Why ri is a Vowel

When I was learning the Bangla alphabet, one thing always bugged me: why is ri (ঋ) listed as a vowel? It is a consonant in every functional sense. Try saying it without the ‘r’ burst at the start. You can’t. The standard answer was: oh, it’s a vowel in Sanskrit. That never satisfied me, even as a kid. We were not learning Sanskrit. We were learning Bangla.

The same uneasiness kept showing up. Spelling & pronunciation rules contradicting each other. Two letters for i (ই/ঈ), two for u (উ/ঊ), three for the sh/s sound (শ/ষ/স), two for n (ন/ণ) - distinctions that exist on paper because Sanskrit had them, and that nobody in the room could actually pronounce apart. Spelling rules were never about sounds you could hear; they were about whether you’d memorized which Sanskrit ancestor the word came from. By the time I got to grammar properly, the feeling sharpened. Most of it felt forced. Like the grammar wasn’t there to organize patterns I could already recognize in the language around me; it was there to enforce rules that fell apart the moment anyone outside a classroom opened their mouth. For my own understanding, the only other grammars I had access to were English and (a little) Arabic. In both, grammar described how the language worked. In Bangla, grammar seemed to legislate how the language ought to work, against what people actually did.

In short, the grammar felt more like a wishlist than an ‘organic’ grammar.

Then there were the pronunciation drills. We spent serious classroom time being corrected on how to say everyday words. I had assumed my “wrong” pronunciations came from my parents (their speech carrying their own regional dialect into mine). But most of my classmates were first-generation Dhaka residents (parents from outside, relocated later), each with a different regional background. Statistically, our pronunciations should have been all over the map. They weren’t. We all sounded the same. And we were all marked “incorrect” the same way.

Whose pronunciation was correct, then? And who decided?

One day on YouTube, I saw a video titled Bangla Bhashar Uponibeshayan O Promitokoron. Normally I wouldn’t have clicked. High school was way behind me, I was in university, no trace of Bangla anywhere in my coursework, no reason to get into it again. But uponibeshayon (colonization) caught my eye. I’d been getting into decolonial studies for a while, pretty into it actually, and I clicked even knowing full well it could be clickbait - these days everyone slaps de/colonial on whatever they’re selling to make it sound serious.

It wasn’t clickbait.

---

The lecture is by Mohammad Azam,¹ professor of Bangla at the University of Dhaka. His PhD is on the colonization and decolonization of Bangla; his book Bangla Bhashar Uponibeshayan O Rabindranath (2014, 2nd edition 2019) is the most direct statement of the argument. The lecture cracked open every confused intuition I had been carrying since school. Below I’ll argue from his thesis directly, instead of stopping every paragraph to write “Azam says.”

The thesis, in one sentence: the version of Bangla I was taught to write is not my language. It was built by the colonial state in the late 18th and early 19th centuries, Sanskritized by Brahmin pandits on East India Company payroll, and its “standard” pronunciation, even today, still belongs to a city across the border that I have never lived in.

 

grammar : by the colonizer, for the colonizer, of the colonizer

In February 1778, Governor-General Warren Hastings stood in front of the India Board in London - the body running the East India Company on the British government’s behalf - holding a printed book, and pitched it. The book was A Grammar of the Bengal Language, written by a 27-year-old Company clerk named Nathaniel Brassey Halhed: the first English-language grammar of Bangla. Hastings wanted the Board to buy a thousand copies at Rs30 each, distribute them to Company servants, and pay the writer and printer up front. The Board took five hundred and paid Rs15,000 in advance.² The grammar was, in everything but name, a state project. Tax-funded infrastructure for governing a colony, with the financial record on file.

Why was the Governor-General of India personally hawking a grammar book in London? Because until 1835, the state language of Bengal was neither English nor Bangla. It was Persian.³ The British arrived in a region where they did not speak the local language and could not yet impose their own. They needed Bangla as a tool of administration, tax collection, and law (the Permanent Settlement of 1793 had to be written in something the zamindars could read). So they decided to learn it. And to learn it, they needed a grammar. Halhed’s book was that slot.

