How Foreign Words Become Chinese

A foreign word entering Chinese is rewritten one syllable at a time out of a small pool of characters picked for how they sound, and whatever those characters mean, and whatever tones they carry, arrives with them uninvited. That is the entire mechanism. The sound of the source word is the constraint; the characters are the only choice anyone gets to make, and almost everything people find strange about Chinese loanwords — the extra syllables, the melody the original never had, the fact that Taipei and Beijing write the same city differently — falls out of how that one choice gets spent.

Two paths, and the rare third

Chinese has two ways to take in a foreign name. 音譯 (sound translation) rewrites the noise: Kul Tiras becomes 庫爾提拉斯, four English syllables becoming five Chinese ones that mean nothing. 意譯 (meaning translation) throws the noise away and translates the sense: Stormwind becomes 暴風城, storm-wind-city, which a Chinese speaker reads as a description rather than a name. Blizzard’s Chinese localizations use both and split them by category — city names mostly go the meaning route (暴風城 Stormwind, 鐵爐堡 Ironforge), while race and region names mostly go the sound route (庫爾提拉斯, 贊達拉).

The third path, 音義兼譯 (sound-and-sense translation), takes both at once: characters that approximate the sound and read as something fitting. It is mostly a branding move, because it needs a coincidence — a string of characters that happens to sound right and happens to mean well. Coca-Cola landing on 可口可樂 is the textbook case. Taiwan’s word for hacker, 駭客, is another: it sounds like the English and 駭 carries a sense of alarm, where the mainland’s 黑客 (black-guest) only does the sound. Sonar became 聲納, sound-take-in. UFO became 幽浮, ghostly-floating. Viagra became 威而剛, roughly power-and-firmness.

Two more circulate constantly in translation teaching and are worth repeating with their status attached. The usual claim about Revlon’s 露華濃 is that it lifts a line from Li Bai — 春風拂檻露華濃 — which would make it both a sound fit and a literary allusion; no first-hand source for the origin story has been located. The usual claim about Carrefour’s 家樂福 is that the French name means “crossroads” while the Chinese was built to read “every household happy and blessed”; Taiwanese official and school material tells it that way. (The Taiwan chain was renamed 萬家福 in June 2026 after 統一 took over, so the example is now historical there.)

The third path is rare because it is expensive. Most of the time nobody finds the coincidence, the meaning slot goes unspent, and the word comes out of the machine as pure sound.

The character pool

Characters used for sound are bleached — worn smooth by use until the meaning stops registering. The standard teaching line for this is 麥克風, microphone, which has nothing to do with 麥子 (wheat) or 風 (wind). A transliterated word only means anything as a whole string; taking it apart produces nonsense, and Chinese readers do not take it apart.

The pool is small, and the assignments are close to fixed. A word-final -l or -r (Bill’s /l/, beer’s /r/) goes to 爾. An /s/ goes to 斯 or 絲. A final /k/ goes to 克, /m/ to 姆, /t/ to 特, /d/ to 德. Syllables starting with a vowel draw on 阿, 埃, 伊, 奧, 烏. The mainland’s tables carry hard rules on top: the letter s is written 斯 whether it is pronounced [s] or [z], which is why Jones is 瓊斯 and not 瓊茲. Cantonese runs a separate pool — Jack’s /k/ takes the entering-tone character 積, Rose’s /s/ takes 士 — and that is why Hong Kong renderings are recognizable on sight as Hong Kong renderings.

Why these characters and not others: the usual explanation is that they are high-frequency, semantically empty, and between them cover the syllable inventory. That is a reasonable account and no formal argument for it has been located, so treat it as folk consensus rather than a finding.

Two systems, one referee

The mainland has a single office. 新華社譯名室, the Xinhua transliteration room, was founded on 24 July 1956 in response to an instruction from Zhou Enlai: 譯名要統一，要歸口於新華社 — renderings must be unified, and they answer to Xinhua. It is the only official body that approves Chinese renderings of foreign names on the mainland, and it runs on two principles stated in eight characters: 名從主人 (the name follows its owner — translate from the pronunciation in the person’s own language) and 約定俗成 (an established rendering stays established, however wrong it looks now).

