saizai

I've now made a Slack for it: unlws.slack.com. Invite link: https://join.slack.com/t/ulws/shared_invite/zt-ewevpfmf-s1r3...

There is now: https://join.slack.com/t/ulws/shared_invite/zt-ewevpfmf-s1r3...

Thanks! People are complex. It's my personal site, not a topical one, so it likewise has on it basically whatever I think would be interesting to put up. ;)

I'd welcome any feedback you have on the rest. Feel free to contact me directly.

We have absolutely no interest in wide-scale usage, let alone if it were to come at the expense of fully exploring what a natively-2D language is like.

Narrative is an utterly different thing in UNLWS. We simply do not have "stories" in the same sense as you're used to. The closest so far is that, rather than the reward/punchline/moral/etc coming from the ordering of things, it comes from understanding the whole, i.e. as an emergent effect of the gestalt.

We don't try to have things that aren't native to 2D. We are, if necessary, able to describe something that happens on a linear timeline - that's pretty easy actually - but it simply isn't the "normal" way to arrange a large utterance. We just aren't all that interested in temporal order. We'd be more likely to represent the underlying structural order, or causality networks.

We do have a quotative, and we've talked about using IPA if we ever needed to literally refer to phonetics for some reason, but so far we simply haven't needed or wanted to do so. (Not for names, either. Names are assigned in a manner that's native to UNLWS; we don't care at all about the phonetic form in another language. Consider ASL, which does this partially [for people who do have sign-names, and under extremely heavy contact with English as a dominant language].)

We never borrow foreign words phonetically. That is just as alien to UNLWS as assigning Chinese hanzi to a name based on their meanings or stroke #s would be to an English-only speaker. We only borrow concepts, and that's only after (sometimes radically) rethinking what those concepts actually are, underlyingly, and what an UNLWS-native perspective of them would entail.

We want to explore two-dimensional language in its own right. Although we do care about expressive power, we really don't care whether it's easily translatable to/from any other languages.

I have two concerns with that, above the (major) technical challenges: 1. we wanted this to be writable using colored pens 2. if we add a dimension, we must add it for real.

#2 is to me the vastly harder problem. Just like in UNLWS, we always have to challenge our own assumptions in order to get a sense for what would be "native to" a two-dimensional written langauge, we'd have to do the same for 3D (or 2.5D). Merely slapping 2D planes together with some links would not come anywhere even close to a robust use of the medium, just like merely having English sentences branch off from each other isn't anywhere close to a robust use of two-dimensionality.

Fully 2D language is already an extremely conceptually challenging problem, at least to me. Hardly anyone has even tried. There are fundamental challenges, like "how do you tell a joke when you don't control the ordering".

I don't think anyone is currently able to do 3D in a way that would truly serve the medium - and the medium would need to be much better defined, since humans' inability to actually see anything 3D instantaneously (we see 2D with an imputed distance and assumptions about what the rest looks like) implies that there must be interaction-in-time of some sort (like, at minimum, moving your head or the object in order to see the whole thing). That's going to be very specific to the medium. Are we talking about arbitrary 3D disconnected polytopes, ants' nest casts, convex hulls, computer-manipulated 2D, VR goggles, ...? Those all have radically different affordances.

Simultaneously, it would trivialize problems in 2D that we have to deal with. E.g. in 3D, there's no circuit wire-crossing problem, whereas in 2D, Borromean rings e.g. have unavoidable collisions. It would be cheating to use 3D in some shallow way to get rid of that issue, without also dealing with 3D's own problems.

Design constraints are a good thing.

I believe that *Pad¹ does what you describe. In multi-user mode, it has an option to show each contributor's writing in a separate color. (I'm not sure if one can pick the color, though.)

¹ eg https://cryptpad.piratenpartei.de/pad/

PS I really like the garden vs stream metaphor for UNLWS-native "conversation". I agree that it's very apt. It nicely captures the spirit of non-linear interaction.

Thank you for this new way of looking at / describing it. We may borrow it for the UNLWS documentation. ;)

Interesting. I've long been interested in ways to make systems more pro-social, but I hadn't previously thought of UNLWS in that context. Thank you for giving me a new idea to mull.

