Conlanging with vec2word

Word2Vec is a family of machine-learning based NLP (Natural Language Processing) algorithms which encode the semantics of words as vectors on a unit hypersphere--that is, the meaning of each word is encoded as a big list of numbers (a vector) whose sum-of-squares is 1. The smaller the angle between any two vectors, the more similar their associated words are in meaning. These kinds of models let you do some kinda neat stuff, like evaluating the semantic similarity between documents (which is useful for fuzzy searches), and doing arithmetic on vectors to complete analogies (e.g., "king - man + woman = queen").

It occurred to me (thanks to a Zoom discussion about a taxonomic philosophical language based on Semitic-style triliteral roots) that one could automatically generate vocabulary with taxonomic structure by starting with a word vector model, sorting the vectors along each dimension, and then mapping each vector entry to a phoneme based on its position along that dimension. Going not from source language words 2 vectors... but from vectors 2 conlang words.

(Side note: in theory, it would make more sense to convert semantic vectors into polyspherical coordinates and factor out the redundant radius dimension first... but in practice, sorting by all-but-one Euclidean coordinate gets you the same groupings, just in a possibly-different order [e.g., think about expressing your latitude and longitude in degrees, vs. miles north or south of the Earth's core and miles under Null Island, respectively; longitude coordinates end up sorting differently, and the scale in not linear, but coordinates that are close in one scheme are still close in the other], and is way more computationally efficient--not because there's anything special about Euclidean coordinates, but because pretty much every model ever already comes in that format, and the fastest conversion operation is the one you never do.)

Of course, decent models tend to have between 100 and 300 dimensions, which would make for really long words... but, we can do a neat thing called Principle Component Analysis (PCA) to figure out which vector components are the most important (i.e., in this case, carry the greatest semantic load), which means we just choose however long we want our words to be, extract that many principle components, and then proceed as before.

Additionally, if we don't allow an arbitrarily large number of phonemes to scatter along every dimension (which we don't, because languages have finite phonemic inventories, and rules about what phonemes can appear in what contexts), we won't necessarily be able to give every vector in the model a unique form. Thus, we will have to group semantically-similar source words together in the output. This is actually a good thing, because we don't want to just create an algorithmic relex of the source language (uh... unless you actually do, in which case, go for it, I guess!); rather, the list of source words associated with any given output word can serve as exemplars of a more general semantic field from which the precise definition of your new conlang word can be picked. It's not completely-automated, definitions-included word generation, but it is a lot easier than coming up with new words and definitions completely from scratch! And, even if you aren't intending to create a proper taxonomic language, this approach to producing semantic prompts for specific word forms can help produce a conlang lexicon which naturally contains discoverable sound symbolism patterns, without obvious taxonomic morphology.

Having realized that we will need to do some clustering, there are a few ways we could go about that. We could try just dividing the semantic space into rectangular regions, by dividing up each dimension individually, either at regular intervals or adjusted to try to get an equal distribution of source words in each cluster (and of course we can do that in Euclidean or polyspherical space)... but natural boundaries in semantic space aren't necessarily rectangular, and forcing rectangular clusters can end up putting weirdly different source words together in the same bucket. If you want that, to give you more flexibility in choosing which way to go with a definition, cool! But, there's another option: K-Means Clustering, which takes a set of points (i.e., vectors) and a number of clusters to group them into, and tries to find the best grouping into that number of clusters, dividing the space into Voronoi regions around the cluster centers. The number of clusters can be determined from the number of possible forms that are available, and then forms can be assigned just based on the locations of cluster centers.

(Another side note: it turns out that K-Means clustering of points in a sphere in Euclidean space produces exactly the same results as clustering points in polyspherical coordinate space--so once again, no coordinate transformation is required!)

If you choose to do K-Means clustering, there is then also the choice to perform clustering before or after doing dimensionality-reduction with PCA. Even though dimensionality reduction with PCA keeps the most important information around, when you are going from 300 dimensions down to, say, 3 (for a triliteral root), or even 10 (for a set of pretty darn long words), the stuff that gets thrown out can still be pretty important, so PCA will smush a bunch of stuff together which genuinely does have some kind of objective semantic relation... but whose relations you might be hard-pressed to actually figure out from the list of exemplars! Again, that could be a plus or a minus, depending on what you are going for (and clustering post-PCA will be more computationally efficient), but if you want the most semantically-coherent categories, clustering should be done prior to PCA.