The apparatus behind that is substantial. 《英語姓名譯名手冊》 first appeared in 1965 and reached its fifth edition in 2018, carrying roughly 74,000 English-language names. 《世界人名翻譯大辭典》 came out in 1993 and was revised in 2007, with about 650,000 entries. Each source language gets its own 譯音表 — a syllable-to-character table, one character per sound, with slightly different tables for personal names and place names. Place names additionally have a national standard, GB/T 17693, covering English, French, German, Russian, Spanish, Arabic, Portuguese, Mongolian and (since 2019) Japanese, whose stated principle is sound translation first with due deference to established forms.

Taiwan has no compulsory standard at all. The system is that publishers and public bodies issue reference wordbooks for anyone who wants them, and long usage settles the rest. The closest thing to an authority is the 國家教育研究院 (formerly 國立編譯館), which publishes 《外國地名譯名》 (fourth edition 2014; the third carried 66,632 place names with English name, Chinese rendering and coordinates) and 《外國學者人名譯名》 (8,912 names across ten source languages), both posted publicly on 樂詞網. Its stated principle is the same 名從主人. The de facto press standard is 中央社, which compiled 《標準譯名錄》 in 1980 and turned it into a database of roughly 122,000 entries. Media do not fall in line: some outlets use CNA’s renderings, some invent their own, and the same outlet can print one person’s name two ways.

The compressed version: the mainland has a table and a referee, Taiwan has reference books and no referee. Neither arrangement is new — a 1924 Republican-era formulation already reads 即西文每一音素，均擇一發音相當之漢字，作為標準譯字: for each Western phoneme, pick one Chinese character of matching sound and make it the standard rendering character.

Where Taipei and Beijing split

The result is visible on names everyone knows. Obama is 歐巴馬 in Taiwan, 奧巴馬 on the mainland and in Hong Kong. Trump is 川普 versus 特朗普. Bush is 布希 versus 布什 (Hong Kong: 布殊), applied identically to both presidents. Reagan is 雷根 versus 里根 (Hong Kong: 列根). Kennedy is 甘迺迪 in Taiwan and Hong Kong, 肯尼迪 on the mainland. Clinton is 柯林頓 versus 克林頓. Sydney is 雪梨 in Taiwan and Hong Kong, 悉尼 on the mainland — a mainland reader meeting 雪梨奧運會 in Hong Kong or Taiwanese news can genuinely fail to identify the Olympics being discussed.

There is a structural explanation for the personal-name half of that, repeated by China Daily and not sourced further than that article: Taiwan inherited a Republican-era habit of making the first character of a foreign personal name a real Chinese surname, which gives 甘迺迪, 雷根, 柯林頓, and even reaches for obscure surnames — 胡笙 for Hussein, 佘契爾 for Thatcher — where the mainland, having no such convention, produces 肯尼迪, 里根, 克林頓, 侯賽因, 撒切爾.

The Obama case is the one time the split was fought in public. In November 2009 the US embassy in Beijing announced that the US government’s standard Chinese rendering was 歐巴馬, on the grounds that it sits closer to the English pronunciation. Xinhua declined to move: Obama is a common Kenyan surname with an established rendering, 奧巴馬 had been in use for years, and head-of-state renderings do not get changed lightly. By February 2015 the embassy’s own press releases had gone back to 奧巴馬. (A local story that Taiwan avoids 奧 because the character carries a sense of underhandedness in Taiwanese — 奧步, a dirty trick — circulates in Hong Kong press with no documentation behind it, and should be treated as folklore.)

The tones ride along

There is no flat-tone rule. Transliterations are not systematically first tone, and the tone that lands on a syllable is whatever tone the chosen character already had.

