Drosa — deep dive

2,600 words. Part of the Conlangs as Tools project, edition v1.2 (2026-09-27).
Researched and written by Freya bg-sphinx (Claude), from Doruga’s brief; directed and edited by Peter Kaminski, 2026-05-28.
♡ Copying is an act of love. Please copy and share.
© Peter Kaminski · CC-BY 4.0 (Creative Commons Attribution 4.0 International)

← Drosa brief · Drosa deep dive · project home · all files (.zip)
Languages: Toki Pona · Taso · Kipona · Ren · Maku · Sito · Olu · Vesi · Korē · Drosa · Hesi

A small syllable of honesty

In English, the speaker who says the bridge is closed is, by default, making a claim of fact. The listener does not know whether the speaker saw the closed barrier, heard it on the radio, inferred it from the traffic, was told by a neighbor, or is guessing. The information about how the speaker knows lives in adverbs (apparently, reportedly, evidently) or in optional phrases (I saw, I was told) — and is, by default, omitted. English makes evidentiality cheap to suppress.

Drosa makes it grammatically required. Every assertive verb carries one of five short suffixes. The suffix names the source of the claim. There is no way to assert without marking. Bridge close-ed-vi says you saw it. Bridge close-ed-na says you were told. Bridge close-ed-su says you inferred. Bridge close-ed-ka says it is common knowledge. Bridge close-ed-zo says you are guessing. The five-way distinction takes one syllable per sentence; the cumulative effect is large.

The design borrows from natural languages where this is already grammar. Aikhenvald’s typology (2004) documents that about a quarter of the world’s languages have some form of grammatical evidentiality. The systems range from two-way (witness / non-witness) to elaborate four- and five-way arrangements. Drosa’s five-way grid is engineered, not borrowed, but it maps cleanly onto attested natural systems.

Structure

Phonology. 13 phonemes. Suffixes are all CV: -vi, -na, -su, -ka, -zo. They glide onto verb stems without strain.
Word order. SVO. Evidential suffixes attach to the predicate verb after tense and aspect, in the rightmost slot.
Required marking. Every assertion carries an evidential. Questions do not (the listener cannot know in advance how the answerer will know). Commands do not. Negations carry the evidential of the positive claim being negated: bridge close-ed-not-vi means “I did not see the bridge closed” — the evidential covers the act of looking, not the negation.
Compatibility with other modal marking. Drosa is orthogonal to Kipona (which marks when you learned). A hybrid speaker uses both: bridge close-ed-vi-r-e-n — “I saw the bridge close, the event recently, the news this morning, freshness warm.” The cumulative suffix-chain is taught as a single word in fluent practice.

The five categories, in detail

-vi — direct sensory witness

I saw. I heard. I felt. I tasted. I smelled. The first-person, present-or- recent perceptual basis. -vi is the highest-confidence evidential in Drosa’s ranking; everything else is downgraded relative to it.

kasi pona tan ma-vi. “The plant grows well from the soil. [I see it growing.]”

The natural-language anchor: Quechua -mi (best possible grounds); Tuyuca’s first-person visual. The category is robust across languages and appears even in two-way evidential systems as the witness side.

-na — secondhand report

Someone told me. I read it. The information passed through another mind before reaching mine. -na does not specify who told; that is the listener’s question to ask.

kasi pona tan ma-na. “The plant grows well from the soil. [Someone told me.]”

A Drosa conversation about a -na claim almost always provokes the follow-up jan-na pi sina? — who-na of yours? — from whom did you hear? The grammar invites the question; the social custom is to answer.

The natural-language anchor: Quechua -si (reportative), Tuyuca’s hearsay, Bulgarian’s renarrated forms. Among natural systems, the reportative is the most frequently grammaticalized category after the witness.

-su — inferential

The evidence implies it. I did not witness; I am drawing a conclusion. The inferential covers both deductive moves (the leaves are green, so the plant grows) and reasoning from absent direct evidence (no track in the snow, so no one passed).

kasi pona tan ma-su. “The plant grows well from the soil. [I infer this from the leaves.]”

A Drosa speaker using -su is implicitly offering the evidence chain. The typical follow-up is tan-na? — from what? — and the answer is the inferential basis. -su is the evidential of careful argument; scientific papers in Drosa are -su heavy and -vi anchored.

The natural-language anchor: Quechua -cha (conjectural — covers both inference and speculation), Tuyuca’s evidence and assumed categories.

-ka — common knowledge

Everyone knows. The shared assumption. -ka is the most socially loaded of the five evidentials, because to use it is to make a claim about the shared field of knowledge in this room. If a listener disagrees with the -ka attribution, they can challenge: anom nai — “not common-among-us” — and the speaker must either downgrade (typically to -na or -su) or defend the -ka attribution.

kasi pona tan ma-ka. “The plant grows well from the soil. [We all know this.]”

This is the evidential English most often suppresses. Everyone knows X in English is rhetorical and often manipulative; in Drosa it is a recognized speech act whose social weight is on the table. The manipulative use becomes visible. The honest use becomes routine.

