Grammatical Agreement

How agreement works

An agreement relation has three parts: a controller (the word that supplies the features), a target (the word that copies them), and a set of features being transmitted. In Spanish las casas blancas ("the white houses"), the noun casas is the controller, supplying [feminine, plural]. The article las and the adjective blancas are targets — both pick up -as to match. Three words, one set of features, copied twice.

Most languages restrict agreement to certain syntactic configurations: a verb agrees with its subject (and sometimes its object), an adjective with the noun it modifies, a determiner with the head of its phrase. The exact configuration is parametric — French past participles agree with a preceding direct object but not a following one (la pomme que j'ai mangée), a quirk that takes native speakers years to fully internalise.

What's left of English agreement

English collapsed nearly all of its inherited Germanic agreement between 1100 and 1400, as final unstressed syllables eroded under Norse and French contact. The survivors:

• 3SG present -s. "She walks" vs "I/you/we/they walk." This is the only verbal agreement that English regularly drills into learners.
• Subject pronouns vs object pronouns. I/me, he/him, she/her, we/us, they/them, who/whom — fossilised case agreement.
• Demonstratives. This book / these books, that book / those books — number agreement on a closed class.
• Reflexive pronouns. Himself, herself, itself, themselves — agreement with antecedent in person, number and gender.
• BE (irregular). I am, you/we/they are, he/she/it is, I was, we were — the most highly inflected verb in the language.

Adjectives, articles, and possessive determiners are completely invariant: red book = red books, the dog = the dogs. A speaker of Old English would find Modern English grammar shockingly bare.

Cross-linguistic data

The amount of agreement a language demands is one of the cleanest typological parameters available. A few benchmarks:

• Spanish: verbs inflect for 6 person-number combinations × 3 main tenses = ~18 distinct present/past/future forms per verb. Adjectives, articles, and participles inflect for gender and number. "Las niñas pequeñas están cansadas" shows 4 separate -as targets controlled by niñas.
• Russian: present-tense verbs agree in person and number; past-tense verbs drop person and instead agree in gender (because the past tense was historically a participle). Ona čitala ("she was reading") vs on čital ("he was reading") differ only in the gendered final vowel.
• Latin: adjectives agree with nouns in case, number, and gender — three feature dimensions multiplying out to dozens of possible forms per stem. Bonus puer / bonum puerum / boni pueri ("good boy" nom/acc/gen).
• Arabic: verbs agree in person, number, and gender. Crucially, agreement is partial when the verb precedes the subject — VS order takes singular agreement even with a plural subject, while SV order takes full plural agreement. Word order changes feature transmission.
• Swahili: noun classes drive everything. The class-1 noun mtu ("person") triggers m-/yu-/a- markers across modifiers and verbs; the class-7 noun kitabu ("book") triggers ki-. A single sentence routinely shows the same prefix on five different words.
• Mandarin: verbs and adjectives are invariant. Wǒ chī, tā chī, tāmen chī — "I eat / he eats / they eat" all use the same verb form. Number and person come from pronouns alone.

The morphological-typology spectrum

Agreement is closely tied to where a language sits on the isolating-to-polysynthetic continuum. Roughly:

Type	Morphemes per word	Agreement load	Example language	Sample form	Gloss
Isolating	~1	None	Mandarin	tā chī fàn	"he eat rice"
Isolating	~1	None	Vietnamese	tôi đi học	"I go school"
Agglutinative	3–6, segmentable	Light to moderate	Turkish	ev-ler-imiz-de	"in our houses"
Agglutinative	3–6, segmentable	Moderate	Swahili	wa-toto wa-dogo wa-na-cheza	"the small children are playing"
Fusional	2–4, fused	Heavy (per/num/gen/case bundled)	Russian / Latin	amabamus	"we were loving" (1PL imperf indic)
Polysynthetic	5–15+, fused and incorporated	Saturating	Mohawk / Inuktitut	tusaatsiarunnanngittualuujunga	"I can't hear very well" (one word)

Agreement is densest in fusional and polysynthetic languages — exactly where each affix bundles multiple features into a single, hard-to-segment morpheme. Isolating languages drop it entirely; agglutinative languages sit in the middle, with cleanly separable but plentiful agreement markers.