The numbers settle it. Eleanor Glewwe’s 2015 UCLA master’s thesis built a corpus of 3,660 syllables drawn from 1,372 English loanwords in Mandarin. Baseline tone frequency across general Mandarin vocabulary runs T1 26.9%, T2 24.7%, T3 18.3%, T4 30.1% — fourth tone, not first, is the commonest tone in the language. In her loanword subset (N=3,047) the figures are T1 28.8%, T2 29.5%, T3 16.0%, T4 25.7%. First tone in loanwords therefore runs about two points above its baseline. That is the whole of the “flat tone” effect, and two points is close enough to noise that no rule can be built on it.

What did condition tone in her corpus was the voicing of the English consonant a syllable starts with. Voiceless obstruents — p, t, k, s, f, the consonants made by blocking or squeezing airflow with the vocal cords switched off — took first tone 38.4% of the time (N=1,957). Sonorants — m, n, l, r, w, y, made with the vocal cords running throughout — took first tone only 11.6% of the time, and second tone 37.2% (N=1,090). Glewwe’s account is F0 perturbation: a voiceless consonant nudges the pitch of the vowel after it upward and a voiced one nudges it down, and the person doing the borrowing hears the raised pitch as high tone and the lowered one as rising.

Her own experiment then contradicted her corpus. Live speakers adapting invented words keyed on English stress instead: stressed syllables took first tone most often, word-final stressed syllables took fourth, and the voicing effect vanished entirely. Glewwe’s conclusion is that the two studies conflict and that tonal adaptation “depends on overall English intonation rather than small F0 perturbations.” The field does not have a settled phonological rule, and we should not write as though it does.

Zhuting Chang’s 2020 CUNY dissertation gives the answer that actually explains the data, by declining to look for a phonological rule at all: tone assignment operates “beyond any usual phonological concerns.” What adapters use is implicit knowledge of tone probability and character frequency in the existing lexicon, plus “seeking of appropriate characters based on their meanings, either to avoid unintended readings of loanwords or to form desired interpretations.” Chang credits the writing system itself, previously underrated in this literature — a script where one character is one syllable, homophony is everywhere, and compounding is free. The load-bearing sentence for everything on this page: you pick the character for its meaning, and the tone comes along for the ride.

One hard constraint sits on top of the choice. Mandarin has tonotactic gaps — syllable-and-tone combinations that simply have no character. There is no dán. Loanword adaptation does not fill those gaps; only combinations already in the native lexicon are available. So a share of the tones in any transliteration were never chosen at all: that syllable comes in one tone, and that is the tone you get.

Three worked examples make the point better than any statistic:

• 可口可樂 kě-kǒu-kě-lè — 3-3-3-4. The most famous transliteration in the language contains no first tone whatsoever.
• 德魯伊 dé-lǔ-yī — 2-3-1. Contains a third tone, the rarest tone in the language.
• 波拉勒斯 bō-lā-lè-sī — 1-1-4-1. Three first tones out of four, which is the shape people have in mind when they say transliterations come out flat.

Three words, three completely different tone shapes, all three verified against 教育部重編國語辭典. And the clincher is the fourth item: Boralus splits across the strait, and the split moves the tone. Blizzard’s Taiwanese localization writes 波拉勒斯, starting on 波 (bō, first tone); the mainland writes 伯拉勒斯, starting on 伯 (bó, second tone). One English source word, two teams, different first character, different tone. If the tone were being computed from the English, that could not happen.

One narrow sub-rule does survive. An earlier claim in the literature (Wu 2006) that inserted syllables beginning with /s/ take first tone holds up in Glewwe’s corpus: of 251 first-tone inserted syllables, 195 are the single syllable sī. That syllable is 斯, and it is the -s at the end of 波拉勒斯 and 庫爾提拉斯.

A practical footnote for anyone saying 可口可樂 out loud: 3-3-3-4 is the citation form, not the spoken one. Two third tones in a row turn the first into a rising tone, so 可口 comes out ké-kǒu and the whole word is ké-kǒu-kě-lè.

Why the source speaker hears nothing familiar

Glewwe’s thesis opens on the fact that drives all of this: English words borrowed into Mandarin must be assigned tones. There is no toneless option. Mandarin has no un-toned syllables, so the borrowing process invents pitch information that was never in the source word.

