Some brief comments on MLL3

[And Rosta]

[It seems that loglangers these days use Discord or the loglangs wiki, but email is much easier for my ingrained consuetudes.]

Mike, https://loglangs.wiki/User:Maiku/MLL3_design_notes

1. What is the rationale for the 'noun'--'verb' distinction?

Rather than the prefinal-MLL3 {my. srana my. mi mlatu my. zo'u xekri my. vau}, why not just {my. srana my. mi mlatu my. xekri my.} (which would be more in the spirit of Neolivagian)? With the implicit first argument rule, that would be {my. srana mi mlatu xekri}.

Some might answer that a kind of subject--predicate structure is needful for quantification, e.g. "All that I no longer find to be beautiful is banished from my home", or "my every cat is black" {ro my. srana my. mi mlatu my. zo'u xekri my. vau}~{ro my. srana mi mlatu zo'u xekri vau}~{ro srana mi mlatu fa xekri vau}. But to this I would riposte that donkey sentences ("most farmers that own a donkey beat it") show that quantificational/conditional meanings require a special protasis--apodosis construction; but there is no need to force that construction onto simple cases like "x: x pertain to me, x are black, x are cats" and arbitrarily shoving some of the predicates into the protasis and some into the apodosis.

2. Implicit first argument. I have two objections to this.

i. For polyadics, if there is an unmarked voice then privileging one argument as subject to unmarked implicitness is sometimes, often or always arbitrary (unless it is the davidsonian event argument that is thus privileged) and forces on us design decisions about the ergonomically optimal choice of default first argument for a given predicate. This objection is more aesthetic than practical.

ii. A more practical problem is that the variable that recurs in multiple predications is not necessarily an argument of the outermost predication, e.g. "x: I no longer think x is beautiful, you think I still want to have x". Of course, you could have two constructions, an unmarked one where some syntactically predictable argument is implicit and a marked one where no argument is predictably implicit, but then you'd need to decide whether the extra grammatical complexity is worth it and whether it is on average conducive to greater sentential brevity.

--And.

[Mike S.]

On Sat, Nov 20, 2021 at 8:43 AM And Rosta <and....@gmail.com> wrote:

[It seems that loglangers these days use Discord or the loglangs wiki, but email is much easier for my ingrained consuetudes.]

Mike, https://loglangs.wiki/User:Maiku/MLL3_design_notes

1. What is the rationale for the 'noun'--'verb' distinction?

Rather than the prefinal-MLL3 {my. srana my. mi mlatu my. zo'u xekri my. vau}, why not just {my. srana my. mi mlatu my. xekri my.} (which would be more in the spirit of Neolivagian)? With the implicit first argument rule, that would be {my. srana mi mlatu xekri}.

Your {my. srana mi mlatu xekri} is in fact the Lojbanic gloss of a valid MLL3 sentence, except for the missing aspect marker.  In essence, what you have done is to move the restriction of {my.} into the assertion.  This movement has a pragmatic effect.

Here's the sentence with restricted constant-assignment (and implicit 1st args):

(1) {my. srana mi mlatu zo'u my. xekri vau}
"M s.t. [M] pertains to me [and] are cats are s.t. M are black."
"My cat is black."

In the probable context of (1), my cat is already in the universe of discourse, but the audience does not know that it's black (which is the assertion).  The properties of pertaining to me and of being a cat are not being asserted here, but rather presupposed and used to identify the entity that I want to talk about.

Moving {srana mi mlatu} results in a sentence with unrestricted constant-assignment:

(2) {my. zo'u my. srana mi mlatu xekri vau}
"M is s.t. M pertains to me, are cats, [and] are black."
"It is my black cat."

In the probable context of (2), the speaker and the audience both know about some entity (perhaps some fast-moving object that just chased a mouse or made a noise--whatever it is, it's not mentioned here), but the audience does not know that it is mine, a cat, and black (which are here all part of the assertion).

The empty pendent {my. zo'u} contributes little and can be dropped, and should be dropped if the speaker wants to continue using a {my.} that has already been assigned, bringing us back to your sentence:

(3) {my. srana mi mlatu xekri vau}
"M pertains to me, are cats, [and] are black."
"M is my black cat."

This is similar to (2), except one might expect that M - which is now free in (3) - was assigned in a previous utterance.  The assertion is the same as in (2).

Besides information structure, another reason for steering towards restricted quantification/assignment is that I see no easy general way to introduce a second assignandum in the middle of the clause without it; having made 1st args implicit, all x1s have to refer to the same entity within a series of predicates.

