awa].

But the contrast-preserving Lojbanist should instead say [ahua], or, more realistically, [aWua] or [axua]. Actually, the contrast-preserving Lojbanist should of course say [aTua], since it is well-established that [h] is not a reliably contrastive realization of /'/.

>
> > The simplest phonological analysis of Lojban is one in which there are
> > no clusters at all. The only phonotactic rules necessary are that a
> > C many be adjacent only to a V, a V may be adjacent only to a C or a
> > glide V of a type other than its own, and /%/ (or however we symbolize
> > the buffer vowel) may be adjacent only to a C.
>
> That would still exclude "a'ua", because there is no valid place to insert a %.

I was assuming that under these phonological rules, /'/ would be a (morphologically irregular) consonant and /u/ a vowel, with nothing counting as both vowel and consonant.

--And.

Jorge Llambías

unread,
Oct 20, 2014, 7:56:25 PM10/20/14
to bpfk...@googlegroups.com
On Mon, Oct 20, 2014 at 8:32 PM, And Rosta <and....@gmail.com> wrote:

On 20 Oct 2014 22:43, "Jorge Llambías" <jjlla...@gmail.com> wrote:
>
> There's no special need to realize /Cia/ as one syllable, but it must be counted as one syllable for penultimate stress rule purposes.

How come? Is it simply that because word-segmentation is sensitive to stress, and stress is sensitive to syllabicity, varisyllabicity risks undermining the consistency of the word-segmentation rules? If so, the stress or word-segmentation rules can be reformulated so that VV counts as a single metrical unit rather than a sequence of two.

Yes. 

> Formulating the rule without mentioning syllables would be more complicated.

It's not yet clear to me that that is so, but that could be because I don't know the rule.


Maybe not terribly more complicated, but at least somewhat more complicated. y-syllables are already exceptions to penultimate stress, as are consonantal syllables and buffer-vowel syllables, so this would be just one more exception.
 

> It also has to be distinguished from "ciia"

How come? Is "ciia" licit? Is it licit even if "cia" isn't?

Yes, there's no constraint against vowel i being followed by glide i.

> "a'ua" is currently not allowed by camxes, but "a'uua" is (it's two words)

Ah. So the two-word version needn't be "a'u.ua"? I had been thinking that all words must begin with a consonant.


Since it causes no ambiguity, camxes doesn't force a glottal stop/pause before a word that begins with a glide. But this need not happen at a word boundary: "braka'uua" is a valid fu'ivla.
 

I would allow "a'ua" (assuming I was not allowed to kill /'/), disallow "uu", and have all words begin with a consonant, but if "a'uua" must be a variant of "a'u.ua" then I see why "a'ua" must be forbidden.


Yes, it should be one or the other, but not both.

John Cowan

unread,
Oct 20, 2014, 8:03:51 PM10/20/14
to bpfk...@googlegroups.com
And Rosta scripsit:

> Ah. So the two-word version needn't be "a'u.ua"? I had been thinking that
> all words must begin with a consonant.

Normatively they must, but the concern is that such glottal stops will
be unstable, since they are not (in the absence of "a'ua"-style words)
necessary for word separation. In any case, I was talking about "a ua"
[a?wa] as hard to distinguish from "a'ua" [ahwa], both tending to become
simple [awa].
When I wrote it I was more than a little febrile with foodpoisoning
from an antique carrot that I foolishly ate out of an illjudged faith
in the benignancy of vegetables. --And Rosta

John Cowan

unread,
Oct 20, 2014, 8:26:11 PM10/20/14
to bpfk...@googlegroups.com
And Rosta scripsit:

> But must Lojban so specify the duration of the /i/ in /ia/? That seems
> unnecessarily pernickety, given the quite proper laxity of all other
> realization rules in Lojban. If, rather, what is settled is that /i/ in
> /ia/ is an onset, then I am asking why that must be so.

Historically, I think it rose out of the introduction of ' to simplify the
Loglan pronunciation rules. In Loglan there is no ', and therefore only
25 VV sequences rather than Lojban's 29 (ignoring "y" in both languages).
These 25 break into four groups according to pronunciation:

ai ei oi ao are always falling diphthongs (in practice, ao is pronounced
[aw] like Lojban au).

ae au ea eo eu oa oe ou are always two syllables; when e is the first,
it is allophonically [e] rather than [E].

aa ee oo are always two syllables *and* one of the syllables is required
to bear the stress (so words like "baarsoa" are invalid, unlike the
Lojban analogue "ba'ars'oa"); ee oo are very rare.

iV and uV may be pronounced either as rising diphthongs, as in Lojban, or
as two syllables: therefore "stomia" may be either ['stomja] or [sto'mia].

> But the contrast-preserving Lojbanist should instead say [ahua],

Easily confused with "a'u ua" [ahu?wa] > [ahuwa].

> or, more realistically, [aWua] or [axua].

The latter is "a xua"; [x] cannot be used as an allophonic fortition of [h].
Not to perambulate the corridors during the hours of repose
in the boots of ascension. --Sign in Austrian ski-resort hotel

And Rosta

unread,
Oct 21, 2014, 4:52:42 AM10/21/14
to bpfk...@googlegroups.com
John Cowan, On 21/10/2014 01:03:
> And Rosta scripsit:
>
>> Ah. So the two-word version needn't be "a'u.ua"? I had been thinking that
>> all words must begin with a consonant.
>
> Normatively they must, but the concern is that such glottal stops will
> be unstable, since they are not (in the absence of "a'ua"-style words)
> necessary for word separation.

