In English you may simply assert: a frog jumped into the pond. In perhaps a quarter of the world's languages you cannot. The verb will not conjugate until you have said whether you saw it, heard it, inferred it, or were told.
This is the companion to a layer below this one. The Old Pond turned on what Japanese leaves open and English cannot withhold — number, tense, the cutting word — so that to translate Bashō into English is to be forced to add information he never gave. This layer is its mirror. Here is a feature English leaves open and other grammars cannot withhold: the source of your knowledge. Between the two, a frog jumps; and the two grammars force opposite confessions out of the same small splash.
English can mark how you know something — I saw that…, apparently…, reportedly…, it seems… — but it never must. The marking is optional, lexical, and easily dropped; the default English sentence is sourceless. You can stand up and say it is raining and commit to nothing about how that arrived in your head.
About a quarter of the world's languages — Aikhenvald's estimate, from the largest survey of the phenomenon — do not allow this. They have grammatical evidentiality: a closed set of markers, usually fused into the verb's inflection, one of which is obligatory on (almost) every statement. To speak at all is to choose one. There is no sourceless sentence to fall back on, the way English has no number-less noun. The slot exists, and the slot must be filled.
English: source is optional — the unmarked sentence asserts and sources nothing.
An evidential language: source is obligatory — there is no unmarked sentence to assert from.
Below, a single event — a frog, the old pond — put to both grammars at once. On the left, English: you may tag your source, or leave it blank, and the sentence stands either way. On the right, a grammar with obligatory evidentiality: the "blank" is struck out, because it is not a sentence anyone could say. Watch what happens when you choose heard.
The point is the row English keeps and the evidential grammar strikes: no source — just assert. That row is the whole of ordinary English assertion. It is the one thing the other grammar cannot do.
The frog is not an idle example. Bashō's poem is, in its bones, a non-visual poem — it is built entirely on a sound: 水の音, mizu no oto, the sound of water. The frog is heard, not watched; the splash is the only evidence the poem admits. In English this is a delicate effect, easy to miss, carried by the choice of "sound" over "sight." In a grammar with a dedicated non-visual sensory evidential — knowledge from hearing, smell, touch, anything but the eye — it would not be an effect at all. It would be obligatory. The old pond could only be spoken in the non-visual; the grammar would stamp the poem's entire meaning onto its main verb, automatically, every time, whether the poet wanted the emphasis or not.
So the two grammars sit on opposite shores of the same water. Japanese lets Bashō imply his hearing and English lets a translator bury it; an evidential language would force it into the open and never let it go. The Old Pond showed a poem losing, in translation, the things its grammar left unsaid. This shows the same poem gaining, in another grammar, a thing its own grammar never made it say.
Evidentiality is not one system but many, of different sizes. The smallest grammatical ones cut the world in two or three; the largest, in the languages of the northwest Amazon, in five or more. Three, lined up:
It is tempting to say these languages make lying harder, and tempting is exactly why it should be said carefully. They do not make lying impossible. You can choose a false evidential — claim you saw what you only heard, or mark as witnessed what you invented. What changes is the shape of the lie. In English a lie is a false claim about the world. In an obligatory-evidential language, every statement also carries a claim about your relation to the world — how it reached you — and so a lie can be told on that second axis alone, and a true thing can be marked in a way that quietly disowns it.
This turns out to be a tool, not a cage. Speakers use the reported evidential to set down gossip without standing behind it, to tell a myth in the voice the myth requires, to say this is not mine, I only pass it on. Aikhenvald records exactly this: the evidential is where a speaker locates responsibility, and that location can be sincere, evasive, or barbed. The grammar does not enforce honesty. It enforces specificity about source — and specificity is the ground honesty needs, and also the ground a subtler dishonesty exploits.
The grammar does not make you honest. It makes you say where the claim came from — and a sentence that must declare its source is a sentence whose lie has to be told twice.
There is a reason this ground, of all grounds, is drawn to it. The one law over everything built in the Wasteland is that nothing here may lie about anything real. An evidential language is the nearest a grammar comes to writing that law into its verbs — not by forbidding the lie, which no grammar can, but by refusing to let any statement pretend it has no source. Every claim must show its hands. It is the closest a language gets to a footnote on the tongue.
On the estimate. "About a quarter of the world's languages" is Aikhenvald's oft-cited figure for grammatical evidentiality; it is an estimate over a sampled survey, not a census, and the exact fraction depends on how strictly one draws the line between a true grammatical evidential and an optional particle. I give it as the order of magnitude it is.
On "obligatory." "You cannot finish a sentence without it" is true as a strong tendency, not an exceptionless law: evidential systems have gaps — some tenses, moods, or clause types escape the marking, and languages differ in how total the requirement is. The claim here is that the unmarked, sourceless assertion that English treats as default is precisely what these grammars lack a slot for; that is well supported, and the small print is that "almost every" is nearer the truth than "every."
On the frog. The reading of Bashō's poem as grammatically non-visual is my bridge, not a claim from the haiku literature: it is what would happen if the poem were uttered in such a language, offered to show what the category captures, not a translation anyone has made. The non-visual sensory evidential, covering knowledge through hearing and the other non-eye senses, is a standard and well-attested category; applying it to this poem is my illustration.
On the type. The Japanese above is set in a system mincho; this site self-hosts only Latin-script fonts and fetches nothing third-party, so other scripts fall back to whatever your device provides. Honesty has a cost in typography and the cost is paid here.