(4) {(my. srana mi mlatu zo'u) (gy. gerku zo'u) (my. prami gy. vau)}
"My cat loves the dog."
Xorban: la je srnaka'a mlta le grke prmake.

Here we see each {zo'u} allows each variable to be restricted by a separate and independent predicate series.


(Side note: {vau} is needed to complete the sentence because we will want a way to delimit sentences as well as subordinate clauses, and I simply prefer mandatory information-bearing terminators to elidable meaningless terminators (and marking aspect is like an anchor IMHO).  The details of these aspect markers are not too important for the current email; here I am using {vau} for everything.)

 

Some might answer that a kind of subject--predicate structure is needful for quantification, e.g. "All that I no longer find to be beautiful is banished from my home", or "my every cat is black" {ro my. srana my. mi mlatu my. zo'u xekri my. vau}~{ro my. srana mi mlatu zo'u xekri vau}~{ro srana mi mlatu fa xekri vau}. But to this I would riposte that donkey sentences ("most farmers that own a donkey beat it") show that quantificational/conditional meanings require a special protasis--apodosis construction; but there is no need to force that construction onto simple cases like "x: x pertain to me, x are black, x are cats" and arbitrarily shoving some of the predicates into the protasis and some into the apodosis.

Your {ro my. srana mi mlatu zo'u [my.] xekri vau} (with {my.} inserted) illustrates restricted universal quantification.  We could rewrite the sentence to illustrate unrestricted quantification if we like (once again, by moving the restriction into the assertion), as follows:

{ro my. zo'u my srana mi mlatu naja xekri vau}

Note though we are forced to insert a conjunction ("not ... or ...", i.e. "implies", connecting {srana mi mlatu} and {xekri}) to make the assertion work.  We are already starting to see a strain when trying to form sentences without RQ, not only syntactically but also in terms of information structure: "Everything is either not my cat or it's black" is an oddly roundabout way to talk about one's pets.

Besides ∃ and ∀ there are other quantifiers such as "most".  It's difficult to translate "most" into logic at all unless we allow a restriction.  "Most" claims something about more than half of a particular kind of thing, not of the whole universe.

I would like to save donkey sentences for another email.

 

2. Implicit first argument. I have two objections to this.

i. For polyadics, if there is an unmarked voice then privileging one argument as subject to unmarked implicitness is sometimes, often or always arbitrary (unless it is the davidsonian event argument that is thus privileged) and forces on us design decisions about the ergonomically optimal choice of default first argument for a given predicate. This objection is more aesthetic than practical.

Indeed.  Case markers and {zo'u} select the x1-places of the governed predicate series, and due to this design, voice-marking suffixes are sometimes needed to access x2+.  Action predicates including speech-act predicates privilege the agent, etc.  This is a design decision.  One could imagine a more neutral design in which the arguments are indicated by ablaut or what-have-you, but MLL3 does not do that.

Then again, many of the predicates covering the people, places, and things that we wish to refer to are one-place.  Some common things like {kerfa} [= hair/fur] has an x2, but pragmatically the x2 of {kerfa} can usually be selected as {mlatu} or whatever. 

Aspect markers, which form the so-called "verb phrase"/"atomic clause", don't pick out referents but rather assert something about them.  These phrases treat x1 and x2 somewhat equally as "inner arguments" (i.e. pronominal arguments placed near the predicate itself), since it's equally easy to place a pronoun to the left or to the right:

{do prami vau}
= You love [someone].

{prami do vau}
= You are loved [or] people love you.

Moreover, clauses treat all "outer arguments" (i.e. those arguments indicated with case markers) equally.

{gerku fa mlatu fe prami vau}
= {mlatu fe gerku fa prami vau}
= The dog loves the cat.

 

ii. A more practical problem is that the variable that recurs in multiple predications is not necessarily an argument of the outermost predication, e.g. "x: I no longer think x is beautiful, you think I still want to have x". Of course, you could have two constructions, an unmarked one where some syntactically predictable argument is implicit and a marked one where no argument is predictably implicit, but then you'd need to decide whether the extra grammatical complexity is worth it and whether it is on average conducive to greater sentential brevity.

The above contains a very interesting example but to handle it I would have to introduce some additional syntax, including that of relative clauses.  I have already figured out relative clauses (they look just like main clauses except that they contain resumptive pronouns, and they can be headless, followed by a case marker), but I need to think more about the nesting of clauses in such an example...  some grist for the mill.

I do hope to have a more complete description of this language sometime in the near future (within months, hopefully).  Thanks for the comments.

 

--And.