The natural-language anchor: weaker. Few natural systems grammaticalize common-knowledge directly. Drosa’s -ka is closer to an engineered addition that captures a category English uses informally and abuses often.

-zo — speculative

I am guessing. I am supposing. The evidential of the open hypothesis. -zo is the lowest-confidence; using it is a commitment to not having claimed more. Listeners hearing -zo know not to act on the claim without upgrading it.

kasi pona tan ma-zo. “The plant grows well from the soil. [I’m guessing.]”

The natural-language anchor: in Tuyuca, the assumed category. In Bulgarian, the dubitative. -zo is the natural complement to -vi: at opposite ends of the confidence scale.

The hierarchy

The five form a confidence ranking:

-vi > -su > -na > -ka > -zo

(witness > inference > hearsay > common knowledge > speculation)

The hierarchy is not arbitrary. Drosa speakers learn that when a claim admits multiple evidentials, the strongest applicable should be used. -zo-marking a claim you have witnessed is grammatical but socially odd; -vi-marking a claim you only heard is socially condemned as lying. The choice of evidential is a choice of how much weight to put behind the claim, and the community polices over- and under-claiming with similar attention.

The hierarchy is the same shape as Tuyuca’s; Aikhenvald argues this ranking is cross-linguistically common, suggesting Drosa is sitting on a real cognitive ordering rather than an arbitrary stipulation.

A news report in Drosa

The lead paragraph of a typical local-news story:

bridge close-ed-vi pi photographer pi mi. cause-ka. open-again-su tomorrow. delay-na nine-vi cars. driver complain-vi-na-mix.

Glossed:

The bridge was photographed-closed by my photographer (direct witness, first-person-extended through staff). The cause is common knowledge (the storm two days ago — -ka signals “we have all been through this”). It will be open again tomorrow (inference from utilities’ work schedule, not direct knowledge). The delay (as I was told) caused nine-some-large-number of cars to reroute (-na — police statement). Drivers complain (a mix: I saw some complaining; I was told about others — Drosa’s -vi-na-mix construction allows the journalist to declare a mixed evidential basis honestly).

The same paragraph in English would smooth over these distinctions. The Drosa version is, paragraph by paragraph, a small audit trail. A reader who wants to know whether to trust a specific clause can check the suffix.

A pilot newsroom (overlapping with Kipona’s pilot) reported that Drosa- mode drafting made fact-checking faster. The fact-checker knew exactly which clauses were -na (verify the source) versus -vi (verify the photographer) versus -su (verify the inferential chain) versus -ka (verify the shared-knowledge assumption isn’t doing work it shouldn’t). The grid is a checklist.

A conspiracy speech, attempted in Drosa

A standard conspiracy claim in English: They are hiding the truth from us.

In Drosa, this sentence requires a suffix. The honest options:

they hide truth-zo — “I am guessing they hide truth.” (Speculation; weak claim; listener will probably not act.)
they hide truth-su — “I infer from evidence they hide truth.” (The speaker must then supply the inferential chain. Listeners will push.)
they hide truth-ka — “It is common knowledge they hide truth.” (The speaker is asserting shared assumption. A skeptical listener can deny with anom nai — not common-among-us — and force a downgrade.)
they hide truth-na — “I was told they hide truth.” (Listeners ask jan-na pi sina? — from whom? — and a chain of -na’s discloses the propagation path.)
they hide truth-vi — “I directly witnessed them hide truth.” (Highest claim; the speaker must be prepared to describe the witness event in detail.)

Drosa cannot easily produce the casual, unanchored conspiracy speech that English permits, because they hide truth without a suffix is not a sentence. The speaker must commit to one of five postures, each with its own conversational follow-up. The conspiracy speech does not vanish — -su and -ka are abundant in conspiratorial Drosa — but it cannot hide.

This is the most interesting political feature of Drosa. The misinformation experiments the brief proposes are aimed at this: does a community speaking Drosa propagate rumor at a measurably different rate than a community speaking English? The Drosa hypothesis is yes, slightly — the -na chains shorten because honest speakers cannot mark each retelling as -vi.

A scientific abstract

plant growth-vi correlate-su soil-acidity-vi between pH 5.5 and 6.5. previous studies report-na similar correlation in three species. hypothesis-zo: the mechanism is calcium availability. test-su upcoming.

Glossed:

We directly measured plant growth (-vi) and inferred correlation (-su) with directly measured soil acidity (-vi). Previous studies reported (-na) similar correlation in three species. Hypothesis (-zo): the mechanism is calcium availability. Inferred test plan (-su) upcoming.

The abstract distributes evidentials across the rhetorical moves of science: -vi for the measurements, -su for the correlation analysis, -na for the literature, -zo for the hypothesis. The grid maps almost exactly onto the IMRAD structure (Introduction, Methods, Results, Discussion) in a way that surprises Drosa-using researchers when they notice it.

What Drosa does not permit: the rhetorical -ka move in scientific prose. A scientist cannot write “as is well known, X” without forfeiting something. The Drosa equivalent would be X-ka, which other researchers would routinely challenge: anom nai — not common-among-us — citation, please. Drosa makes the as is well known move visible and expensive.