Is the concern that because /./ is elidable when its presence is not morphologically contrastive, the risk is that through habit it would end up being elided even when it is? There are various solutions to that. One, as with Dotside, is to not elide it at all. Another is to elide it willynilly, and worry about distinguishing the unelided and elided versions only when the addressee might misunderstand; though, a problem with that is that it is hard to carefully and deliberately show that one is using a properly /./-less form.

> In any case, I was talking about "a ua" [a?wa] as hard to distinguish
> from "a'ua" [ahwa], both tending to become simple [awa].

Specifically for L1 English speakers, you must mean, rather than for people in general. Does it really make sense to base the rules of Lojban on the specific needs of L1 English speakers?

Many L1 English speakers would tend to hear /a.ua/ as /at ua/.

If /'/ is to be kept distinct from /x/, /'/ must be [T], giving [aTua] for /a'ua/, which is unlikely to become [awa].

>> or, more realistically, [aWua] or [axua].
>
> The latter is "a xua"; [x] cannot be used as an allophonic fortition of [h].

It would be an assimilation rather than a fortition. As I've said before, [h] is articulatorily impossible as a realization of /'/ in some environments, e.g. /i'i/, at ordinary speech rates, and the /'/:/x/ is not robustly sustainable. Under the current rules, /./ must be [T] (contrastive voiceless continuant). OTOH, usage has, I think, hitherto enshrined a posterior voiceless fricative as the realization of /'/, so overlapping with /x/. With rule changes, /'/ could be /G/, or got rid of entirely, which would be my preference, tho it would eliminate Lojban's most distinctive (if egregiously noisome) feature; to eliminate it would -- for many -- be like a well-loved friend having cosmetic rhinoplasty -- they might emerge more beautiful, but not with the face one has loved so dearly.

John Cowan, On 21/10/2014 01:26:> And Rosta scripsit:
>
>> But must Lojban so specify the duration of the /i/ in /ia/? That seems
>> unnecessarily pernickety, given the quite proper laxity of all other
>> realization rules in Lojban. If, rather, what is settled is that /i/ in
>> /ia/ is an onset, then I am asking why that must be so.
>
> Historically, I think it rose out of the introduction of ' to simplify the
> Loglan pronunciation rules.
>
> In Loglan there is no ', and therefore only
> 25 VV sequences rather than Lojban's 29 (ignoring "y" in both languages).
> These 25 break into four groups according to pronunciation:
>
> ai ei oi ao are always falling diphthongs (in practice, ao is pronounced
> [aw] like Lojban au).
>
> ae au ea eo eu oa oe ou are always two syllables; when e is the first,
> it is allophonically [e] rather than [E].
>
> aa ee oo are always two syllables *and* one of the syllables is required
> to bear the stress (so words like "baarsoa" are invalid, unlike the
> Lojban analogue "ba'ars'oa"); ee oo are very rare.
>
> iV and uV may be pronounced either as rising diphthongs, as in Lojban, or
> as two syllables: therefore "stomia" may be either ['stomja] or [sto'mia].

The Loglan treatment of /ia/ is clearly better (because it doesn't require rules of syllabification). /ii, uu/ are problematic either way, and should be forbidden. The Loglan way is problematic if it must contrast /ia/ and /iia/.

Obviously it was the glideless /ae, ea, aa/ type that led to Lojban's "'". That in itself was not so bad a move, tho the choice of realization was, but making it contrastive with zero between other vowels gives greater headaches. I'd have just forbidden them altogether; going all Livagian on their ass, I'd allow i to be followed by any vowel but i, u to be followed by any vowel but u, e to be followed by no vowel but i, o to be followed by no vowel but u, and a to be followed by no vowel but i and u.

--And.

And Rosta

unread,
Oct 21, 2014, 5:06:13 AM10/21/14
to bpfk...@googlegroups.com
Jorge Llambías, On 21/10/2014 00:56:
>
>
> On Mon, Oct 20, 2014 at 8:32 PM, And Rosta <and....@gmail.com <mailto:and....@gmail.com>> wrote:
>
> On 20 Oct 2014 22:43, "Jorge Llambías" <jjlla...@gmail.com <mailto:jjlla...@gmail.com>> wrote:
> >
> > There's no special need to realize /Cia/ as one syllable, but it must be counted as one syllable for penultimate stress rule purposes.
>
> How come? Is it simply that because word-segmentation is sensitive to stress, and stress is sensitive to syllabicity, varisyllabicity risks undermining the consistency of the word-segmentation rules? If so, the stress or word-segmentation rules can be reformulated so that VV counts as a single metrical unit rather than a sequence of two.
>
> Yes.
>
> > Formulating the rule without mentioning syllables would be more complicated.
>
> It's not yet clear to me that that is so, but that could be because I don't know the rule.
>
>
> Maybe not terribly more complicated, but at least somewhat more complicated. y-syllables are already exceptions to penultimate stress, as are consonantal syllables and buffer-vowel syllables, so this would be just one more exception.

