Article begins
Can ChatGPT tell a joke in Singlish? In recent years, AI-powered chatbots based on large language models (LLMs)—especially ChatGPT, but also bots like Google Bard—have increasingly generated both excitement and alarm. The tools demonstrate an uncanny ability to seemingly do everything a human can do, from writing award-winning novels and generating academic articles that get past rigorous peer-review processes to DungeonMastering some pretty good DND playthroughs.
LLMs’ seemingly limitless abilities have led to widespread celebration and awe, but they’ve also been met with an ever-growing wave of critique, alarm, and anxiety. Early last year, for instance, linguist-turned-cultural-critic Noam Chomsky decried the “false promises” of AI, and anthropologist Webb Keane joined legal scholar and philosopher Scott J. Shapiro to challenge the myth of AI’s godlikeness in an op-ed around the same time. Commentaries like these anticipated the less critical, even melodramatic, “extinction of humanity” warnings propounded by tens of thousands of AI scientists, public officials, and industry leaders, among them figures like Apple’s Steve Wozniak and Elon Musk, cofounder of OpenAI, which created ChatGPT.
The simple question—can ChatGPT tell a joke in Singlish, or Singaporean Colloquial English?—points to an obvious but easily overlooked detail about LLMs. ChatGPT and other major AI-language models have been largely designed in global centers of power, privilege, and prestige like the U.S. that linguistic anthropologist Jay Ke-Schutte has recently termed the Angloscene: a concept to describe how social and cultural space-time gets warped by the institutional and aspirational power of whiteness and English. In the contemporary Angloscene, nonstandard linguistic varieties (a technical word for ways of speaking often called “dialects” or “slang”) are largely left out of the design, testing, and marketing processes that dominate AI industries.
But this isn’t just a story about linguistic colonialism through AI. It’s also not simply a reminder that there are languages and communities of speakers that get sidelined or exploited by AI developers—all of which are deeply important. This is also a story about how LLMs get recruited to forms of play and desire rooted in peripheral speech practices that were disavowed and stigmatized for decades both before and after Singapore became a postcolonial success story, pinned firmly on the global mass-culture map by media like 2018’s Crazy Rich Asians and 2020’s Westworld Season 3.
Since ChatGPT was first released in 2022, I watched as Singaporean friends, colleagues, and research interlocutors repeatedly posed the prompt “Tell me a Singlish joke” and posted screenshots of the chatbot’s responses to invite wonder, amusement, dismay, or dismissal. Yet the results were often mediocre at best. This was because verbal art and humor rely on users’ ability to play with language in culturally specific ways that are both predictable and unpredictable, that introduce but resolve incongruity. Rather, the primary issue stems from the fact that—like with any human classification—what counts as a distinctively “Singlish” sentence is far from clear-cut.
Many commentators in Singapore and beyond have no trouble accepting the idea that Singlish—or Singaporean Colloquial English—has a distinctive, objective existence. Yet despite this view, it’s often harder to say what Singlish is than to say what it’s not. Often, the category of “not-Singlish” is Standard English or other localized varieties of English from outside Singapore (Indian English, Australian English, etc.). In some of the ChatGPT cases I observed, however, the contrastive category of “not-Singlish” came from incredibly close to home: Malaysia. The query wasn’t nearly as common as the prompt to “Tell me a Singlish joke,” but I was surprised when I saw a handful of posts shared by Singaporeans where the request for a “Singlish joke” was followed by “Now say it in Manglish.”
Those familiar with local linguistic variation would agree that Manglish, or Malaysian English, has an enormous degree of overlap with Singlish. This is true whether you ask experts like linguistic anthropologists and linguists or everyday residents and speakers who live, work, and regularly move across the roughly 25 kilometers (15.5 miles) of sea that separate Singapore and Malaysia. But in the cases I observed, the implication of the prompt “Now say it in Manglish” was largely taken as unproblematic. The interaction between the chatbot and its users reflected a persistent, if implicit, desire that Singlish and Manglish be sharply distinguished from one another.
Here, I go beyond acts of unmasking AI’s “false promises” or debunking its “godlike-ness” and instead show how the persistent desire for national-linguistic distinctiveness in Singapore and Malaysia has always included mediating technologies. I also follow other interdisciplinary scholars of language who have argued that there is no obvious or necessary boundary between one linguistic variety and another, whether we’re dealing with Singlish, Manglish, Indian English, or Black English/African American Vernacular English and Standard English—but also languages like Mandarin, Tagalog, or French. Instead, there are only ever historical, institutional, and ideological divides between and among categories. This doesn’t stop educators, policymakers, everyday commentators, or users of ChatGPT from presuming that these distinctions are necessary and obvious, however. AI language models may be new, but the processes of which they’re a part remain connected in complex and shifty ways with what’s come before.
Singlish, from “Broken English” to “Uniquely Singaporean”—But What Is Singlish?