Here’s the fun part - at the time of writing, Halhed didn’t really know Bangla. He had come to Calcutta in 1772 as a Company Persian translator, and was close friends with Sir William Jones, the man who a few years later would launch the field of Indo-European philology. He knew Persian, Arabic, Hindustani, and (eventually) some Sanskrit. So how do you write a grammar of a language you barely speak? You hire a Brahmin pandit (?). The pandit knew Sanskrit & some Bangla, and Sanskrit grammar became the template. In Halhed’s defence, this wasn’t an unreasonable call, epistemologically. He knew English grammar had been written on Latin’s template, and Persian grammar on Arabic’s; he knew both Persian and Arabic, so he’d seen how that kind of templating works in practice. And Sanskrit grammar was, by any technical measure, well constructed - a fine candidate to template against. The problem was that he didn’t actually know Bangla. So when the Sanskrit template was applied, the actual spoken Bangla of 1778 (with its enormous regional variation, alive in millions of mouths from Sylhet to Midnapore) dropped out of the picture entirely. Debesh Ray’s account of this period puts it sharply: the (colonizer) sahib and the (Brahmin) pandit fused into a single figure, and that fused figure wrote the grammar of a language neither of them felt connected to. Rabindranath later marked sahib-pandit collaboration as Bangla’s ‘original sin’. He used the christian reference on purpose to represent one moment that stained everything downstream.

Underneath the grammar there is another layer. The book had to be printed, and the printing was its own colonial project. Halhed had almost nothing on the page to work from. Compiling the grammar in 1778, he could trace only six extant Bangla books, all in manuscript, the great religious epics included. Two decades later William Carey, hunting through Nabadwip (the supposed cultural and religious center of Bengal), turned up forty handwritten works in total. The corpus the sahib-pandit team was about to standardize, in print, was that thin.

The type to print it didn’t exist either. Joseph Jackson in London had a partial Bangla font in his 1773 inventory, filed under “modern Sanskrit” and glossed as “a corruption of the older characters of the Hindoos, the ancient inhabitants of Bengal.” Five years before Halhed wrote a word, the type founders’ trade catalogue in London was already classifying Bangla as a degenerate Sanskrit. Hastings turned to a young Company clerk named Charles Wilkins, posted at the Hooghly factory and already experimenting with Persian and Sanskrit type. Wilkins, with help from a Bengali engraver named Panchanan Karmakar, cut the first usable Bangla font from scratch. An Indian script runs to over six hundred characters once you count the conjuncts and vowel signs (a roman case needs about a hundred), and a composing room for it takes seven cases against roman’s two. Hastings’ own line on the achievement, in print: Wilkins was “obliged to charge himself with all the various occupations of the Metallurgist, the Engraver, the Founder and the Printer.” The grammar was printed in 1778 at a Mr Andrews’ press in Hooghly, fifteen miles outside Calcutta, on Wilkins’ types. Whatever letterforms he fixed in lead became the printed Bangla shape; the rest fell out of the alphabet by attrition. And in 1799 Wellesley issued a wartime order that no printing was to happen outside Calcutta. Until 1818 the entire infrastructure for putting Bangla on a page sat in one city.⁴

Worth pausing on what kind of class the pandit was. Bangla had patrons long before the British showed up, and they were Muslim. Sultan Ghiyasuddin Azam Shah is the one credited with commissioning Krittivasa’s Bangla Ramayana around 1400; Hossein Shah’s court a century later supported Shah Muhammad Sagir’s Yusuf-Zulekha. Court patronage of Bangla literature, across Hindu and Muslim subject matter, was the working arrangement of the period. The Brahmin establishment was on the other side of that arrangement. Orthodox opinion held that a Brahmin who translated the Sanskrit puranas into Bangla earned eleven thousand years in Raurava hell, and early Hindu translators like Maladhar Basu got around the embargo by claiming a god or goddess had appeared in a dream and ordered the work - divine command being harder to argue with than a Brahmin elder.⁵ So when Halhed’s people went looking for a pandit in 1778, the candidate pool wasn’t a neutral scholarly class. It was the one social formation that had spent centuries trying to keep Bangla as close to Sanskrit as possible, and out of reach of anyone who couldn’t read it. Sanskritization at scale wasn’t invented in Fort William. It was state-funded there.

