This is a response to an open letter from Joshua Hartshorne entitled “Fantasy TV in the service of science,” posted as a guest blog by Scientific American about Dothraki, a language created by the Language Creation Society for the HBO adaptation of George R. R. Martin's A Game of Thrones. David Peterson (creator of Dothraki) and Sai Emrys (LCS president) emailed him immediately after its posting. David's response below takes into account what was discussed during a fairly lengthy conversation.

Zhey vichomer Joshua Hartshorne,

First, thanks for the shout out! It’s nice to get some love from someone in academia (and my mother would be thrilled to know that my name was mentioned in something having to do with Scientific American).

As some of the commenters on your initial post noted, there are a few practical concerns regarding the implementation of “non-human” processes or elements in the Dothraki language which I’ll outline below. For example…

It’s Too Late!

The Language Creation Society was contacted about creating the Dothraki language for HBO’s adaptation of George R.R. Martin’s A Game of Thrones in the summer of 2009. The application process was conducted in the months of September and October of 2009, and by November, I had produced a more or less complete grammar of Dothraki along with a vocabulary of 1,700 words (many of which had become “canon”, in that they have been used in the pilot and can no longer be altered). By the time I read your open letter, the grammar had been pretty much solidified, and the word count exceeded 2,000.

To give a concrete example, in your letter you mention that “action verbs” (like the change-of-state verb “to break”) commonly have an agent (the “do-er”) as a subject and a patient (“do-ee”) as an object. If one were to flip those roles around, so that the agent was the object and the patient was the subject, it certainly would set the stage for an interesting experiment in learnability. And while I wouldn’t be opposed to including a class of verbs like this in a language, the first thing that popped into my head while reading this section of your letter was, “Darn! I already have a word for ‘break’.” Actually, there are a couple. Here are some illustrative examples:

Arakh samvo (mahrazhoon).

/arakh-nom break1- pst (man- abl )/

“The arakh broke (and the man broke it).”

Mahrazh assamve arakh.

/man-nom break2- pst arakh- acc /

“The man broke the arakh.”

The verbs above differ in how they assign case (a grammatical role) to their arguments. In the first example, the subject is the thing that breaks, and the one that does the breaking is a non-necessary element which can be added in kind of as an afterthought or explanation. The second sentence works just like the English translation does.

Unfortunately, even the first sentence differs crucially from “shbreak”, as the Dothraki verb samvolat is intransitive and doesn’t assign accusative (or object) case to its reintroduced agent. As such, they wouldn’t fit the criteria for the “action verb” experiment—and there are already a whole slew of verb pairs that work exactly the same way (drivolat/addrivat “to die/to kill”; nithat/annithat “to feel pain/to cause pain”, raggat/arraggat “to choke on/to choke”, etc.). I suspect that the same is true of many potentially linguistic universal-bending aspects of language in Dothraki.

That said, this series is young, and there may be a need for other languages (or expansions to Dothraki) as things progress. Now that I know there’s interest, if an opportunity to intentionally bend a few choice linguistic universals arises in the future, I’ll certainly consider the possibility.

Artistic Integrity

As you suggest, the experiments you propose would only work with a somewhat sizable—and at least semi-fluent—speaker community. In that respect, Dothraki may turn out to be a good test case.

However, as Dothraki is an artlang, its primary goal is to remain true to the printed material about the Dothraki in George R.R. Martin’s books and to create a desirable product for David Benioff and Dan Weiss. There’s little wiggle room when it comes to implementing non-natural linguistic oddities.

I don’t believe it was Mr. Martin’s intention for the humans in his books—vicious Dothraki horse warriors included—to be radically different from humans in our world (at least with respect to their higher cognitive functions). As such, I made it my goal to create an artlang that would look and feel just like any natural language (English, Turkish, Hawaiian, ASL, etc.), exemplifying the kind of variability that one would find in natural human languages. As linguistic universals tend to describe the way natural languages behave in the real world, it shouldn’t be surprising to find that Dothraki doesn’t break many.

That isn’t to say that it breaks none, though, or that no natural languages can. In fact, the very first universal in the Universals Archive (one of Greenberg’s originals) lists no fewer than 30 natural language counterexamples (though, wouldn’t you know it, Dothraki doesn’t violate this one). If someone goes over Dothraki with a fine tooth comb, I’m sure they will find it breaks a number of linguistic universals (there are over 2,000 described; I’ll take those odds).

Even so, linguistic universals are notoriously fickle things.

