Modality

Three flavours, one verb

The English sentence You must be home by ten sounds like a parental order; You must be tired sounds like a guess. Both contain must. The difference is not in the modal — it is in what kind of background the modal is being evaluated against. Linguists call the first reading deontic (Greek deon "duty"), the second epistemic (Greek epistēmē "knowledge"). A third flavour, dynamic, encodes capabilities or dispositions of the subject: She can swim says nothing about what is permitted or what the speaker infers — it asserts an ability inherent in her.

Flavour	What it consults	Example	Paraphrase
Deontic	Rules, laws, norms, obligations	You must wear a seatbelt.	The traffic code requires it.
Epistemic	Speaker's evidence, perception, inference	It must be raining.	Given the wet streets, I infer rain.
Dynamic (ability)	Subject's inherent capabilities	She can swim.	Her physical ability includes swimming.
Dynamic (disposition)	Subject's regular tendencies	This printer can be temperamental.	The printer's disposition includes malfunctioning.
Bouletic	Speaker's wishes or desires	You must try the cake.	I want you to try it.
Teleological	Goals or means-end reasoning	To reach the airport, you must take Route 9.	Given the goal, that route is necessary.

Kratzer's framework treats all six as instances of one mechanism: each is a quantifier over possible worlds, with the conversational background telling you which set of worlds to quantify over. The same modal element does the quantifying; the difference between flavours is which body of facts you consult to fix the worlds.

Kratzer's possible-worlds semantics

Angelika Kratzer's analysis works as follows. A modal expression like must takes two arguments. The first is a modal base — a function from the world of utterance to a set of relevant propositions (the laws, the speaker's evidence, the subject's abilities, etc.). The second is an ordering source — another set of propositions that ranks the worlds in the modal base by how stereotypical or ideal they are. The modal then quantifies over the best worlds in the modal base.

"Must p" is true if p holds in all the best worlds. "May p" is true if p holds in some best world. Necessity and possibility are universal and existential quantifiers over the same domain. Changing flavour is changing the modal base: a circumstantial modal base plus a deontic ordering source gives a deontic reading; an epistemic modal base plus a doxastic ordering source gives an epistemic reading.

The two-parameter design solves a classic puzzle. You should help her is true even though there are clearly worlds compatible with the law where you do not help her. The trick: the deontic ordering source ranks worlds where you obey the law above worlds where you do not, so we quantify only over the law-respecting worlds. Modality is universal but only over the best worlds, not all worlds.

The English modal class

English has a closed class of ten modal auxiliaries: must, may, can, will, shall, should, would, could, might, ought (to). They share peculiar syntactic properties:

• No agreement. He must leave, not *He musts leave.
• Bare-infinitive complement. must leave, not *must to leave — except for ought to, which retains the historical particle.
• No non-finite forms. *to must, *musting, *musted.
• Auxiliary syntax. They invert directly with the subject (Must you leave?) and host negation (You must not leave).

The class is a fossil. In Old English, must's ancestor motan was a regular verb with full inflectional paradigm. Through Middle English the modals lost their non-finite forms and stopped grammaticalising new ones. David Lightfoot's Principles of Diachronic Syntax (1979) treats this as a reanalysis: the modals migrated from V to T, and learners stopped acquiring them as full verbs. The closed-class-ness of English modals is unusual cross-linguistically — German müssen, können, dürfen are full verbs that inflect normally and have non-finite forms.

Cross-linguistic encoding

Language	Encoding strategy	Necessity example	Possibility example
English	Closed-class modal verbs	You must leave.	You may leave.
German	Inflecting modal verbs (full verbs)	Du musst gehen.	Du darfst gehen.
Mandarin	Modal particles preceding the verb	Ni bixu zou.	Ni keyi zou.
Turkish	Verbal suffixes -mAlI- (necessity), -Abil- (ability)	Gitmelisin.	Gidebilirsin.
Quechua	Evidential suffixes -mi (direct), -si (reportative), -chá (conjectural)	(epistemic via evidential)	(epistemic via evidential)
Korean	Sentence-final endings + auxiliaries (-eya hada, -l swu issta)	Ka-ya hae.	Kal swu isse.