That is the process that I have currently implemented in the vec2word Python program, hosted in the conlang-software-dev GitHub organization. And for this Lexember, I have been using prompts generated from a filtered (numbers and proper nouns removed) version of a word2vec model generated from a 2017 Wikipedia dump to create vocabulary for 2 new conlangs. This process might not work for everybody, but it's been my most consistent and productive Lexember to date! The exact process I have developed around this tool is not quite what I thought it would be when I first conceived of it, and is in fact different for each language, so I might write up some more on that later. But for now, the software is there for other people to try out, and I wanna get more some more experience through the end of Lexember to solidify my process thoughts before putting them out here for the world.

Additional Thought: Useful semantic relations could be automatically extracted by taking the differences of vectors (e.g., "woman - man = queen - king") and looking to see how often applying that difference to some other vector yields another known word. Such vectors are not guaranteed to correspond to any existing regular morphological process in the source language used to build the model, but that's just perfect for providing inspiration for new stuff that you could do in a conlang!

I have not implemented this yet, partly because it would be extremely computationally expensive, but I am very tempted to.

And, as always, just in case you feel like giving me money: you can do that right here.

The Transgalactic Guide to Solar System M-17

I have been meaning to review the next three books in the Steerswoman series for months now, and continually failing to actually do so. And when I started reading The Transgalactic Guide, I had no idea that it would present me with any reason to review it here. But then it went and had alien language content (and what was I supposed to do? Not blog about that!?), so here we go...

The Transgalactic Guide to Solar System M-17 (beware the Amazon Affiliate link!) by Jeff Rovin is satirical science-fiction travel guide, supposedly published by the Transgalactic touring company to describe the tourist attractions and accommodations available in the eponymous solar system M-17 (but actually published by the Perigee imprint of the Putnam Publishing Group in 1981). As science fiction, it is decidedly archaic; the author doesn't really give a crap about physical, chemical, or ecological plausibility, such that aside from the genre trappings of spaceships and alien planets, it's really more a work of fantasy--think C.S. Lewis's Space trilogy with less plot, less allegory, and more description of the bad science. Rovin makes repeated use of the "make them alien by not giving them eyes" trope (which Wayne Barlow has applied to much greater effect), which is sometimes justified and sometimes... not so much. Nevertheless, for the modern author it may provide a decent source of inspiration for weird and interesting environments and creatures, if you are willing to do the work to clean them up a bit for modern audiences.

But the reason I am bothering to review it here is that each of the 5 worlds of M-17 has at least one, and sometimes several, native alien languages which are represented in the text, along with brief tourist glossaries. As conlangs go, they are also.. not great, although there are some neat ideas. There is some decent effort put into the Alladis logography (which is supposedly tactile in nature), and the basic idea of an Oleran scent-based language (a concept which is developed in more detail in the Semiosis duology).

Excerpts of alien languages are frequently integrated into the text to refer to alien concepts or proper nouns. For the most part, a very straightforward translation strategy, using appositives or parenthesized translations, is employed--and that's really all you would expect from something presenting itself in the style of a travel guide! However, I found it notable that chapter 3, on the planet Morana, actually attempts to Teach The Reader, employing italicized alien words untranslated to refer to objects and locations after they are first introduced; the attempt is not particularly skillful, but it is there!

If you liked this post, please consider making a small donation.

The Linguistically Interesting Media Index

The Steerswoman & the Wood Gnome

The Steerswoman is the first of what is currently a four-book series by Rosemary Kirstein, with more entries apparently planned. It was originally published in 1989, with the latest book so far, The Language of Power, coming out in 2004. (See also my reviews of books 2, 3, and 4.)

The setting is what appears to be a low-fantasy world slowly making its way to a scientific revolution via the actions of the eponymous Steerswomen to collect knowledge and explore the world. Several items and events seen by the characters as "magical" are easily (or less than easily) identifiable to an attentive reader as entirely mundane and technological in nature, which gives the story a somewhat similar feeling to other stories like the earlier Dragonriders of Pern, or the more contemporaneous Black Trillium. As of the end of the first book, however, it is not entirely clear whether this will turn out to be a full-on sci-fi in disguise, like Pern, or if some of the magic will remain unexplained and fundamentally magical. (As an aside, all three of these series happen to be written by women and have female main characters--which is unfortunately notable when it comes to "classic" sci-fi and fantasy!)

The plot threads dealing with physics and chemistry make it feel almost like a mystery novel at times, as the characters investigate mysterious events, although there is an extra level of dramatic irony present given that we, the readers, exist in a post-Isaac-Newton society which allows for solving these mysteries must faster than the characters can! And I think the author is to be commended for not falling victim to presentism and allowing the characters to realistically struggle with their existing physical preconceptions, rather than having genius-level flashes of insight about things that are only obvious to us because we have been taught the answers.

