Re: [engelang] Xorban: Termsets - engelang messages

On Mon, Oct 8, 2012 at 7:56 AM, And Rosta <and.rosta@hidden.email> wrote:

Mike S., On 05/10/2012 20:34:
> No one has really been using them, but that's to be expected given
> that we are all novice users at this point. Intuitively, I suspect
> that parentheticals will find great use among fluent speakers, e.g.
> as a way of adding interjectives, vocatives, topic switching,
> nonrestrictive clauses and similar things. All of these things would
> be useful right now if we were to move past the toy-sentence stage.

I agree, but each of these things can be considered in their own right. For example, interjections should be able to unproblematically occur anywhere, because they don't combine syntactically with anything. Nonrestrictive clauses could be signalled simply by a sentence-internal illocutionary-operator marking a formula as outside the scope of main illocutionary force. And so on.

All of that we already have under the BNF grammar.

>>> 2. It would be good to adopt a sounder less improvised formalism,
>>> separating syntax from phonology, and able to accommodate
>>> phonologically-null and semantically-interpreted syntax. What we
>>> have at present is a constituent structure imposed on a
>>> phonological string -- everything wrong with Lojban pseudogrammar
>>> is wrong with this, albeit to a far lesser degree.
>>
>>
>> I am not sure how separating phonology (which is really just the
>> segment-terminals in these rules) from syntax would help anything,
>
> It would remove cruft and avoid the delusion that syntactic terminal
> nodes consist of phonological material. If there happened to be
> perfect homomorphism between syntax and phonology, the annoyance
> would be purely conceptual, but given the goal of usability and
> brevity, there will not be this homomorphism; & we already are in
> agreement about this.
>
> You may have to jar my memory -- can you give an illustration
> involving a hypothetical design in which it would be desirable for
> that homomorphism not to hold?

The homomorphism affords simplicity of rules, but as a design criterion, concision is far more important than simplicity of rules -- I think we agree on that. We also agree that there may be elements in logical form that are implicit, i.e. present in logical form but not in phonological form.

Such elements clearly exist, e.g. the implicit restrictions for "a'a" and "e'e". I'd handle them in the formal semantics though, not the grammar.

> Jorge's grammars have been complete formal grammars under this
> definition.

I think this is a good characterization of the difference between where you and I are coming from.

It's my impression that formal grammars are a staple of introductions to Computer Science and are considered important by Computer Scientists. But they don't feature in introductions to linguistics, or anywhere in linguistics apart from its computational corners, not because linguistics doesn't like formalisms but because it's inapplicable to language -- that is, it's not suited for application to language, because language doesn't work like that; "formal grammar" in the sense under discussion seems, like Chomsky's 1950s work that is foundational to it, a naive, crude and erroneous attempt to model language formally (-- as any initial attempt at the birth of the discipline is likely to be). I have met linguists with a kind of 1950s-ish view of the mechanics of language, but they were rather unreflective by nature, or worked in other subdisciplines.

I see now that there are two ways of making a loglang.

(I) You start with the notion of "formal grammar" and create a set of rules that implement it.

(II) You consider how language works, and create a loglang version of it.

It had never occurred to me that anybody would have (I) as a goal, and hence when I saw evidence of (I), e.g. in Lojban, I had attributed it to naivety and ignorance, on the assumption that (I) would merely be a misguided attempt at (II). I suppose the attraction of aiming for (I) in the first place is that it is well-understood and readily formalizable, but I don't know how it would avoid the garbage-in-garbage-out problem (that is, is there virtue in formalized garbage?).

