Module 40: Language

about this module

Language is the most sophisticated thing human beings do. Every time you speak or read, you are performing a feat of combinatorial computation that no other species can replicate and that linguists have spent a century struggling to explain. There are roughly 7,000 living languages, structured in wildly different ways, yet they share deep organisational principles that suggest something profound about the human mind. This module gives you the conceptual toolkit to think clearly about what language actually is.

Six parts cover: the features that make language unique (A), its structural levels from sound to meaning (B), how we acquire and vary it (C), language and thought (D), major world languages in depth (E), and language in the digital age (F).

design features of human language

The linguist Charles Hockett identified 13 "design features" that characterise human language, three of which are found in no other communication system. Displacement lets us talk about things not present in time or space — you can describe yesterday, tomorrow, or Alpha Centauri. Productivity (also called openness) means there is no upper limit on new sentences: every utterance you produce has almost certainly never been said before in exactly that form. Duality of patterning means meaningless units (sounds) combine into meaningful units (words) which then combine into larger meaningful units (sentences) — a two-level combinatorial system that gives language its explosive expressive power. Bee dances have displacement. Birdsong has productivity of a kind. No other system has all three together.

top 10 languages by total speakers (L1 + L2, 2024 estimate)

Bars show total speakers (L1 native + L2 fluent). English leads on this measure; Mandarin leads on native speakers alone (~940 million).

the major language families

Languages descended from a shared ancestor are grouped into families. The comparative method — matching regular sound correspondences across languages — lets linguists reconstruct ancestor languages (proto-languages) spoken thousands of years ago, long before writing existed. The Indo-European family is the most studied: its proto-ancestor, Proto-Indo-European (PIE), was spoken on the Pontic-Caspian steppe around 4,000 BCE and eventually gave rise to over 400 languages including English, Hindi, Russian, and Greek.

share of world's L1 speakers by language family

Source: Ethnologue 2024. "Other" includes ~100 smaller families plus language isolates (languages with no known relatives).

language isolates — the ones with no known relatives

Some languages resist all attempts at family classification. Basque, spoken by 750,000 people in the Pyrenees, is the oldest living language in Europe — predating the arrival of Indo-European languages from the steppe by thousands of years. Korean is likely an isolate (the Altaic hypothesis linking it to Japanese has been largely abandoned). Zuni in New Mexico and Ainu in Japan have no established relatives either. These languages are windows into utterly separate evolutionary lineages of human thought.

phonemes, allophones, and the IPA

A phoneme is the smallest unit of sound that distinguishes meaning in a language. English has roughly 44 phonemes despite only 26 letters — one reason English spelling is notoriously unreliable. The International Phonetic Alphabet (IPA), standardised in 1888, gives a unique symbol to every possible human speech sound across all languages. A key insight: phonemes are not physical sounds but abstract categories. The "p" in "pin" and "spin" are acoustically different (the first is aspirated, with a puff of air; the second is not), yet English speakers hear them as the same sound. In Hindi, these two sounds are distinct phonemes — they can distinguish meaning, as in "phal" (fruit) vs "pal" (moment). Languages divide the continuous acoustic space of sound into different categories.

tonal languages — more common than you think

Mandarin uses 4 tones; Cantonese has 6; Vietnamese has 6; Hmong has 8. English and most European languages are non-tonal. Tone languages are concentrated in sub-Saharan Africa, Southeast Asia, and Mesoamerica.

morphological types — a comparison scale

Turkish "evlerinizden" (from your houses) packs five morphemes into one word. The Inuktitut word "ᑕᑯᒃᓴᐅᓂᕐᒧᑦ" can encode what English needs an entire sentence to express.

the Inuktitut example — a case study in polysynthesis

A famous Yupik (related to Inuktitut) example: "Tuntussuqatarniksaitengqiggtuq" means "He had not yet said again that he was going to hunt reindeer." That is a single grammatical word composed of many morphemes, each adding a layer of meaning. This is not just "packing more in" — it reflects a fundamentally different logic of how meaning is organised. No approach is cognitively superior; they are equally expressive but differently structured. Languages tend to drift between types over centuries: English was far more inflected in Old English (870 CE) than today, having shed most of its case endings.

basic word order types across world's languages

English and Mandarin are SVO ("She eats rice"). Japanese, Turkish, and Korean are SOV ("She rice eats"). Classical Arabic and Welsh are VSO ("Eats she rice"). Free word-order languages (Latin, Russian, Hungarian) use case endings rather than position to signal grammatical roles.