Three transformations then stack, and a fourth runs alongside them.

The word gets longer. Mandarin does not allow consonant clusters, so English dr- cannot survive intact — it splits across two syllables, 德 + 魯, and two English syllables come out as three. Mandarin also limits what can end a syllable to -n, -ng and -r; every other final consonant is either dropped or promoted to a syllable of its own. Boralus’s final -s becomes 斯, and three syllables become four. Glewwe’s term for the inserted ones is epenthetic syllables, and her clearest case is lace arriving as lěisī: one English syllable, two Chinese ones.

The word gains a tune. Every syllable now carries a pitch contour the original never had, and those contours came from the characters, not from the English.

Every sound gets rounded to the nearest legal Mandarin one, because a syllable that does not exist in Mandarin cannot be written.

And underneath all three, per Chang, the adapter is shopping for characters whose meanings will not embarrass the result, so the sounds get bent further to reach a character that reads acceptably. Where a tonotactic gap bites, there was no choice to make in the first place.

By the time Boralus comes out the other side it is a syllable longer, it is carrying a melody, every sound in it has been rounded, and at least one of its syllables was chosen for reasons that have nothing to do with English. An English speaker not recognizing it is the expected outcome, not a failure of the translation.

Reading the results literally shows how empty the pool is. 德魯伊 is 德 (moral norm, virtue), 魯 (dull-witted, clumsy; also the ancient state of Lu, Confucius’s home), 伊 (a literary third-person pronoun, today mostly a pure transliteration syllable — 伊朗 Iran, 伊拉克 Iraq). Virtue-dull-he. 德 is a high-prestige character carrying zero meaning here; it is present because it sounds like d-, the same way it does in 德國 Germany and 德州 Texas. Meanwhile the English word means “oak-knower” — Proto-Celtic *dru-wid-s, a reading Pliny the Elder endorsed — and a meaning translation exists on record, 橡木賢者, oak sage. Nobody uses it.

波拉勒斯 reads the same way: 波 (wave), 拉 (to pull, drag), 勒 (a bridle or bit), 斯 (to split; as a literary word, “this”). Wave-pull-bridle-this. That 波 happens to land on a harbour city is coincidence — 波 is the default character for foreign bo-/po- syllables, as in 波士頓 Boston, 波蘭 Poland, 波音 Boeing.

Worth knowing about 德魯伊 specifically: Blizzard did not coin it. Druid is the priestly class of the pre-Christian Celtic religion, and the older Chinese rendering is 督伊德 (dū-yī-dé, tones 1-1-2 — the same source word with a completely different tone shape). It was in games before World of Warcraft: Diablo II: Lord of Destruction added the Druid class on 29 June 2001, with the Taiwanese release writing 德魯依 and both spellings still alive in the community; upstream of that it is a classic Dungeons & Dragons class, uniformly 德魯伊 across Taiwan’s Chinese D&D 5E material. Blizzard is the company that pushed the word into general vocabulary, not the one that made it. (A widely repeated claim holds that 德魯伊 came into Chinese via Japanese ドルイド rather than straight from English, which would explain its distance from the English sound. Chinese Wikipedia’s article does not say this and no primary source has been found. The syllable mapping does fit ドルイド more tidily than it fits /ˈdruːɪd/, which is suggestive and is not evidence.)

The Coca-Cola name, and the story about it

The true history is undramatic: the name evolved in the Shanghai market across the 1920s, and nobody won a prize for it.

祝淳翔’s research traces the lineage back to 可口汽水 in 《時報圖畫周刊》 no. 160, dated 6 August 1923. Watsons (屈臣氏) ran 可口露 advertisements in 《申報》 on 1 May 1924 and 《新聞報》 on 2 May 1924, and 可口水 appears in 《工商新聞》 on 10 October 1924. The inaugural issue of 《商業雜誌》 in November 1926 states the mechanism outright — 可口係 coca 之譯音, 可口 is the transliteration of coca — so the sound-mapping was explicit at the time. 葉新 located a May 1930 advertisement in 《商業雜誌》 reading 屈臣氏汽水可口可樂. The Coca-Cola Company applied for the trademark on 26 March 1930, received approval on 24 June 1931, and had it published in 商標公報 no. 54 on 15 August 1931; the agent of record was the American lawyer Norwood F. Allman (阿樂滿), who practised in Shanghai from 1916 to 1950 and served as Coca-Cola’s China counsel.