Thanks for reporting the broken link. Alex has fixed it in the document. The content is now at https://000024.org/conlang/relay19/index.html

Google "conlang relay" to learn more about the context for which this was made.

Another way to express the latter is that if something becomes too popular, upper socioeconomic classes will want to adopt something else, so that they can preserve a class distinction of "higher register" speech. Meanwhile, lower ones will try to adopt the style of the higher register, since it's perceived better. That's an endless cycle.

Same thing for euphemisms, insults, etc. And uniformity was far less common before mass media and centralized schooling…

Linguists have a much more descriptivist way of looking at language use and change.

Nuance is not a feature of education; focus is. E.g. AAVE (AKA Ebonics) is currently considered a lower class register, but it has grammatical distinctions that standard English doesn't (like whether an action is part of a pattern, vs a one off).

ETA: also, an influx of non-native speakers tends to result in borrowing of both words and grammar from their native languages. Not just in the development of creoles. What's "correct" is purely defined by consensus; if there are lots of new speakers, that's of course going to shift towards their preferences.

Can we even distinguish those in practice? AFAICT its main problem was being so general-purpose that it was hard to describe as a compelling use case.

Of course, I also don't subscribe to the belief that the popularity or memetic prowess of an idea defines its "success" except commercially. I thought it was a very nifty thing. To me, that counts as a success. Alas, that is not the metric used by companies. :p

D'aww. /blush

That seems to me to be more like multithreaded linear than non-linear per se.

There are multiple linear components, and it's not obvious how they emergently interact — but they don't quite have the interconnectedness and non-directedness properties.

(I think that it's impossible for anything that's time-embedded to really be non-linear, short of time travel, since time is so fundamentally a directed linear vector.)

It's certainly a related concept, and these are aspects of non-linearity, but to my view it lacks some essential Buddha-nature of the full thing.

Actually IIRC, that was our inspiration for it. ;)

You're describing Google Wave¹, which (sadly) was killed as unsuccessful.

¹ https://en.wikipedia.org/wiki/Google_Wave

… or because of pure happenstance, i.e. it just didn't come up?

Extant natural languages (including dead ones) do not come anywhere close to coming the entire ambit of what could happen in a language.

My interest as a conlanger is precisely that — exploring the boundaries of what language can be, not merely what it is.

As for expressiveness, I think that, within the things we've gotten to, it's just a expressive as English — and actually more so, for the things that are natural to a non-linear or recursive way of thinking.

Obviously we haven't gotten to everything. The lexicon, and grammar, are nowhere near complete. So, yes, we'll need to make lots more to cover everything.

Though also, we deliberately don't make every distinction English does, and we make distinctions English doesn't. It's a very noob move to make a conlang lexically (or even conceptually) 1:1 with a natlang.

It took us several months to figure out a non-linearly-natural way to express "if", for example. A noob would have just chucked in a new glyph and called it a day. We had to rethink fundamentals (like the scoping of irrealis); the result, IMO, is far more interesting.

As for optimality, that has nearly nothing to do with how natlangs spread. You learn a language because you need to — economically, socially, or militarily — not because it's better.

And how they evolve also has almost nothing to do with optimality in any hackery sense. That's an entirely different, stochastic process. It has some tendencies towards eg simplicity, regularity, etc, but just as many against.

In any case, the result is that natural languages don't have teleology; they don't become "better" over time, just different.

FWIW, we deliberately decided to have no phonetic representation at all, to avoid any linearizing influence.

Yes, it would need a large number of glyphs. Not so many as speculated above, though — e.g. we have one glyph that covers communicate, say, tell, hear, speech (all senses), etc. (with some mix-ins for sensory modality); another one that covers give, receive, gift, sell, etc.

Linear languages with European structure need way more words, because the syntax isn't very able to represent the shared concept between these with a single word. We can.

And yes, common structures do get turned into glyphs, or reduced forms.