-Mike

[And Rosta]

On Sun, 21 Nov 2021, 14:25 Mike S., <mai...@gmail.com> wrote:

On Sat, Nov 20, 2021 at 8:43 AM And Rosta <and....@gmail.com> wrote:

[It seems that loglangers these days use Discord or the loglangs wiki, but email is much easier for my ingrained consuetudes.]

Mike, https://loglangs.wiki/User:Maiku/MLL3_design_notes

1. What is the rationale for the 'noun'--'verb' distinction?

Rather than the prefinal-MLL3 {my. srana my. mi mlatu my. zo'u xekri my. vau}, why not just {my. srana my. mi mlatu my. xekri my.} (which would be more in the spirit of Neolivagian)? With the implicit first argument rule, that would be {my. srana mi mlatu xekri}.

Your {my. srana mi mlatu xekri} is in fact the Lojbanic gloss of a valid MLL3 sentence, except for the missing aspect marker.  In essence, what you have done is to move the restriction of {my.} into the assertion.  This movement has a pragmatic effect.

Here's the sentence with restricted constant-assignment (and implicit 1st args):

(1) {my. srana mi mlatu zo'u my. xekri vau}
"M s.t. [M] pertains to me [and] are cats are s.t. M are black."
"My cat is black."

In the probable context of (1), my cat is already in the universe of discourse, but the audience does not know that it's black (which is the assertion).  The properties of pertaining to me and of being a cat are not being asserted here, but rather presupposed and used to identify the entity that I want to talk about.

Moving {srana mi mlatu} results in a sentence with unrestricted constant-assignment:

(2) {my. zo'u my. srana mi mlatu xekri vau}
"M is s.t. M pertains to me, are cats, [and] are black."
"It is my black cat."

In the probable context of (2), the speaker and the audience both know about some entity (perhaps some fast-moving object that just chased a mouse or made a noise--whatever it is, it's not mentioned here), but the audience does not know that it is mine, a cat, and black (which are here all part of the assertion).

The empty pendent {my. zo'u} contributes little and can be dropped, and should be dropped if the speaker wants to continue using a {my.} that has already been assigned, bringing us back to your sentence:

(3) {my. srana mi mlatu xekri vau}
"M pertains to me, are cats, [and] are black."
"M is my black cat."

This is similar to (2), except one might expect that M - which is now free in (3) - was assigned in a previous utterance.  The assertion is the same as in (2).

Okay, I see. I tend to think of (3) as the basic, default case, with (1) warranting a special construction, which seems to be how it is in MLL3 too -- I think I was misled by the rapidly composed exposition focusing only on (1). In Neolivagian the basic syntax is more like (2) than (3) (to the extent that it is possible to compare the syntaxes), but its meaning is (3).

Besides information structure, another reason for steering towards restricted quantification/assignment is that I see no easy general way to introduce a second assignandum in the middle of the clause without it; having made 1st args implicit, all x1s have to refer to the same entity within a series of predicates.

(4) {(my. srana mi mlatu zo'u) (gy. gerku zo'u) (my. prami gy. vau)}
"My cat loves the dog."
Xorban: la je srnaka'a mlta le grke prmake.

Here we see each {zo'u} allows each variable to be restricted by a separate and independent predicate series.

How does the implicit argument rule work? Is it filled by the nearest prenex, hence why {my} is not implicit before {prami}?

(Side note: {vau} is needed to complete the sentence because we will want a way to delimit sentences as well as subordinate clauses, and I simply prefer mandatory information-bearing terminators to elidable meaningless terminators (and marking aspect is like an anchor IMHO).  

I would have the same preference between these two, but I would be wanting to seek a terminatorless solution, such as predicate-last clauses, or head-marking on the predicate, or (the more typical Neoliv move) marking the dependents for finality/nonfinality (like Lojban {be, bei}, IIRC). For something in the style of MLL3, I think I'd have expected predicate-last clauses. Ah, but then you lose your way of showing whether the first arg is bound from the prenex -- I see. 

 

Some might answer that a kind of subject--predicate structure is needful for quantification, e.g. "All that I no longer find to be beautiful is banished from my home", or "my every cat is black" {ro my. srana my. mi mlatu my. zo'u xekri my. vau}~{ro my. srana mi mlatu zo'u xekri vau}~{ro srana mi mlatu fa xekri vau}. But to this I would riposte that donkey sentences ("most farmers that own a donkey beat it") show that quantificational/conditional meanings require a special protasis--apodosis construction; but there is no need to force that construction onto simple cases like "x: x pertain to me, x are black, x are cats" and arbitrarily shoving some of the predicates into the protasis and some into the apodosis.

