Evidentiality: Grammar That Asks How You Know

English lets me say “he played football” without revealing whether I watched, heard the match, saw muddy boots, received a report, or guessed. Tuyuca requires speakers to mark that distinction in many statements. Janet Barnes’s 1984 description identifies five sets of verbal suffixes that mark the source behind a statement: visual, nonvisual, apparent, secondhand, and assumed.

Five answers to one claim

Tuyuca is an Eastern Tukanoan language spoken in Colombia and Brazil. Its verbal morphology marks tense, person, number, and gender, while its evidential suffixes identify the speaker’s information source.

Category	What licenses the claim
Visual	I saw the event
Nonvisual	I heard or otherwise sensed it
Apparent	I inferred it from perceptible traces
Secondhand	Someone reported it
Assumed	I reasoned from prior knowledge

The suffix does not certify truth. A witnessed event can be misperceived, while a reported claim can be correct. Evidentiality records the route by which information reached the speaker.

Grammar marks information source

Ferdinand de Haan’s 2013 WALS sample contains 418 languages: 181 lack grammatical evidentials, 166 mark only indirect evidence, and 71 distinguish direct from indirect evidence. English can express the same ideas with phrases such as “I saw,” “apparently,” and “they say,” but usually makes them optional.

That difference resembles concept information theory. Two sentences may carry the same proposition while one also identifies how the information reached the speaker. Tuyuca encodes that source in the verb.

What is contested

Linguists disagree about where evidentiality ends and epistemic modality begins. Evidence source answers “How do you know?” Confidence answers “How certain are you?” Real languages often let the two meanings bleed together, especially around inference.

A second dispute concerns cognition. Obligatory marking proves that speakers must select a grammatical form; it does not prove they are more truthful or reason differently outside speech. The claim that evidential grammar rewires thought needs more evidence than the grammar itself.

An open question

Could an AI trained to mark every sentence as observed, retrieved, reported, or inferred become easier to audit, or would it merely learn five new ways to sound certain? That question sits between concept bayesian inference and concept information theory.

Key Sources

Janet Barnes, “Evidentials in the Tuyuca Verb,” International Journal of American Linguistics 50(3), 1984, pp. 255–271. DOI
Alexandra Y. Aikhenvald, Evidentiality, Oxford University Press, 2004. Publisher record
Ferdinand de Haan, “Semantic Distinctions of Evidentiality,” The World Atlas of Language Structures Online, 2013. WALS Chapter 77
Ferdinand de Haan, “Coding of Evidentiality,” The World Atlas of Language Structures Online, 2013. WALS Chapter 78

Abhishek's take

The grammar interests me as interface design, not as an exotic language fact. I want machine answers to make “observed,” “retrieved,” and “inferred” as difficult to omit as tense. A confidence score tells me how loudly a system speaks; an evidential tells me where the claim came from.

Abhishek