Then came Fort William College in 1800.⁶ Founded by Lord Wellesley, ostensibly to train Company administrators in “the languages of command” (Bernard Cohn’s phrase). In practice, it was the factory where modern written Bangla was manufactured. Up to 1799, six British scholars produced translations of twenty-one legal works, all done with Brahmin-pandit collaboration. With Brahmin-pandit collaboration, again and again. M. Siddiq Khan, looking back on the period from a 1962 essay, summed it up without ornament: the English orientalists - Halhed, Carey, Nathaniel Pitts Forster - “stoutly promoted the teaching of Bengali in its pure Sanskritized form. The Islamic languages, including Muslim Bengali, were under attack.”⁴

And here is where the numbers get hard to look away from. In Early Bengali Prose (1960s), Sisir Kumar Das tabulated the percentage of Sanskrit-derived vocabulary in pre-colonial vs. Fort William-era Bangla writing. The Charyapada (the earliest surviving Bangla literature, ~10th century) sits at about 5% Sanskritic. Bharatchandra and Alaol’s Padmavati, working closer to colonial contact, run up to ~33%. Then come the Fort William books, all heavily pandit-supervised: Mrityunjay Vidyalankar’s Rajabali (1808) and Prabodhachandrika (written 1813), William Carey’s Itihasmala (1812), Ramram Basu’s Raja Pratapaditya Charitra (1801) on the lighter end. The Sanskritic share climbs through the lot. By the time you get to Vidyasagar’s Sitar Banabas (1860), the figure is roughly 92%. Ninety-two percent. A near-complete substitution of vocabulary in the span of a few decades, conducted by a small group of state-paid scholars. You can’t explain this with ‘natural drift’.

While the Sanskritic share was climbing toward ninety, the Arabic-Persian share was being squeezed in the opposite direction. Persian was the state language of Bengal until 1835; Persian and Arabic words had been seeping into Bangla through Hindustani for centuries. Words like adalat, kanun, dokan, kagoj, baksho, dorja, kissa, jamin are still in every Dhaka kitchen. In Fort William prose they vanish. Mrityunjay Vidyalankar, the First Pandit of the College, actively instructed his colleagues to drop Persian and Arabic words and replace them with Sanskrit ones. The Persianized literary register that had developed for Muslim readers, Dobhashi, was walled off as a separate, lower-prestige stream at the same moment the new “standard” was being built. The codification was, in plain terms, an Islamophobic project. The Brahmin pandit class doing the language work for the Company was constructing a written Bangla for the Hindu elite, and constructing it by stripping out the Muslim half (actually more than half) of the inherited vocabulary. The Sanskritization filling the resulting hole was the same project from the other side. A quiet partition of vocabulary, decades before anyone drew a partition on the map. In 1838 the state stamped it in law: a Company statute prohibited Arabic and Persian in any law court under its jurisdiction, three years after the English Education Act had taken Persian out of the state register.⁴ Salimullah Khan, looking back at Halhed’s volume, doesn’t flinch on the diagnosis: “a product of European Orientalism and a reactionary attempt to introduce communalism.” The book, in Khan’s reading, “claimed that Bengali was the child of Sanskrit, which was destroyed by infiltration of Arabic and Farsi vocabulary”² - and that single framing, Sanskrit-pure mother and Muslim loanword as corruption, became the working assumption of every downstream codification, even after nobody said it out loud anymore.

