AI Translation and "The Language of Liars" by S.L. Huang

 

 

 

 

R.L. Huang's "The Language of Liars" is an anthropological parable for our times. But in ways that exceed anthropology and linguistics. First, it is a clear indictment of colonization and appropriation and that, I think, was what Huang intended. Ro is a linguist from a small colony of vaguely mammalian creatures (the Ponto). He spends his days in formal linguistic study--as any grad student in linguistics might, with one exception: the Ponto's high level of empathy occasionally enables the most gifted linguists among them to "leap" into a host body, literally becoming the Other. At the top of the list for leaping: the most enigmatic species in the galaxy--the 'Star Eaters,' spherical, tentacled creatures with the ability to sense and interact with a hyperspatial element (meridian) that, Dune-like, enables the expansion of life across the galaxy. 

In the midst of his grad student doubts about his own abilities and his purpose, Ro makes the leap into a Star Eater who toils in a meridian mining facility along with their fellow Star Eaters, all under the watchful eyes of pugilistic AI robots (the Overseers) and "the Conglomerate," a hyper-version of the British East India Company. Luckily for Ro, Star Eaters are prone to periods of dissociative amnesia, providing a handy deus ex mchina for Ro's leap into a Star Eater body. Accordingly, he makes many mistakes and needs to learn to mine meridian, among other social and cultural mores. All of his cultural missteps come to a head, however, when he discovers a stolen cache of meridian on the mining vessel, destroys its containment and then sets off a chain reaction that ports the mining ship several light years away. Brought before the Conglomerate at the inevitable hearing, Ro connects the dots from his time among the Star Eaters and draws the Far Side comic-ish conclusion--all of the Star Eaters have been replaced by linguists and there are no real Star Eaters left! 

WE”RE ALL FROM OTHER WORLDS, AREN’T WE. Ro flung it forth before another order could be given, before any automaton could pounce. His words tripped over each other, a jerking thrash against the tableaux. THE MORPHOSYNTAX, LOOK AT THE MORPHOSYNTAX . . . .THE LOST INFLECTIONS, THE DISCOURSE MARKERS, NONE OF US ARE NATIVE SPEAKERS, TELL ME I’M WRONG, TELL ME I'M WRONG . . . (p. 138)     

There are some bumps in this narrative road. For example, it's hard to sympathize with Ro (or anyone else) that would willingness extinguish the consciousness of another creature in order to replace it with their own. But, as a parable, it works great, and gestures to the extractive logic that unites both anthropology and capitalism. 

There are some great moments. Ro's supervisor tells him, "we understand their culture perfectly” (p. 84), a veiled(?) reference to the hubris of early, twentieth-century ethnographers. Malinowski, after all, laid out the extractive imperative in his introduction to "Argonauts of the Western Pacific":

Though it may be given to us for a moment to enter into the soul of a savage and through his eyes look at the outer world and feel ourselves what it must feel to him to be himself — yet our final goal is to enrich and deepen our own world’s vision, to understand our own nature and to make it finer, intellectually and artistically.  

As Vine Deloria, Jr. complains decades later, the cultural appropriation appears complete when indigenous peoples read anthropology in order to practice their own cultural beliefs:  “Many Indians have come to parrot the ideas of anthropologists because it appears that the anthropologists know everything about Indian communities” (1988:82).  Still, Franklin Hamilton Cushing aside, I'm not sure anthropologists really wanted to wholesale replace native peoples with themselves--although, as Rebecca Roanhorse writes in her short story, "Welcome to Your Authentic Indian Experience" (2017), that certainly seems like the next step after "entering the soul" of the cultural other. In any case, there's plenty here to stimulate self-critique in anthropological worlds. 

Actually, I find "The Language of Liars" to be a better parable for another form of cultural and linguistic extraction--generative AI. It's easy to see the appropriation here. Foundation models have been trained on massive data sets appropriated from our lives, the data that we continuously extrude as a byproduct of the digital surveillance state in which we live. And, of course, "low-resource" languages have also been sucked up into training data, albeit to a degree considered inadequate for model training. As with other attacks on data sovereignty, there are various forms of resistance to this wholesale appropriation.  

Another source of danger, however, lies in the apparent fluency of LLMs--even in languages marginalized on the internet. That fluency--heir to decades of research in natural language processing--appears to seamlessly shift between source and target languages. While literary translation seems a reach for LLMs, it will still produce what some would find to be adequate translations, even if zero context is given in the prompt. This is not to minimize the many problems of low-resource translation in LLMs (Elsner, 2024). But the problems exceed questions of accuracy. As Barkin reveals in a recent essay, generative AI translation utilizes English as a pivot language, even when source and target languages are not in English. That is, the source language is translated into English, with the translation into the target taking place at a subsequent layer.  

This isn't particularly novel--Google translate has worked in the same way. The difference, here, 

is in the context itself--the ways that LLMs appear to understand social and cultural life in addition to 

language. All of that extraction has led to some fairly convincing simulacra of cross-cultural understanding.

This creates “a powerful illusion of locality: by speaking and writing in users’ languages through a translation 

layer that is continuously fine-tuned by user input, they appear culturally situated even as they organize 

meaning through a a relatively stable set of epistemological assumptions inherited from overwhelmingly 

U.S.-based, English-language training data” (Barkin, 2025: 95). In other words, asking for information in 

a low-resourced language may yield a response, but it is one premised on an English-language 

epistemological frame, together with assumptions about the individual, the family and identity. 