The Nature of Linguistic Universals

As you noted, “linguistic universals” are observations about the way human languages documented so far by linguists happen to be. They’re not necessarily observations about the way language must be. Their definitions tend to also be nebulous. Even linguistic universal 2000, with no confirmed counterexamples, is worded thus: “If an agreement target can agree (in gender, number, or whatever), then typically it must agree…”

“Typically”? Even though it’s more likely to be accurate, that’s not as strong as one would expect from a linguistic universal.

Moreover, many linguistic universals only apply under very specific conditions. To produce a true counterexample to them would take some careful planning, and probably a new language—not one whose parameters have been set (like Dothraki).

I should note, though, that there are a number of universals that describe human languages, but say (at least, as I see it) nothing about learnability. Consider for example the fact that the vast majority of languages have base 10 or base 20 number systems, corresponding to the number of fingers (or fingers and toes) we have. This is surely merely a contingent property; one can easily imagine a language that uses a different number system (say, base 8). And indeed, there are a few dozen languages that do have different systems; no doubt their speakers (and Tom Lehrer) have no difficulty counting.

Base 8 number systems will probably never be predominant in our world (unless some genetic mutation causes us all to have fewer fingers). But to suggest that a language exhibiting a pattern like this is unlearnable seems silly—to me, at least.

So, in terms of breaking universals, Dothraki may not be very interesting. But…

Morphologically speaking…

I’d be lying if I said I didn’t have a theoretical axe to grind when it comes to creating languages, but it’s a morphological axe, as opposed to a typological or syntactic one. That is, when it comes to innovation, my main focus is on how to represent a given set of categorical elements in a unique way, as opposed to which categorical elements to include.

Many of the conlangs that have come to national and international attention (i.e. those used in film, television, novels, etc.) have interesting or quirky features here and there. For example, Klingon famously has basic OVS word order and both subject and object marking on the verb, and Na’vi has an ejective series. Nevertheless, their structure is pretty straightforward: a series of elements lined up one after another like cars on a train.

No natural language is that transparent (no, not even Turkish), and it’s that analytical transparency present in a lot of first-time conlangs that I’ve tried to avoid. For example, in Dothraki, it’s a simple matter to separate affixes from stems, but it’s not always simple to attach a meaning to a given affix. Compare the following:

  • asafazhan “hotter”
  • asafazhanaz “hottest”

  • osafazhan “less hot”
  • asafazhanoz “least hot”



Looking at the forms above, the stem afazh (“hot”) can be picked out fairly easily, but assigning a single meaning to each affix becomes either a very difficult process, or a trivial process (i.e. simply restating what’s there). However, it would seem a mistake to treat each form as entirely unrelated to the rest (i.e. as if each one had an entirely different affix).

Towards the end of my tenure at UCSD, I became fascinated with various instantiations of Word and Paradigm morphology thanks to Prof. Farrell Ackerman, and that’s where the bulk of my conlinguistic experimentation lies (cf. my talk at LCC1). As a result, it’s no wonder that most of the innovation in Dothraki lies in the morphology.

Not that I don’t find other avenues of experimentation interesting—quite the contrary! It’s just that creating a language from scratch is a monumental undertaking, and with all the other constraints present (not the least of which, I might add, was time), certain things are bound to take precedence over the others.

Looking Ahead

Where some of the bigger and more contentious linguistic universals are concerned, Dothraki may not prove as interesting as an enterprising researcher might hope. But I think the message to take away from this discussion is: stay tuned! More languages may be on the horizon, and while neither I nor anyone else can make any guarantees, you have certainly gotten our attention.

Fonas chek,

David J. Peterson

Language Creation Society

P.S. I’d be remiss if I didn’t mention that there are lots of great engelangs that do tackle a number of generally-accepted linguistic universals head on. While they may not come with large speaker communities, I think they would prove useful in experimentation.


Sai Emrys twice created and taught a semester long class on conlanging at the University of California, Berkeley, founded the nonprofit Language Creation Society, ran three Language Creation Conferences, and is co-host of the interview series of the Language Creation Society Podcast. His personal conlanging interests are primarily in engelangs, particularly in novel ways of using language, such as non-linear writing systems and tactile language.


David J. Peterson is a professional language creator hired to create the Dothraki language for HBO's television adaptation of George R.R. Martin's A Game of Thrones. He received his BA in linguistics from UC Berkeley, and his MA in linguistics from UC San Diego. He's been creating languages since 2000, and is especially interested in nominal morphology, lexicon creation and writing systems. Aside from Dothraki, his best-known and most fully-developed language is Kamakawi.


The views expressed are those of the authors and are not necessarily those of Scientific American.