The Structure
of Language

Prelude

The Hidden Architecture

Every sentence a person produces is a small act of engineering. It involves selecting sounds from a finite inventory and combining them into words, selecting words from a mental lexicon of tens of thousands of items, arranging those words according to rules that the speaker has never consciously learned, and packaging the whole into a structure whose meaning is not simply the sum of its parts but something richer and more complex. This engineering is so automatic, so deeply embedded in the machinery of the brain, that the architecture is invisible to the people using it.

Linguistics is, among other things, the project of making that architecture visible. The field divides the structure of language into four levels of analysis, each describing a distinct layer of organization and each possessing its own technical apparatus and its own foundational questions.

Phonology asks: what sounds does a language use, and how do those sounds pattern? Morphology asks: how are those sounds assembled into meaningful units, and how do those units combine into words? Syntax asks: how are words arranged into sentences, and what rules govern the possible arrangements? Semantics asks: what do sentences mean, and how does meaning arise from the combination of parts?

These four levels are not simply convenient academic divisions. They correspond to genuinely distinct types of structure, with their own organizing principles, their own kinds of universals, and their own characteristic ways of varying across languages.

The ambition of this artifact is to lay out each level with precision, to show what is known about the universal properties that hold across all human languages, and to convey something of the astonishing diversity that exists within those universals. The range of structures that human languages have been found to use is wider than most people imagine, and the range of structures that no human language has ever been found to use is more constrained than any a priori argument would predict.

I

Phonology
The Inventory of Sounds

The first thing a language must do is carve a set of distinct sounds out of the continuous acoustic space that human vocal anatomy can produce. A child born anywhere on Earth can learn any human language; the articulatory and perceptual apparatus is species-wide. But different languages make different selections from this shared repertoire, and the range of selections is extraordinary.

At one extreme sits Rotokas, spoken by approximately 4,000 people in Papua New Guinea. Rotokas has eleven phonemes: six consonants and five vowels. At the other extreme sits the Taa language (also called !Xóõ), spoken in Botswana and Namibia. Depending on the analysis, Taa has between 112 and 141 distinct phonemes, including approximately 80 different click consonants produced by drawing air backwards into the mouth and releasing it in different ways against different parts of the palate. Between these extremes, English has approximately 44 phonemes (varying by dialect), Mandarin has around 35, and Arabic has around 28.

The International Phonetic Alphabet

The diversity of sounds across languages created an immediate practical problem for linguists in the nineteenth century: there was no notation system capable of representing all of them. Standard orthographies were designed for individual languages and reflected historical conventions rather than phonetic reality. English spelling, notoriously, represents the pronunciation of the fourteenth century more faithfully than the pronunciation of the twenty-first.

The International Phonetic Alphabet (IPA) was developed beginning in 1886 by the International Phonetic Association, founded by the French linguist Paul Passy. Its organizing principle is strict: one symbol, one sound; one sound, one symbol. Every phoneme in every described human language has a designated IPA symbol. The IPA uses a combination of modified Latin and Greek letters, and the familiar letters where their phonetic values in major European languages are consistent (so /p/ represents the sound in English "pin" and French "pain"), with purpose-designed symbols for sounds not covered by European orthographies.

The International Phonetic Alphabet, 2020 revision. The chart organizes consonants by place of articulation (columns) and manner of articulation (rows), and vowels by tongue position and lip rounding. Shaded cells represent sounds that are physiologically impossible or near-impossible for the human vocal tract.

Place and Manner of Articulation

The IPA chart reveals the underlying logic by which the enormous diversity of consonant sounds is organized. Every consonant can be described in terms of two primary parameters. The first is place of articulation: where in the vocal tract the airstream is obstructed or modified. Bilabial sounds (/p/, /b/, /m/) involve both lips. Alveolar sounds (/t/, /d/, /n/, /s/) involve the tongue tip against the alveolar ridge just behind the upper teeth. Velar sounds (/k/, /g/) involve the back of the tongue against the soft palate. The Taa click consonants add places that do not appear in any non-click language: the extreme front of the palate, the back of the alveolar ridge, and the sides of the mouth simultaneously.

The second parameter is manner of articulation: how the airstream is modified. Stops (/p/, /t/, /k/) involve a complete closure followed by a release. Fricatives (/f/, /s/, /ʃ/) involve turbulent airflow through a narrow constriction. Nasals (/m/, /n/, /ŋ/) involve complete oral closure with airflow through the nasal cavity. Liquids (/l/, /r/) and approximants (/w/, /j/) involve partial constriction without turbulence.

Vowels are organized differently: by the height of the tongue body (high, mid, low), its position from front to back, and whether the lips are rounded or spread. All known human languages have both consonants and vowels, and all have at least three vowel contrasts. The minimum vowel system appears to be a three-way distinction between high front, high back, and low central, approximately the system of classical Arabic. The maximum documented vowel inventory belongs to some dialects of Danish, with as many as 25 distinct vowel phonemes.

Tone: When Pitch Carries Meaning

In approximately 70 percent of the world's languages, pitch is used to distinguish the meanings of otherwise identical syllables. These are called tonal languages, and they represent the majority of documented human languages rather than an unusual subtype. Mandarin Chinese has four tones: a high level tone (first tone), a rising tone (second tone), a dipping tone that falls then rises (third tone), and a falling tone (fourth tone). The syllable ma means "mother" in the first tone, "hemp" in the second, "horse" in the third, and "to scold" in the fourth. Cantonese has six tones (some analyses find nine). Vietnamese has six. Hmong has eight.

