What a map samples
A typological map places coded language descriptions at approximate locations. It is not a census of every speaker, period or dialect. WALS chapters draw on published descriptions and their authors’ feature-specific samples. Its introduction describes a core set selected to increase geographic and genealogical breadth; individual maps can contain different languages and different amounts of missing information. A dot on a map normally represents one coded variety, not a continuous territorial boundary.
A global percentage calculated from a map is a percentage of that map’s coded sample. It is not automatically a percentage of all human languages. WALS chapter 13 gives an instructive warning: the sample includes much denser representation of some European groups than of Niger-Congo languages, so a raw non-tonal proportion underrepresents the global spread of tone. The problem is not repaired by making a larger map with the same imbalance.
Genealogy and region
Closely related languages inherit many structural traits from a common ancestor. Neighbors may share traits through contact. If a sample includes many members of one family or region, those points are not independent repetitions of the same universal fact. A balanced comparison can take one language per genus or assign weights by family and area, then test whether the pattern survives other plausible samples. There is no single balancing method suitable for every question; report the method chosen.
APiCS is explicitly an atlas of 76 pidgins, creoles and mixed languages. That targeted scope makes it valuable for contact-language comparisons, but it cannot be read as a random sample of all languages. Grambank offers a wider structural database with feature definitions and source records. The datasets answer different questions and use different coding schemes, so a matching column name is not proof of identical criteria.
Unknown is not absent
A language omitted from a map may have no description for that feature, may not be in the chosen sample, or may be coded under a different period or variety. “Not attested,” “not reported,” and “structurally absent” are distinct conclusions. For ancient languages the missingness can be extreme: a royal inscription may preserve dozens of names yet no question, subordinate clause, or ordinary conversation. Mark the untestable feature as unknown.
Exercise: open a WALS map, read the chapter’s definition and sample notes, then inspect three language data points from different families. Find each point’s source. Change one coding rule and predict which cases would become uncertain. This makes the colors on the map an auditable argument.
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.