Almost no language grew up alone. Every sentence you spoke today contains borrowed material; it is just disguised well enough that you cannot pick it out. And borrowing is never only a matter of extra words. Sometimes what comes across is grammar, and sometimes two entirely unrelated languages end up looking like relatives.
Three things below: how borrowed material stacks into layers, why a language's words and its grammar can come from two different directions, and how a set of unrelated languages can end up living as one family. The last one matters most — it pushes the craft described in the episode on finding a language's relatives right up against a wall.
Back to those three readings. The earliest layer travelled with Buddhist texts from the south of China and is read myō, still audible in 光明 (kōmyō, "radiance"). The second came back with envoys from the Tang capital and is read mei, as in 明治 (Meiji, the era name) and 説明 (setsumei, "explanation"). The latest, min, survives in 明朝体 (minchōtai, a typeface). 行 works the same way: 修行 (shugyō, "religious practice"), 旅行 (ryokō, "travel"), 行灯 (andon, "paper lantern").
The point is that no language clears out its old loans before taking new ones. The new layer simply settles on top, and the old one stays put in whatever words it landed in back then. So a "messy" set of readings is not disorder — it is history that never got erased.
English has the same stratigraphy. Native shirt and borrowed skirt come from one Germanic root: the native line turned that consonant into sh, while Old Norse — the language the Vikings spoke — did not, so both versions survive and split the meaning between them. Even pronouns can be borrowed: they, them and their are from Old Norse, and the native third-person plural set was pushed out. Chinese has three visible layers too: 刹那 (chànà, from Sanskrit kṣaṇa, "an instant") and 塔 (tǎ, a clipping of stūpa, "pagoda") arrived with Buddhist scripture; 文化 (wénhuà, "culture") and 革命 (gémìng, "revolution") were coined in Meiji Japan out of Chinese characters and borrowed back.
What borrows easily and what resists is ranked. The World Loanword Database traced the source of 1460 everyday meanings in each of 41 languages; borrowed shares run from 1.2% in Mandarin to 62.7% in Selice Romani, the variety of a Romani community in Slovakia. Nouns go first, verbs next, function words last — a function word has to slot into a whole set of grammatical positions that are already arranged, while a noun only needs somewhere to sit. And that spread is mostly not a property of the languages: at the Mandarin end, speakers who rarely need a second language day to day and a powerful written standard; at the Selice Romani end, universal multilingualism, minority status, no written standard. Borrowing a lot does not make a language impure or inadequate — that percentage is a reading off a social situation, not a score on the language.
Tok Pisin is a lingua franca of Papua New Guinea with millions of speakers. Pull its words apart one by one and they are almost all English: mi "I", yu "you", gutpela man "a good man" (from good fellow), kukim "to cook something", where the ending -im (from him) marks a transitive verb.
But it has two words for "we". Yumi (you + me) includes the person you are talking to; mipela (me + fellow) does not. English has no such distinction at all. The Melanesian languages of the region do — the pronoun system of Kuanua, also called Tolai, spoken in eastern New Britain, divides things exactly that way. The shells are English; the categories are local. Haitian Creole is the same story: ap marks an action in progress and pral marks one about to happen, and both take their sound shape from French après and après aller, yet the arrangement and division of labour across the whole tense-aspect system line up with Fongbe, a Gbe language of West Africa, not with French.
This creates a problem. The craft described in the episode on finding a language's relatives rests on an assumption: that vocabulary and grammar point back to the same ancestor. Here they point in two directions. Which is why these languages are still not slotted into ordinary family trees — not because they are exotic, but because the tree can only draw a single source.
Whether they form a type of their own has been argued for over twenty years. McWhorter proposed a "creole prototype": very little inflectional affixation, no use of tone to distinguish words, transparent derivation — on the grounds that these languages are only a few centuries old. The other side (DeGraff, Mufwene and others) calls that creole exceptionalism: these languages went through the same processes as any other, and singling them out has historically travelled alongside contempt for the people who speak them. Strength of evidence: rich description on both sides, and recent quantitative work testing whether grammatical features are transmitted normally during creole formation came down on the side of "normally"; the argument is not over. What matters is that "simple" here is a countable technical measure — how many inflectional affixes there are — not poverty of expression. Any creole can say anything.
There is a larger backdrop that makes this argument less isolated. Lupyan and Dale surveyed over 2000 languages in 2010 and found inflectional complexity negatively correlated with population size, degree of contact and geographic spread: languages of small communities that few outsiders learn tend to carry more morphology, while those spread wide and learned by many adults tend to carry less. If that holds, then simpler morphology under heavy contact is a position on a continuum, with creoles at one end rather than off the scale. As for how such a grammar gets assembled within a single generation, the episode on children inventing grammar supplies the other half of the answer.
Something on the Balkan peninsula puzzled people for a long time. Albanian, Romanian, Bulgarian and Macedonian all attach the definite article to the end of the noun — where English puts the in front, they hang it off the back: Romanian lup ("wolf") becomes lupul ("the wolf"), Bulgarian masa ("table") becomes masata ("the table"). These four sit on different branches of Indo-European, only distantly related; and their own closest relatives, Italian and Russian, do nothing of the kind.
Nor is that the only one. The infinitive has left the area as a group: Macedonian and Modern Greek no longer use it at all, Bulgarian almost never, replacing it with clauses of the shape "I want that I go". The future tense is generally built out of the verb meaning "want", which is where the future markers of Greek, Romanian, Bulgarian and Albanian all come from. Greek missed the postposed article but took these two — features travel one at a time, not in bundles.
This situation is called a Sprachbund, a linguistic area: a set of languages that gradually converge in grammar because they sit next to each other, intermarry and trade. What carries the material across is bilingual people — someone running two systems in one head takes a construction from one side and lets it into the other, while the words themselves need not travel.
Areal features can also mislead badly. Vietnamese has tone, mostly monosyllabic words, and classifiers before nouns, which makes it look like Chinese or Thai; on that basis Maspero assigned it to the Tai group in 1912. In 1954 Haudricourt showed that Vietnamese tones grew out of consonants lost from the ends of syllables: those once ending in -h fell into one tone class, those ending in a glottal stop — the catch in your throat when you cut off "ah" abruptly — into another, somewhere between the sixth and eighth centuries, and the correspondences match Mon-Khmer material exactly. Vietnamese is Austroasiatic; its closest relative is Mường. Tone was not handed down by its ancestor. It is a piece of regional outfit shared across mainland Southeast Asia. The lesson is blunt: the loudest feature is very often the borrowed one.
There are cases close to that limit. In Michif, spoken in Canada, the noun phrases are essentially French — including the agreement between adjective and noun — while the verb system is essentially Cree, an Indigenous North American language. Each half is internally complete, and they are joined together. What it challenges is not whether such a thing can happen, but whether the question "whose descendant is this language" has an answer at all.
They are not answering the same question. A family asks about origin; an area asks what the language looks like now. A language can belong to one family and sit typologically with another set entirely — Vietnamese does. The trouble is that we habitually draw language relationships as a single tree, and one tree cannot hold both "it inherited from these" and "it lives among those" at the same time.
It is not circular, but it is not safe either. Body parts, numerals and pronouns are empirically less often borrowed, yet that is only a probability: English they, them and their came from Old Norse, and pronouns are supposed to be among the hardest things to borrow. A basic wordlist lowers the noise; what actually carries the weight is systematic correspondence, which is where loans tend to give themselves away.