I am not sure by what reasoning you have concluded that there is a profound and inherent dichotomy between (I) and (II). My approach, and I might have hoped that it would be obvious by now, holds (I) and (II) equally indispensable. Although my CS educated / language-hobbyist background is probably obvious to everyone, and I am indeed capable of looking at Xorban as a set of strings mechanically produced by BNF rules when it's apt to do so, nevertheless I also have felt no qualms about calling (with appropriate qualifications) "mlta" occurring in a restriction a noun, "je xkra" quite similar to an adjective, "ju glka" quite similar to an adverb, and "ltpkcma" occurring in a predication a verb. Despite being misguided by the BNF grammar, I have been drawing attention to these connections between ostensible FOPL constructions and garden-variety natural language constructions all along, and frankly, I am delighted to informally -- and hopefully plausibly -- map Xorban structures to natural language ones (as I see them in Latin, English, German, etc.) because it strengthens the idea that the gulf between formal logic and natural language is not as unbridgeable as it first might have seemed, and that a humanly speakable loglang is truly possible. It appears to me that you wish to poohpooh these seemingly straightforward but presumably "naive" connections, coming up with an (IMHO) intuition-defying analysis that posits a binder like "la" as the head of a sentence. I don't shake my head at that because I think there's anything wrong with looking at things from a different point of view. It just seems weirdly incongruous to me to rail against the "formalized garbage" of BNF production rules and invoke "how language works" one moment and then the next moment take positions on the language so abstract, (underlyingly) mathematically formal, and metalinguistic, rather than naturalistic.

Under approach (II), syntax would be a structure of Argument and Binding relations, with lexical and inflectional rules mapping these structures to the sentential phonological form and with semantic rules (i) defining semantic equivalences between syntactic structures (-- something restricted to loglangs, I think) and (ii) defining semantic interpretation (truth-conditions, etc.). And above the word-level, phonological structure would probably be a mere concatenation of phonological words, certainly without any of the elaborately (and pointlessly) patterned structure of Lojban and its less egregious Xorban counterparts.

Does your objection to the use of BNF boil down to anything more than not being able to handle things like binding rules? Would a modification of BNF be acceptable if it could effectively bar from the language (mostly) silly things like bindings that don't bind anything?

In the rest of what I say, I am supposing Xorban to be taking Approach (II), but if anyone wanted to take Approach (I) instead, I guess we could take a step back, agree to disagree in principle, maybe work on parallel projects, and then look to see where we have common ground.

There is no harm in anyone at any time presenting a variant description of the language as that person understands it or visualizes it. I plan on fleshing out my own pages a little more. It's positively helpful for everyone to get a snapshot of the language above and beyond this ever growing thread tangle on Engelang. Your snapshot would be welcome if you created it.

> But we can't intelligently use that as the basis for serious
> discussion. Jorge's sketch is innocuous because the rules of syntax
> are so simple that one grocks them quite independently of the BNF
> stuff. But for more complicated proposals it simply won't do. As
> formulated, these termset rules merely impose pointless patterns on
> phonological strings, & it's somewhat shameful that we are so
> blithely recapitulating the inanities of Lojban. If the syntactic
> rules don't generate logical formulas then throw them away; they're
> junk.
>
> If the grammar for formulas without termsets (i.e. trees) generate
> logical formulas, then assuming no mistakes have been made I assure
> you that the proposed grammar for formulas with termtrees also
> generate logical formulas, because every formula that contains
> termtrees can be interpreted unambiguously as a rewriting of a
> certain formula that does not contain termtrees.

I'm certain that in your head there exists a real functioning grammar that generates formulas. What I was objecting to was the current explicit formulation, which doesn't. The current formulation appears to divide the phonological form of a sentence into labelled parts, which looks like junk to me, and says nothing at all about the essential stuff, which is the Argument and Binding relations.

To the extent that I glork that, that's all part of the semantic module, which indeed exists mostly in my head (and others' heads) at this point. That can be remedied but it'll take time.

> The language with termtrees is a superset of the language without
> them, but under the accompanying proposed semantics, the language
> with termtrees has precisely zero greater expressive power than the
> language without them. It has merely greater abbreviatory power.

Yes, I understand this.

> There was a fair deal of work over two weeks coming up with a design
> that is grammatically unambiguous. Given the skills that I have, I
> would not have been able to contribute towards a meaningful result
> without relying on the formalism of BNF or something similar.