It sounds like there are all these exceptions because the rule is wrongly formulated. If syllables are discarded and the metrical units are instead AEIOU clusters, might the rule become exceptionless?

>
> > It also has to be distinguished from "ciia"
>
> How come? Is "ciia" licit? Is it licit even if "cia" isn't?
>
> Yes, there's no constraint against vowel i being followed by glide i.

I would forbid /ii, uu/, but if you were set on allowing them, then things could be kept in order by a rule that requires every phonological string to be parsable as a CV sequence, which would rule out /cia/.

> I would allow "a'ua" (assuming I was not allowed to kill /'/), disallow "uu", and have all words begin with a consonant, but if "a'uua" must be a variant of "a'u.ua <http://u.ua>" then I see why "a'ua" must be forbidden.
>
> Yes, it should be one or the other, but not both.

And then <a'ua> could be an orthographic variant of <a'uua>, <cia> of <ciia>.

--And.

John Cowan

unread,
Oct 21, 2014, 2:08:02 PM10/21/14
to bpfk...@googlegroups.com
And Rosta scripsit:

> Is the concern that because /./ is elidable when its presence is not
> morphologically contrastive, the risk is that through habit it would
> end up being elided even when it is?

Just so.

> a problem with that is that it is hard to carefully and deliberately
> show that one is using a properly /./-less form.

Indeed.

> >In any case, I was talking about "a ua" [a?wa] as hard to distinguish
> >from "a'ua" [ahwa], both tending to become simple [awa].
>
> Specifically for L1 English speakers, you must mean, rather than for
> people in general. Does it really make sense to base the rules of
> Lojban on the specific needs of L1 English speakers?

Both [W] and the cluster [hw] are rare in the world's languages compared
with [w], so it's not too surprising that most varieties of English have
lost them. ("Sir, it is not so much to be lamented that Old England is
lost, as that the Scots have found it." --Sam: Johnson)

> Many L1 English speakers would tend to hear /a.ua/ as /at ua/.

Lojban /t/ is problematic for anglophones in general, given the North
American (i.e majority) tendency to voice it between vowels and to
glottalize it between a vowel and a syllabic consonant. What is worse,
all anglophones tend to hear [t] (as opposed to [t_h]) as /d/. I don't
think we can do anything about this.

> If /'/ is to be kept distinct from /x/, /'/ must be [T], giving [aTua]
> for /a'ua/, which is unlikely to become [awa].

I'm not sure if this is meant to be an anglophone or a universal claim.
Anglophones tend to render [x] as [k], as in _loch, bach, Bach_, and
Germans have no problem distinguishing /h/ and /x/ systematically,
though it's arguable that there are no [h] : [x] minimal pairs, as [h]
is only in onsets whereas /x/ in onsets is realized (in the standard
accent, at least) as /C/.

> It would be an assimilation rather than a fortition. As I've said
> before, [h] is articulatorily impossible as a realization of /'/
> in some environments, e.g. /i'i/, at ordinary speech rates,

I articulate /i'i/ as [iCi], /u'u/ as [uWu], /ii/ as /j\i/ (with a voiced
palatal fricative like Spanish-Spanish "y"), and /uu/ as [wu].

> Obviously it was the glideless /ae, ea, aa/ type that led to Lojban's
> "'". That in itself was not so bad a move, tho the choice of realization
> was, but making it contrastive with zero between other vowels gives
> greater headaches. I'd have just forbidden them altogether; going all
> Livagian on their ass, I'd allow i to be followed by any vowel but i,
> u to be followed by any vowel but u, e to be followed by no vowel but i,
> o to be followed by no vowel but u, and a to be followed by no vowel
> but i and u.

This would, of course, involve a complete discarding of the cmavo list and
starting over.
Linguistics is arguably the most hotly contested property in the academic
realm. It is soaked with the blood of poets, theologians, philosophers,
philologists, psychologists, biologists and neurologists, along with
whatever blood can be got out of grammarians. - Russ Rymer

John Cowan

unread,
Oct 21, 2014, 2:12:00 PM10/21/14
to bpfk...@googlegroups.com
And Rosta scripsit:

> It sounds like there are all these exceptions because the rule is
> wrongly formulated. If syllables are discarded and the metrical units
> are instead AEIOU clusters, might the rule become exceptionless?

It would get worse: we would have to have an exception for words like /stabaa/
explaining why it is /sta'baa/ rather than /'stabaa/.
"The exception proves the rule." Dimbulbs think: "Your counterexample proves
my theory." Latin students think "'Probat' means 'tests': the exception puts
the rule to the proof." But legal historians know it means "Evidence for an
exception is evidence of the existence of a rule in cases not excepted from."

