The attestation filter
A surviving corpus is a sample produced by preservation, excavation, decipherment, editorial choice and digitization. Administrative tablets, funerary formulas and monumental dedications are not interchangeable with spontaneous conversation. A language represented mainly by personal names cannot be coded for basic clause order; one represented by unvocalized writing may not support a full vowel inventory. State the genre and size of the evidence before assigning structural values.
The dates of composition, surviving copy and edition may differ. A manuscript copied a millennium after composition may preserve archaic forms, introduce newer ones, or both. Cross-language comparisons should specify a stage rather than collapsing “Ancient Egyptian,” “Greek” or “Latin” into one timeless data point. This is particularly important when a feature changes across well-attested periods.
Layers of inference
For each proposed feature, separate four layers: what the object visibly records; how an editor reads or restores it; the linguistic analysis of the attested form; and the typological code. A missing sign restored to produce an object–verb clause is not independent evidence for order. A grammatical parse chosen partly because the language is believed to be SOV cannot then be counted as new proof of SOV without circularity.
A useful profile carries a status for each feature: directly supported by multiple attested constructions, supported but limited, disputed, reconstructed from comparative evidence, or not assessable. Attach line references, edition and date. “Not assessable” is a positive scholarly conclusion when the required construction is absent. It protects readers from mistaking a blank map cell for a negative finding.
Comparison without filling gaps
Related languages and neighboring languages can suggest hypotheses about an ancient variety. They cannot automatically fill unattested cells. A daughter language may have innovated; a neighbor may have influenced another stage; a writing system may hide a distinction. State the hypothesis and the independent evidence that would test it. Typological commonness is a weak prior, not a substitute for an inscription.
Exercise: create a five-feature profile for one ancient corpus: basic order, noun case alignment, article marking, consonant inventory and evidentiality. For each, write one cited attestation, the relevant feature definition and a confidence status. If the corpus does not show the necessary contrast, enter “not assessable” and explain which evidence is missing.
Modern-language examples on this page clarify a linguistic pattern. They do not establish unattested sounds or forms for an ancient language. Follow the cited sources below and keep editorial readings separate from the surviving trace.