Across languages, modality recruits one of four strategies: independent modal verbs (Germanic, Romance), particles or adverbs (Mandarin, some Polynesian), inflectional affixes (Turkish, Algonquian), or evidential systems (Quechua, Tibetan, Tariana). The four strategies are not mutually exclusive — German and Mandarin have both modal verbs and modal adverbs; Korean stacks final endings on top of auxiliary constructions.

Evidentials: modality's neighbour

Languages like Cuzco Quechua, Standard Tibetan, Cherokee, and Tariana grammaticalise the speaker's source of evidence — direct perception, inference, hearsay, reported speech — independently of any modal force. Quechua marks every assertion with one of three suffixes: -mi (direct evidence), -si (reportative), -chá (conjectural). Failing to use a suffix at all is ungrammatical.

Evidentials and epistemic modals share territory: both signal the speaker's epistemic stance toward the proposition. The relationship is debated. Martina Faller's 2002 dissertation argued that Quechua evidentials are not modals — they are illocutionary modifiers, contributing to the speech act rather than the propositional content. Speas (2004) and Aikhenvald (2004) argue that some evidentials shade into modals, particularly the inferential and conjectural varieties. Either way, the existence of evidentials shows that the territory English carves up with epistemic must can be carved up by an entirely different grammatical category in other languages.

Why does epistemic "must" feel weaker than a bare assertion?

Karttunen (1972) noticed that It must be raining sounds weaker, not stronger, than It is raining. The puzzle: must is supposedly a universal quantifier (necessity), so on a naive analysis the modal claim should entail the unmodalised one. Why does adding a "necessity" make the claim feel hedged?

Kai von Fintel and Anthony Gillies's 2010 paper "Must... Stay... Strong!" argues that must retains its universal force but adds a presupposition: the speaker's evidence is indirect. It is raining is appropriate when you can see the rain through the window. It must be raining is appropriate when you have only indirect evidence — wet streets, dripping umbrellas — from which you infer the rain. The apparent weakness is not weakening of force; it is the indirectness presupposition giving a hedge-flavour. The same pattern shows in German muss, French doit, and Italian deve.

Disambiguating the flavour

English modals are systematically ambiguous between deontic and epistemic readings. Several heuristics resolve them in practice:

• Predicate type. Epistemic readings prefer stative or generic complements (be tired, own a car). Deontic readings prefer dynamic, agentive complements (leave, pay the fine).
• Tense and perfect. You must have left is unambiguously epistemic — past-time deontic obligations cannot be retroactively imposed.
• Subject animacy. Inanimate subjects of must almost always trigger epistemic readings (The package must be in the warehouse).
• Adverbs. You must absolutely reads deontic; You must surely reads epistemic.
• Negation scope. Deontic must not is "obligation not to" (You must not smoke). Epistemic can't takes over the slot — ?It mustn't be raining is degraded; It can't be raining is fine.

Variants and related notions

• Subjunctive mood. Romance languages use a dedicated verb form for irrealis contexts — wishes, hypotheticals, embedded modal contents. Spanish quiero que vengas (subjunctive vengas); French il faut que tu partes.
• Conditional clauses. If-then constructions are modal in disguise — Kratzer's 1986 analysis treats if-clauses as restrictors of a covert modal operator (Lewis's 1975 "if-clauses are restrictors").
• Modal concord. Some languages stack multiple modals with apparent semantic redundancy: Catalan has de poder "you must can" can mean simple necessity, not double-modalised.
• Modal subordination. A modal in one sentence can license anaphora into a non-existent referent: A wolf might come in. It would eat you first. The pronoun it picks up a wolf that exists only in the modal worlds (Roberts 1989).
• Performative modality. I promise to come uses no modal verb but encodes deontic commitment performatively (Searle 1969 on speech acts).

Common pitfalls

• Don't equate "must" with logical necessity. Linguistic must is restricted to a contextually-supplied set of best worlds, not all metaphysically possible worlds.
• Don't conflate negation of necessity with necessity of negation. You don't have to leave (¬□p) is very different from You must not leave (□¬p).
• Don't read "can" as ability only. Can I leave? is deontic permission, not a question about ability.
• Don't ignore evidentials when working on epistemic modality typologically. Many languages encode the territory differently — what looks like an "epistemic modal gap" in Quechua is actually filled by evidentials.
• Don't assume one-to-one cross-linguistic mappings. German dürfen is "may" only deontically; epistemic possibility uses können. English-to-German translation breaks if you map by single equivalents.