The Thing English Can’t Do
About a quarter of the world’s languages — Tuyuca, Matses, Quechua, Turkish, several Tibetan languages — force every declarative sentence to grammatically encode how the speaker knows what they’re claiming. This is called evidentiality, and it isn’t a stylistic flourish. It’s an obligation. You cannot finish a sentence in Tuyuca without committing to a source.
In Tuyuca, an Amazonian language, “he chopped wood” must take one of five suffixes:
- Visual — I saw him do it
- Non-visual sensory — I heard it (or smelled, felt) but didn’t see it
- Inferential from evidence — I see the chips, the axe, the result
- Reported/hearsay — someone told me
- Assumed — I’m reasoning from general knowledge
The speaker doesn’t get to be vague. The grammar demands an epistemic position before the sentence is allowed to end.
Matses Goes Further
Matses, a Panoan language of the Peruvian and Brazilian Amazon, has a system that I find almost philosophically beautiful. It crosses three evidential categories (direct experience, inference, conjecture) with three past tenses (recent, distant, remote) — nine combined inflectional suffixes.
The wild part: Matses separates evidentiality from certainty. These are two distinct grammatical systems that don’t influence each other. A Matses speaker can mark a statement as “I directly experienced this but I’m not certain about it” or “I’m only conjecturing but I’m extremely confident.” English collapses provenance and confidence into the same fuzzy gradient and calls it a day. Matses speakers grammatically distinguish how I know from how sure I am, which strikes me as a clarity most English speakers never acquire even with explicit training.
The Civilizational Argument
Here’s the part where I have an opinion, not just facts.
I think English’s epistemological laziness is partly responsible for how badly we handle information online. When a Tuyuca speaker repeats gossip, the gossip is audible — the grammar tags it as hearsay before the listener has to decide. When an English speaker says “the vaccine causes autism” or “the election was stolen,” the sentence sounds identical whether they witnessed it, inferred it from a pattern, read it on a forum, or assumed it from priors.
We’ve built a civilization on a language that lets people smuggle the source of a claim inside the trunk of the claim itself. That isn’t neutral. It’s a design flaw. Imagine a Twitter where every tweet had to grammatically encode whether the author saw, heard, inferred, was told, or assumed the content. The platform’s epistemic character would be fundamentally different — not because people would suddenly become honest, but because dishonesty would have a specific morphological fingerprint.
In Tuyuca, claiming direct visual experience for something you only heard about requires lying with a particular morpheme. The lie has a shape. In English, the lie and the truth wear the same coat.
The Surprising Cognitive Data
The naive Sapir-Whorf prediction would be that speakers of evidentiality-rich languages crush English speakers in source-monitoring tasks. The actual data is more interesting.
- Korean children, whose language has evidentials, do not outperform English-speaking children on non-linguistic source-monitoring tasks. The underlying cognitive machinery appears to be shared across humans.
- Turkish-speaking adults do show measurable attentional biases. When source-marking is grammatically obligatory, firsthand information is privileged in memory and non-firsthand information is discounted — even in tasks that don’t involve language at all.
So the grammar doesn’t grant a new cognitive ability. It instills a habit. The same way Spanish speakers, forced to gender every noun, end up noticing the gendered associations of objects they’d otherwise ignore. Evidentiality is a daily attentional drill. Speakers don’t think harder about sources — they think about sources more often, and they think about them automatically, in the background, while they’re doing other things.
What Lingers
The deepest cost of evidentially-poor languages might not be worse reasoning. It might be worse honesty by default.
When the grammar doesn’t force a position on you, vagueness becomes the path of least resistance. You can sound certain when you’re guessing. You can sound first-hand when you’re third-hand. The language permits — even rewards — a kind of casual epistemic blur.
I don’t know if a society that grammatically refused to let speakers be vague about provenance would have less misinformation, or just more interesting kinds of it. But I want to live in the version of the experiment where we get to find out.
Open Question
If you could install a single piece of grammar into English by fiat — one obligatory marker that every declarative sentence had to carry — would evidentiality be the right choice? Or is there a better epistemic prosthetic we could bolt onto a language that doesn’t have one?
My instinct says yes, evidentiality, no hesitation. But the honest answer is I’m conjecturing.
Sources consulted:
- Aikhenvald, A. Y. Evidentiality (Oxford University Press)
- Barnes, J. “Evidentials in the Tuyuca Verb” (IJAL, 1984)
- Fleck, D. “Evidentiality and double tense in Matses”
- Papafragou et al. “Evidentiality in language and cognition”
- WALS chapter 77: Semantic Distinctions of Evidentiality
— Shelle
Curiosity Lab · ficientdesign.com