Your {ro my. srana mi mlatu zo'u [my.] xekri vau} (with {my.} inserted) illustrates restricted universal quantification.  We could rewrite the sentence to illustrate unrestricted quantification if we like (once again, by moving the restriction into the assertion), as follows:

{ro my. zo'u my srana mi mlatu naja xekri vau}

Note though we are forced to insert a conjunction ("not ... or ...", i.e. "implies", connecting {srana mi mlatu} and {xekri}) to make the assertion work.  We are already starting to see a strain when trying to form sentences without RQ, not only syntactically but also in terms of information structure: "Everything is either not my cat or it's black" is an oddly roundabout way to talk about one's pets.

Besides ∃ and ∀ there are other quantifiers such as "most".  It's difficult to translate "most" into logic at all unless we allow a restriction.  "Most" claims something about more than half of a particular kind of thing, not of the whole universe.

I would like to save donkey sentences for another email.

I take it pretty much for granted that quantification must be restrictive. Finding a syntax for donkey sentences is quite daunting; but once found, it serves for all quantification and conditionals. (I can't remember what the final/current Neolivagian solution is, just the pain of seeking it.)

Looking forward to seeing further info on MLL3 -- is there more on it on the wiki already?

--And.

[Mike S.]

On Sun, Nov 21, 2021 at 12:51 PM And Rosta <and....@gmail.com> wrote:

On Sun, 21 Nov 2021, 14:25 Mike S., <mai...@gmail.com> wrote:

On Sat, Nov 20, 2021 at 8:43 AM And Rosta <and....@gmail.com> wrote:

[It seems that loglangers these days use Discord or the loglangs wiki, but email is much easier for my ingrained consuetudes.]

Mike, https://loglangs.wiki/User:Maiku/MLL3_design_notes

1. What is the rationale for the 'noun'--'verb' distinction?

Rather than the prefinal-MLL3 {my. srana my. mi mlatu my. zo'u xekri my. vau}, why not just {my. srana my. mi mlatu my. xekri my.} (which would be more in the spirit of Neolivagian)? With the implicit first argument rule, that would be {my. srana mi mlatu xekri}.

Your {my. srana mi mlatu xekri} is in fact the Lojbanic gloss of a valid MLL3 sentence, except for the missing aspect marker.  In essence, what you have done is to move the restriction of {my.} into the assertion.  This movement has a pragmatic effect.

Here's the sentence with restricted constant-assignment (and implicit 1st args):

(1) {my. srana mi mlatu zo'u my. xekri vau}
"M s.t. [M] pertains to me [and] are cats are s.t. M are black."
"My cat is black."

In the probable context of (1), my cat is already in the universe of discourse, but the audience does not know that it's black (which is the assertion).  The properties of pertaining to me and of being a cat are not being asserted here, but rather presupposed and used to identify the entity that I want to talk about.

Moving {srana mi mlatu} results in a sentence with unrestricted constant-assignment:

(2) {my. zo'u my. srana mi mlatu xekri vau}
"M is s.t. M pertains to me, are cats, [and] are black."
"It is my black cat."

In the probable context of (2), the speaker and the audience both know about some entity (perhaps some fast-moving object that just chased a mouse or made a noise--whatever it is, it's not mentioned here), but the audience does not know that it is mine, a cat, and black (which are here all part of the assertion).

The empty pendent {my. zo'u} contributes little and can be dropped, and should be dropped if the speaker wants to continue using a {my.} that has already been assigned, bringing us back to your sentence:

(3) {my. srana mi mlatu xekri vau}
"M pertains to me, are cats, [and] are black."
"M is my black cat."

This is similar to (2), except one might expect that M - which is now free in (3) - was assigned in a previous utterance.  The assertion is the same as in (2).

Okay, I see. I tend to think of (3) as the basic, default case, with (1) warranting a special construction, which seems to be how it is in MLL3 too -- I think I was misled by the rapidly composed exposition focusing only on (1). In Neolivagian the basic syntax is more like (2) than (3) (to the extent that it is possible to compare the syntaxes), but its meaning is (3).

From the perspective of standard predicate logic, I would say (3) is indeed the basic case. It is not a usual practice to introduce "presupposed" constant symbols within the SPL object language itself; one finds no exact analog of Xorban's L- operator or Lojban's {la}/modern {lo} or MLL3's {VAR PREDS zo'u} syntax within SPL.  Instead, the logician writes "j = John; m = Mary; c = the black cat" (or whatever is needed) using English as a metalanguage, and then proceeds to use those constant symbols so defined.  Loglangs have* to find a way to define their constants without the prop of metalanguage.  It turns out that this can be done relatively easily using a syntax quite similar to restricted quantification (note that Xorban's S-, R- and L- all have the same syntax, despite having somewhat different semantics).  Low and behold, we have reinvented the noun phrases of natural languages.  There is much more to the semantics of noun phrases than this, but this is the starting point, I think.