Alex Burka

unread,
Oct 21, 2014, 2:57:10 PM10/21/14
to bpfk...@googlegroups.com
To make this more concrete, I went and did the search. Using an XML dump from yesterday, there are 342 valsi in jbovlaste (2 zei-lujvo, 2 experimental cmavo, 28 cmevla, 310 fu'ivla) that contain C[iu]V. If I restrict C to sibilants [jczs], there are only 52 (1 experimental cmavo, 4 cmevla, 47 fu'ivla).

mu'o mi'e la durka

Alex Burka

unread,
Oct 21, 2014, 4:50:29 PM10/21/14
to bpfk...@googlegroups.com
Side note: half of those fu'ivla (156 out of 310) have one or more of kingdom/phylum/class/order/family/genus/species in their definitions, suggesting they are transliterated from Latin (I have not looked at the full list of words). So those may be fixable by simply changing CiV to CiiV, etc.

mu'o mi'e la durka

Jorge Llambías

unread,
Oct 21, 2014, 6:30:48 PM10/21/14
to bpfk...@googlegroups.com
On Tue, Oct 21, 2014 at 6:06 AM, And Rosta <and....@gmail.com> wrote:

I would forbid /ii, uu/, but if you were set on allowing them, then things could be kept in order by a rule that requires every phonological string to be parsable as a CV sequence, which would rule out /cia/.

It would have been better to not allow them from the start, disallowing them now would be a lot of work. 

We could still salvage both the CV rule and /cia/ if the buffer vowel is allowed between consonant and glide. 
 
And then <a'ua> could be an orthographic variant of <a'uua>, <cia> of <ciia>.

I doubt the isomorphicists would be very happy with two words written together like that. It would also be somewhat inconsistent that we can't write ".iia" as ".ia", unless ".ia" must be interpreted as ".iia" and "ia" can't be written with a dot in front. But then that would interfere with dotside and names like ".iakob."...

And Rosta

unread,
Oct 21, 2014, 6:37:14 PM10/21/14
to bpfk...@googlegroups.com
John Cowan, On 21/10/2014 19:11:
> And Rosta scripsit:
>
>> It sounds like there are all these exceptions because the rule is
>> wrongly formulated. If syllables are discarded and the metrical units
>> are instead AEIOU clusters, might the rule become exceptionless?
>
> It would get worse: we would have to have an exception for words like /stabaa/
> explaining why it is /sta'baa/ rather than /'stabaa/.

Is /stabaa/ an alternative notation or analysis for /staba'a/? If not, what is it? If it is, then we just need a suitable definition of AEIOU cluster: either we say that /'/ is phonologically visible, in which case it can't be cluster-internal (since it is not an AEIOU), or we take 'AEIOU cluster' as a primitive rather than derived term, and have a rule of apostrophe-insertion between phonologically distinct but contiguous AEIOU clusters.

>> Obviously it was the glideless /ae, ea, aa/ type that led to Lojban's
>> "'". That in itself was not so bad a move, tho the choice of realization
>> was, but making it contrastive with zero between other vowels gives
>> greater headaches. I'd have just forbidden them altogether; going all
>> Livagian on their ass, I'd allow i to be followed by any vowel but i,
>> u to be followed by any vowel but u, e to be followed by no vowel but i,
>> o to be followed by no vowel but u, and a to be followed by no vowel
>> but i and u.
>
> This would, of course, involve a complete discarding of the cmavo list and
> starting over.

A certain revision, rather than a complete discarding. You could convert to new forms by rule, /e'V/ to /eiV/, /o'V/ to /ouV/, /i'V/ to /iV/, /u'V/ to /uV/, some other rule for /a'V/, and sort out the newly created homophones, perhaps by making use of a /aiV/:/auV/ contrast.

But to be more realistic and hence more conservative, my reading of what xorxes said camxes does, namely make every string analysable as a sequence of CVs, sounds like the best rule.

John Cowan, On 21/10/2014 19:08:> And Rosta scripsit:
>
>> Is the concern that because /./ is elidable when its presence is not
>> morphologically contrastive, the risk is that through habit it would
>> end up being elided even when it is?
>
> Just so.
>
>> a problem with that is that it is hard to carefully and deliberately
>> show that one is using a properly /./-less form.
>
> Indeed.

This is a wider problem with /./, isn't it. A solution would be to make glottal-stop elision illicit.

>>> In any case, I was talking about "a ua" [a?wa] as hard to distinguish
>> >from "a'ua" [ahwa], both tending to become simple [awa].
>>
>> Specifically for L1 English speakers, you must mean, rather than for
>> people in general. Does it really make sense to base the rules of
>> Lojban on the specific needs of L1 English speakers?
>
> Both [W] and the cluster [hw] are rare in the world's languages compared
> with [w], so it's not too surprising that most varieties of English have
> lost them.

I think [hw] is virtually inarticulable. I don't know if anybody knows the frequency of [W] or any other phone in the world's languages. The fact that it's rare as the primary allophone of a phoneme doesn't mean it's rare as a phone. (E.g. bilabial trills are rare as primary allophone of a phoneme in world's languages, but the phone is still to be heard in English words for quite a few speakers.)

But anyway, rather than [aWa] tending to become [awa] and hence neutralized with /aua/, it could instead become the far more innocuous [aWua].

>> Many L1 English speakers would tend to hear /a.ua/ as /at ua/.
>
> Lojban /t/ is problematic for anglophones in general, given the North
> American (i.e majority) tendency to voice it between vowels and to
> glottalize it between a vowel and a syllabic consonant. What is worse,
> all anglophones tend to hear [t] (as opposed to [t_h]) as /d/.

(Most but not all.)

> I don't think we can do anything about this.

We could specify aspirated realizations for /ptk/ and voiced for /bdg/.