Among African languages, tonal systems are nearly universal. Yoruba, spoken by over 40 million people in Nigeria and Benin, has three contrastive tones: high, mid, and low. Bantu languages across sub-Saharan Africa use tone to mark not only lexical distinctions but grammatical ones: the same verb root in the same tense can mean different things depending on the tonal pattern applied to it. In some Bantu languages, the difference between a declarative and an interrogative is marked by tone alone.

The existence of tonal languages matters for the study of language universals because it complicates the idea that speech sounds and musical pitch are separate domains. For a speaker of a tonal language, pitch is as much a part of the phonological system as consonants and vowels; the distinction between language and music that seems obvious to a speaker of English or French is not a feature of human language in general but a feature of non-tonal languages in particular.

II

Phonemes, Allophones,
and Distinctive Features

A phoneme is not a sound. It is a category of sounds. This distinction, seemingly pedantic, is one of the most important insights in the history of linguistics, and it is the contribution of the Prague School linguists working in the 1920s and 1930s, above all Nikolai Trubetzkoy and Roman Jakobson.

Minimal Pairs and the Phoneme

The standard method for identifying phonemes in a language is the minimal pair test. A minimal pair consists of two words that differ in meaning and differ in exactly one sound in the same position. "Pin" and "bin" differ only in the initial consonant: /p/ versus /b/. Since substituting one for the other changes the meaning, /p/ and /b/ must be distinct phonemes in English. "Ship" and "sheep" differ only in the vowel: /ɪ/ versus /iː/. These are distinct phonemes. "Pin" and "spin" do not form a minimal pair in the relevant sense: "spin" begins with a consonant cluster, not a single sound in the same position.

The phoneme concept immediately reveals something non-obvious: the same phoneme is not always pronounced the same way. In English, the phoneme /p/ is pronounced with a burst of air (aspiration) at the beginning of a word ("pin": [pʰɪn]) but without aspiration after /s/ ("spin": [spɪn]). These two physically distinct sounds, [pʰ] and [p], are both realizations of the single phoneme /p/. They are called allophones of that phoneme: variants whose distribution is predictable from the phonological environment and which never distinguish meaning in English.

In Hindi and other South Asian languages, aspiration is phonemically contrastive. The pair [pÊ°al] ("fruit") and [pal] ("moment") are different words distinguished solely by aspiration. What English treats as non-meaningful variation, Hindi treats as a distinction between different phonemes. The physical sounds are the same; the phonological systems organize them differently.

Distinctive Features

Jakobson's insight was that the phoneme could be decomposed into smaller units still: distinctive features. A phoneme like /b/ can be described as [+voice] (vocal cords vibrating), [+labial] (lips involved), and [+stop] (complete oral closure). A phoneme like /p/ shares all these properties except voicing: it is [-voice]. A phoneme like /d/ shares voicing and stop manner with /b/ but has an alveolar rather than labial place of articulation.

The value of feature theory is that it explains why phonological rules tend to affect natural classes of sounds rather than arbitrary collections. Languages frequently have rules that apply to "all voiced stops" or "all fricatives" or "all sounds produced at the back of the mouth." These are natural classes defined by shared features. A rule that applied to, say, /p/, /n/, and /Ê’/ (which share no features) would be unnatural, and such rules are not found. Feature theory predicts this: the structure of the phonological system is not arbitrary but constrained by the universal feature inventory from which all phonological contrasts are drawn.

Syllable Structure and Its Constraints

Beyond individual sounds, languages also organize those sounds into larger units: syllables. The syllable, while intuitively obvious to speakers of most languages (people can clap the syllables in a word without being taught to do so), is surprisingly difficult to define phonetically. What is clear is that every language places constraints on which sequences of sounds can form a syllable.

The most common syllable type across the world's languages is CV (consonant-vowel). Hawaiian, one of the world's phonologically simplest languages with only 8 consonants and 5 vowels, requires a vowel at the end of every syllable and allows at most one consonant at the beginning: words like humuhumunukunukuapuaa (the name of the state fish) are entirely composed of open CV syllables. At the other extreme, Georgian allows consonant clusters of extraordinary complexity: the word gvprtskvni (you are peeling us) begins with six consonants before the first vowel. English stands in the middle, allowing clusters like "strengths" (which, in careful pronunciation, has two consonants before and three after the vowel).

The observation that CV is the most universal syllable type and that languages tend to avoid complex clusters wherever possible reflects a general tension in phonological systems between the articulatory ease of open syllables and the acoustic distinctiveness that consonantal contrasts provide. This tension, operating across thousands of generations of language change, produces typological patterns that look almost like laws.

III

Morphology
The Architecture of Words

A word is not the smallest unit of meaning, and it is not the largest unit of sound. It sits at the intersection of two systems: the phonological system that organizes sounds into syllables and sequences, and the syntactic system that organizes words into sentences. Morphology is the study of what happens inside the word: how sounds are assembled into meaningful units, and how those units are combined to create the words that speakers actually use.