*: well, I suppose, for the sake of experimentation, you could design a loglang such that the presupposed/asserted distinction is ignored and the properties of entities are, in effect, continually reasserted (and in fact the new loglang Eberban does this), but I suspect the users of such a language would eventually reimpose the presupposed/asserted distinction by some other means, perhaps by word order, given the way humans track old and new information.

 

Besides information structure, another reason for steering towards restricted quantification/assignment is that I see no easy general way to introduce a second assignandum in the middle of the clause without it; having made 1st args implicit, all x1s have to refer to the same entity within a series of predicates.

(4) {(my. srana mi mlatu zo'u) (gy. gerku zo'u) (my. prami gy. vau)}
"My cat loves the dog."
Xorban: la je srnaka'a mlta le grke prmake.

Here we see each {zo'u} allows each variable to be restricted by a separate and independent predicate series.

How does the implicit argument rule work? Is it filled by the nearest prenex, hence why {my} is not implicit before {prami}?

Here's a way to think of it. (I write at length with the intention of eventually incorporating these comments in a proper language description.) In Xorban, a clause is composed of formulae which can be nested to any depth without any special clause-marking devices; everything starting with verbs ("atomic formulae") is in effect a clause. MLL3 is different:

- A clause is composed of one or more predicate series each delimited by some sort of postposition: case/pendent postpositions mark NPs; aspect/mood postpositions mark VPs. A VP ends a clause. Predicate series can't be nested except indirectly by means of subordinate clauses (themselves full clauses ending with a VP).

- The x1 pronominal argument of any predicate series, if explicit, is always placed in front of the predicate series.

- If the predicate series stands in a NP, then
 - 1. a preposed x1 variable functions as an assignandum and can be used later in the clause (including as x2+ inside subsequent NPs)
 - 2. the predicate series expresses a restriction on the x1-variable

- If the predicate series stands in a VP, then
 - 1. a preposed x1-variable functions merely as an assignatum
 - 2. the predicate series expresses an assertion

Despite all these differences, you can see the reflection of Xorban if you divide the Xorban sentence into "NP" formulae prefixed with determiners and bare "VP" formulae:

_la je srnaka'a mlta / le grke / prmake_
{my. srana mi mlatu zo'u / gy. gerku zo'u / my. prami gy. vau)}

NP / NP / VP

{my.} reflects the variable _a_ wherever it appears in the Xorban, but collapses the first arguments of {srana} and {mlatu} into Xorban's assignandum (found in _la_), all of which must therefore be the same.  Meanwhile, {my. prami gy. vau} contains no assignandum because Xorban did not have a corresponding determiner in the first place.   It's a complete clause by itself, however, just as _pramake_ is (containing free variables, though).

 

(Side note: {vau} is needed to complete the sentence because we will want a way to delimit sentences as well as subordinate clauses, and I simply prefer mandatory information-bearing terminators to elidable meaningless terminators (and marking aspect is like an anchor IMHO).  

I would have the same preference between these two, but I would be wanting to seek a terminatorless solution, such as predicate-last clauses, or head-marking on the predicate, or (the more typical Neoliv move) marking the dependents for finality/nonfinality (like Lojban {be, bei}, IIRC). For something in the style of MLL3, I think I'd have expected predicate-last clauses. Ah, but then you lose your way of showing whether the first arg is bound from the prenex -- I see.

I am not sure that a loglang can avoid terminators entirely, but it's possible to get rid of some, and give others a useful role beyond merely showing structure.  Among the ones that seem unavoidable to me: the optional but sometimes needed left and right "parentheses" of conjunctions; something to mark the end of a nested clause (in MLL3, this goes on the left opposite the verb, while in Toaq and Morneau's Latejami, it goes on the right.  (Some people are surprised to learn that Latejami had terminators, but it did.)

 

 

Some might answer that a kind of subject--predicate structure is needful for quantification, e.g. "All that I no longer find to be beautiful is banished from my home", or "my every cat is black" {ro my. srana my. mi mlatu my. zo'u xekri my. vau}~{ro my. srana mi mlatu zo'u xekri vau}~{ro srana mi mlatu fa xekri vau}. But to this I would riposte that donkey sentences ("most farmers that own a donkey beat it") show that quantificational/conditional meanings require a special protasis--apodosis construction; but there is no need to force that construction onto simple cases like "x: x pertain to me, x are black, x are cats" and arbitrarily shoving some of the predicates into the protasis and some into the apodosis.