This was noticed at the time. Bankim Chandra Chatterjee, not exactly a decolonial firebrand but the most institutionally central literary figure of late-century Calcutta, wrote a piece called Bangala Bhasha in 1878 in which he ruled point blank that no word ought to be rejected on grounds of origin. Arabic, Persian, rustic, bucolic - if it served the writer’s purpose, it was Bangla. In Sahaj Rachana Siksha he recommended ishtehar (Perso-Arabic) over bijnapan (Sanskritic) for “announcement”. In an English-language piece on Bengali literature, he singled out the Sanskrit College authors, Vidyasagar named among them, for a “dull, pompous array of high-sounding Sanskrit words.”⁵ So the 92% number wasn’t invisible. It was being pushed back on, from inside the Calcutta establishment itself, by the man who at that moment more or less was that establishment. It just didn’t take.

The colonizers moved on. The pandits kept writing. By 1926, Suniti Kumar Chatterji’s The Origin and Development of the Bengali Language had become the canonical scholarly description of Bangla; almost every later grammar derives from it. But the language Chatterji was describing was already the post-Sanskritized one. Centuries of intervention had been baked in.

 

spoken vs written

The Bengalis of the time noticed. Rajnarayan Basu (to whom Madhusudan Dutt⁷ wrote letters - one of them including asking for comment on his own Meghnadbadh Kabya) reaches back twenty years and describes sitting in a gentleman’s house in Calcutta as a young man: “the elaborately Sanskritized Bangla words you find in any magazine or anyone’s writing, we never used them in speech.”

That is the gap. The language people wrote was not the language they spoke. By the second half of the 19th century, Ramendrasundar Tribedi, Haraprasad Shastri, and Shyamacharan Ganguli were noticing the same thing in Bangadarshan and at the Bangiya Sahitya Parishad. Pramatha Chaudhuri later gave it the line that still stings: in the normal life of a language, words flow from the mouth to the pen. In Bangla, after colonization, they flowed from the tip of the pen to the mouth. The literature dictated the speech, instead of the other way around.

And for a stretch, the diagnosis looked like it might actually translate into reform. Pramatha Chaudhuri’s Sabuj Patra, launched in 1914, ran the experiment in print: contributors tried to write in chalit bhasha, the language people actually spoke, instead of the Sanskritized sadhu bhasha. Rabindranath came over to the project, cautiously at first and then with decisive weight, and the field tipped. The implication was concrete and, for the keepers of shuddho, uncomfortable: bringing the written register back to the mouth meant making room for the Persian and Arabic vocabulary that had been part of everyday Bangla for centuries (adalat, kanun, dorja, kissa, jamin - the layer the Fort William pandits had been bleaching out). For roughly two decades, Hindu and Muslim writers were arguing in the same magazines for a joint reconstruction of the standard. It was, briefly, a real conversation.

Then the political weather changed. Through the 1920s and 30s, communal tension hardened across Bengal. The 1926 Calcutta riots bled into the years leading up to partition; the question of which loanwords belonged in Bangla stopped being a literary question and turned into a religious one. By the late 1920s, parts of the Muslim intelligentsia had given up on the Bangla project and started arguing for Urdu instead. The Hindu literary establishment, anxious about Muslim assertion, quietly closed ranks back around the Sanskritized standard. By the time Bangladesh came into existence and Dhaka was its capital, the cleanup had been done decades earlier. What we inherited as “standard” was a settlement we never sat at the table for, and the alternative - the reform program drafted by the people who had named the colonization while it was still happening - had been quietly written out of the textbook.

Bangla’s written form had been frozen into a Sanskritized facsimile that nobody’s mother had ever spoken. You can argue this also gave us the Bengal Renaissance. May be. But it also created a permanent class divide between people who could perform the artificial register and peop le who couldn’t, and that divide has never gone away.

 

the capital of language - outside of state

Eventually, chalit Bangla (the spoken-form prose standard) was itself standardized, mostly through Pramatha Chaudhuri and (more decisively) Rabindranath.⁸ The standard tracked the speech of the Calcutta region, which historically drew its elite from the Nadia-Shantipur belt - the West-Central dialect, modeled, in the canonical phrasing, on the variety spoken in the Shantipur and Shilaidaha region of Nadia and Kushtia. Why Calcutta won out over (say) Dhaka or Chittagong is a longer story; Azam declines to get into it in the lecture, and I’ll follow him there. But it won.