The famous story, which is false, runs like this: in the mid-1930s Coca-Cola advertised in The Times offering £350 for a Chinese name, and the London-based scholar 蔣彝 (Chiang Yee, the “Silent Traveller”) entered 可口可樂 and won. Taiwanese press including CNA still prints it. It dies on dates — 蔣彝 did not reach London until 1933, and 可口可樂 is in print in 1930; in 1930 he was serving as county magistrate of 九江 in Jiangxi, in China, and cannot have submitted anything to a London contest. Two more objections stand against it: The Times is a British newspaper with no reason to run a Chinese-naming competition, and no such advertisement has ever been found.

The companion legend, also false, is that Coca-Cola’s earlier Chinese name was 蝌蝌啃蠟, “tadpoles bite wax.” It was never an official name. Its documented origin is Allman himself, writing it up in a 1957 issue of Coca-Cola Overseas as a negative example — a 荒誕字眼, an absurd rendering, illustrating merchants who chase sound and ignore meaning. There is a real phenomenon underneath the legend: Coke reached China around 1927–28 before the company had registered a Chinese name, so individual shopkeepers wrote their own characters on in-store placards and some of them were nonsense. So “bite the wax tadpole” signage plausibly existed on shop walls; 蝌蝌啃蠟 as the official predecessor brand name traces to a 1957 cautionary anecdote, not to a 1927 label. Chinese Wikipedia files the whole package — the tadpoles, the £350, 蔣彝 — as 傳聞, hearsay, and 備受質疑, heavily doubted.

None of that diminishes the name. 可 means to affirm or to be suitable; 口 is the mouth; 樂 as lè is joy. 可口 kěkǒu is a real pre-existing Chinese word meaning tasty or palatable, not something invented for the brand, and 可樂 reads as “can be enjoyed” — and has now been so completely captured that in Taiwan 可樂 simply means cola. (樂 is a three-reading character — lè happy, yuè music, yào to be fond of — and a standing trap for learners; here it is lè.) The name tracks ko-ka-ko-la in sound and reads “tasty + joy” in meaning, both at once. The ranking of it as the best brand translation ever made into Chinese is advertising-industry consensus rather than a measured finding, but the double fit is real and it is why the example is unkillable.

What stays open

Two gaps are declared rather than papered over. No Chinese-language study of 音譯詞聲調 using Taiwan-specific data has been located, so every solid finding on loanword tone assignment above is English-language and based on mainland or general Mandarin. The tone values for 可口可樂, 德魯伊 and 波拉勒斯 are identical in Taiwan and on the mainland regardless, so the worked examples hold, but the general claims rest on sources that never looked at Taiwan. And the claim that 德魯伊 arrived through Japanese ドルイド is unsourced past aggregator sites; the syllable mapping fits, which is a reason to keep looking and not a reason to believe it.

Neither gap moves the mechanism. The sound is the constraint and the characters are the choice, and the choice is the only place a translator has any freedom: spend it on meaning and you get 可口可樂, tasty and joyful and phonetically close, a name that outlived every story invented to explain it. Leave it unspent and you get 德魯伊 and 波拉勒斯 — virtue-dull-he, wave-pull-bridle-this — words whose characters were picked for sound alone and whose tones are simply the tones those characters happened to have. The unspent case is the ordinary one. That is why the tones look random, why two localization teams can hand back the same city with different pitch, and why the person who owns the original word rarely hears it in the result.

Grounding, with every fact’s confidence marked: Transliteration Into Chinese (音譯). Adjacent in the character cluster, on why tone is learned as a separate fact from the character that carries it: Sound Components.