>> If /'/ is to be kept distinct from /x/, /'/ must be [T], giving [aTua]
>> for /a'ua/, which is unlikely to become [awa].
>
> I'm not sure if this is meant to be an anglophone or a universal claim.

A universal phonetic claim.

> Anglophones tend to render [x] as [k], as in _loch, bach, Bach_, and
> Germans have no problem distinguishing /h/ and /x/ systematically,
> though it's arguable that there are no [h] : [x] minimal pairs, as [h]
> is only in onsets whereas /x/ in onsets is realized (in the standard
> accent, at least) as /C/.

For phonetic reasons, [h] is possible only between vowels sufficiently open that the narrowest constriction of the vocal tract is at the glottis, so e & a but not i, u, o. Frication (turbulence) occurs at the locus of narrowest constriction.

Any posterior fricative will tend strongly to assimilate to [C] in environment [i _ i] and to [W] (or labialized [x_w]) in environment [u _ u].

>> It would be an assimilation rather than a fortition. As I've said
>> before, [h] is articulatorily impossible as a realization of /'/
>> in some environments, e.g. /i'i/, at ordinary speech rates,
>
> I articulate /i'i/ as [iCi], /u'u/ as [uWu], /ii/ as /j\i/ (with a voiced
> palatal fricative like Spanish-Spanish "y"), and /uu/ as [wu].

How about /ixi/ and /uxu/?

--And.

Jorge Llambías

unread,
Oct 21, 2014, 7:12:54 PM10/21/14
to bpfk...@googlegroups.com
On Tue, Oct 21, 2014 at 7:37 PM, And Rosta <and....@gmail.com> wrote:

But to be more realistic and hence more conservative, my reading of what xorxes said camxes does, namely make every string analysable as a sequence of CVs, sounds like the best rule.

We can postulate that as the underlying rule, if the four diphthongs ai, au, ei, oi are instances of V, and ., ', i and u are instances of C, and if we allow the buffer vowel to be inserted between consonants and i/u glides, and after the final consonant of a cmevla at the end of a string.

But camxes doesn't know anything about the buffer vowel, so the surface rules are somewhat more complicated. A syllable for camxes can be as long as CCCVVC.

John Cowan

unread,
Oct 21, 2014, 7:53:09 PM10/21/14
to bpfk...@googlegroups.com
And Rosta scripsit:

> Is /stabaa/ an alternative notation or analysis for /staba'a/?

Yes. That's what I thought you were going after.

> I think [hw] is virtually inarticulable. I don't know if anybody knows
> the frequency of [W] or any other phone in the world's languages. The
> fact that it's rare as the primary allophone of a phoneme doesn't
> mean it's rare as a phone.

Fair enough. Wikipedia reports it as phonemic in Cornish and Hupa, and
allophonic in Nahuatl (before voiceless consonants) and Taiwanese
(as an articulation of /hw/), aside from its special status in English.

> How about /ixi/ and /uxu/?

Unless I am very careful, I always articulate [x] as [X] in any language.
I have no trouble with either [iXi] or [uXu].
Where the wombat has walked, it will inevitably walk again.
(even through brick walls!)

Jorge Llambías

unread,
Oct 22, 2014, 5:19:26 PM10/22/14
to bpfk...@googlegroups.com
On Tue, Oct 21, 2014 at 5:50 PM, Alex Burka <dur...@gmail.com> wrote:
Side note: half of those fu'ivla (156 out of 310) have one or more of kingdom/phylum/class/order/family/genus/species in their definitions, suggesting they are transliterated from Latin (I have not looked at the full list of words). So those may be fixable by simply changing CiV to CiiV, etc.

Changing  CIV to CIIV will work in all cases except when CIV is at the beginning of the word. In those cases, the CI will drop off, e.g.: cionmau -> ciionmau = ci ionmau. 

Alex Burka

unread,
Oct 22, 2014, 5:24:49 PM10/22/14
to bpfk...@googlegroups.com
Indeed. 46 of those: ["cia'o'e", "xua'ai", 'bianfu', 'biorka', 'ciencia', 'cionmau', 'dianzi', 'fiakre', 'fiesta', 'fiorso', 'guanci', 'kuadranta', 'kuaidzi', "kuardicka'u", 'kuargo', 'kuarka', 'kuinke', 'liante', 'liunko', 'luodna', 'mianma', 'niengatu', 'niutni', 'nuansa', 'suenska', 'suomne', 'tiatro', 'tiotka', 'tuitsku', 'violbasu', 'violna', 'violni', 'violtcelo', 'xuandi', 'bie,uaRUC', 'buenosaires', 'guaspis', 'guonJAUS', 'kuadragesim', 'kuadril', 'LIEtuvas', 'lietuvos', 'suomen', 'suomis', 'tienjin', 'tuityr']

Jorge Llambías

unread,
Oct 22, 2014, 5:31:09 PM10/22/14
to bpfk...@googlegroups.com
On Wed, Oct 22, 2014 at 6:24 PM, Alex Burka <dur...@gmail.com> wrote:
Indeed. 46 of those: ["cia'o'e", "xua'ai", 'bianfu', 'biorka', 'ciencia', 'cionmau', 'dianzi', 'fiakre', 'fiesta', 'fiorso', 'guanci', 'kuadranta', 'kuaidzi', "kuardicka'u", 'kuargo', 'kuarka', 'kuinke', 'liante', 'liunko', 'luodna', 'mianma', 'niengatu', 'niutni', 'nuansa', 'suenska', 'suomne', 'tiatro', 'tiotka', 'tuitsku', 'violbasu', 'violna', 'violni', 'violtcelo', 'xuandi', 'bie,uaRUC', 'buenosaires', 'guaspis', 'guonJAUS', 'kuadragesim', 'kuadril', 'LIEtuvas', 'lietuvos', 'suomen', 'suomis', 'tienjin', 'tuityr']

cmevla don't have that problem, cmavo won't fall off of them. so it's 34.