Worked example: Spanish concord across a phrase

Take the Spanish noun phrase aquellas dos viejas amigas mías ("those two old friends of mine"). The controller is the head noun amigas, which is feminine and plural. Now read the targets left to right:

aquellas    dos     viejas    amigas       mías
DEM-FEM-PL  NUM     ADJ-FEM-PL HEAD-FEM-PL POSS-FEM-PL
[FEM, PL]   [PL]    [FEM, PL] [FEM, PL]    [FEM, PL]

Four targets pick up the controller's features. Change the head to masculine plural amigos and every -as flips to -os: aquellos dos viejos amigos míos. Change to feminine singular amiga and they all become -a: aquella vieja amiga mía. Spanish concord is mechanical, but you have to remember every word is part of the same circuit.

Subject-verb agreement across Romance

Latin had six distinct present-tense forms per verb (1/2/3 SG and PL). All Romance descendants kept the system, with predictable phonological wear:

Person/Number	Latin amare	Spanish	Italian	French	Portuguese
1SG	amō	amo	amo	aime	amo
2SG	amās	amas	ami	aimes	amas
3SG	amat	ama	ama	aime	ama
1PL	amāmus	amamos	amiamo	aimons	amamos
2PL	amātis	amáis	amate	aimez	amais
3PL	amant	aman	amano	aiment	amam

Notice French: aime, aimes, aime, and aiment are all pronounced identically [ɛm]. French agreement is preserved orthographically but largely lost in speech, which is why the language has compensated by making subject pronouns (je, tu, il, ils) effectively obligatory — they carry the agreement load that the verb endings used to.

Agreement asymmetries

Real agreement systems are messier than tidy paradigms suggest. Some classic asymmetries:

• Conjunct-sensitive agreement. Arabic, Russian, and many others show different agreement with VS vs SV word order. Arabic jaa'a-t l-banaat-u ("came-FEM.SG the-girls") uses singular even though l-banaat is plural — the verb only "sees" a default 3SG slot when it precedes the subject.
• Resolution rules for coordination. "John and Mary are" is plural — but what about gender? In Spanish, masculine wins by default: Juan y María están cansados. In Russian, conjoined masculine + feminine takes plural with no gender, since past plural verbs don't gender-agree.
• Honorific agreement. Japanese and Korean encode social relations on the verb without true person agreement; conversely, Spanish usted ("you" formal) takes 3SG verb forms. The morphological category is person, the function is politeness.
• Long-distance agreement. Hindi-Urdu allows verbs to agree across a clause boundary with an embedded object — a feature that has launched a small library of theoretical syntax papers.

Agreement as a syntactic diagnostic

The most useful thing about agreement is that it makes hidden structure visible. Wherever you see two words sharing features, you have evidence that they sit in a particular structural relation. Linguists exploit this constantly:

• If a verb agrees with a phrase, that phrase is the subject — even if it does not appear in the canonical subject position.
• If two coordinated nouns trigger plural agreement, the coordination is at the phrasal level, not at the word level.
• If a moved wh-phrase still triggers agreement on the original verb, you have evidence the phrase originated lower in the structure.
• If agreement fails to obtain across a particular boundary (e.g. an island), that boundary is a real syntactic barrier.

Generative syntax treats agreement as a probe-goal operation: a syntactic head with unvalued features searches its domain for a goal that can value them. The empirical core is older than the theory — Pāṇini described concord rules for Sanskrit two and a half millennia ago.

Common pitfalls

• Confusing agreement with government. Agreement copies features from controller to target. Government imposes a specific feature value on a complement. Most prepositions govern case; most adjectives agree with nouns.
• Treating English plural -s as agreement. The plural -s on nouns is inflection on the noun itself, not agreement with anything. The 3SG -s on verbs is agreement.
• Assuming gender is semantic. Most "gender" is grammatical, not semantic. German Mädchen ("girl") is neuter; Spanish la persona is grammatically feminine even when referring to a man.
• Ignoring attraction. Native speakers regularly produce "the keys to the cabinet is..." — an attraction error where the verb agrees with the closer noun rather than the structural subject. Style manuals deplore it; cognitive science treats it as a working-memory window.
• Reading number off the noun rather than the agreement. Collective nouns (team, family, jury) can take singular or plural agreement in British English depending on whether the speaker conceives of the group as a unit or as individuals. American English is stricter.

Why grammatical agreement matters

• Typology. Agreement load is a primary parameter for cross-linguistic classification.
• L2 pedagogy. Learners spend disproportionate effort on conjugation tables and adjectival concord.
• Theoretical syntax. Agreement diagnostics motivate every modern theory of clause structure.
• NLP. Tagging, parsing, and machine translation pipelines must reproduce agreement to look natural.
• Sociolinguistics. Variant agreement (collective-noun number, "they/them" with singular antecedent) is a sensitive marker of register and identity.
• Diachrony. Loss or gain of agreement is a long-arc signal of grammatical drift.