The Steerswoman has very little surface-level linguistic content, which puts it outside the realm of things I would usually review and analyze, but there is a great deal of recognition of linguistic knowledge; while the Steerswoman's world seems to have a single universal human oral language, that fact is not entirely taken for granted by the Steerswomen, and several characters do in fact comment on the dialectal and accent variations that exist throughout the world, and how they can be used to identify someone's origins and social affiliations. And it turns out that there is some surface-level linguistic content... exhibiting a sign language!

I have been looking for examples of books with sign language representation, so finding this entirely by accident in a book I was reading for other reasons was kind of nice! In particular, while they appear only incidentally in this first book, the Steerswoman's world contains multiple seemingly-intelligent non-human cultures, the most salient of which are the Wood Gnomes--small humanoid creatures who communicate via sign language with humans. (It is unclear if they also have their own oral language or not.)

In my initial research on how sign language might be represented in print, I came across this article specifically about ASL language and culture for writers; while the central focus of that article is not about portraying the language itself, it does have one very straightforward bit of advice: whatever you do, don't gloss. To quote:

Glossing, a tool that is often used in ASL textbooks and courses to help students remember ASL syntax, uses the words that most closely align to ASL signs and puts them in ASL order.  Words in gloss are always written in the present tense and in capital letters.  For example, the gloss of the ASL translation of the English sentence, “Where is your car?” would look like this:

YOUR CAR WHERE?

Glossing can be a valuable tool, but it is extremely limited because it does not show use of space or nonmanual signals (for example, eyebrow and mouth movements and body shifts, all of which serve a grammatical function in ASL).

Worse, when glossing appears in fiction, it gives an incomplete picture of the language and makes deaf characters sound primitive and limited in communication.  What’s wrong with using standard dialogue conventions and replacing “said” with “signed”?:

“Where is your car?” she signed.

And you know what? That seems like pretty darn good advice. Almost obvious advice. I certainly wouldn't recommend word-for-word translations of oral languages, either--and I have yet to ever come across that as a secondary-language incorporation technique in the wild, which suggest to me that most-if-not-all authors also see that as an obviously bad idea. Yet there does seem to be a temptation to do it with signed languages, as that is the only exposure to sign language examples in print that many people have ever had. So what else are you gonna do? In the absence of standard sign orthographies, it doesn't seem like there are really any good options other than not actually showing the sign language at all, and relying entirely on free translation with language-specific dialog tags as suggested above.

So, what does The Steerswoman do?

A typical example of Wood Gnome conversation can be seen in this excerpt from the beginning of chapter 9:

    Rowan turned back to the wood gnome and addressed him in the language of hand signals that his people shared with humans. "Where woman?" she gestured.
    "Woman in bed," he replied, obviously meaning Rowan.
    "No. Other woman." She pointed to the bed with Bel's clothing.
    "Fur-woman. Noisy woman gone. Throw rock at me."

This isn't quite free translation, and it's not entirely clear what the broken-English representation is really meant to represent. Is it effectively word-for-word gloss? I don't know, but at least it is easy to read, not-all-caps block text!

Later on, Rowan encounters a deaf man who appears to communicate through some combination of iconic gestures and home sign. We are introduced to him and his communication style as follows:

    [The woman] tapped her assistant on the shoulder as he made to unload another crate. Pausing in his work, he watched intently as she indicated Bel and Rowan and pointed from the boxes to the door; then he nodded pleasantly at the pair; he was deaf.
    [...]
    Attracting the disheveled man's attention again, she attempted to give him a more difficult, complicated instruction. Eventually comprehending, he led the way.

And a little later, we get some Wood Gnome Sign Language again:

    At the top was a second landing, and there the man put down his crate, indicated those carried by the women, then indicated the floor. When they complied, he pointed at Rowan and Bell, back down the stairs, pointed at himself, and made a motion toward a short corridor behind him.
    Without thinking, Rowan replied in the wood gnome language of gestures. "I understand. We go down now."
    These particular phrases were simple and obvious, easily comprehensible to an intelligent person; but the formality of the gestures, and the fluid naturalness of their use, surprised him. It was more than pantomime, it was language, and he seemed to recognize something of this. 
    With a look of surprise and concentration, he repeated a phrase, pointing at himself, then extending his index finger near his right temple. "I understand." He said it twice, testing the moves.

Notably, although the signs cannot be directly written, there is some attempt in this passage to represent the actual diagetic language through partial descriptions of the forms of the signs. The last paragraph could even be analyzed as a classic example of appositive subtitling. And on general principle, I appreciate the explicit acknowledgment than sign language is language, at the same level as oral language, and distinct from non-linguistic gesturing. This deaf man has no real significance to the larger plot (although the context of this interaction with him does provide some additional clues for the reader in solving some scientific mysteries), so I can only assume that he was included specifically to allow the author to make this point!