We also have a barely-explored concept about glyphs being fractal (i.e. if you zoom in, it's actually the shape of an underlying utterance whose syntax "draws" the higher order glyph). This presumably would be a sort of literary or poetic form. (I have composed one very simple example.¹)

¹ https://s.ai/poetry/distant_love

Anguilla permits them. I had to buy off a squatter.

But Sai is literally my full name (I'm mononymous), at the time it was affordable to buy off the squatter, and the per year registration fee isn't too high. So I think it's certainly been worth it.

Not really. Programming is very linear in all forms I know¹.

1. There's a start point. UNLWS has none.

2. There's a mandatory canonical traversal order of the text (the compiled instruction pointer, at base). UNLWS has none.

3. Structurally — i.e. even ignoring execution — it's mainly a tree, not a graph (parent class > child class > instance > function, e.g.). There are some parts that are directed graphs, like mix-ins and some RDBs. True multigraph structure is fairly rare. UNLWS is pervasively a fully connected multigraph with some directed edges (like irrealis).

…

Programming is, however, similar to UNLWS in that both are isomorphic to predicate calculus (e.g. functions, like UNLWS glyphs, are n-ary predicates). I think that's what you're probably catching on.

It is an important feature… but I think you're missing the more essential nature of non-linearity.

(It took me several years to grasp enough to start making UNLWS, so don't feel too chastised. It's a very difficult concept. :p)

I believe that complex circuits (as in wiring) and flowcharts do have the non-linear nature, though. Perhaps it'd help to think of those?

¹ I don't know functional programming or constraint satisfaction systems. It's possible those are different.

UNLWS can definitely be described as being based on "predicate calculus" (as in formal logic). I personally have never felt it to be a very apt or helpful description — I think of it more in terms of Lakoff-style "frames". However, every glyph in UNLWS, including the all-pervasive minimal glyph (a dot, which is implicitly in any line), is a predicate.

FWIW, no this isn't how Chinese works — and it isn't ideographic. Hanzi (& Japanese kanji) have both semantic and rhyme hinting components, and aside from a few words that are nearly minimal in their use of radicals (the components), are generally vastly underspecified to get the meaning of the grapheme unless you know it.

Could you elaborate? How do you see this as relating to NLP?

FWIW, I very strongly recommend Rick Morneau's monograph Lexical semantics of a machine translation interlingua ¹. It was a major influence for me, and is squarely intended for NLP model-internal usage.

¹ http://www.rickmor.x10.mx/lexical_semantics.html

Yes, we're still working on it. (E.g. we were discussing the concept "only" a couple days ago.)

Right now, our main blocker is technical: we want a JavaScript based method for rendering it from a DSL, because making, loading, & maintaining images in the doc is a huge pain.

We've provisionally decided on using Cytoscape.js, but it needs some extensions to work for us — e.g. cubic splines (it only has quadratic) and relative angles of attachment for edges to nodes. My areas of expertise are mainly back end web dev & analytics, and Alex's are mainly academic math & coding competitions, so JS isn't particularly easy for either of us.

We'd appreciate help with this. It's on my GitHub¹; get in contact personally² if interested.

As for a speaker community, we don't have one currently, though over the years we've had around 3-4 people express interest. We never really considered it something that others would want to do (though we've always been quite interested in getting feedback, suggestions, etc., and we'd be willing to engage in in-UNLWS dialogue).

However, this seems like roughly the critical mass needed for such a community, so I'll try to set one up. Please contact me directly², and LMK any platform preferences.

¹ https://github.com/saizai across a couple recent repos

² https://s.ai/contact

Tracking: if you're asking a question or otherwise anticipating where the other will add, there are relatively few places to look.

Also, we use different colors per author in multi-author utterances, which makes it easy to spot another's writing.

Of course, unless you have an animated or history-preserving medium (which we assume we don't), then reconstructing the order of conversation might be difficult. That is entirely in keeping with our sense of what is "natural" to a non-linear language. The order in which something was written is temporal, and everything temporal is linear, so it's disfavored semantically & grammatically.

We prefer to challenge concepts like "conversation" at the root on such issues, rather than try to adapt UNLWS to afford them.