Jorge Llambías

unread,
Oct 22, 2014, 5:43:53 PM10/22/14
to bpfk...@googlegroups.com
On Wed, Oct 22, 2014 at 6:24 PM, Alex Burka <dur...@gmail.com> wrote:
"cia'o'e", "xua'ai", 'bianfu', 'biorka', 'ciencia', 'cionmau', 'dianzi', 'fiakre', 'fiesta', 'fiorso', 'guanci', 'kuadranta', 'kuaidzi', "kuardicka'u", 'kuargo', 'kuarka', 'kuinke', 'liante', 'liunko', 'luodna', 'mianma', 'niengatu', 'niutni', 'nuansa', 'suenska', 'suomne', 'tiatro', 'tiotka', 'tuitsku', 'violbasu', 'violna', 'violni', 'violtcelo', 'xuandi'

Most of these can be fixed by changing CIV -> CI'V. The only ones that can't be fixed that way are: 'fiakre', 'fiesta', 'kuadranta', 'tiatro'.

Jorge Llambías

unread,
Oct 22, 2014, 5:47:19 PM10/22/14
to bpfk...@googlegroups.com
On Wed, Oct 22, 2014 at 6:43 PM, Jorge Llambías <jjlla...@gmail.com> wrote:

Most of these can be fixed by changing CIV -> CI'V. The only ones that can't be fixed that way are: 'fiakre', 'fiesta', 'kuadranta', 'tiatro'.

Actually, "ku'adranta" is fine, because "dranta" is a slinku'i so "ku'a" stays put.  That only leaves three.

Jorge Llambías

unread,
Oct 22, 2014, 5:53:56 PM10/22/14
to bpfk...@googlegroups.com
But "kuardicka'u" gives a lujvo, so it's back to four.

Alex Burka

unread,
Oct 22, 2014, 6:01:56 PM10/22/14
to bpfk...@googlegroups.com
Just make 'em type 3, I would say, if we were to end up going this route -- they're just cognates with no recorded usage. marcrfiakre, salcrfiesta, dijrtiatro (or marcrfi'akre, salcrfi'esta, dijrti'atro).

As for kuardicka'u (which doesn't show up in the corpus either), if we don't allow the "qu-" sound at all, it's going to be difficult to make a satisfying fu'ivla for "quark". The diabolical solution is just to use the wrong hyphen and turn it back into a zi'evla: ku'andicka'u. Another solution would be to reformulate it as a lujvo: ratslepaudicka'u, or something.

mu'o mi'e la durka
--

Bob LeChevalier, President and Founder - LLG

unread,
Oct 23, 2014, 10:24:15 AM10/23/14
to bpfk...@googlegroups.com
On 10/21/2014 2:08 PM, John Cowan wrote:
>> If /'/ is to be kept distinct from /x/, /'/ must be [T], giving [aTua]
>> for /a'ua/, which is unlikely to become [awa].
>
> I'm not sure if this is meant to be an anglophone or a universal claim.
> Anglophones tend to render [x] as [k], as in _loch, bach, Bach_, and
> Germans have no problem distinguishing /h/ and /x/ systematically,
> though it's arguable that there are no [h] : [x] minimal pairs, as [h]
> is only in onsets whereas /x/ in onsets is realized (in the standard
> accent, at least) as /C/.
>
>> It would be an assimilation rather than a fortition. As I've said
>> before, [h] is articulatorily impossible as a realization of /'/
>> in some environments, e.g. /i'i/, at ordinary speech rates,
>
> I articulate /i'i/ as [iCi], /u'u/ as [uWu], /ii/ as /j\i/ (with a voiced
> palatal fricative like Spanish-Spanish "y"), and /uu/ as [wu].
>
>> Obviously it was the glideless /ae, ea, aa/ type that led to Lojban's
>> "'". That in itself was not so bad a move, tho the choice of realization
>> was, but making it contrastive with zero between other vowels gives
>> greater headaches. I'd have just forbidden them altogether; going all
>> Livagian on their ass, I'd allow i to be followed by any vowel but i,
>> u to be followed by any vowel but u, e to be followed by no vowel but i,
>> o to be followed by no vowel but u, and a to be followed by no vowel
>> but i and u.
>
> This would, of course, involve a complete discarding of the cmavo list and
> starting over.