Our final narrative exposure to him is as follows:

    When Rowan delivered [the crate], she could not help speaking to the man again. "Work finished." Those signals were more abstract, and she amplified them with gestures indicating the stairs, the box, herself, and the man, and a negative shake of her head.
    He watched in fascination. Then, with the crate precariously tucked under one arm, he replied, "I understand." He paused, thinking, the hesitantly added, "You go down now."

Although it is rather limited, I was also quite pleased by the representation of sign language, and it should provide a decent basis for comparison with other works.

If you liked this post, please consider making a small donation!

The Linguistically Interesting Media Index

Living in a Hyperbolic Universe

 A few years ago, there was a question on the Worldbuilding StackExchange site about living on a hyperbolic plane. This turns out to be a pretty fascinating setting, but it runs into all the same sorts of problems as infinite flat Euclidean worlds do--how does the sun work? If there's gravity, why don't mountains destabilize the whole thing and fragment it into ball-worlds? Etc.

So, I started thinking; what if you just had a planet in a hyperbolic space?

It turns out, the surfaces of spheres in hyperbolic spaces are not themselves hyperbolic manifolds; just like Euclidean spheres, they end up having positive curvature all over. So, you don't get all the interesting features of living on a hyperbolic plane, like the entire surface area of the Earth being within walking distance and the navigational problems that that implies--but there are other interesting features of such worlds!

First, let's talk about exactly how we're going to construct a hyperbolic universe. It turns out, you can do this in General Relativity, but a relativistic universe with hyperbolic space turns out to also be an exponentially expanding universe as well. The reason for this is that parallel lines in a hyperbolic space diverge, and in a relativistic universe you can't distinguish objects in motion from objects at rest--which means that parallel world lines of some collection of objects that are at relative rest to each other must diverge, no matter how that collection is moving relative to anything else. Since geodesics diverge in space, they must diverge in time as well, and the result is an expanding universe. If the curvature is strong enough to be interesting on human scales, that means the universe blows up before anything interesting can happen, so we won't be putting out planet in a relativistic universe after all!

Instead, we'll use a 3-dimensional hyperbolic space with Newtonian absolute time. Now, in a Euclidean Newtonian world, we still have Galilean relativity; any object moving inertially can be declared to be at rest, and physics is unchanged. But in this hyperbolic universe, that is no longer true; motion implies geodesic divergence, which means moving objects will feel strain from pseudoforces trying to blow them apart perpendicular to their direction of motion, so it's easy to determine a universal absolute state of rest, and absolute measurement of velocity. Travel speeds won't be limited by the speed of light, but they will be limited by the tensile strength of materials--move too fast, and you will blow up!

This also applies to a planet orbiting a sun. At different parts of its orbit, a planet will have different absolute velocities, producing seasonally-varying strains trying to distort it, and the orientation of those strains relative to the surface features will vary as the planet rotates throughout its day--meaning that there will be tides with no need for a moon! Incidentally, this also puts limits on where planets can exist and how big they can be, lest the divergence forces of orbital motion tear them apart! Fortunately, gravitational force drops much more rapidly with distance in hyperbolic space than it does in Euclidean space (exponentially, in fact), so orbits can be rather slow, which will keep our planet from blowing up even if it orbits very close to a star--which it will need to do to get enough exponentially-decaying light!

Now, the characteristics of the planet itself: we want the hyperbolic curvature to be strong enough to be interesting and noticeable, but not so strong that it makes human life impossible. Suppose we choose a hyperbolic scale of 10km. On that scale, a planet with about 55 thousand square kilometers less surface area than Earth would have a radius of a mere 71.4998km. Why am I using so many decimal places? Because if you increase the radius by a mere 10cm, to 71.4999km, the surface area jumps to 9,695 square kilometers bigger than the Earth! Fortunately, the circumference of the world changes by only about 0.02% over the 2-meter height of a tall human, so you can in fact walk on such a world without your feet leaving your head behind!

So, humans can certainly walk around on that kind of world; but what if we make the curvature more extreme? At a hyperbolic scale of 1km, a planet with a radius of a mere 9.45257km exceeds Earth in surface area, and the variation in arc lengths over a 2-meter height is still only about 0.2%. Meanwhile, if you dig down 50 meters (a typical subway tunnel depth), all distances contract by 5%. And if you travel upwards by even 150m (not even quite meeting FAA requirements for minimum flight altitudes), distances expand by 16%. So, unlike living on a hyperbolic plane, the entire surface of the world is not within walking distance over that surface, and navigation on the surface isn't too different from navigation on Earth... but flying is significantly disincentivized, and everything is within 20 km if you are willing to take the most direct route tunneling through the interior of the planet! So, subways rather than airlines turn out to be the way to go for fast travel!