Your {ro my. srana mi mlatu zo'u [my.] xekri vau} (with {my.} inserted) illustrates restricted universal quantification.  We could rewrite the sentence to illustrate unrestricted quantification if we like (once again, by moving the restriction into the assertion), as follows:

{ro my. zo'u my srana mi mlatu naja xekri vau}

Note though we are forced to insert a conjunction ("not ... or ...", i.e. "implies", connecting {srana mi mlatu} and {xekri}) to make the assertion work.  We are already starting to see a strain when trying to form sentences without RQ, not only syntactically but also in terms of information structure: "Everything is either not my cat or it's black" is an oddly roundabout way to talk about one's pets.

Besides ∃ and ∀ there are other quantifiers such as "most".  It's difficult to translate "most" into logic at all unless we allow a restriction.  "Most" claims something about more than half of a particular kind of thing, not of the whole universe.

I would like to save donkey sentences for another email.

I take it pretty much for granted that quantification must be restrictive. Finding a syntax for donkey sentences is quite daunting; but once found, it serves for all quantification and conditionals. (I can't remember what the final/current Neolivagian solution is, just the pain of seeking it.)

Looking forward to seeing further info on MLL3 -- is there more on it on the wiki already?

--And.

More is scattered on Discord, but I could move parts to the Wiki. The SSM strategy might be interesting to discuss.

-Mike

[And Rosta]

On Tue, 23 Nov 2021 at 13:57, Mike S. <mai...@gmail.com> wrote:

On Sun, Nov 21, 2021 at 12:51 PM And Rosta <and....@gmail.com> wrote:

On Sun, 21 Nov 2021, 14:25 Mike S., <mai...@gmail.com> wrote:

On Sat, Nov 20, 2021 at 8:43 AM And Rosta <and....@gmail.com> wrote:

(3) {my. srana mi mlatu xekri vau}

"M pertains to me, are cats, [and] are black."
"M is my black cat."

This is similar to (2), except one might expect that M - which is now free in (3) - was assigned in a previous utterance.  The assertion is the same as in (2).

Okay, I see. I tend to think of (3) as the basic, default case, with (1) warranting a special construction, which seems to be how it is in MLL3 too -- I think I was misled by the rapidly composed exposition focusing only on (1). In Neolivagian the basic syntax is more like (2) than (3) (to the extent that it is possible to compare the syntaxes), but its meaning is (3).

From the perspective of standard predicate logic, I would say (3) is indeed the basic case. It is not a usual practice to introduce "presupposed" constant symbols within the SPL object language itself; one finds no exact analog of Xorban's L- operator or Lojban's {la}/modern {lo} or MLL3's {VAR PREDS zo'u} syntax within SPL.  Instead, the logician writes "j = John; m = Mary; c = the black cat" (or whatever is needed) using English as a metalanguage, and then proceeds to use those constant symbols so defined.  Loglangs have* to find a way to define their constants without the prop of metalanguage.  It turns out that this can be done relatively easily using a syntax quite similar to restricted quantification (note that Xorban's S-, R- and L- all have the same syntax, despite having somewhat different semantics).  Low and behold, we have reinvented the noun phrases of natural languages.  There is much more to the semantics of noun phrases than this, but this is the starting point, I think.

There's one family of loglangs, which includes Lojban, Toaq and, I gather, MLL3, that has noun-phrasey things, and another family, which includes Livagian and Unlws, that doesn't. The former builds on the patterns of natlangs; the latter cleaves more closely to the underlying logical form,

*: well, I suppose, for the sake of experimentation, you could design a loglang such that the presupposed/asserted distinction is ignored and the properties of entities are, in effect, continually reasserted (and in fact the new loglang Eberban does this), but I suspect the users of such a language would eventually reimpose the presupposed/asserted distinction by some other means, perhaps by word order, given the way humans track old and new information.

In Livagian, there are illocutionary predicates, and a predication not within the scope of any illocutionary predicate would be presupposed; i.e. 'presupposed' = outside scope of illocution.

(4) {(my. srana mi mlatu zo'u) (gy. gerku zo'u) (my. prami gy. vau)}
"My cat loves the dog."
Xorban: la je srnaka'a mlta le grke prmake.

How does the implicit argument rule work? Is it filled by the nearest prenex, hence why {my} is not implicit before {prami}?