Morphemes: Free and Bound

The morpheme is the minimal unit of meaning in language. It differs from the phoneme in a crucial way: phonemes are units of sound that distinguish meaning but carry no meaning themselves, while morphemes are the smallest units that actually carry meaning or grammatical function.

Some morphemes can stand alone as words. "Book," "run," "happy," "the": these are free morphemes, capable of occurring independently. Others cannot stand alone but must attach to other morphemes. The English suffix "-ness" (happiness, darkness, sadness) is a bound morpheme: it is meaningful and productive, but it cannot appear in isolation. Similarly, the prefix "un-" (unhappy, unkind, unfair) is bound.

Bound morphemes subdivide further into inflectional and derivational types. Inflectional morphology modifies a word's form to express grammatical information without changing its core lexical category. The English suffix "-s" on a verb (she walks) marks third-person singular present tense. The suffix "-ed" marks past tense. The suffix "-s" on a noun marks plural. These inflections are required by grammatical context: a speaker cannot choose whether to mark plural on a noun; the grammar mandates it in the appropriate circumstances.

Derivational morphology, by contrast, creates new words, often changing their grammatical category. Adding "-ness" to "happy" (an adjective) creates "happiness" (a noun). Adding "-ize" to "modern" creates "modernize" (a verb). Adding "un-" to "happy" creates "unhappy," still an adjective but with reversed meaning. Derivational processes are typically optional and often not fully productive: "-ize" can be added to many adjectives ("modernize," "finalize," "digitize") but not all ("*happize," "*tallize").

Inflectional Morphology and Grammatical Categories

The grammatical categories that inflectional morphology marks vary considerably across languages. English inflectional morphology is relatively impoverished by cross-linguistic standards: verbs inflect for tense and agreement, nouns for plural and possessive, and pronouns for case. Compare this to Russian, where nouns inflect for six grammatical cases (nominative, accusative, genitive, dative, instrumental, prepositional), each indicating a different syntactic role. Or to Finnish, which has fifteen grammatical cases, including the elative (indicating motion out of), the inessive (location inside), the illative (motion into), and the abessive (the absence of something).

The category of grammatical gender is particularly striking for speakers of languages that lack it. Languages like Finnish and Turkish have no grammatical gender whatsoever. Languages like French and Spanish have two genders (masculine and feminine), which apply to all nouns including inanimate objects: a table is feminine in French (la table) but masculine in Spanish (el vaso for "glass"). German has three genders (masculine, feminine, neuter), with assignments that often conflict with natural gender: the word for "girl" (das Mädchen) is grammatically neuter. Bantu languages such as Swahili have noun classes that function like grammatical gender but number between eight and fifteen, organized not by natural gender but by categories such as animate/inanimate, abstract/concrete, and mass/count.

Perhaps the most grammatically distinctive category from a cross-linguistic perspective is evidentiality: grammatically obligatory marking of how the speaker knows what they are claiming. Languages like Turkish, Quechua, and Tibetan require speakers to indicate, through suffixes or particles, whether they witnessed an event directly, inferred it from evidence, or heard about it from someone else. In Turkish, the past-tense suffix -di indicates direct witness, while -miÅŸ indicates non-witnessed or reported information. The sentence "John came" requires a different verb form depending on whether the speaker saw John arrive or merely learned of it. English speakers can express these distinctions voluntarily ("I heard that John came") but are not grammatically required to do so. For speakers of evidential languages, the requirement is categorical.

IV

Language Types
How Words Carry Meaning Differently

One of the oldest questions in linguistic typology concerns how languages differ in their overall morphological strategy: how much information do they pack into individual words, and by what means? In the nineteenth century, Wilhelm von Humboldt and August Wilhelm Schlegel developed a classification of languages into morphological types that, with significant refinement, remains useful today. The four principal types represent different solutions to the same fundamental problem of how to organize the relationship between sounds and meanings.

Agglutination in Turkish

Turkish provides the clearest illustration of agglutination. In Turkish, morphemes stack onto a root in a predictable order, each contributing exactly one grammatical feature, with clear boundaries between each. The word evlerinizden decomposes as follows: ev (house) + ler (plural) + iniz (your, second person plural possessive) + den (from, ablative case). The result: "from your houses." Each morpheme has a stable, transparent meaning; none fuses with its neighbors in ways that obscure the analysis. The word is long but entirely predictable from its parts. The system is, in this sense, maximally compositional at the morphological level: the meaning of the whole is the sum of the meanings of the parts, in a way that is far less reliable in fusional languages like Latin or Russian.

Fusion in Latin

Latin demonstrates the fusional strategy. The word amavi (I loved) decomposes as: am- (love root) + -av- (perfect aspect marker) + -i (first person singular). But that final -i simultaneously encodes person (first), number (singular), tense (perfect), mood (indicative), and voice (active). A single morpheme, five grammatical features. Attempts to isolate the individual carriers of those features within the suffix fail: there is no separable piece that means "first person" and a different separable piece that means "singular." They are fused into a single, undivided form whose meaning is irreducibly complex.

Polysynthesis in Inuktitut

Polysynthetic languages take the logic of inflectional complexity to an extreme that speakers of European languages find initially difficult to conceptualize. In Inuktitut, spoken across Arctic Canada and Greenland, a single word can express what would require an entire sentence in English. The Inuktitut word tusaatsiarunnanngittualuujunga translates as "I can't hear very well." This is not a phrase; it is a single word, a single morphological unit. The components: tusaa- (hear) + -tsiaq- (well) + -runnann- (to be able to) + -nngit- (not) + -tualuu- (very much) + -junga (first person singular indicative). All of this constitutes a single verbal word.

