Abhishek S.
Shipping in public. Listening in private.

Abhishek

I lead women’s Indo-Western & Premium at Max Fashion. I also wrote the AI that runs the buying floor.

Rare profile. Category operator who ships production code.

Senior Buying Leader · Max Fashion Women’s Indo-Western & Premium · 530+ India stores NIFT ’12 · Twelve years on the floor

abhishek@bengaluru ~ %
>role: senior buying lead
>dept: women’s indo-western + premium
>floor: 530+ stores india

Evidentiality: The Grammar of Evidence

In Tariana, “the dog bit him” is grammatically unfinished until the verb also says how the speaker knows. The recent-past ending can mark visual evidence, nonvisual sensation, inference from a specific trace, assumption from general knowledge, or another person’s report. Evidence is part of the sentence’s machinery, not a disclaimer attached afterward.

How the receipt enters the sentence

Tariana is an Arawak language of Brazil’s Vaupés region. In ordinary non-future declarative clauses, tense and evidence source fuse into one marker: -ka for something seen, -mahka for something heard or otherwise sensed, -nihka for a specific inference, -sika for an assumption, and -pidaka for a report in the recent past.

The categories are not universal. Tuyuca, documented by Janet Barnes in 1984, also sorts statements by evidence source, while Cuzco Quechua uses three markers whose meanings reach into speaker commitment and conversational authority. English can express the same distinctions with phrases such as “I saw,” “apparently,” or “she told me,” but it usually permits the provenance to disappear.

What is contested

Evidentiality does not automatically measure truth. A reported statement can be trusted completely; a visual statement can be mistaken. The live dispute concerns where information source ends and epistemic stance begins: Alexandra Aikhenvald treats them as distinct grammatical domains, while work on Cuzco Quechua shows that particular evidentials can also alter the force of an assertion.

The stronger psychological claim remains unproved. Obligatory evidentials make speakers encode provenance, but they do not by themselves show that speakers reason more accurately or lie less often.

Grammar as information architecture

A Tariana proposition travels with a source label that cannot be removed without damaging the clause:

message = proposition + provenance

That resembles concept information theory, except the extra signal identifies a channel rather than reducing uncertainty directly. A listener still has to decide how much weight to assign sight, inference, or testimony, the same problem that links probability to codes in concept probability cipher connection. Grammar has made the first step of source criticism compulsory.

An open question

If generated text becomes a routine source of assertions, should machine provenance remain optional metadata, or become part of the grammar of digital communication?

Key Sources

Further Reading

Abhishek's take

I care less about asking English to copy Tariana than about the defect Tariana exposes. A generated answer without provenance is like a Tariana verb missing its final clitic: fluent on the surface, structurally unfinished.

Tags: #evidentiality #grammar #epistemology #linguistic-typology #information-source

See Also