A repeatable reading procedure
Start with the chapter or feature definition, not its map color. Identify the unit of analysis, sample rule and allowed values. Open the language data point and trace its cited grammar or corpus. Determine whether the point describes a language variety, a standardized register or a broad name covering several stages. Check whether the relevant source actually provides the diagnostic constructions. Finally, compare a second independent analysis if one exists.
When two atlases differ, do not average their values. They may define the feature differently, code different varieties, privilege different grammatical domains, or draw on different descriptions. Write a short reconciliation note: feature definition, time and variety, source used, and what changed. The disagreement can identify a genuinely interesting linguistic question.
Tables and statistics
A cross-tabulation of two features can suggest correlation, but rows may share genealogical history or regional contact. Missing data can be nonrandom: the best-described literary languages often have more coded features than fragmentary ones. At minimum, report the number of coded languages, how many are missing, which families and regions dominate, and the effect of excluding closely related languages. A p-value alone cannot correct a bad feature definition or dependent sample.
Keep denominators local to each dataset. A WALS tone map and a Grambank order feature need not use the same set of languages. Joining them requires stable language identifiers and period choices, and even then the intersection may be small or biased. A table of “all ancient languages” cannot silently treat a missing code as “no.”
Rights and attribution
Use the databases’ own citation and licence pages before reusing records, maps or illustrations. This course links to source descriptions and gives short attributed examples; it does not import entire datasets or reproduce atlas maps. If a future Language Help tool derives a coded data table, preserve record-level provenance, version information and any required licence notice. The public lesson prose is original teaching analysis, while cited forms and definitions remain attributable to their sources.
Exercise: choose one language and one feature in each of two databases. Record the source title, feature definition, coded value and date accessed. Explain whether the values can be compared directly. If the answer is no, specify the differing criterion rather than calling either database wrong.
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.