Sunday, January 31, 2016

Schools of artlanging

Its [conlanging's] development to perfection must none the less certainly be prevented by its solitariness, the lack of interchange, open rivalry, study or imitation of others' technique.
"A Secret Vice", J.R.R. Tolkien, circa 1931.
The important thing is that conlanging start to have a critical apparatus within which the artistic merits of conlangs can be evaluated and where different schools of thought can define and defend themselves. [...] I therefore announce the founding of the Naturalist school of conlanging, which regards the following three things as values: [...]
— "Lighting Some Flames: Towards conlang artistry" (commonly known as "The Artlanger's Rant"), Jesse Bangs, Conlang Mailing List, March 12, 2002.
If naturalistic conlanging is the equivalent of realist painting, then what is an impressionist conlang?  A surrealist conlang?  What's the conlang equivalent of Guernica?  Who is the conlanging equivalent of Gauguin or Dalí?
The Art of Language Invention (Penguin Books, 2015), David J. Peterson, p. 264.

I've some things to say here about conlangs as works of art, with particular attention to the relationship between artist and audience.

The special occasion for this is that I've recently read David J. Peterson's new book The Art of Language Invention.  It's kind of mind-boggling to realize how far conlanging has come since the term was coined 25 years ago, as isolated practitioners of Tolkien's "secret vice" found each other through the internet and began to draw together into a community.  Peterson, of course, has played a central part in the conlanging community during its emergence from the shadows; he was involved in the establishment of the annual Language Creation Conference series and the Language Creation Society, and is probably most broadly recognized atm as the architect of the Dothraki language for the HBO series Game of Thrones.

Artlang appreciation
Small stuff
Artlang appreciation

Early in the Postscript of his book, Peterson asks "how an audience appreciates a conlang" (p. 260).  Seems to me there are two ways:  by observing its use — spoken or written — and by studying its description.  (Descriptions can include examples of use, of course, and it seems awfully common for critiques of conlang descriptions to either praise the plentiful examples or complain that there aren't enough of them.)  Peterson expresses doubts about studying a reference grammar, on grounds that doing so takes an awful lot of work.  On the other hand, against the former method, way back in his Introduction (p. 6) he suggests that if the actors made a mistake speaking a conlang in a TV program, only about a tenth of a percent of viewers would notice.

I think the one-in-a-thousand number gives something of a false impression of how much benefit works of fiction can accrue from using conlangs.  The number may be in the right ballpark, but all Peterson actually claimed for it was that it was how many viewers consciously notice a specific mistake in using a conlang.  One might ask, instead, how many viewers would subliminally notice if the alien (in its broadest sense) speech were faked with nonsense rather than using a conlang.

Granted, the benefits of using a conlang may be amplified in written fiction, because there alien speech appears in a very stable form inviting close and protracted scrutiny (with the complicating factor that it'll almost certainly have to appear in a romanized form).  For alien speech/writing in a TV or movie setting, though, the more of it there is the harder it is to produce an air of authenticity if there's really no language behind it.  (Peterson does allude to this effect in passing on page 2.)

Much has been made of the use, in Star Wars: The Force Awakens, of physical models in preference to computer-generated imagery (CGI).  I recall one of the actors remarking about this (all part of the pre-release publicity, of course) that it's possible to give a more authentic performance when interacting with a model that's really there than when pretending to interact in a blue room with something that isn't there.  The difference, both indirectly through the actors' performances and directly through the feel of the set, is meant to make the movie seem real to an audience subliminally.  I think the same principle applies to use of conlangs:  for a truly authentic feel there has to be a language.

Studying a reference grammar is a lot of work.  One thing we should be asking, in that regard, is how to make descriptions of language more accessible.  In fact it's not at all clear how a language ought to be described even for linguistic purposes, a shortcoming likely to become gradually clearer to a conlanger as they get into the more advanced levels of the subject.  For me the first major warning sign came with studying "The Blue Bird of Ergativity" (which I was set onto by the Conlangery Podcast) — a 2005 linguistics paper arguing, rather convincingly imho, that ergativity isn't really a thing, in the same way that blueness isn't a thing for birds:  a completely superficial property with no biological significance.  Before long I'd lost faith in things like the dative case.  Keep in mind, natural language translation has been just a few years away for the past half century; it's a dreadfully hard problem, which seems to require a thinking mind who actually understands the text being translated and is fluent in both languages.  Describing a language seems unlikely to be less difficult than translation.