That's fine. I don't object to BNF as an ad hoc aid for thinking or recording ideas. But if we actually had to start discussing Xorban syntax seriously, which I think the complications added by the termtrees mean we do, we should translate the BNF stuff into the real rules that generate the real structure of Argument and Binding relations.

Yes, the trees add complexity to the syntax, but they're nothing to worry about wrt semantics because every arborized formula corresponds to just one core formula. All we'd need to do is get the core formulas defined in the way that you consider proper; it will follow demonstrably that the arborized formulas are also so defined.

> The termset proposals themselves seem to me to be essentially a good
> idea - a useful abbreviatory device. I mean not to criticize the
> attempt to find workable abbreviatory devices, but only to criticize
> the idea that these BNF rules have merit. The ideas that the BNF
> rules are a misguided attempt to formalize surely do have merit, tho
> until I understand them I can't comment on their adequacy.
>
> The way I see it: Xorban is founded on FOPL, or is possibly a modest
> extention of FOPL;

I agree with that.

> the language of FOPL is aptly defined by BNF. Therefore, Xorban is
> aptly defined by BNF.

My only objection to BNF per se is the weak objection that it is too unrestrictive. You can use BNF to write good rules and to write bad ones. I can think of vaguely how to do FOPL in BNF with two orthogonal trees, one for Argument structure and one for Binding structure, on the presumption that there is some version of BNF neutral with regard to the linear ordering of parts, and on the presumption that mere use of BNF doesn't imply confusion between phonology and syntax, tho I can't see how to tie the trees together. Doing something similar for Core Xorban, for the non-binding tree, we'd want something like:
complement := argument-terminal | phrase
phrase := head complement*
[where the number of complements varies according to the identity of the Head]. And for binding, we'd want:
Binding := binder bindee*
But I can't see how to capture the fact that a binder is a head and its bindees are contained within its complement. "*Aptly* described by BNF" seems like a bit of a stretch, for FOPL and for Xorban: "Described by BNF clumsily or perhaps not at all" seems closer to the mark, tho I'm basing that only on my own half-arsed attempts.

There is no simple way of accomplishing using BNF what you wish to accomplish wrt binding rules, and that is no fault of your attempts here. However, what you're saying about FOPL is erroneous; FOPL does allow vacuous binding and while pointless vacuous binding does no harm to that language's semantics. Whether Xorban should follow suit is a matter of opinion. Cowan and Jorge both seem to say yes; Clifford (I think) and myself would probably impose an interpretation rule by extending the restriction and possibly the predication with something like "je psa" or "je sma". You would obviously declare it ungrammatical. To be honest I am wavering on this question. Cowan makes a case that keeping Xorban and FOPL isomorphic requires permitting vacuous binding. I really can't deny that.

> Nevertheless, I am not opposed to exploring other formalisms. I
> definitely will revisit your "reformulating" thread if you continue
> fleshing out your formal description. I am very interested in what
> you come up with.

I think I'd done the rules for core Xorban. Possibly the binding rules are incomplete: I can't remember whether the rules I gave said, as they probably need to, that is X is subordinate to Y and Z is subordinate to X and Y binds Z but not X, then the form of X must be differently inflected from the forms of Y and Z.

The only real options are to put one's ideas together in one place on the Web, or let them slip into obscurity.

> One thing that you have made clear to me: In the long run, we will
> need more intuitive and informal ways of describing the language to
> our fellow humans than BNF. Plentiful illustrative usage examples,
> which I apologize for being short on, are a must. The formal
> definition is just that, a formal definition.

But hopefully the formal description describes the actual mechanics of the syntax too.

> No one is meant to think in it when they use the language.

Given the goal of being ergonomic, I think it would be good to base the formal definition on what speakers are in fact meant to think in when they use the language. Well, I don't mean go all cognitive and mentalist, but at least base the formal definition on something consistent with a very simple incremental parsing algorithm.

--And.