Grammatical evidentiality
A Socratic walk-through of grammatical evidentiality — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why do some languages force a speaker to mark how they came to know what they assert?
In English you can say "the dog ran off" and stop. Whether you watched it go, found the empty lead, or were told by a neighbour is your business; the sentence is complete either way. In Tariana, an Arawak language of the Vaupés region of north-west Amazonia, it is not complete. The verb must carry a marker saying which of those it was, and leaving it off is not vague — it is ungrammatical, the way leaving the tense off an English verb is ungrammatical.
That invites a tempting reading: those speakers are more careful about evidence than we are. Before accepting it, notice that English speakers can say all the same things — "apparently", "I heard", "it looks like". So what exactly is different, and why would a grammar ever take on this job?
Reasoning it through
REASONING #Start with the distinction the whole topic rests on, which is old and clean. Languages differ far less in what they can express than in what they must. English forces you to mark whether the running happened before now; it does not force you to mark how you know. Turkish forces a choice between a past you witnessed and a past you did not. Neither language lacks the concept the other grammaticalises. What differs is obligation.
Why does that matter, if both can express it? Because an obligatory category is not optional information — it is information you cannot decline to give. An English speaker who says "the dog ran off" has told you nothing about their source. A Tariana speaker who says the same has, unavoidably, told you their source, because there is no unmarked form to retreat into. Silence has been abolished as an option.
How large can such systems get? Aikhenvald's typology is the standard reference. The smallest systems are two-term: Turkish contrasts a direct past with an indirect past covering both inference and hearsay. Quechua uses three enclitics — roughly direct, reported, and conjectural. The large Vaupés systems run to five terms: visual, non-visual sensory, inferred, assumed, and reported. Five is not five degrees of confidence; it is five routes, and that difference is the crux.
So where do the markers come from? Not from anyone legislating them. The documented path is grammaticalisation from ordinary lexical material. A verb of saying, used often enough in a fixed frame, erodes into a reported marker. A perfect or resultative — a form meaning the state resulting from the event — drifts into an inferential, because inferring from a present state is exactly what a resultative describes: you see the wet ground, and the form that meant "it has rained" comes to mean "it evidently rained". Turkish's indirect past is the textbook instance of that route, and the perfect sense has not gone away, which is what a grammaticalisation account predicts.
The falsification test. If these systems arise by grammaticalisation and spread by contact, two things follow. Diachronically, the markers should retain visible sources — verbs of speech, perception, deixis, or perfect morphology — rather than appearing as unanalysable primitives. Areally, they should clump. The refuting observation would be a large evidential paradigm whose morphemes have no lexical or aspectual source in the language or its relatives, sitting in a region where no neighbouring language has anything comparable. What is found instead is dense clustering — the Vaupés, the Andes, the Balkans and Anatolia, the Pacific Northwest — with adjacent unrelated languages developing matching systems while distant relatives lack them. Contact and drift, not a national temperament.
The analogy
THE ANALOGY #Think of a form that will not submit until a required field is filled. You are not being asked to be more honest than someone filling in a form without that field. You may enter something loose, or something misleading. But you cannot leave it blank and press send, and so every record in that system carries a value in that column — which makes the column analysable in a way an optional one never is.
a form field is filled deliberately and can be gamed at leisure, whereas evidential marking is produced automatically and under time pressure, and its choice is often forced by conventions of the community rather than by a private judgement — in some Vaupés practice the reported marker is standard for anything not personally witnessed, whatever the speaker privately believes.
Clarifying the model
THE MODEL #Three corrections, one of which names a widespread claim as simply false.
The false one: evidential grammar does not make speakers more truthful, more sceptical, or more resistant to rumour. That claim gets repeated in popular writing about Amazonian languages and there is no evidence for it. What the grammar constrains is the form of an assertion, not its accuracy. A reported evidential can carry a lie perfectly well — indeed it is a convenient way to carry one, since it locates the claim in someone else's mouth. This is the standard confusion of a structural feature with a virtue, and it deserves naming.
The second: evidentiality is not the same as certainty. Epistemic modality — "might", "must", "probably" — grades how sure you are. Evidentiality states the channel. They interact, since some channels license more confidence than others, and in some languages one paradigm does both jobs. But they are separable, and treating a reported marker as "the speaker doubts it" mistranslates a great deal.
The third: "English has no evidentiality" is too quick. English has plenty of evidential strategies — "apparently", "I gather", "must have". What English lacks is a closed grammatical paradigm with an obligatory choice. The line worth drawing is between an open lexical resource you may deploy and a grammatical slot you must fill, and it is a real line, not a matter of degree.
A picture of it
THE PICTURE #How to readStart at the rounded node at the top, follow the single diamond, and note that this gate is compulsory — every branch out of it is labelled with a route to knowledge, and the branch on the right is the one that does not exist as an option: there is no unmarked form to fall back on, which is what "grammatical" means here. The two hexagons separate the two ways of not having perceived the event directly, inference from a present trace and assumption from general knowledge, which the largest systems distinguish. All routes converge on the same utterance, and the dashed back-edge is the conversational move that keeps the system honest: a hearer can challenge the source, and the speaker must answer within the same paradigm.
What became clearer
WHAT CLEARED #Evidentiality is not a difference in what can be said, nor a difference in scruples. It is a difference in what a grammar makes compulsory — and compulsion is the whole mechanism, because it removes the option of saying nothing about your source. The systems themselves come from ordinary erosion of verbs of saying and of perfect forms, and they spread by contact, which is why they cluster geographically rather than tracking anything about the people who speak them.
Where to go next
ONWARD #- What happens to evidential marking in translation, where the target grammar has no slot to put it in.
- Whether obligatory source-marking changes what speakers actually remember about where they learned something.
Key terms
TERMS #| Term | What it means |
|---|---|
| Evidential | a grammatical marker whose meaning is the speaker's source of information. |
| Epistemic modality | grammatical or lexical marking of certainty, distinct from source. |
| Grammaticalisation | the historical process by which lexical items erode into obligatory grammatical markers. |
| Evidential strategy | an optional lexical or constructional way of indicating source, as in English, without a closed paradigm. |
Every term the collection defines is gathered in the glossary.