This is not a curiosity or a complexity introduced by careless design. It is a genuine alternative architecture for the relationship between morphology and syntax. Languages like Inuktitut have very little syntax in the sense that European linguists typically study it: the combinatorial work that English does by arranging words in a sentence is done by Inuktitut at the level of morpheme concatenation within a single word. The word is the clause.

Modern linguists treat these types not as fixed categories but as positions on continua. No language is purely isolating or purely polysynthetic; every language mixes strategies to some degree. Mandarin, often cited as the paradigm isolating language, nonetheless has some bound morphemes. English, treated as analytic in many comparisons, has more inflectional morphology than Mandarin. The typological classification captures tendencies, not absolutes. And it provides a framework for one of the central questions in historical linguistics: how do languages shift from one typological profile to another over time?

V

Syntax
The Rules of Combination

Syntax is the study of how words combine into phrases and sentences. Its central insight is that sentences are not simply sequences of words but hierarchically organized structures in which words form groups (phrases) that function as single units at a higher level of organization. This hierarchical organization is not visible in the spoken or written form of a sentence, which is necessarily linear; it is an abstract structure that speakers compute and listeners recover.

Constituency: The Hidden Groupings

The sentence "The old man met the young woman on the bridge" contains ten words. But those ten words are not equally grouped: "the old man" forms a unit (a noun phrase), as does "the young woman," and "on the bridge." These groupings are not arbitrary conventions but real structural facts, evidenced by a battery of tests that linguists use to identify constituents.

One such test is substitution: a constituent can be replaced by a single word of the same category without disrupting grammaticality. "The old man met the young woman on the bridge" can become "He met her there": "he" replaces the entire noun phrase "the old man," "her" replaces "the young woman," and "there" replaces "on the bridge." The substituted words are single items; the replaced strings are constituents.

Another test is movement: constituents, but not arbitrary substrings, can be moved to other positions in the sentence. "On the bridge, the old man met the young woman" is grammatical (the prepositional phrase "on the bridge" has been fronted). "Bridge the old man met the young woman on" is not: "bridge" alone is not a constituent and cannot be fronted. "Old man met" is not a constituent and cannot be moved as a unit.

These tests converge on a hierarchical description of sentence structure. At the top level, a sentence (S) consists of a noun phrase (NP) and a verb phrase (VP). Within the VP, the verb takes objects and prepositional phrases. Within each NP, a determiner ("the") and possibly adjectives precede the head noun. Each of these units is itself a phrase with its own internal structure. The whole is a nested, hierarchical tree of constituents, with the sentence at the root and individual words at the leaves.

A phrase structure tree (after Chomsky's early generative grammar) showing how a sentence is organized not as a linear string of words but as a hierarchical structure in which words group into phrases that function as units at a higher level. The tree structure is the syntactic representation; the string of words at the bottom is what speakers actually hear or read.

Phrase Structure and X-Bar Theory

The observation that all phrases have a similar internal organization, whatever their category, led to the development of X-bar theory in the 1970s, emerging from work within the generative tradition associated with Chomsky. The central claim of X-bar theory is that every phrase, regardless of category (noun phrase, verb phrase, adjective phrase, prepositional phrase), has the same internal structure: a head word that determines the category, optional specifiers that precede the head, and optional complements that follow it.

This generalization is strikingly powerful. It means that a child learning language does not need to learn a separate structural template for noun phrases, verb phrases, prepositional phrases, and so on. A single schematic template applies uniformly across all phrasal categories. The specific realizations differ (the head of an NP is a noun, the head of a VP is a verb, and so on) but the abstract structural pattern is identical.

Grammatical Relations: Subject, Object, and Ergativity

All known human languages distinguish between different participants in an event and mark those distinctions grammatically. The concepts of subject and object, familiar from European languages, capture one way of organizing this: the subject is typically the agent of an action or the entity that a sentence is "about," and the object is the entity acted upon.

But not all languages organize grammatical relations in this way. Approximately one-quarter of the world's languages are ergative rather than nominative-accusative in their case-marking system. In a nominative-accusative language like English, the subject of an intransitive verb (She ran) and the subject of a transitive verb (She hit him) are treated the same way; both are marked (if marked at all) as nominative. In an ergative-absolutive language, the subject of an intransitive verb and the object of a transitive verb are treated the same way (absolutive), while the subject of a transitive verb is marked differently (ergative).

Basque, spoken in northern Spain and southern France, is the best-known ergative language in Europe. The agent of a transitive verb takes a suffix (-k in many forms) while the subject of an intransitive verb and the object of a transitive verb take no case suffix. This is the reverse of what speakers of nominative languages intuitively expect, and it illustrates a general point: the way European languages organize grammatical relations is not the only logical possibility, and it is not even the most common one cross-linguistically when languages are weighted by structural type rather than by number of speakers.