I haven't been following this discussion, and I'm not sure what is
written in CLL about it, but I think what we said originally was that
the apostrophe could permissibly be realized as ANY unvoiced consonant
sound not otherwise found in Lojban (but hopefully being fairly
consistent at using the same sound all the time). One of our original
Lojban students liked to tease everyone by using unvoiced "th" as his
realization of all apostrophes. It sounded funny and probably caused
everyone else to try harder to use 'h'.

lojbab

Jonathan Jones

unread,
Oct 23, 2014, 3:17:42 PM10/23/14
to bpfk...@googlegroups.com


On Oct 23, 2014 8:24 AM, "Bob LeChevalier, President and Founder - LLG" <loj...@lojban.org> wrote:
><snip>


> I haven't been following this discussion, and I'm not sure what is written in CLL about it, but I think what we said originally was that the apostrophe could permissibly be realized as ANY unvoiced consonant sound not otherwise found in Lojban (but hopefully being fairly consistent at using the same sound all the time).  One of our original Lojban students liked to tease everyone by using unvoiced "th" as his realization of all apostrophes.  It sounded funny and probably caused everyone else to try harder to use 'h'.
>
> lojbab

I'm fairly certain I remember reading the bit about " ' " as well. IIRC, it was either in the bit about "strange" pronunciations (c ,j ,x, etc.), or the bit about being allowed to for example pronounce {mlatu} as "milatu" using the "I" in "fit".

guskant

unread,
Dec 11, 2014, 1:12:52 AM12/11/14
to bpfk...@googlegroups.com
I think the decision should be done as soon as possible. As I mentioned in my second post to the thread, allowing CgV to ma'ovla is critical in parsing. {jiebroda} is a fu'ivla for the current jbofi'e+yacc, while it consists of two words for the current camxes. While we are discussing, CgV cmavo are increasing in la jbovlaste.

If we have enough discussed it, it is time for voting. As a voting system, I would recommend approval voting in order to reduce a spoiler effect. 
How and where can we hold an election under an approval voting system?
Or do you recommend another system? 


Here are the candidates that I consider. If you noticed any lack, please inform us of it.

1.1.1
allow CgV in cmevla/fu'ivla/ma'ovla (current camxes)

1.1.2
allow CgV except [sibilant or dental C + iV] in cmevla/fu'ivla/ma'ovla

1.2.1
allow CgV in cmevla/fu'ivla, disallow it in ma'ovla (current jbofi'e+yacc, regarding CLL3.4 as a rule)

1.2.2
allow CgV except [sibilant or dental C + iV] in cmevla/fu'ivla, disallow all CgV in ma'ovla

1.3.1
allow CgV in cmevla, disallow it in fu'ivla/ma'ovla (not discussed, but my preference)

1.3.2
allow CgV except [sibilant or dental C + iV] in cmevla, disallow all CgV in fu'ivla/ma'ovla

2
disallow CgV in cmevla/fu'ivla/ma'ovla (ban it in Lojban word)


mu'o

Gleki Arxokuna

unread,
Dec 11, 2014, 1:28:15 AM12/11/14
to bpfk...@googlegroups.com

2014-12-11 9:12 GMT+03:00 guskant <gusni...@gmail.com>:
1.3.1
allow CgV in cmevla, disallow it in fu'ivla/ma'ovla (not discussed, but my preference)

I vote for this. + I prefer it to be just a synonym (dialect variation of the same sequence but with a {'} inside. co'o zo guaspi

John Cowan

unread,
Dec 11, 2014, 8:16:18 AM12/11/14
to bpfk...@googlegroups.com
Gleki Arxokuna scripsit:

> I vote for this. + I prefer it to be just a synonym (dialect variation of
> the same sequence but with a {'} inside. co'o zo guaspi

I can live with this, but I'd prefer to ban it altogether.
Clear? Huh! Why a four-year-old child could understand this report.
Run out and find me a four-year-old child. I can't make head or tail
out of it. --Rufus T. Firefly on government reports

Gleki Arxokuna

unread,
Dec 11, 2014, 8:41:35 AM12/11/14
to bpfk...@googlegroups.com
2014-12-11 16:16 GMT+03:00 John Cowan <co...@mercury.ccil.org>:
Gleki Arxokuna scripsit:

> I vote for this. + I prefer it to be just a synonym (dialect variation of
> the same sequence but with a {'} inside. co'o zo guaspi

I can live with this, but I'd prefer to ban it altogether.

It depends on in what way should it be banned. 
I vote for banning it from the formal grammar however, I vote for the preprocessor to be able to autocorrect {jie} back to {ji'e}. Speech recognition must be able to recognize both variants.


--
John Cowan          http://www.ccil.org/~cowan        co...@ccil.org
Clear?  Huh!  Why a four-year-old child could understand this report.
Run out and find me a four-year-old child.  I can't make head or tail
out of it.        --Rufus T. Firefly on government reports

Alex Burka

unread,
Dec 11, 2014, 11:30:49 AM12/11/14
to bpfk...@googlegroups.com
The preprocessor thing doesn't make a ton of sense to me. Does anyone else support this "dialectal variation"? We don't preprocess {ie} to {i'e}, so I don't see the point in changing ?{jie} to {ji'e}. They're quite different sounds, and moreover different syllable counts, so you'll be complecting the stress and word segmentation rules as well as breaking audiovisual isomorphism.

As for the larger question, I don't have a very strong opinion. I lean towards consistency (that would be banning it or allowing it everywhere). If we do allow it, maybe the sibilants should still be an exception for ease of pronunciation (I don't have trouble with ?{guaspi}, but ?{jiespi} and ?{ciaspi} are difficult).