Azam calls this the capital theory of standard language. The standard of a language tends to be whatever the capital speaks. In 1900, the capital was Calcutta. So the standard was Calcutta’s. Fine, perhaps inevitable.

What is not inevitable is what came next. After 1947, after 1952, after 1971, after Dhaka became the capital of an independent Bangla-speaking state, the standard did not move. And it did not fail to move by accident. Look at the institutional record. The Bangla Academy, an idea Muhammad Shahidullah had been pushing since 1948 and the body finally chartered in Burdwan House in 1955, was the institution that should have done the work. After liberation, the Central Board for Development of Bengali was folded into it; a Textbook division was opened; the post of Director was upgraded to Director General. The Academy now had the mandate, the staff, and the political weather. What it produced, across the next fifty years, was a string of standardization documents that codified the inherited register: the Promito Bangla Bananer Niom (Standard Bangla Spelling Rules) drafted in 1992 and adopted by government in 1994, the Bangla Uchcharan Abhidhan (Dictionary of Pronunciation), the Bangla Banan-Abhidhan (Dictionary of Orthography), revised editions in 2000 and 2012. The NCTB was instructed to follow the Academy’s spelling in every primary and secondary textbook. None of these documents broke from the Calcutta-derived prestige model. They locked it in.

What does that lock-in look like in a Dhaka classroom? Concretely: when you say শরীর (shorir, body) without a separate palatal sibilant for শ - as nearly every Dhaka speaker does, collapsing শ and স into one /s/-like sound - you are corrected. When you say ফুল (ful, flower) with the bilabial fricative that is the natural Bangladeshi articulation, you are corrected toward phul. When you produce the alveolar /t/ and /d/ that the Dhaka mouth produces, instead of the retroflex stops the Calcutta mouth uses, you are corrected. Vowel nasalization, the rhotic merger of r/ṛ/ṛh into a single sound, the eastern lexicon (pola for chhele, lobon for nun) - all marked toward Calcutta. The Academy’s 2016 dictionary still rules parbatya (mountainous) malformed by Sanskrit derivational rules, even though it is the form Bangla speakers actually use.

And the diagnostic comes from outside the polemic. Sameer ud Dowla Khan’s 2010 paper Bengali (Bangladeshi Standard) in the Journal of the International Phonetic Association⁹ - a peer-reviewed phonological description, not an opinion piece - notes that Bangladeshi Standard Bengali is influenced by Kolkata Standard “as well as by the (nonstandard) Eastern dialect spoken in and around the capital.” Read that again. In the language of an international academic journal, the dialect spoken in the capital itself is the nonstandard one. The standard is whatever arrived through media, textbook, and pandit. Even the scholars building the institution from inside have worked within this frame. Shahidullah’s foundational work juxtaposed “standard forms” against “dialectal forms” without asking why those labels had landed where they had. Monsur Musa, a later Director General of the Academy, devoted a whole book - Bangladesher Rashtrabhasha (1995) - to urging the middle class to adopt Promito in administration, education, and public life, which is to say, to internalize the inherited measure. Humayun Azad, the most descriptively oriented Bangla linguist of his generation, did his work inside a standard he did not propose. Field studies of attitudes in Bangladeshi classrooms report what you would predict: students absorb the standard as proper and learn to feel their own grandparents’ speech as coarse or uneducated, and a generation of Dhaka-born speakers can understand Puran Dhakaiya but cannot produce it.

The pronunciation we all shared in that classroom was the Dhaka dialect (not the Dhakaiya dialect, by the way). Not regional drift. Not parental contamination. The dialect of the city we were sitting in, the capital of the country whose name literally means “land of the Bangla.” Marked wrong. Corrected toward something else.

