Why a single language score cannot be built
Combining interface, document and voice coverage into one language score requires a weighting. No weighting is defensible, because the readers who care about each column want different vendors at the top. As of 2026-09-12.
| Column | Matters most to | Matters least to |
|---|---|---|
| Interface languages | A team buying a tool to work in | A studio dubbing a catalogue |
| Document languages | A non-English production crew | A single-language studio |
| Voice languages | Anyone releasing in several markets | A subtitle-only release |
| All three | Nobody in particular | Nobody in particular |
Inclusion rule. The three columns, with the reader each one is decisive for. The last row is the combined score, which is decisive for nobody. Order. Cheapest column first.
1The columns are filled by different companies
Among the vendors read for this atlas, the group with the longest interface declarations and the group with the longest voice lists overlap on very few companies. Half the field fills one and not the other.
So a combined column would not smooth out noise; it would put the wrong vendors at the top for either question. A reader with a specific question would get a worse answer from the combined figure than from either column alone.
2Any weighting encodes a buyer nobody is
Weight the interface heavily and the score describes a company shopping for software to work in. Weight voices heavily and it describes a distributor. Weight them equally and it describes nobody at all.
The weighting would also be invisible in the output. A reader seeing a single number has no way to know which buyer it was built for, which is worse than a table that makes them choose a column.
3The units do not even match
One column holds counts of declared pages. One holds a rule. One holds sets, counts and lower bounds in at least three different units. Adding them would produce a figure in no unit at all.
That is a formal objection rather than a taste one. Before a score can be argued about, the quantities have to be commensurable, and these are not.
4What replaces it
Three columns, most cells empty, no ranking, and a reader who has to decide which question they are asking. That is less convenient than a ranked list and it is the only arrangement that does not mislead.
A reader arriving from a ranked comparison will find it frustrating before they find it useful. Saying so up front is better than pretending the table answers a question it does not.
An explainer about method and structure, not a count. No vendor number appears here that is not also in the register. The sourced material is on the register. Related: Check one yourself, Regions nobody addressed.