Grice's cooperative principle and its four maxims

The philosopher Paul Grice argued in 1975 that conversation works because speakers tacitly agree to be cooperative — and that we infer meaning from apparent violations of this agreement. His four maxims: be truthful (Quality), say as much but not more than needed (Quantity), be relevant (Relation), and be clear (Manner). When someone says "Can you pass the salt?" they are technically asking a yes/no question about physical ability. But because we understand that a cooperative speaker would not ask something so trivially answerable, we infer the real intent: please pass the salt. This gap between sentence meaning and speaker meaning — what Grice called "conversational implicature" — explains most of how language actually functions in social life.

milestones in first language acquisition

These timelines are approximate norms; there is enormous individual variation. What is remarkable is that every neurotypical child, regardless of intelligence or parental coaching, passes through the same stages in the same order. Children cannot be taught to skip stages.

the nativist vs. empiricist debate

Chomsky's nativism holds that children are born with an innate language acquisition device (LAD) containing the principles of Universal Grammar — which is why acquisition is so fast, uniform, and robust to impoverished input ("poverty of the stimulus"). The behaviourist alternative (Skinner's "Verbal Behaviour," 1957) held that language is learned through reinforcement and imitation. Chomsky's 1959 review of Skinner is one of the most cited academic papers ever written and largely demolished the behaviourist account. The modern debate is more nuanced: statistical learning and input frequency clearly matter, but so does something innately linguistic in the human brain. The strongest evidence for the nativist position is the emergence of new sign languages — when deaf children in Nicaragua were brought together for the first time in the 1980s, they spontaneously created a new, fully grammatical sign language (Nicaraguan Sign Language) without any adult model.

critical period hypothesis — explore the effect of age on ultimate attainment

Adjust your age of first exposure to a second language:

dialect vs accent vs register

An accent is a difference in pronunciation only. A dialect involves differences in vocabulary and grammar as well as pronunciation. There is no linguistic basis for calling one variety "the" language and another a dialect — the old joke is that "a language is a dialect with an army and a navy." Mandarin and Cantonese are officially dialects of Chinese but are mutually unintelligible; Serbian and Croatian are officially different languages but are largely mutually intelligible. The distinction is political, not linguistic. Register is the style variation within a single speaker's repertoire: the same person uses different vocabulary, sentence complexity, and politeness markers when talking to their employer versus their friends — code-switching not between languages but between modes of the same language.

William Labov and the social stratification of language

William Labov's 1963 study of Martha's Vineyard showed that islanders who strongly identified with their community raised the vowels in words like "right" and "house" — an unconscious phonological marker of local identity. His 1966 New York City study found that the pronunciation of post-vocalic "r" (the r in "car" or "fourth") correlated with social class and, crucially, with social aspiration: lower-middle-class speakers were more hypercorrect in formal speech than upper-middle-class speakers, overcorrecting toward the prestige form they associated with upward mobility. Language encodes social meaning at every phoneme.

mechanisms of language change

Languages change constantly through four main mechanisms. Sound change is regular and systematic: Grimm's Law (c. 500 BCE) shifted all Proto-Indo-European stops in Germanic — every PIE "p" became an "f," which is why Latin "pater" corresponds to English "father." Semantic shift changes word meanings: "nice" once meant foolish (from Latin "nescius," ignorant); "awful" meant inspiring awe; "bimbo" was a common term for a male idiot in 1920s American slang. Borrowing is the largest source of new vocabulary: English borrowed wholesale from French (after 1066), Latin (Renaissance), and now borrows from global English back-formations. Grammaticalisation turns content words into function words: the English future auxiliary "will" derives from Old English "willan" (to want).

strong vs weak Whorfianism

Benjamin Lee Whorf, a fire prevention engineer turned amateur linguist, proposed that the language you speak determines the thoughts you can have — what you cannot name, you cannot think. This strong version is almost certainly false: thought is not identical with language, as demonstrated by deaf people who think without spoken language, mathematicians who think in notation, and the fact that new words can be coined for previously unnamed concepts (which you could presumably already think). The weak version — that language influences (not determines) certain kinds of cognition — is supported by good evidence. Russian speakers, who have separate basic words for light blue ("goluboy") and dark blue ("siniy"), are measurably faster at discriminating these colours in the region of perceptual space between the two categories. The effect is real but modest — it biases perception, it does not imprison thought.

the Pirahã challenge — a language without numbers or recursion?

Daniel Everett's decades-long fieldwork with the Pirahã people of Brazil produced startling claims: that Pirahã has no numbers beyond "one/two/many," no colour terms, no creation myths, and possibly no recursion in its syntax. The Pirahã apparently could not learn to count even after eight months of training. If this is right — and many linguists dispute Everett's analyses — it suggests that the content of a culture's language and the cognitive capabilities emphasised in it co-evolve. Lera Boroditsky's cross-linguistic studies offer cleaner experimental evidence: Kuuk Thaayorre speakers, who use absolute compass directions rather than relative terms like left/right, have extraordinarily fine-grained spatial orientation abilities, always knowing which direction is north even in windowless rooms.

Lakoff and Johnson's conceptual metaphor theory

In "Metaphors We Live By" (1980), George Lakoff and Mark Johnson argued that metaphor is not a poetic decoration but the basic structure of conceptual thought. We do not merely speak of arguments using war metaphors — we actually conceptualise arguments as war: we attack positions, demolish arguments, shoot down ideas, defend our views. An alternative conceptual metaphor — ARGUMENT IS A DANCE — would produce entirely different linguistic and cognitive responses: we would seek harmony, find satisfying moves, create something together. Political language exploits conceptual metaphors systematically. "Tax relief" frames taxation as an affliction from which citizens need rescuing. "Illegal alien" frames undocumented immigrants as threats from outside. The frame shapes what policy options seem natural.