This is the part of Azam’s argument I think Dhaka cannot keep waving away. A postcolonial capital, importing another capital’s speech as its measure of correctness, calling that import the protection of mother-tongue tradition. That is not preservation. That is a self-administered second colonization. It is also, structurally, exactly what the Brahmin pandit and the Halhed sahib did in 1778: take an external prestige variety and impose it on the speech of the people actually in the room.

 

the reckoning (or lack thereof)

You’d think there would be a reckoning by now. A national institution noticing it has been guarding the wrong thing. Even the start of that question. There hasn’t been one. A 2016 dictionary out of the Academy is, in substance, a 1930 dictionary, that is in substance a 1778 dictionary. The corrections it issues, the words it strikes through, the rulings it codifies, are still being made on Sanskrit’s terms, on the grounds that some Bangla word is malformed in a language nobody in the country speaks as a mother tongue. The colonial grammarian’s chair, freshly upholstered, still occupied.

Meanwhile the discourse the Academy actually engages with is the wrong one. Endless hand-wringing about Banglish, about kids corrupting the language with English loanwords, about social media flattening the prose. Some of that is fine; lazy mixing is in fact lazy. But it is not the central problem of this language, and the noise around it provides cover for the conversation the standard-keepers are least interested in having: that the standard they guard was never theirs to guard.¹⁰

The shallow ask is obvious. Dhaka needs its own standard. A register built from the city’s actual speech, codified, taught, defended in the dictionaries the Academy publishes.

The deeper point is that this is one symptom of something much larger, and I don’t think we take that seriously enough. The colonized mind doesn’t only show up in our grammar books. It shows up in which English accents we treat as smart, in which schools we treat as ‘good,’ in the dress codes we still measure ourselves against, in the buildings we copy. Bangla is just the corner of it I happen to be staring at this week. Decolonization, if we are actually serious about it (and I am not yet sure we are), cannot keep being confined to a panel discussion once in a while. There needs to be research, policy and deep uncomfortable personal audits, all at the same time, for a long time.

And there is a deadline I had not been thinking about until recently.

The written corpus any language model can scrape for Bangla is, almost by definition, the Sanskritized one. Fort William output, Bangla Academy publications, half a century of NCTB textbooks in promito. Dhaka speech leaves almost no written trace, by design. So a Bangla LLM trained on that data learns to produce the colonial settlement.

This is not a Bangla-specific worry. There is a growing literature on standard-language ideology in LLMs, the bias toward ‘an abstracted, idealized, homogenous spoken language imposed from above’ - almost word for word the Academy’s project. In English, models privilege the American variant over the British one and assign lower-prestige occupations to speakers of African American English. Those are well-resourced languages. Bangla isn’t. The largest open Bangla pretraining corpus, Bangla2B+, is around 35 GB; English sets run to tens of terabytes. Small corpus, skewed register, sharpened bias.

Then the feedback loop. The share of web text that is itself model-generated is climbing fast enough that researchers are openly worrying about model collapse, where each new generation trains on the previous generation’s output. Translate that to Bangla. Every government circular auto-drafted in promito, every school essay written by a chatbot in promito, every news copy paraphrased by an LLM in promito, becomes training data for the next model. The colonial standard compounds, in a register that was never anybody’s mother tongue, at a volume Halhed and Mrityunjay Vidyalankar could not have dreamt of.

So this is the shape of the deadline. 1778 to 1930 to 2016 was two-and-a-half centuries of the same settlement being copied by hand and then by press. The next loop runs at machine speed. A few product cycles in, after the models have eaten their own output once or twice, the standard may be impossible to back out of: the language the models produce becomes the language the public reads, which becomes the next generation’s training data, and the bottom of that stack is still a Brahmin pandit on Company payroll, telling Dhaka what its mother tongue is supposed to be like.

I don’t have a clean ending for this. I think we are, conservatively, one product cycle away from the reform window narrowing in a way that is hard to reverse. I also think most of us, including me, aren’t yet treating that as urgent.