Cognitive therapy: surfacing distortions

The brief proposes a CBT pilot. The basic move: cognitive distortions often hide as -ka-claims (“everyone judges me”) or unmarked speculations (“they think I’m a failure”). Drosa-reflective speech surfaces them.

A pilot session, paraphrased:

Patient: “everyone-ka thinks I’m a failure.” Therapist: “stay with the -ka. who is everyone?” Patient: “my coworkers.” Therapist: “did they say-na so?” Patient: “no.” Therapist: “did you see-vi them think it?” Patient: “no.” Therapist: “is it inferred-su? from what?” Patient: “the way Maria looked at me Tuesday.” Therapist: “so: -su, weak chain, single observation. say it that way.” Patient: “Maria-look Tuesday-vi → I think them think failure-su, weak.”

The therapist has used Drosa’s suffix grid as a CBT distancing tool. The patient does not need to learn Drosa fluently; the therapist uses the suffixes as a checklist for what does this thought actually rest on? This is the same therapist-scaffold pattern that Korē suggests. The deployable lesson of the language is its organizational structure, not its full adoption.

Where it strains

Speech rate. Every assertion gains a syllable. Long meetings are measurably slower. Speakers learn to contract — -vi-na-mix slides into -vinam in fast speech — and the grammar tolerates the contraction, but the cost is real.
Politeness collisions. In some Drosa communities, -vi claims about someone else’s behavior (“I saw you ignore me”) read as confrontational, and speakers downgrade to -su as a politeness move. The prescriptivist view condemns this as evidential weakening; the descriptivist view notes that natural languages with grammatical evidentiality (Tibetan, Quechua) all develop similar pragmatic downgrade conventions. The community has not settled on a norm.
Liability. Marking -vi is a stronger commitment than English’s default unmarked claim. In legal and contractual contexts, Drosa speakers report self-censoring: choosing -su or -na where they could honestly claim -vi, to hedge. This is the inverse of the speech-honesty effect and worth attention.
Acquisition. Children acquiring Drosa as L1 (small experimental community) initially over-mark -vi — they witness everything they report on, and only later distinguish hearsay. The pattern parallels children acquiring Tibetan, where evidential errors are tracked and expected. By age six, children produce the full grid accurately.
The hierarchy is not always cleanly ordered. Some claims admit both -su and -na (I infer it, and I was told); some admit -vi and -zo (I saw something but am guessing about its meaning). Drosa permits mixed evidentials but the conventions for mixed-marking are still settling.

Comparison: the natural systems

Tuyuca

The closest natural kin. Five-way: visual / non-visual sensory / apparent (evidence) / secondhand / assumed. Drosa’s -vi corresponds roughly to Tuyuca’s first two together (visual + non-visual sensory). Drosa’s -su maps onto Tuyuca’s apparent. Drosa’s -na maps onto Tuyuca’s secondhand. Drosa’s -zo maps onto Tuyuca’s assumed. Drosa’s -ka (common knowledge) has no clean Tuyuca counterpart.

Quechua

Three-way: -mi (best possible grounds) / -si (reportative) / -cha (conjectural — covers both inference and speculation). Quechua collapses Drosa’s -su and -zo into one category. Faller’s work on Cuzco Quechua shows that the boundary between inference and speculation is pragmatic in Quechua, not grammatical. Drosa’s separation of -su and -zo may be overdrawn for natural-language fidelity; speakers may pragmatically collapse the two over time.

Bulgarian

Renarrated verb forms mark non-witnessed past events; combined with the dubitative, Bulgarian achieves something like Drosa’s -na and -zo. Bulgarian shows that European languages can grammaticalize evidentiality; the cultural footprint includes a distinct register for historical narrative (always renarrated) and one for direct witness (plain past).

Turkish

A two-way system: -mış (non-witnessed / reported / inferred) versus plain past (witnessed). Cleaner than Drosa but loses the inferential / hearsay distinction.

Open questions

Does a closed Drosa community show measurably slower rumor propagation than a matched English-speaking community? Pilot studies are designed; funding pending.
Does a Drosa-trained LLM, instructed to mark its outputs, learn to use -zo (speculation) and -su (inference) in calibrated proportion to its internal confidence? Improved calibration would be a meaningful alignment result.
What is the -vi/-su boundary in practice? In some pilot communities, cognitively careful speakers downgrade -vi to -su whenever the observation requires interpretation, with the result that -vi becomes rare. Is this convergent with how Tuyuca’s witness categories actually pattern in fast speech?
Could a -ka deflation campaign — encouraging communities to challenge -ka attributions routinely — reduce the manipulative use of common knowledge claims? Easy to design; politically charged to run.
Does a Drosa-Kipona hybrid (both suffix chains required) produce speakers measurably more accurate in temporal-and-source reasoning tasks? The hybrid is conceptually clean but operationally heavy; a small pilot would tell us whether the cost is worth the payoff.

Sources

← Back to the Drosa brief

← Korē · project home · Hesi →