Orwell's diagnosis — and its limits

George Orwell's 1946 essay "Politics and the English Language" argued that political language is designed to make lies sound truthful and murder respectable. His examples remain instructive: "pacification" for bombing villages; "transfer of population" for mass forced displacement. The appendix to 1984 describes Newspeak — a language engineered to make dissent literally unthinkable by eliminating the vocabulary required to formulate it. The insight is real but partly overstated: people subjected to Newspeak-like language control (Soviet Russia, Maoist China) managed to think and communicate resistance anyway, through irony, private language, and subtext. Language constrains thought at the edges; it does not fully determine it.

major periods and their defining influences

English has borrowed from over 350 languages. Its vocabulary is roughly 29% French, 29% Latin, 26% Germanic, and 6% Greek — a unique hybrid that explains both its richness and its orthographic irregularity.

Arabic diglossia — one language, two registers

Arabic is the clearest living example of diglossia: the coexistence of two varieties of the same language used in different social contexts. Modern Standard Arabic (MSA/Fusha), derived directly from Classical Quranic Arabic, is used in newspapers, formal speeches, legal documents, and education across all 22 Arab-majority countries. But nobody grows up speaking MSA at home. Every Arab child learns a regional dialect as their first language — Egyptian, Moroccan Darija, Gulf Arabic, Levantine — which differ from MSA and from each other as much as Spanish differs from Portuguese. An educated Egyptian and an educated Moroccan can communicate in MSA but may struggle to follow each other's everyday dialect speech. The Quran's status means Classical Arabic has been uniquely preserved for 1,400 years — it is still largely comprehensible to modern Arabic speakers in a way that Chaucer's English is not to modern English speakers.

evolution of machine translation

what machine translation still fails at

Google Translate and DeepL produce impressive output for high-resource language pairs (English-French, English-Spanish) but still fail in characteristic ways. Idiom and metaphor are handled by pattern-matching, not understanding — "kick the bucket" may or may not be correctly translated depending on whether the idiom appeared frequently in training data. Pragmatic register — the difference between a formal and an informal tone in Japanese honorifics — is often flattened. Low-resource languages (Yoruba, Navajo, Tibetan) translate poorly because there is little training data. Ambiguity resolution requiring world knowledge is still problematic: "I saw the man with the telescope" has two readings that require context to disambiguate.

word embeddings and distributional semantics

Modern language AI builds on the distributional hypothesis, articulated by linguist John Firth in 1957: "You shall know a word by the company it keeps." Word2Vec (2013) and its successors represent words as vectors in high-dimensional space, where semantically similar words cluster near each other. In this geometry, the vector for "king" minus "man" plus "woman" is approximately the vector for "queen." Large language models extend this to context-sensitive representations — the same word "bank" has different vector representations in "river bank" and "bank account." The crucial question debated since Emily Bender and Timnit Gebru's 2021 "stochastic parrots" paper is whether this distributional knowledge amounts to anything like understanding, or whether it is extraordinarily sophisticated pattern matching without grounding in the world.

how long before a language community reaches crisis level?

Enter a language community's current speaker count and annual decline rate to estimate when it reaches 50 speakers (the UNESCO threshold for "critically endangered"):

the Hebrew miracle — the only successful language revitalisation

Hebrew is the only language in history to be revived from a purely liturgical state to a full first language of a nation. By 1880, Hebrew had not been a spoken vernacular for approximately 1,800 years — it existed only in religious texts and scholarly correspondence. Eliezer Ben-Yehuda, who immigrated to Palestine in 1881, decided that a Jewish state required a Jewish language and raised his son Ben-Zion as the first native Hebrew speaker in millennia. Modern Hebrew required over 30,000 new words for concepts that did not exist in Biblical Hebrew. Today, 9 million people speak it as a first language. Linguists consider this achievement unrepeated and perhaps unrepeatable: it required a unique combination of ideological motivation, a concentrated immigrant population, and deliberate institutional support that no other revitalisation effort has fully replicated.

Welsh and Irish — partial success stories

Welsh is the most successful modern language revitalisation still underway. From a low of about 20% of the Wales population speaking Welsh in 1991, active government policy — bilingual education, S4C Welsh television (launched 1982), Welsh Language Acts, and a cultural renaissance — has stabilised speakers at about 29% (2021 census). Welsh is now required in all schools to age 16. Irish is a more cautionary example: despite being a co-official language of Ireland since 1922, it has continued to decline as a community language, with only about 73,000 native daily speakers in the Gaeltacht regions, despite 40% of the population claiming some knowledge. The difference is that Welsh policy actually created new speakers; Irish policy largely failed to.

share of endangered languages by world region

The Americas and Pacific together account for over half of all endangered languages despite relatively small total speaker populations. Colonial language imposition in the Americas and island isolation in the Pacific both contribute to this concentration.