The conversation may be in Uyghur, but the underlying logic will be from English. And as AI trains on 

AI slop, so the language itself will become more English in its logic. The "AI Slop Loop" is a race to the 

bottom, the death knell for accuracy, but also the hollowing out of low-resourced languages, now sock 

puppets with the hand of English language imperialism inside. 

 

The call for LLMs trained on indigenous language and culture is certainly pertinent here, but the hegemony 

on English in training data continues (Zoli et al, 2025). So, while plenty to beat one's chest about regarding the ethical failings of anthropology, Huang's 

novella tells us more about the replacement of language by its simulacrum: a kind of post-anthropological 

nightmare from which marginalized languages might never awake.  

 

References

Barkin, Gareth. "EPISTEMOLOGICAL PERSISTENCE IN MULTILINGUAL AI: THE ILLUSION OF LOCALITY IN LARGE LANGUAGE MODELS." INTERNATIONAL REVIEW OF MODERN SOCIOLOGY 51, no. 1-2 (2025). 

Deloria, Vine. Custer died for your sins: An Indian manifesto. University of Oklahoma Press, 1988. 

Elsner, Micha. "Shortcomings of LLMs for Low-Resource Translation: Retrieval and Understanding are Both the Problem." In Proceedings of the Ninth Conference on Machine Translation, pp. 1332-1354. 2024. 

Micha. "Shortcomings of LLMs for Low-Resource Translation: Retrieval and Understanding are Both the Problem." In Proceedings of the Ninth Conference on Machine Translation, pp. 1332-1354. 2024. 

Zoli, Carlo, Greta Mazzaggio, and Neri Binazzi. "Small Codes: a platform for digital resources and tools for minority languages and dialects." In DIGITAL HERITAGE (2025), pp. 1-9. Eurographics-The European Association for Computer Graphics, 2025 

Relooted and Its Predictable Controversies

In early 2026, the South African independent game company, Nyamakop, released "Relooted," a horizontal, side-scrolling heist game with lots of cutscenes. In this Afrofuturist game, players recover looted artifacts from western museums in order to return them to their rightful owners, along the way learning about the meanings of these objects in a variety of different countries and cultures.

 

 

 

In the screenshot above, players are briefed on Djenne terracottas, objects looted from the Djenne-Djenno archaeological site that date from 250 BCE to 900 CE. Gameplay involves making it past defenses, battling drones, and grabbing the artifacts. 

For a relatively modest game released by a heretofore little-known independent game company, Relooted generated a great deal of commentary and playthrough videos on YouTube. Partly because I wanted to try the new "topic analyzer" options on Communalytic, I downloaded 8000 comments from a number of playthrough videos and let Communalytic's algorithm generate clusters of related topics--relatively easy in YouTube since comments and replies can yield rich, linked data. 

 

 

The topic visualizer has an interesting '3D' interface that was less helpful than  a 2D for me. Out of the 107 clusters, I generated some general themes based on cluster. Here are the first 12:

1. Africans can't handle their own artifacts.

2. No one is playing this game.

3. Colonization was a good thing.

4. The game is racist.

5. 'Woke' and DEI issues.

6. Sluggish speed and bad design.

7. Western nations are the real thieves.

8. Developer was South African.

9. The trouble with the 'modern audience'

10. Can't believe this game even came out.

11. Games versus real life.

12. Poor game play and story.

Pretty predictable, really, with people presumably fine with Tomb Raider or GTA outraged by museums being robbed for the artifacts they stole. It is similarly astounding that people will default to 'white man's burden' racism so quickly--really just days after the game's release! Gamers suddenly sounding like E.B. Tylor guarding the collections at the Pitt Rivers! Also interesting--imputations of racism in that characters in the game are African--with African voice actors used.  

Anthropology's Sad AI Archive

 

There are 3 approaches to generative AI in the classroom: 1) an outright ban on it; 2) a limited use policy that covers certain assignments or parts of assignments, and 3) an open approach that allows students to do what they would. None of these are fool-proof, whatever the intentions of the professor. Ultimately, generative AI are third-party, black-boxed products–more tempting to students, perhaps, than Wikipedia, but also more treacherous. I feel for my colleagues in the humanities attempting to wrest essays from students on Shakespeare or Aristotle: generative AI is all too good at producing a mediocre essay on these subjects. I also understand my colleagues in the computer and information sciences, who utilize these chatbots to help with their instruction.

 

But with anthropology, there are several caveats. 99.99% of writings on other peoples of the world are drenched in ethnocentrism, colonialism and racism. The internet is awash in complete nonsense about “tribes'' and their “traditional culture,” and, in generative AI, all of this is ground up and, like sausage, pumped into prompt-driven content. Yet typically, students don’t know enough to be able to distinguish a “good” and “bad” response from ChatGPT or Gemini. 

 

This is a somewhat longer way of saying that students often tried to utilize generative AI in my introductory assignments and take-home exams, and their grades suffered for it. Not because I was penalizing them for cheating; proving that they’ve used AI is almost impossible, and generative AI detectors are unreliable at best. Instead, the questions that I asked were all about the anthropology I’ve taught in classes, and generative AI is, unfortunately, only too willing to spit out all manner of palaver. Only someone who knows what to ask can minimize the racism and colonialism inherent in generative AI engines. The default is ideology. And hallucinations. 

 

One thing I want to include next year is some process of education. I really think that students don’t really know any better. The least I can do is show them that it’s not so easy and explain why that is–that generative AI is not giving them the “truth.” Or, rather, it is: the truth of colonialism and racism that underlies Western thinking about non-Western peoples. Anthropology’s sad archive. But to someone who’s never taken anthropology before, this stuff looks correct to them, and the temptation is too strong, especially in the panicked moments before a deadline.