And don't forget, the core of the planet is only 9 kilometers down. The deepest mines on Earth are not-quite-4km deep--a significant fraction of that distance. Which means this hyperbolic planet is much easier to mine than Earth--almost all of it is extremely close to the surface. Precious (and mundane) metals won't all be inaccessible in a deep core.

And you know what that means: hyperbolic space is pretty much made for dwarfs!

Shadowscent: The Darkest Conlang

Shadowscent: The Darkest Bloom and Shadowscent: Crown of Smoke together compose the Shadowscent Duology (beware the Amazon Affiliate links!), the debut work of author P. M. Freestone. Shadowscent is a secondary-world low fantasy adventure set primarily in the Empire of Aramtesh, whose culture puts huge emphasis on scent, olfaction, and perfumery technology.

This would seem like the perfect sort of novel to incorporate a conlang, and I first found out about it via the Lingthusiasm podcast because one of the hosts, Lauren Gawne, actually created a conlang for it, which she gave a presentation on at the latest Language Creation Conference!

To my minor disappointment, there is not much actual Aramteskan language shown in the text, apart from proper names, so there's not a whole lot to analyze in my usual manner. We do, however, get one sentence of Old Imperial, the mother language from which modern Aramteskan descends, which comes up during a debate over what its proper translation into the modern language actually is:

masaat asytaa amidak snalu masaat kiregtaa traalapaame

While the full translation would potentially constitute a spoiler, we get a little bit of explanation with more repetitions of the Old Imperial words in this excerpt:

"We still use marsat for 'cloud', but the other terms have no modern equivalent. If we were to be literal, masaat asytaa and masaat kiregtaa translate to 'wet cloud' and 'dry cloud', more or less."

This is just plain old diagetic translation, but it does a nice job of almost-lampshading the implicit translation convention between the modern Aramteskan that the characters are "actually" speaking and the English that the reader can see. (As well as demonstrating phonological evolution between the archaic words and a modern equivalent.) And this causes me to wonder how direct that translation should be assumed to be, at least in this instance, because the speaker almost certainly would not have actually said the equivalent of "we still use the word 'cloud' for 'cloud'". Are there perhaps multiple modern words for "cloud", such that a different one was substituted in the latter case? Or did he really just say "we still use the word 'cloud'" with the extra "for 'cloud'" being added in "translation"? Something to keep in mind when crafting dialog that's supposed to be a translation--similar to the "it just happens to rhyme in English" trope. (This excerpt also tells us something about noun-adjective order in Old Imperial... but since syntax as well as phonology can change over time, we can't be sure that the same applies to modern Aramteskan! So you'll just have to wait for the grammar to be published. UPDATE: The grammar of Aramteskan is now available through Fiat Lingua.)

While I don't have a whole lot of Aramteskan text to analyze in context, though, I did manage to get an interview with Lauren Gawne for another behind-the-scenes look at how the language influenced the books.

Q: There was very little Aramteskan that actually made it into the text, aside from proper nouns. Was simply using it as a naming language the originally planned extent of its usage? If so, do you know the reasoning behind that? If not, and there was a plan to exhibit more of the language, what caused the change?

A: There’s one phrase-length string of Old Aramteskan in the first book (see above). Other than that, the language appears either directly as a naming language or indirectly in a variety of in-world structures, such as the way people greet each other, insults, socio-dialectal dynamics and the way the currency is structured.

This was the extent the language was always intended to feature in the books. The additional depth to the language was something I worked on in part for my own interest. PM Freestone is a long-time colleague and collaborator, and they brought me into the worldbuilding on this project really early. I don’t normally get the chance to work with authors so early in the process, normally there’s a world, characters and/or plot already in place. Being on board from the start was a chance for me to build something more structured while also helping to build out the world for the books.

While of course I always like to see the actual language in a secondary-language situation, I think this is an excellent addition to the list of Reasons Why You Need A Conlang; a simple naming language already adds a lot, and Shadowscent is a demonstration of how thinking about the language, even if it never makes it to the page, can feed back into informing the culture and character interactions.

Q: Was there any consideration of including a pronunciation guide in the books?

A: Not that I’m aware of. I’m very sanguine about the fact that there are names and words that people will bring their own pronunciation to. I try to structure the language and the names to minimise the likelihood of that, but it happens and it’s ok.

My conlanging for other people’s work always accounts for the medium and the audience when it comes to the sound and the structure of the language. A conlang is there to enrich the experience, not to pull people out of the world.

A related excerpt from the forthcoming Aramteskan grammar:
Aramteskan was created with the aim that it be easy to read for English speakers, using the Latin alphabet. I have deliberately avoided using diacritics or other markings. Aramteskan mostly exists in the written medium and so I have avoided using sounds beyond English phonology, since people will encounter the language as written rather than spoken.
...
Stress is free, based on how English speakers find the words best to pronounce. A more consistent stress system can be developed if this needs to become a spoken language.