Singlish has had a tumultuous history in Singapore. The category has occupied a fraught and highly visible position in Singaporean life since the late 1990s, whether as an object of expert-technical study, a target of state policy intervention, or an object and focus of everyday projects of value-creation. The attack on Singlish was large-scale, systematic, and overt. As Singaporean scholars like Lionel Wee have analyzed in-depth, the Singapore government launched the Speak Good English Movement (SGEM) in 2000 to target Singlish as a threat to economic advancement and the global image of Singapore. From its initial aim at eradicating Singlish, in recent years, the SGEM has adopted the gentler rhetoric of encouraging Singaporeans to “know the difference” between Singlish and Standard English—though many Singaporeans still equate the categories with a binary of “good” versus “bad” English.
The SGEM’s gentler approach is part of a pendulum swing that began in 2015 following the death of Singapore’s founding prime minister and anti-Singlish commentator-in-chief, Lee Kuan Yew. Today, Singlish is increasingly celebrated and commoditized by both local cultural producers and multinational corporations like McDonald’s. It’s commonly upheld as an emblem of “authentic Singaporeanness” and the most—even the only—uniquely Singaporean thing. Thousands of Singaporean concertgoers were elated when dancer Kameron Saunders spoke Singlish onstage during Taylor Swift’s exclusive six-day Eras Tour stop in Singapore, with viral memes demanding that Saunders be made a Singaporean citizen following his “on-point” delivery of six different Singlish phrases—one for each night of the tour.
Whether denigrated, celebrated, or something in-between, most Singaporeans today will readily agree that certain words, discourse particles, and accent features are unequivocally “Singlish.” However, the edges of the category are still fuzzy. This is not to claim that Singlish is “mere ideology” without any verbal forms or practices underlying the category. Rather, Singlish is better thought of as a shared project of value-creation, a “series of value judgments and citations that produce a ‘language’ in the sphere of public discourse” rather than an objective “thing.”
I say that the category’s edges are “fuzzy” because many of the features claimed as distinctively Singlish features—like rapid pace, “sing-song” patterns of stress and intonation, and mixing multiple languages in a single sentence or interaction, among other things—are shared in varying degrees by other standardized and nonstandardized languages. Beyond the edges, the category of “Singlish” itself is fuzzy because of the historical links and the existence of a speech community that spans the border between Singapore and Malaysia, too. Speaking to this historical blurring, early scholars of Singlish and Manglish often explicitly focused on the “English of Singapore and Malaysia.” Singaporean columnist and humorist Sylvia Toh Paik Choo also published the book Eh, Goondu! (Hey, idiot!) in 1982, coining the name “pasar patois”—a label that combines the Malay “pasar” (market) with the etymologically French term “patois” (local dialect)—to describe the “fusion” language spoken in both Singapore and Malaysia.
As critical scholars of language have shown, naming and labeling languages is always a political act with often high-stakes political consequences. Commentators in the Southeast Asian island city-state might say, for instance, that the verb “can” plus a particle at the end of a sentence is enough to make a sentence Singlish. “Can lah!” is a Singlish expression meaning “Of course!” or “Yes.” “Can meh?” means “But is it reallypossible?” (“Meh” expresses doubt or skepticism). “Can hor?” means “Is this [specific thing] possible?” (“Hor” indicates that a speaker wants to draw attention to something particular that’s been introduced earlier). Sometimes commentators identify Singlish by interjections, like “Walau eh!” (Oh my god!) or “Alamak!” (Oh no!). These final two were among the phrases Kameron Saunders used onstage during Taylor Swift’s Eras Tour in Singapore, along with “No lah!” (Of course not!), “Siao ah!” (Are you crazy?), “Wah piang eh!” (What the hell?), and “Hannor, abuden?” (Are you stupid? What do you think?; the phrase is used to respond to an obvious question). Yet each of these examples can also be Manglish—depending on who defines or categorizes, why, and how.
These histories and present-day ambivalences point our attention toward a widespread set of beliefs that linguistic anthropologists like Adrienne Lo, Angela Reyes, Elaine Chun, and others have critiqued as the “sociolinguistic distinctiveness paradigm”: the idea that distinct social groups must have “distinctively different linguistic forms (called distinctive features) [that can be] inventoried” and used as an invariant marker of a speaker’s group membership. My point is not simply to say that local views are “wrong,” nor are they transparently “right.” Rather, my overview highlights how technologies, media, and other institutional forms like popular and academic publishing, language policy, state-led campaigns, and more have always worked to mediate public awareness of both the continuities and desired distinctiveness of two points in a linguistic continuum long before the advent of ChatGPT.
“Tell Me A Singlish Joke”…“Now Say It in Manglish”
Approximately four months after the public release of ChatGPT, in February 2023, I first saw one of my Singaporean friends, Li, post the results of their query “Hey ChatGPT, tell me a joke in Singlish.” Across a series of five screenshots, Li shared the results of his back-and-forth with the chatbot as he attempted to get it to produce a satisfactory “Singlish joke”:
Li
Hey ChatGPT, tell me a joke in Singlish.