VI

Word Order
and Greenberg's Universals

The arrangement of subject, verb, and object in a basic declarative sentence is one of the most studied typological properties of language, largely because it is easy to identify and because it correlates with a remarkable number of other structural properties. In 1963, the American linguist Joseph Greenberg published a paper that transformed the study of linguistic universals: "Some Universals of Grammar with Particular Reference to the Order of Meaningful Elements." Working from a sample of 30 languages, Greenberg identified 45 cross-linguistic tendencies and absolute universals, many of which have since been confirmed and extended across larger samples.

The Distribution of Basic Word Orders

Six logically possible orders of Subject (S), Verb (V), and Object (O) exist. Greenberg found that these six possibilities are not equally distributed across the world's languages. Subject-Object-Verb (SOV) order is the most common, attested in roughly 44 to 47 percent of languages. Subject-Verb-Object (SVO) order, the order of English, French, and Mandarin, is the second most common at around 28 to 30 percent. Verb-Subject-Object (VSO) order, found in classical Arabic, Welsh, and many Polynesian languages, accounts for around 9 percent. The other three orders (VOS, OVS, OSV) are quite rare, each attested in only a handful of language families.

The dominance of subject-initial orders (SOV and SVO together account for roughly 75 percent of languages) is not accidental. Cross-linguistically, subjects tend to be topical (the entity the sentence is about), definite, and high in animacy. Verbs tend to be complex predicates that anchor the sentence. Objects tend to be new information. The ordering biases reflect systematic patterns in how information is structured across languages, not merely surface conventions.

Implicational Universals: The Harmony Principle

Greenberg's most influential discovery was not the frequency distributions themselves but the implicational relationships between word order and other structural properties. These are universals of the form "if a language has property X, it will also have (or strongly tend to have) property Y."

Greenberg's Universal 3 (approximately): VSO languages overwhelmingly use prepositions (before the noun phrase) rather than postpositions (after it). SOV languages overwhelmingly use postpositions. Universal 4: languages with normal verb-final order almost always have postpositions; languages with normal verb-initial or verb-medial order almost always have prepositions.

The explanation for this pattern lies in what linguists call head directionality. In English (SVO, prepositions), phrases are generally head-initial: the head of the phrase precedes its complement. The verb "hit" precedes its object; the preposition "on" precedes its noun phrase complement; the complementizer "that" precedes its clause. In Japanese (SOV, postpositions), phrases are generally head-final: the verb follows its object; postpositions follow their noun phrase complements; complementizers follow their clauses. Languages tend toward consistency in head directionality across all their phrase types, creating a coherent typological "package." Greenberg's implicational universals document this tendency, and the typological regularity is strong enough that knowing a language's basic word order allows confident predictions about its preposition/postposition preference, the position of adjectives relative to nouns, the position of relative clauses, and several other properties.

Absolute Universals

Some of Greenberg's universals are not statistical tendencies but apparent absolutes: no known exceptions have been found. Universal 1: in declarative sentences with nominal subject and object, the dominant order is almost always one in which the subject precedes the object. Universal 38: no language has a trial number (a grammatical category meaning "exactly three") unless it also has a dual (a grammatical category meaning "exactly two"). Universal 40: when a language has case-marking on nouns, it almost always marks the direct object of a transitive verb differently from the subject of an intransitive verb.

These are not logical necessities. There is no logical reason why a language could not have trial number without dual, or why objects could not precede subjects as a statistical majority across languages. That these patterns hold is an empirical discovery, and explaining why they hold requires engagement with the cognitive and communicative pressures that shape language structure over historical time.

VII

Semantics
Sense, Reference, and Compositionality

Meaning is what language is for, and yet it is the most philosophically treacherous of the four levels of linguistic analysis. It is easy enough to say that words mean things, but what is the nature of that relationship? What is it for a word to mean something? And how does the meaning of a sentence arise from the meanings of its parts? These questions were posed with the greatest precision not by a linguist but by a mathematician and logician: Gottlob Frege, whose work in the late nineteenth century established the foundations on which formal semantics was eventually built.

The Puzzle of Identity Statements

Frege's starting point was a puzzle about identity. The statement "the morning star is the morning star" is trivially true: it is true by definition, and no one learns anything from it. The statement "the morning star is the evening star" is also true, but it is not trivially true; it is a significant astronomical discovery (that both expressions refer to the planet Venus). Yet if meaning were simply reference, these two statements should have the same meaning, since both "the morning star" and "the evening star" refer to the same object. The cognitive difference between them demands explanation.

Frege's solution: distinguish between Sinn (sense, or meaning) and Bedeutung (reference, or the object in the world that a term picks out). "The morning star" and "the evening star" have the same Bedeutung (both refer to Venus) but different Sinne (they pick out Venus by different descriptions: the star that appears in the morning versus the star that appears in the evening). The cognitive difference between the two identity statements is explained by the difference in sense, even though the reference is the same.

This distinction has wide application. Proper names present a parallel puzzle: "Samuel Clemens is Mark Twain" is informative, even though both names refer to the same person. Frege would say they have the same Bedeutung but different Sinne (or at least, that there is a descriptive sense associated with each that differs). Definite descriptions like "the current president of France" pick out different individuals at different times even though the description (the sense) remains constant. The reference varies; the sense does not.

Compositionality

Frege's other foundational contribution to semantics is the principle of compositionality, sometimes called Frege's principle: the meaning of a complex expression is a function of the meanings of its parts and the way those parts are syntactically combined. This principle is what makes it possible to understand sentences one has never heard before. A competent speaker of English who has never encountered the sentence "The fluorescent iguana ignored the hypothesis" can immediately understand it, because they know the meanings of the individual words and they know the compositional rules by which those meanings combine.