All of this seems to me to imply another, deeper question, though.  From How do we describe a language? we're led to What does it mean to construct a language?, and then we find that the real starting point should be What is a language?.  A living natlang lives in the collective understanding of its speakers.  An extinct natlang is presumably defined by the past collective understanding of its speakers, which we try to reconstruct.  For a conlang?  There's a good chance a conlang mightn't be spoken fluently even by its creator; so either the conlang is its description, or the conlang is some sort of natural resonance point that its description tries to lead us to.

Small stuff

One early passage in the book stirred a minor pet peeve of mine.  On page 34, expanding on a remark that IPA represents sound independent of spelling, he goes off on a several-sentence rant about the absurdity of English spelling.  As a fluent speaker and writer of English, I find his ridicule of English spelling perfectly reasonable (and I enjoyed the way he said it).  What's striking to me, though, is that he has nothing similar to say, afaics, against IPA (the International Phonetic Alphabet).  Presumably because he's a very highly trained linguist.  It's possible to get so used to almost anything, that one stops noticing its oddness; and it seems as if Peterson may have done so with the IPA.  Avoiding that fate myself, I've retained my awareness that the IPA is utterly insane.

There is, of course, no likely cure for the IPA, for pretty much the same reason there's no likely cure for English spelling.  The ASCII-only alternative to the IPA, X-SAMPA (with its variant CXS), has the advantage of being ASCII-only and essentially nothing else to recommend it.  I exaggerate less than one might think.

The prospects for getting an alternative adopted certainly wouldn't stop a conlanger from trying to design such a thing.  So far my own efforts aren't worth blogging about.  If they ever reach the level of being interesting, perhaps I will blog about them.  In order to actually be useful for presenting conlangs, they'd have to be significantly easier than IPA for a non-linguist to understand, which seems a very high bar indeed given the challenge a non-linguist will face no matter what the notation (but this high bar also offers some small degree of guidance for efforts in that direction).

Another low-level point that bugs me is glossing, where an example of the conlang is juxtaposed with its translation on a separate line, and between them is a gloss explaining the features of each conlang element.  Here I think the bar for improving utility is considerably lower.  Glossing, like the examples it's attached to, is encouraged by conlang critiques.  I used glossing myself in my blog post a while back on Lamlosuo, e.g.

 losutu   li   susua 
the sometime-speaker sleeps
The trouble here may even be clear to a linguist (whereas a linguist probably wouldn't see the problem if the language being glossed were less peculiar).  What do those abbreviations stand for?  Hint:  NEUT does stand for neutral, but V doesn't stand for verb, and SUB doesn't stand for subject.  Even I do double-takes on STR, which in the circles I move in would ordinarily stand for string (but not in this gloss).  Linguistics documents tend to have a list somewhere of abbreviations used; and that's awful for conlanging, because memorizing those stupid abbreviations is one more obstacle to learning about the language, which is the end goal.  Your target audience for a conlang description isn't limited to professional linguists who consider learning a whole new system of analysis for this new language a more-or-less normal part of their profession.  I actually regretted, after the fact, having indulged common practice by introducing gloss abbreviations; the above should probably have been something more like
 losutu   li   susua 
the sometime-speaker sleeps

It would also be great for non-linguists if everything had sound files attached so you could hear what all the examples are supposed to sound like; but that is so difficult (it's not uncommon to create a conlang you can't yourself pronounce) one never hears complaints when it isn't done — just praise when it is well done.


Almost all conlangs have reference grammars; it's hard to define a conlang without a grammar, although I recall hearing of a collaborative project in which people proposed sentences in the language and the community voted on which ones to allow.  However, defining a conlang via a reference grammar has, in my experience fwiw, three major drawbacks to keep in mind.