Q: What was the editor/publisher's opinion on developing a conlang for the books? Did they even have one?

A: I think what is immediately apparent in PM Freestone’s work is the general depth of worldbuilding that underpins the fast-paced narrative. The language building is just one element of the cultural and physical world building that they bring to their work, so I don’t think it was something that was singularly noticed in the work process. The publisher trusted them to deliver a rich and fulfilling story and I’d like to hope that the language helped them accomplish that.

Q: At what point in the writing process was the language "finished"? Did it develop more between books 1 and 2?

A: There are still so many things I’d tinker on, but there are more projects to do then there are hours in the day. Putting together the materials I have to publish with Fiat Lingua (forthcoming) is part of trying to draw a line under Aramteskan now that the books are finished. The structure of the language was broadly in place by the end of book one, and then for book two it was mostly about adding vocabulary and tweaking a few things, including fleshing out the pronoun paradigm to make sure Rakel’s chat with another character (spoilers!) about pronouns and gender make sense within the structure of the language.

Q: Are you aware of reader feedback specifically about the linguistics content? Is it practically nonexistent, generally positive, or generally negative?

A: I haven’t seen too much, but then I am happy to not spend time delving into the bookwebs (either as a creator or a reader!). A lot of the positive feedback has come from linguists and authors. When it comes to authors they love hearing about how building a rich conlang can help give motivation to narrative, characters and the world they’re building. Linguists enjoy hearing how conlanging can be a creative way to highlight the constraints of what we know about existing language typologies, for example almost a quarter of the world’s languages have a grammar feature that shows the source of evidence for a statement (in English we use additional, optional phrases like ‘I saw’ or ‘I heard’), but none of those languages have a distinct category for scent-based evidentials. I always like to find one or two ways to push against my own tendency to be very focused on naturalness, and deliberately build in features unattested in natlangs. Linguists appreciate the humour and creativity in that.

So there you go! It's just not just me who likes linguistic content in books! But if you, current reader, happen not to be a linguist or another author, but you have opinions about the linguistic content in Shadowscent, leave a comment or @ me on Twitter! More data is always good!

And finally, a bit of con-historical linguistic nerdery:

Q: Have I correctly interpreted the grammar as indicating that "ph" represents an aspirated p in the language of Doskai, but was loaned as /f/?

A: That’s pretty much how it goes. This is one of those moments where the conlang builds itself around the constraints inadvertently created by the author. Aphorai was a place that was named by PM Freestone in early drafts, and I never managed to budge that <ph>, even when they were happy with the phonology of the language in general. So, /f/ as <ph> became a sound in the language specifically in that region, and specifically loans from a long lost older language. You’ll also notice that it pops up in names from that area, such as Sephine. This kind of chaos would stress me out when I started creating languages, now I appreciate the way it builds in more texture and natlang style layers.

And I think there's a useful lesson there in how artistic creativity can be sparked by constraints!

If you liked this post, please consider making a small donation!

The Linguistically Interesting Media Index

The Mandalorian & Tusken Sign Language

A long time ago, in a language community far, far away, a bounty hunter negotiated with some desert nomads in a sign language....

Two episodes of Disney's The Mandalorian feature on-screen portrayals of a constructed sign language: Tusken Sign Language or Tusken Raider Sign Language. Since Star Wars is a Disney property now, and it is in fact Disney's The Mandalorian, I suppose this makes TSL Disney's fourth actual conlang, after Atlantian, Kumandran, and Barsoomian. The Star Wars franchise, however, is not really known for its linguistic sophistication (Greedo, whom Han shot, for example, just speaks random phrases of garbled Quechua; portrayal of alien languages in Star Wars generally has been the domain of sound designers rather than linguists or conlangers), and this bit of conlang representation came about essentially by accident: sign language was specified in the script, and someone on the crew suggested that, if they are going to portray signing, they really ought to get an actual Deaf person to consult, and that Deaf person (one Troy Kotsur, who also plays one the Tusken Raiders) decided to come up with a conlang!

Essentially all of the publicly-available information about this process comes from one almost-8-minute video interview (with English transcripts) done with Troy by The Daily Moth, a dedicated ASL news site. Of course, information about the language itself is also extractable from the scenes in which it occurs, and a Tusken Raider Sign Language Facebook group has been established to study and decipher it; as of now, very little deciphering has actually occurred, but Troy is a member of that group, which gives it some legitimacy, and leaves open the possibility that one might actually get authoritative answers if one asks questions there.