Compositionality is not guaranteed by anything in the nature of language; it is a property that the semantic system has, and it has it because it must. Any language in which the meaning of complex expressions was not systematically determined by the meanings of their parts could not be learned: the learner would have to memorize the meaning of every possible sentence separately, an infinite task. The finiteness of human learning capacity and the productivity of human language jointly force the semantic system to be compositional.

Truth Conditions and the Nature of Sentence Meaning

Frege proposed that the meaning of a declarative sentence is a function from possible states of the world to truth values: a sentence is true or false depending on whether the world is in a state that satisfies the conditions the sentence describes. The meaning of "Snow is white" is the condition under which it is true: when snow is, in fact, white. This truth-conditional approach to sentence meaning became the foundation of formal semantics, developed through the twentieth century by philosophers and linguists including Richard Montague, who combined the tools of formal logic with Chomskyan syntactic theory to produce the first comprehensive formal treatment of natural language semantics in the 1970s.

The truth-conditional approach handles compositionality naturally: the truth conditions of a complex sentence are computed from the truth conditions of its component clauses and the meanings of the connectives that link them. "It is raining and the streets are wet" is true when both conjuncts are true; "It is raining or the streets are wet" is true when at least one conjunct is true. The logical operators provide the compositional rules for building complex meanings from simpler ones.

VIII

Prototype Theory
and the Colour Term Universals

The formal, truth-conditional approach to semantics handles logical structure elegantly but struggles with the messiness of natural language categories. When is something a "chair"? When is a body of water a "lake" as opposed to a "pond"? When is something "red" as opposed to "orange"? Classical theories of meaning, which define categories by necessary and sufficient conditions (a bachelor is, by definition, an unmarried adult male), cannot accommodate the fact that categories in natural language have fuzzy edges and internal structure: some members are better examples of a category than others.

Prototypes and Graded Membership

Rosch's experiments revealed that when speakers judge how well an instance fits a category, their judgments are not binary but graded. A robin is a very good example of BIRD; a penguin is a poor example. A kitchen chair is a good example of CHAIR; a beanbag is a marginal one. These goodness-of-example judgments are consistent across speakers and predictive of response times in categorization tasks: people identify prototypical members of a category faster than peripheral ones.

The prototype is the most representative member of a category: the instance that shares the most features with other members and the fewest features with members of contrasting categories. It is not an average of the category; it is the best example. And crucially, it is not defined by a checklist of necessary properties. There is no single property that all chairs share and that no non-chairs share; the category is organized instead around a family resemblance (Wittgenstein's term) in which members share overlapping but not identical sets of features.

Prototype effects have consequences for compositional semantics. The meaning of "red car" should, on a compositional account, be simply the intersection of the set of red things and the set of cars. But prototype effects complicate this: a pink or orange car may be called "red" in context (because it is the reddest thing in the relevant set), while the same color on a stop sign would not be "red" but rather a specific shade. Context, comparison sets, and pragmatic factors influence category membership in ways that resist purely compositional treatment.

Berlin and Kay: The Colour Term Universals

The study of colour terms across languages provides one of the most striking cases of the interplay between linguistic universals and cultural variation. The naive assumption, reinforced by the Sapir-Whorf tradition (which Artifact IV will examine), is that colour terms are arbitrary cultural constructs: different languages carve up the colour space in entirely different and equally valid ways.

Berlin and Kay's study identified basic colour terms in each language: terms that are monolexemic (not composed of other words), of high frequency, not restricted to a narrow domain, and whose meaning is not included in any other colour term. English has eleven basic colour terms: black, white, red, green, yellow, blue, brown, purple, pink, orange, and grey.

Their central finding: languages with fewer basic colour terms do not choose which colour distinctions to make arbitrarily. If a language has only two basic colour terms, they will be roughly equivalent to "dark/cool" (black, blue, green, purple) and "light/warm" (white, red, yellow). If it has three, the third will be red. If four, either green or yellow is added. The sequence is: black and white; then red; then either green or yellow; then the other; then blue; then brown; then purple, pink, orange, and grey in some order.

This sequence, confirmed and refined by the World Color Survey (Paul Kay et al., 1997), which examined 110 languages, implies a universal hierarchy: no language has a basic term for blue without also having basic terms for black, white, red, and at least one of green or yellow. No language has a basic term for brown without having terms for all of the above plus blue. The colour space is not carved up arbitrarily; there are universal anchors around which colour terms crystallize, corresponding to the most visually salient regions of the colour space.

The implication is that even a domain as seemingly arbitrary as colour naming is constrained by universal features of the human perceptual system. The Sapir-Whorf position that colour categories are purely linguistic conventions is incompatible with this evidence. But the strong universalist position is also too simple: languages genuinely differ in where they draw boundaries between colour categories, and these differences have measurable cognitive effects (which Artifact IV will examine carefully). The colour term universals occupy a middle ground: universal pressures operating on a domain that also shows real cultural and linguistic variation.