Sketching: we assume an infinite plane writing canvas. In practice, this means eg that one can use scaling to fit anything anywhere. However, we generally do not make use of images, unless it's purely quotative.

We do have e.g. graphs grammaticalized, though (inspired by Tufte's sparklines, but with a few more affordances from not being in a linear embedding.)

3d+: We considered that early on — and it's addressed in my essays on language design — but we rejected it as too under-constrained and difficult to work with in practice.

However, I've recently been thinking about a UNLWS-ish tactile 2.5D concept, which would use texture and a shallow height dimension — like a topographic map, not like an ants' nest. That's still in early conceptual stage, and not documented anywhere except a couple posts on CONLANG-L.

See also Ouwi¹ by Schuyler Duveen, which is one of the very few other non-linear written languages. It takes a somewhat different approach, and (IIRC) has some examples of conversational output.

(Ouwi & UNLWS were developed without either of us knowing about the other, though we got to know each other later on.)

¹ https://ouwi.org

Part of the inspiration for UNLWS was Heptapod B from Ted Chiang's Story of Your Life, from which Arrival was made.

See also my essays Non-Linear Fully Two-Dimensional Writing System Design ¹ and On the design of an ideal language ², which I wrote before we started to make UNLWS.

The written "language" in Arrival is purely babble (though it has some nice aesthetics at the fine level). It also violates explicit features described in the story, e.g. that an utterance is composed of unbroken strokes (i.e. a connected graph).

FWIW, I asked Chiang years ago whether he had any actual implementation or visualization in mind for Heptapod B. He didn't; it was purely an abstract set of properties for him.

¹ https://s.ai/essays/nlf2dws

² https://s.ai/essays/on_the_design_of_an_ideal_language

I'm Sai, co-creator of UNLWS. AMA(relevant).

Jake's response: http://www.twitlonger.com/show/n_1soorlp /via https://twitter.com/ioerror/status/739731362404536320

I note that he does not deny any of Nick's factual claims. The denial is only of "criminal sexual misconduct", which is a legal claim (what misconduct counts as criminal?) rather than a denial of the specific factual allegations made @ http://jacobappelbaum.net/ (however they may be legally construed).

I can understand why Jake may not want to engage with them in detail — doing so would almost certainly raise more questions than it answers — but AFAICT, this is functionally a non-denial denial.

One thing I don't understand from Nick's piece (etc.) is whether the referenced "plagiarism" means only Jake insisting on alphabetical author order (thus making it seem like he did most of the work when he didn't) or something else.

<s>It's also saddening to read of yet another of us committing suicide, whatever the circumstances.</s> Having read other comments, I now realize what this referred to. I suppose it's fortunate that it is not yet another, but that loss is no less saddening for the time passed since. (FWIW, I've stayed with that person's partner at one point. They were incredibly kind to me at time that was difficult for both of us.)

…

Repeating two comments I made on Nick's Medium article:

1. Nick: in my limited experiences with you, you’ve been one of the most warm & welcoming people in the hacker community. I hope that at some point you get back your health and manage to resume attending conferences. Take care of yourself.

2. Jake espouses very hard line principles of source protection, right to speak, fighting bad speech with more speech, etc. I generally agree with him on those principles. (I’d rather not go into my opinion on his personality.)

If Jake had gotten the kind of threats from Nick that Nick describes getting from Jake, there’s no way in hell he’d comply. He’d probably have an aneurism, irretrievably commit the documents to public view after thoroughly anonymizing them, and make it very, very public.

Kudos to Nick for both protecting his source and trying to handle this issue with tact. It’s indefensible that Nick was treated this way for doing so, and a loss to the community for him to have come to harm for it.

See http://conlang.org/axanar/#press for more articles on it.

(Will update soon for stuff that came out over the weekend.)

That's very neat. Could you email me (eg press@conlang.org) so we can collaborate / keep updated?

I think it would be grossly inappropriate for me to comment either way on an issue about our counsel, especially one that he disputes.

What I can say is that he is a brilliant First Amendment lawyer and we are very satisfied with his services.