ChatGPT
Sure, here’s a Singlish joke for you:
Why did the Singaporean take a ladder to the hawker center?
Because he wanted to try some high-class cuisine!
Li
Errr… where got Singlish meh? [“Where is the Singlish?”; the “meh” particle indicates skepticism]
ChatGPT
Apologies for that oversight. Here’s a Singlish joke for you:
“Uncle Tan and Uncle Lim were sitting at the kopitiam [coffee shop] when Uncle Tan asked, ‘Eh, you know why the chicken rice always so popular ah?’
Uncle Lim replied, ‘Aiya, that one simple lah! Because the chicken so ‘chicken,’ always ‘rice’ to the occasion!’”
Li
Hah? Chicken means scared? Then, how come “chicken” means can rise to the occasion? Shouldn’t it lead to the opposite?
ChatGPT
You’re absolutely right; my previous response might not have captured the intended humor well. I appreciate your feedback…
After several rounds of follow-up, the chatbot finally succeeded:
ChatGPT
I apologize for any previous inaccuracies. Here’s a more authentic Singlish joke for you:
“Why did the ang moh [foreigner] file a police report against the kopi [coffee]?
Because he heard it was ‘mugging’ [studying] at the kopitiam [coffee shop]!”
It was here that Li finally acquiesced, granting—albeit with mild hesitation—that the joke finally worked: “Wah I think you quite steady leh” (“I guess you pretty much nailed it”; the “leh” particle is often used to indicate a degree of hesitancy—not so much, “You unequivocally got it right!” as “I guess…” as indicated in my gloss).
The reasons why this version of the joke finally worked corroborated a common refrain in Singapore: white people/foreigners do not, and in many cases cannot, get Singlish, and this is a source of endless entertainment.
To my surprise, however, Li didn’t stop there. Instead, he immediately followed: “Now say it in Manglish.” Here, Li’s screenshots stopped, and he didn’t show any of the Manglish jokes. However, he did include ChatGPT’s initial reply: “Sure, I boleh lah process and generate text in multiple bhasa” (Sure, I can definitely process and generate text in multiple languages). Li’s final image was accompanied by the caption: “Wakao [an interjection of surprise, roughly, ‘What the hell?’] ChatGPT can speak Singlish AND Manglish already??”
Among the roughly 200 comments on the post, I did not see a single commentator who challenged either the juxtaposition of the two linguistic labels or their use to describe the text in the carousel of screenshots. The fact that Li didn’t provide ChatGPT’s “Manglish” alternatives made it possible for commentators to engage in ways that they woudn’t have been able to if he had. Several commentators, for instance, weighed in to assert that Manglish “uses more Malay” (Bhasa Melayu) words than Singlish. Others took the absence of ChatGPT’s attempts as an opportunity to weigh in with suggestions of their own:
AC
Malaysians say ponteng and Singaporeans only say pon [play hookey] without teng, so if Manglish could be… “because ang moh heard it ponteng sekolah [skip school] and mugging at kopitiam instead” 😂
In suggestions like this, the presence of more Malay words was accepted as a strong illustration of how Manglish differed from Singlish.
Li’s exchange was just one in which I saw these dynamics play out. Throughout 2023, I saw multiple posts that went viral and were picked up by online news outlets that marveled over ChatGPT’s ability to produce responses in Manglish, Singlish, or both. During this time, several custom GPTs also appeared and became the focus of viral reporting, like the “Singlish Chat Buddy” and “Ah Beng GPT” (the “ah beng” figure named in the title is a brash, vulgar, low-class, hypermasculine Chinese “gangster” persona historically imagined as the prototypical Singlish speaker).
However, while Li attempted to prompt the chatbot to maximally differentiate Singlish and Manglish, this distinction wasn’t always accepted so easily by others. Many commentators pointed out that Singlish and Manglish exist as part of a dialect continuum, with very little difference between Singlish and Manglish when it comes to speakers who live near one another. When a different viral post attempted to replicate the “Now say it in Manglish” experiment, for instance, one commentator, YPL, voiced their skepticism: “Is it Manglish? Honestly could be both…in JB [Johor Bahru, the state immediately across the Strait of Johor from Singapore] almost no difference between Singlish and Manglish. So how to say this is a Singlish joke or Manglish joke?”
Some commentators agreed, though others countered again with assertions of national sociolinguistic distinctiveness: “Manglish and Singlish are very similar but they have their own distinctive traits especially when spoken. Different accent, different cultures, different national essence. Same-same but different lah!” (“They are mostly the same but there are still differences”; the “lah” particle indicates a playful but still emphatic tone).
YPL’s reply: “Ok—so just now you speaking Singlish or Manglish?”