Here's a way to think of it. (I write at length with the intention of eventually incorporating these comments in a proper language description.) In Xorban, a clause is composed of formulae which can be nested to any depth without any special clause-marking devices; everything starting with verbs ("atomic formulae") is in effect a clause. MLL3 is different:

- A clause is composed of one or more predicate series each delimited by some sort of postposition: case/pendent postpositions mark NPs; aspect/mood postpositions mark VPs. A VP ends a clause. Predicate series can't be nested except indirectly by means of subordinate clauses (themselves full clauses ending with a VP).

- The x1 pronominal argument of any predicate series, if explicit, is always placed in front of the predicate series.

- If the predicate series stands in a NP, then
 - 1. a preposed x1 variable functions as an assignandum and can be used later in the clause (including as x2+ inside subsequent NPs)
 - 2. the predicate series expresses a restriction on the x1-variable

- If the predicate series stands in a VP, then
 - 1. a preposed x1-variable functions merely as an assignatum
 - 2. the predicate series expresses an assertion

Despite all these differences, you can see the reflection of Xorban if you divide the Xorban sentence into "NP" formulae prefixed with determiners and bare "VP" formulae:

_la je srnaka'a mlta / le grke / prmake_
{my. srana mi mlatu zo'u / gy. gerku zo'u / my. prami gy. vau)}

NP / NP / VP

{my.} reflects the variable _a_ wherever it appears in the Xorban, but collapses the first arguments of {srana} and {mlatu} into Xorban's assignandum (found in _la_), all of which must therefore be the same.  Meanwhile, {my. prami gy. vau} contains no assignandum because Xorban did not have a corresponding determiner in the first place.   It's a complete clause by itself, however, just as _pramake_ is (containing free variables, though).

Useful info, but have I missed in it an answer to my question? AIUI (& I can be slow on the uptake), {my srana mi mlatu zo'u my xekri} = {my srana mi mlatu zo'u xekri}, with first argument of xekri implicit and bound by {my srana mi mlatu}. What is the rule governing binding of implicit arguments? 

 

(Side note: {vau} is needed to complete the sentence because we will want a way to delimit sentences as well as subordinate clauses, and I simply prefer mandatory information-bearing terminators to elidable meaningless terminators (and marking aspect is like an anchor IMHO).  

I would have the same preference between these two, but I would be wanting to seek a terminatorless solution, such as predicate-last clauses, or head-marking on the predicate, or (the more typical Neoliv move) marking the dependents for finality/nonfinality (like Lojban {be, bei}, IIRC). For something in the style of MLL3, I think I'd have expected predicate-last clauses. Ah, but then you lose your way of showing whether the first arg is bound from the prenex -- I see.

I am not sure that a loglang can avoid terminators entirely, but it's possible to get rid of some, and give others a useful role beyond merely showing structure.  Among the ones that seem unavoidable to me: the optional but sometimes needed left and right "parentheses" of conjunctions; something to mark the end of a nested clause (in MLL3, this goes on the left opposite the verb, while in Toaq and Morneau's Latejami, it goes on the right.  (Some people are surprised to learn that Latejami had terminators, but it did.)

It certainly can avoid terminators entirely, tho at the cost of alternative devices. In the case of a conjunction with potentially indefinitely many conjuncts, the conjuncts can be marked for initial--noninitial or final--nonfinal. In Livagian syntax, nodes form a binary or unary branching tree and the nodes are marked for whether the mother precedes or follows and for how many daughters it has and whether they precede or follow -- that marking suffices to recover the tree structure. Again there's a contrast between the natlangy strategy and the unnatlangy (Livagian), tho terminators in the sense of something the marks the end of a phrase whose other end is also marked, are not really natlangy -- it's fairer to say that terminators are an unnatlangy solution to disambiguation under a natlangy approach, I think.

Looking forward to seeing further info on MLL3 -- is there more on it on the wiki already?

More is scattered on Discord, but I could move parts to the Wiki. The SSM strategy might be interesting to discuss.

Yes, let's. A useful distinction can be drawn between SSM strategies that operate over a sentence and those that operate on any arbitrary phonological string. Livagian is unambiguous when you parse the phonological string from the start of the sentence, but breaks completely if you start the sentence midway through, e.g. because the first part is missing; Livagian is not really designed for coping with use cases where the first part of sentences might be missing, but that also means that it might be hard to start reading a sentence from halfway through. Most SSM strategies I know of can cope fairly well with part of the sentence being missing (even if it remains syntactically unparsable).

--And.

[Mike S.]

Snipping most of the older conversation...