So if we're approval-voting: 1.1.2 + 2. But I'm open to being convinced otherwise.

mu'o mi'e la durkavore

Alex Burka

unread,
Dec 11, 2014, 12:48:34 PM12/11/14
to bpfk...@googlegroups.com
mu'onai One more thought: along with sibilants/dentals, what about liquids? Those are more of a problem for me: without a preceding vowel, ?{lua} is comes out as {lu ua} or {ly ua}. Unfortunately ?{.influenza} is in jbovlaste.

mu'oja'ai

Gleki Arxokuna

unread,
Dec 11, 2014, 1:01:00 PM12/11/14
to bpfk...@googlegroups.com
There wasnt any reasonable consistency and there isnt. Take {mz} but {nz}, {ms} and {ns}.

Alex Burka

unread,
Dec 11, 2014, 1:02:54 PM12/11/14
to bpfk...@googlegroups.com
I mean consistency as in not having sounds that are allowed in one word class but not another (zifcme aside).

Gleki Arxokuna

unread,
Dec 11, 2014, 1:03:07 PM12/11/14
to bpfk...@googlegroups.com
This will be {influ'enza} and for me {influenza} will be an acceptable pronunciation.
Just like {klamygau} is the same as {klagau}

Jorge Llambías

unread,
Dec 11, 2014, 4:23:48 PM12/11/14
to bpfk...@googlegroups.com
On Thu, Dec 11, 2014 at 3:12 AM, guskant <gusni...@gmail.com> wrote:

1.3.1
allow CgV in cmevla, disallow it in fu'ivla/ma'ovla (not discussed, but my preference)

This would be my preference too, noting that it'd be allowed in cmevla in the same way that .ktktkt. is a valid cmevla, i.e. as a zifcmevla, not as a jbocmevla, which must consist of lojbanic syllables.

John Cowan

unread,
Dec 11, 2014, 8:02:05 PM12/11/14
to bpfk...@googlegroups.com
Alex Burka scripsit:

> mu'onai One more thought: along with sibilants/dentals, what about
> liquids? Those are more of a problem for me: without a preceding vowel,
> ?{lua} is comes out as {lu ua} or {ly ua}. Unfortunately ?{.influenza}
> is in jbovlaste.

I continue to think they should all be banned, except possibly in cmevla.
The phonology is what it is, and there is no need to extend it further.
If I have seen farther than others, it is because I am surrounded by dwarves.
--Murray Gell-Mann

Gleki Arxokuna

unread,
Dec 12, 2014, 2:33:23 AM12/12/14
to bpfk...@googlegroups.com
I change my vote. I vote for all such combinations to be banned everywhere, even in cmevla however the existing words are to be assumed to have {'} inside illegal CVV combinations. Jbovlaste should redirect obsolete words to their apostrophe-enabled counterparts. This will retain many existing words.

guskant

unread,
Dec 12, 2014, 2:40:37 AM12/12/14
to bpfk...@googlegroups.com
OK, I added [liquid C+uV] to the relevant options, and created an approval-voting form on Google Docs:

You are required to be logged in to your Google account in order to vote. You can modify your response after voting. I think of closing the poll on 2014-12-27 at 0:00 UTC, if you have no problem about the date and time. 

If I don't misunderstand, everybody interested should have a voting right. Enjoy voting!

Gleki Arxokuna

unread,
Dec 12, 2014, 2:59:40 AM12/12/14
to bpfk...@googlegroups.com
This doesn't include an option to retain existing words that can be preprocessed into apostophe-enabled or assumed to be their synonyms.

--

Alex Burka

unread,
Dec 12, 2014, 3:07:31 AM12/12/14
to bpfk...@googlegroups.com
Under this scheme, mu'a {guaspi} becomes {gu'aspi}. Do we keep it as a tcizbaga word or change it?

Gleki Arxokuna

unread,
Dec 12, 2014, 3:11:02 AM12/12/14
to bpfk...@googlegroups.com
that would be a separate poll with the options:
1. i approve of shutting down the whole brivla morphology distinction
2. i approve of shutting down the brivla morphology distinction except for lujvo with jvajvo-generating suffixes/prefixes like -gau, cma- and similar
3. i dont approve of changing it

etc.

guskant

unread,
Dec 12, 2014, 3:20:11 AM12/12/14
to bpfk...@googlegroups.com


Le vendredi 12 décembre 2014 16:59:40 UTC+9, la gleki a écrit :
This doesn't include an option to retain existing words that can be preprocessed into apostophe-enabled or assumed to be their synonyms.


You should create another poll for that problem. The current poll is to decide if CgV is allowed in all or some Lojban words. Your problem is to be decided _if_ the current poll results in other than 1.1.1.

By the way, if all existing CgV forms are preprocessed into CV'V forms, some of fu'ivla will become lujvo. It is not very desirable for me. 

Gleki Arxokuna

unread,
Dec 12, 2014, 3:21:54 AM12/12/14
to bpfk...@googlegroups.com
Yes, this interacts with the tcizbaga problem noi drata nabmi 

selpa'i

unread,
Dec 12, 2014, 5:30:37 AM12/12/14