IX

What Is Universal,
What Varies

Having traced the four levels of linguistic structure and the key findings at each, it is possible to take stock of what appears to hold across all human languages and what varies. The picture that emerges is neither one of pure universality (all languages are fundamentally the same) nor of pure arbitrariness (languages can be organized in any way at all). It is something more interesting: a constrained variation, in which universal pressures operate on domains that also show genuine and sometimes startling cross-linguistic diversity.

What Appears Universal

• Consonants and Vowels Every known human language has both consonants and vowels. The distinction between full closure or narrow constriction (consonant) and open approximation with resonance (vowel) is universal. No language is composed entirely of one or the other. Every language has at least three vowel contrasts.
• Duality of Patterning All languages organize sounds into two distinct layers. At the lower layer, meaningless sound units (phonemes) combine to form higher-level meaningful units (morphemes and words). This "duality of patterning" is not found in any other known communication system. It is what gives language its combinatorial power without requiring a separate phoneme for every word.
• Noun and Verb Distinction All languages appear to have grammatical categories that function like nouns (referring to entities) and verbs (predicating properties and relations of those entities). The precise formulation of this universal is contested: Evans and Levinson argue that some languages do not draw the noun-verb distinction in the way European languages do. But all languages have ways of referring to objects and ways of predicting things of those objects.
• Negation and Interrogation All languages have grammatical mechanisms for negating propositions (asserting that something is not the case) and for forming questions (requesting information). The specific forms vary enormously: negation can be a separate particle, a verb inflection, a reduplication pattern, or a change in word order. But every language has them.
• Personal Pronouns and Deixis All languages have pronouns distinguishing at least first person (the speaker) from second person (the addressee). All languages have deictic expressions capable of indicating proximity or distance (this/that, here/there). These are universal because communication is always situated: speakers always occupy a position in space and time relative to the things they talk about.
• Displacement All languages can refer to things not present in the immediate context: past events, future plans, hypothetical situations, absent individuals. No known language is restricted to referring only to what is immediately perceptible.

What Varies

The range of structural variation across languages is wider than most speakers of European languages are likely to assume. The following properties are not universal: they vary substantially across the world's languages.

Phoneme inventory size ranges from 11 (Rotokas) to over 100 (Taa). Some languages use tones phonemically; some do not. Some use click consonants; most do not. Some languages (like Arabic) have pharyngeal consonants produced in the far back of the throat; most do not.

Morphological complexity ranges from the near-zero inflectional morphology of Mandarin to the single-word clauses of Inuktitut. Grammatical gender ranges from absent (Finnish, Turkish, Mandarin) to two genders (French, Spanish) to three (German, Latin) to fifteen or more (Bantu languages). Evidentiality marking is obligatory in some languages and entirely absent in others.

Grammatical tense is absent or not grammatically obligatory in some languages. Mandarin has no tense morphology: the time of an event is conveyed through temporal adverbs or context, not through obligatory verb inflection. The same is true of Burmese, Yoruba, and others. This does not mean speakers of these languages cannot talk about the past or future; it means they are not grammatically required to mark tense on the verb.

Word order varies from near-rigid (Mandarin word order is quite fixed) to extremely free (in Latin, the grammatical case suffixes carry the relational information and the word order is used for pragmatic rather than syntactic purposes). Russian, which retains a rich case system, allows all six possible orderings of subject, verb, and object, all grammatical, with different pragmatic nuances.

Ergativity organizes the expression of grammatical relations differently from the nominative-accusative pattern of European languages, and is found in roughly one-quarter of the world's languages, including Basque, the Caucasian languages, most Australian Aboriginal languages, Tibetan, and many Mayan languages.

The coexistence of universals and variation is not paradoxical but expected. The universals reflect constraints that arise from the shared biology of human cognition and communication: the vocal anatomy, the auditory system, the memory and processing capacities that all humans share, and the communicative purposes that all languages must serve. The variation reflects the fact that within those constraints, many different solutions are possible, and historical contingency, contact between languages, and cultural pressures have led different speech communities to different solutions.

X

The Architecture of Variation
Typology, Implication, and Explanation

The discovery that cross-linguistic variation is structured rather than random raises an immediate explanatory challenge: why are these particular universals universal? Why do head-initial languages tend toward prepositions while head-final languages tend toward postpositions? Why does the colour term sequence proceed in the specific order that Berlin and Kay identified rather than some other order? The answers are not given by the typological data themselves; they require engagement with the cognitive, communicative, and processing factors that shape language structure over time.

Functional Explanations for Typological Tendencies

The head-directionality correlations identified by Greenberg are explicable in terms of processing efficiency. When a speaker encounters a head at the beginning of a phrase (in a head-initial language), they can immediately begin computing the structure of the phrase: they know what category of complement to expect. In a consistently head-final language, the structure is also predictable, because the head at the end resolves the structure retrospectively. Mixed systems, in which some phrases are head-initial and others head-final, require more complex and less efficient parsing strategies. The typological tendency toward internal consistency in head-directionality reflects, on this account, pressures for processing efficiency that operate across generations of language change.

The colour term sequence of Berlin and Kay reflects the structure of the human visual system. The foveal region of the retina, responsible for high-resolution color vision, is most sensitive to wavelengths in the red-yellow region of the spectrum; green and blue are processed somewhat differently. The universal focal colors identified by Rosch (the "best example" of each color in any language) cluster around the same visually salient regions regardless of what color terms a language has. The fact that black and white are always the first color terms to emerge reflects the fundamental importance of luminance contrast to visual processing; red is consistently next because of the salience of the long-wavelength region to foveal processing.