On Tue, Nov 23, 2021 at 5:02 PM And Rosta <and....@gmail.com> wrote:

There's one family of loglangs, which includes Lojban, Toaq and, I gather, MLL3, that has noun-phrasey things, and another family, which includes Livagian and Unlws, that doesn't. The former builds on the patterns of natlangs; the latter cleaves more closely to the underlying logical form,

I agree with your sorting of languages under this categorization.  I am not sure what you mean by "logical form" - is this a reference to Chomsky's idea of a sort of "Ur-sentence" that is subsequently modified through transformational processes into the sentence actually uttered? (Sorry if that is a mangling of the idea.)  I suppose that NPs could be thought to arise from movements of old-information-bearing predicates to phrases orbiting a central rump VP.  It's an interesting idea, though perhaps not the one you intend.

At any rate, I would say that although the patterns of natlangs arise largely out of the demands of convenience and utility, they are also ultimately constrained by logics that are only relatively recently coming to be appreciated -- plural logic being an example.

 

In Livagian, there are illocutionary predicates, and a predication not within the scope of any illocutionary predicate would be presupposed; i.e. 'presupposed' = outside scope of illocution.

I almost think we're pointing at the same thing from two different angles, though I am not quite sure.  Someday I would like to be able to skim through the Livagian reference grammar (don't feel hurried; I too have been taking my time).

 

Useful info, but have I missed in it an answer to my question? AIUI (& I can be slow on the uptake), {my srana mi mlatu zo'u my xekri} = {my srana mi mlatu zo'u xekri}, with first argument of xekri implicit and bound by {my srana mi mlatu}. What is the rule governing binding of implicit arguments? 

I misunderstood the question to be specifically about the implicit x1. Empty arguments in MLL3 are generally filled by unnamed unrestricted constants, somewhat similar to the Lojban {zo'e} rule.  I think it would be inapt to change {my srana mi mlatu zo'u **my** xekri} into {my srana mi mlatu zo'u **0 [=zo'e]** xekri}.  Firstly, the main point of assigning a constant symbol is to use it later (and the language is designed to make it easy to insert pronouns before or after any predicate--no argument-linker, à la Lojban's {be}, is needed for pronouns). Secondly, if you are seeking brevity, then you could have just as easily changed the {zo'u} to {fa} and kept the linkage explicit.  The sentence without a subject in fact would seem to signal to me that the cat is the topic but not the subject (perhaps: "As for the cat, [the tail of it] is black.").

 

It certainly can avoid terminators entirely, tho at the cost of alternative devices. In the case of a conjunction with potentially indefinitely many conjuncts, the conjuncts can be marked for initial--noninitial or final--nonfinal. In Livagian syntax, nodes form a binary or unary branching tree and the nodes are marked for whether the mother precedes or follows and for how many daughters it has and whether they precede or follow -- that marking suffices to recover the tree structure. Again there's a contrast between the natlangy strategy and the unnatlangy (Livagian), tho terminators in the sense of something the marks the end of a phrase whose other end is also marked, are not really natlangy -- it's fairer to say that terminators are an unnatlangy solution to disambiguation under a natlangy approach, I think.

Conjunctions: I have two suffixes, call them {ke} and {ke'e}, i.e. Lojban's parentheses.  {jeke} is optionally placed before a coordination containing {je}, and {jeke'e} optionally after.  I might have had this idea myself, but Morneau had it long before me.

Ultimately, our various contrivances to achieve total monoparsing across a whole human language can never be perfectly natural.  We loglangers might as well bite the bullet and admit it.  But it's worth trying to keep the unnaturalness to a minimum.

 

Looking forward to seeing further info on MLL3 -- is there more on it on the wiki already?

More is scattered on Discord, but I could move parts to the Wiki. The SSM strategy might be interesting to discuss.

Yes, let's. A useful distinction can be drawn between SSM strategies that operate over a sentence and those that operate on any arbitrary phonological string. Livagian is unambiguous when you parse the phonological string from the start of the sentence, but breaks completely if you start the sentence midway through, e.g. because the first part is missing; Livagian is not really designed for coping with use cases where the first part of sentences might be missing, but that also means that it might be hard to start reading a sentence from halfway through. Most SSM strategies I know of can cope fairly well with part of the sentence being missing (even if it remains syntactically unparsable).

--And.

This property that you are alluding to, regarding the ability to correctly parse a broken stream, is called self-synchronization.  (The UTF-8 encoding of Unicode is robustly self-synchronizing, if you want an example.  I think Ceqli and Toaq are probably among the best loglang examples.)  This property has been discussed in the LLL; it is a nice property to have, but I don't know of anyone who thinks it's as critical to loglanghood as the properties of monoparsing and monosemicity.

I will try to present the SSM (wordstream monoparsing) strategy of MLL3 in the next few days on the Wiki.

-Mike