Of course, none of this would've happened if some sort of sign language had not been specified in the script in the first place, which raises the question of why anybody thought that signing should be in the script! George Corley of Conlangery suggested to me that it might be because the Tuskens' vocalizations were already established as the sounds of braying donkeys, which doesn't really lend itself to a proper language--but that presupposes that the writers actually cared about portraying a plausible language in the first place, an assumption that I find heavily suspect given the portrayal of Frog Lady's supposedly-linguistic vocalizations in the same show. Wookiepedia claims that the Tusken language is a multi-modal combination of vocalizations and sign, but that seems to be a retcon to explain the appearance of sign language in The Mandalorian, not a reference to a common source that The Mandalorian may have drawn from, and the questionably-canonical Star Wars: Absolutely Everything You Need to Know (updated edition; note that these are, as usual, Amazon Affiliate links) already claimed consistent English translations for particular (transcriptions of) Tusken vocalizations. The most plausible explanation to me comes from this tweet by Star Wars Autograph News, which claims that it is "undoubtedly" a reference to Plains Native American Sign Language.

So, how is this new language actually used? As with Disney's previous use of Kumandran and The Dragon Prince's portrayal of ASL, The Mandalorian completely eschews the use of subtitles in favor of a combination of Making it Obvious and diagetic translations / explanations (with the shortage of direct translations in the on-screen source material making fan decipherment rather difficult!)

Unlike Raya and The Dragon Prince, The Mandalorian is clearly not aimed at a child audience, so they don't need to worry about the literacy level of their viewers. Why then, would they avoid subtitling? I expect a large part of it is just the biased expectation that audiences won't like them, but whether or not it was actually on purpose, I think the decision actually worked out for the best in this context. One could argue that subtitling of Tusken would be appropriate because Mando speaks Tusken and Mando is our viewpoint character--but, Mando is not a first-person main character, and there is essentially no dramatic irony in the series as far as Mando is concerned. If we, the audience, know something, then Mando also knows it, but the converse is not true; we do not know everything that Mando knows, and that is on purpose. He is supposed to be a mysterious figure. If The Mandalorian were a serial novel rather than a TV show, it would be told in limited third person, not first. Thus, we shouldn't necessarily have subtitles for the same reason that we don't get voiceover narrations in this show: we are not supposed to be inside Mando's head. If someone else in the scene is confused, we should be confused.

That said, let's get finally get into the nitty-gritty stuff!

Tusken Sign Language first shows up briefly in Chapter 5: The Gunslinger, with this conversation:

[Mando signs to the Tuskens.]
Calican: "What are you doing?"
Mando: "Negotiating."
[More back-and-forth signing.]
Calican: "What's going on?"
Mando: "We need passage across their land."
[More back-and-forth signing.]
Mando: "Let me see the binocs."
Calican: "Why?"
[Mando tosses the binocs to one of the Tuskens.]
Calican: "Hey! What?"

This already shows that mix of diagetic explanation and Making it Obvious. While some of the signs are pretty iconic, Troy is probably the only person who can give us a complete literal translation, but the context makes it easy to reconstruct a basic outline. Mando is negotiating, he hands over the binocs, they conclude--he must have agreed to trade the binocs as payment for passage across Tusken land.

While there's not much here, this minimal interaction is narratively necessary to establish in a low-stakes setting the fact that Mando knows the Tusken language and has reasonably amicable dealings with them, so that that skill can be employed with much higher stakes later on, and there is indeed considerably more representation of Tusken Sign, along with Tusken vocalization, in Chapter 9: The Marshal.

Initially, we see some signing with Irrelevant content during the desert crossing montage shot, while Mando sitting around campfire with Tuskens. This is just showing us the language for flavor, and to remind the audience that Mando has that skill.

For the most part, the remaining Tusken language is handled with minimal diagetic explanation, with the audience taking the point of the view of Cobb Vanth. E.g.,

[Mando engages in multi-modal communication with Tuskens]
Cobb: "Hey partner, you wanna tell me what's goin' on?"
Mando: "They want to kill the krayt dragon, too."

and later: 

[Tuskens add more tokens to their battle map.]
Cobb: "That's more like it. Where are they gettin' the reinforcements?"
Mando: "I volunteered your village."

with a few instances of what may be more direct translation in the form of indirect reported speech:

Mando: "He says your people steal their water, and now you insult them by not drinking it."
"They know about Mos Pelgo, they know how many Sand People you killed."
Cobb: "They raided our village! I defended the town!"
Mando: "Lower your voice."
Cobb: "I knew this was a bad idea."
Mando: "You're agitating them."

and direct reported speech:

Cobb: "What are you telling them?"
Mando: "The same thing I'm telling you. If we fight amongst ourselves, the monster will kill us all."
Mando, SimComming with English, slowly: "Now, how do we kill it?"