The Limits of Functional Explanation

Functional explanations have real force, but they have limits. Many universal properties of language do not have obvious functional explanations. Why do all languages have a subject-before-object preference (Greenberg Universal 1)? A plausible answer: agents (typically subjects) are cognitively more salient than patients (typically objects), and linguistic order tends to track cognitive salience. But this is a tendency, not a logical necessity, and the exceptions are real languages with real speakers for whom the alternative ordering evidently works perfectly well.

Some universals appear to be genuine formal properties of grammar that resist reduction to function. The constraint on "parasitic gaps" (the conditions under which certain types of missing arguments are permitted in English and other languages) does not follow from any obvious processing or communicative pressure; it appears to reflect a formal property of how movement operates in syntactic derivations. Chomsky's program of seeking universal formal properties of grammatical systems is motivated by the observation that some universals seem to be formal rather than functional in character.

What Typology Reveals About the Language Faculty

The cross-linguistic study of language structure offers a perspective on the language faculty that is unavailable from the study of any single language. When a property is universal, it is either a logical necessity (impossible for any language to lack), a biological constraint (something the language faculty cannot fail to have), or a functional consequence (something that every language independently converges on because it solves a universal problem).

When a property varies, it is a parameter of the language faculty: a point at which the biological endowment underdetermines the outcome and the specific language learned in childhood sets the value. The Chomskyan notion of "parameters" (introduced in Artifact I in connection with Universal Grammar) maps onto the typological observation that languages vary in systematic, constrained ways. The question of why variation is constrained in the specific ways it is remains one of the deepest questions in linguistic theory, connecting the study of the external, cross-linguistic variation in human languages to the internal, biological structure of the language faculty itself.

XI · Synthesis

Four Levels, One System
The Architecture of Language

The four levels of linguistic structure described in this artifact are not independent systems bolted together. They are interacting layers of a single system, each level providing the units that the next level organizes. Phonemes are organized by phonology into the syllable patterns that morphology assembles into morphemes. Morphemes are assembled into words, which syntax organizes into phrases and sentences. Phrases and sentences receive interpretations from semantics, which maps syntactic structures onto meanings via compositional principles.

The interaction between levels is as important as the properties of any single level. The word "cats" is phonologically a monosyllable with three phonemes (/k/, /æ/, /s/), morphologically a complex word composed of two morphemes (cat + plural suffix), syntactically an NP that can serve as subject or object, and semantically a predicate true of all pluralities of cats. The full description of a linguistic form requires all four levels simultaneously.

What the cross-linguistic study of all four levels reveals is the same thing from different angles: human languages are vastly more diverse in their surface forms than speakers of any single language are likely to appreciate, and considerably more uniform in their underlying organizational principles than the surface diversity might suggest. The diversity is real; the Taa click system and the Inuktitut polysynthetic morphology and the Turkish postpositional system and the Arabic root-and-pattern morphology are genuine structural alternatives, not superficial variants of a single underlying scheme. And yet the range of possible human languages is much narrower than the space of logically possible communication systems. The constraints on that range are the subject of the universals program; the explanation of those constraints is the deepest question in linguistic theory.

What is most remarkable is not that languages are different from each other. It is that they are not more different. Given the number of human communities, the length of their separation, and the freedom that the arbitrary nature of the sign would seem to allow, languages constrain each other in ways that demand explanation from the structure of the minds that create them.

The next artifact in this curriculum takes up the question of how meaning works in practice: not the formal semantics of truth conditions and compositionality, but the gap between what a sentence literally says and what a speaker actually communicates by uttering it. That gap is where most of the work of everyday language happens, and it is mediated by processes (inference, context, shared knowledge, implicature) that lie between the four levels examined here and the social world in which language is actually used.

Key Figures in This Artifact

Paul Passy (1859–1940): Founded the International Phonetic Association, 1886; developed the International Phonetic Alphabet. · Nikolai Trubetzkoy (1890–1938): Principles of Phonology, 1939. The phoneme as a functional unit; Prague School phonology. · Roman Jakobson (1896–1982): Distinctive feature theory; phonological universals; markedness. · Leonard Bloomfield (1887–1949): Language, 1933. The morpheme; American structural linguistics. · Wilhelm von Humboldt (1767–1835): Early morphological typology; the relationship between language structure and thought. · Ian Maddieson (b. 1942): Patterns of Sounds, 1984. The PHOIBLE phoneme database; systematic cross-linguistic phonology. · Noam Chomsky (b. 1928): Phrase structure grammar; X-bar theory; Government and Binding theory. · Joseph Greenberg (1915–2001): "Some Universals of Grammar," 1963. Word order typology; implicational universals; language classification. · Gottlob Frege (1848–1925): "On Sense and Reference," 1892. Sinn and Bedeutung; the principle of compositionality. · Alfred Tarski (1901–1983): Truth-conditional semantics; the semantic theory of truth. · Richard Montague (1930–1971): Montague grammar; formal compositional semantics for natural language. · Eleanor Rosch (b. 1938): Prototype theory; basic level categories; graded category membership. · Brent Berlin and Paul Kay: Basic Color Terms, 1969. The universal colour term hierarchy; the World Color Survey.