First, grammars as used for natlangs are, or should be, descriptive rather than prescriptive.  The classificatory schemes of such grammars are analytic rather than synthetic:  even though a reference grammar is meant to describe the language (whereas a pedagogical grammar teaches its use), still it seeks to tell how the language works rather than why.  This is close to the point mentioned earlier about the superficiality of ergativity.  If such classifications are indeed superficial, someone carefully studying a natlang based on the reference grammar may still manage to glom on to the deeper nature of the language by using the grammar as a jumping-off point; but a conlang defined thus superficially may actually lack that deeper nature that its author, if pursuing naturalism, would like to have.

Second, grammars may tend to make languages seem much more orderly than they are in practice, especially spoken practice.  Anyone who has transcribed an oral presentation is aware of this:  we speak haltingly, with all sorts of false starts, typically with lots of low-level discourse particles apparently serving to help the listener navigate through the speaker's on-the-fly revisions in assembling their syntax (in English, um and er might be used for this, varying by dialect).  Perhaps this would be mitigated in fiction, both written and on-screen, because dialog would tend to be somewhat unrealistically orderly anyway for the sake of exposition; but even so, fully explaining a paragraph of fluently written English (or whatever is the primary natlang of the work) tends to involve some remarkably esoteric grammatical gymnastics.

Third and likely most problematic, grammars tend to describe things in terms of a standard set of principles for organizing sentences and clauses.  Why is that a problem?  Because very often the organization of sentences and clauses isn't the point, but merely a tool used in service of some other purpose — often a purpose that can only be properly appreciated by studying longer narratives and discourses.

Consider verb–argument alignment.  The point, apparently, is to specify which arguments to a verb play what part in the verb.  Most commonly you have nominative-accusative alignment, where the agent of a transitive verb is marked the same as the subject of an intransitive verb, while the transitive patient is marked differently.  Alternatively there's ergative-absolutive alignment, where transitive patient is marked as intransitive subject while transitive agent is marked differently.  Why this difference matters may vary from language to language, but often has partly to do with the way verbs are valency-reduced — as with the passive voice in nominative-accusative languages, anti-passive in ergative-absolutive — when fitting together larger narratives.  But then there's Tagalog, the dominant language of the Philippines.  Tagalog uses Austronesian alignment, which means there's marking on the verb indicating what system is used to mark the arguments, with a variety of options available — and the reason for doing so has to do with specificity.  That's right, specificity.  How specific a noun is.  Don't expect to get a really good explanation of how this works in Tagalog, either, because as best I can tell, as of now the linguistic community hasn't altogether figured it out yet.

What does seem clear to me, though, is that alignment in Tagalog is a particular instance where understanding what the feature is being used for is essential to making sense out of it (linguistically, at least; there's no need to consciously understand it in order to use it, otherwise Tagalog native speakers would simply explain it to the linguists and it would all be settled).  That is, figuring out which argument does what in the verb isn't really the "point" of alignment, it's just the mechanical act we use to delineate which facets of the language we've chosen to put in the box labeled "alignment".  I suspect some degree of the same effect applies to many of these seemingly low-level language features:  their apparent mechanical operation isn't necessarily their primary purpose, and their primary purpose may be impossible to see clearly until one gets up to the narrative scale.  (Appeal to the narrative scale is a recurring theme in the Conlangery Podcast, btw, even though they've never actually devoted a podcast to it, as such.)

What we're facing here is the difference between the "universal" — Chomskyist — approach to grammar, and the "functional" approach.  In essence, the universal approach supposes that grammar is hardwired into the human brain, so that the shape of language is determined by that wiring, whereas the functional approach supposes that grammar is shaped by the functions people want language to serve.  Whatever you think of the philosophies underlying these positions, there's a bunch of things going on in language that tend to get missed by the usual grammatical approach, and the functional approach seems like it ought to provide those missing elements.  Except, apparently it doesn't.  Functional grammar advocates tend to take the principle to its extreme and look at everything from the narrative perspective.  Whatever you think of that philosophically, it doesn't seem very useful for conlanging since it doesn't provide local structure — less a problem for functional grammarians than conlangers, thanks again to the descriptive/prescriptive distinction.