And in one instance, they actually exploit the unique capacities of the visual modality to allow Mando to report on someone speaking in sign at an extreme distance, via binocs:

[Tusken signing at long distance]
Cobb: "What'd he say?"
Mando: "He says it's sleeping. If we listen carefully we can hear it breathing."

I somehow doubt that the show's writers planned ahead to introduce a sign language just to allow that one scene to work (if they hadn't, a walkie-talkie would've served just as well), but hey, now you know that that's something you can do! 

So, that's Tusken Sign Language in The Mandalorian! If you liked this post, please consider making a small donation!

The Linguistically Interesting Media Index

Into the Night of Language Diversity

Into the Night is a 2020 Belgian Netflix-original series (apparently inspired by the Polish sci-fi novel The Old Axolotl [beware the Amazon Affiliate link!], although I see very little resemblance myself; it feels to me more like an SCP Tale in the 001 Daybreak canon, just without the shambling blob monsters), which has an ensemble cast of characters from a wide range of linguistic backgrounds--and it fully embraces that fact! There is no single language shared by all of the characters (although between French and English we have them all covered), and no on-screen translation convention; everyone is portrayed as speaking the language they would really be speaking at any given moment. The official language of the series as indicated by Netflix is French, and that does seem to be the most common language spoken throughout the series (which is sensible, given that the story starts in Belgium), but I'm not actually sure it's a majority--and if it is, it's clear that other languages are not treated as second-class. If you don't turn on the subtitles, you have to just know every language that's spoken in order to follow the plot; like War & Peace, it does not have a primary language with affordances to make secondary languages accessible--it just is fully multilingual. As such, there is not much to say about the techniques it uses for language affordance; it's just subtitled! You could maybe make an argument for a couple cases of Making It Obvious, but they're not really trying; and I can think of one case of diagetic translation, but that's because the original speaker was mumbling and would've been unintelligible otherwise regardless of the audience's linguistic prowess.

Unlike War & Peace, however, the sheer diversity of languages represented on screen (French, English, Flemish, Russian, Arabic, Italian, Turkish, German, and a throwaway bit of Mandarin right at the beginning) means that you can't actually expect any significantly-sized audience to actually know all of them. No matter who is watching, you will need subtitles at some point--so Into the Night just goes all out and subtitles everything, uniformly. So, given a sufficient number of subtitle tracks, it does not need to assume its audience speaks any of the featured languages--the viewer can pick whatever linguistic representation they need, without changing the video. At the moment, Netflix only offers subtitles in English, French, and Chinese, but all three language groups are in pretty much the same boat. Which leads me to the intriguing conclusion that this is a bit of story-telling media with no primary narrative languages.

Now, there are ways of telling stories with no spoken or written language at all--wordless picture books, mime shows, LEGO assembly instructions--and that's a whole interesting area of study to get into on its own. Heck, you could probably do interesting stuff with a series of statuary forming 3D "frames", although I don't know of any examples myself. And if that interests you, I must recommend the work of Dr. Neil Cohn, the Visual Linguist, and his Visual Language Lab, which looks at conventions of graphical representation and how they support storytelling either in conjunction with or in absence of written language across different cultures.

But that's not really my thing. My thing is explicitly analyzing the narrative use of secondary languages. And while, as I have just explained, Into the Night doesn't really do the secondary language thing, the way that it manages to not do it is fascinating. I can't imagine that this is the first bit of media to do something like this, but it's the first one I've noticed, and it just blows my mind, even if for no other reason that helping to better define the boundaries of my area of interest. And what it is doing is something that I just cannot imagine working very well in any other medium. You could imagine, for example, a multilingual opera with the libretto printed in a variety of languages in the play book for the audience to follow along, but that suffers from dividing the audience members' attention between page and stage. I am told that modern opera productions sometimes display the translations on screens for the audience, but then you are implicitly choosing a privileged language for at least that performance--the whole audience, regardless of background, is forced to rely on a single language chosen for the cinematic captions. And since subtitling relies on multimodal perception, or at least multichannel perception--you can read the subtitles while listening to the actors' spoken dialog, or read subtitles in your foveal vision while remaining aware of signing on another part of the screen (although Into the Night in particular does not happen to feature any sign language), you really can't do it in print or audio. (Although, perhaps the War & Peace trick of having a audience-language foreground track superimposed over a uniform diagetic language background track might work for audio format? I don't know how annoying that would get if used so continuously; someone should try it!) This form of storytelling is something that is just fantastically well suited to the very specific medium of home-viewed video with selectable subtitles. It's not specifically a streaming video thing; you could imagine this being done on DVD, for example. But has anyone?

So, anyway. That's a thing you can do with video, apparently.

If you liked this post, please consider making a small donation!

The Linguistically Interesting Media Index