Ideal for conlanging — or so it seems to this conlanger — would be a modified variant form of traditional grammars, with comparable degree of local structure but free of traditional assumptions about purpose and mechanical classification, avoiding implicit bias toward conjectured universal restrictions (conlangs don't have to obey any rules hardwired into the human brain, if there are any, and some conlangs deliberately try not to);  and with high-level functionality integrated into it in a way that naturally meshes with the more low-level-mechanical facilities.  Both universal and functional grammarians would seem to have philosophical reasons for not trying to construct a grammar like that; but conlangers should have no such disincentive.


It's easy to write software that forces its human users to fit what they're doing into a rigid structure.  Doing so can have negative consequences for society, as it reduces users' opportunities to do what human sapience is actually good at; but to the current point, limiting the user's options is definitely counterproductive when what you're trying to support is creative artlanging.  It's not so easy to write software that supports the best of what humans can do (creative thinking) together with the best of what computers can do (rote memory and purely mechanical manipulation).  That sort of software must by nature be general-purpose and inherently flexible.  Two of the best artifacts we have so far in this genre — TeX and wikis — are both plausible candidates to support conlanging, but they both have flaws for the purpose.  In brief, TeX is overly technical, wikis are computationally weak.

What we need is the sort of simple basic elements one finds in a wiki;  with TeX-like computational power available at need — if the opportunity arises I'd like to try a fexpr-based extension strategy rather than the macro-based strategy shared by TeX and wikis;  and an interactive element with enough power to let the computer provide advice and mechanical aid while aggressively preserving the user's flexibility of form as well as substance.  Is all that even possible?  I suspect so.  The limiting factor isn't computational power (we've likely had enough of that since before the word "conlang" was coined) or brute force programming power (throwing more people at a project makes it take longer), but finessing the design of the software.  I've got a few lines I'm pursuing, some of which I've been pursuing for decades; note that I've named just two artifacts distributed across a half century of development, an average of less than one every other decade.  Patience is indicated.


Jesse Bangs's notion in the Artlanger's Rant was that Naturalism — which he defined as valuing a naturalistic, complex, creative conlang — would be a "school" of conlanging.  There's some history there.  The conlanging community that had formed around the CONLANG mailing list in 1991 underwent a schism in 1993, with a subgroup leaving to form the AUXLANG mailing list.  You can't very well use an auxlang to unite the world if you can't get everyone to agree on which auxlang is best for that purpose, so it seems natural that the auxlangers were inclined to advocate their favorite auxlang; after the schism, the CONLANG mailing list adopted a policy against language advocacy (though afaict auxlangers who respect the advocacy ban are more encouraged than required to use the other list).  When the Artlanger's Rant was written, that schism was still recent community history, and thinking of conlanging in terms of that sort of big goal would have been pretty natural (irony not intended).

I suspect Bangs's proposed definition of a Naturalist school may be a touch heavy on goals and light on means.  He does mention complexity/completeness, which is about content; and creativity, i.e. not imitating your own native language, which is sort of a technical criterion.  But I'm thinking of the nature of the description; like, the form of the reference grammar.  If you're trying for something that evokes naturalism using a very traditional and rather sketchy grammar, might you end up with a sort-of cubist conlang?  It seems that might modify the notion of completeness, shifting emphasis to a more abstract side of naturalism; not sure quite how that would work, but the nature of cubism suggests something of the sort; while on the purely technical side it might give a sense of simplification.  Perhaps some other non-traditional form of grammar would produce a sort-of impressionist conlang.  The goal of course is to nurture the conlangers' art, not to imitate schools of painting; but I do take from these pseudo-examples an encouragement to explore different forms of grammatical description.

What about the next part of Peterson's question?  Can a conlang convey a powerful message like that of Guernica?  Not to overplay the analogy, but can a conlang come across to its audience as anti-war?  Can it carry profound ideas, or strong emotions?  By what vector?  The mere phonaesthetics of a conlang surely can't carry so much, though it can presumably contribute to it.  Perhaps more can be be carried, for those who learn about them, by the vocabulary of the language, and the ideas/attitudes/ways of thinking that the language fosters.  Improved lucidity of language description for non-linguists would seem to help here; and ideas/attitudes might be affected less by traditional, relatively low-level mechanical considerations than by the sort of narrative-level things that functional linguists emphasize.  At any rate, I suspect that making some headway on the first round of challenges I've been discussing here will put us in a better position to tackle this second round of more abstract, expressive challenges.