What counts as a writing system

A system that writes a language directly, in marks a reader can see. That is the whole test, and it does most of the work of deciding what is here.

It admits things that are not alphabets: logographies, syllabaries, abjads, abugidas, and the featural design of Hangul, where the shape of a letter encodes how the mouth makes the sound. It admits systems nobody can read, because being undeciphered is a fact about us rather than about them. It admits scripts invented by a single named person, several of whom could not read any existing writing when they began, and two of whom said the letters came in a dream.

It excludes four things that are often listed alongside writing and are not writing. Braille, Morse and shorthand re-encode a script that already exists rather than write a language. Musical, mathematical and choreographic notation record something other than language. Scripts built for invented languages are left to the fiction they belong to. And a national orthography that is the Latin or Cyrillic alphabet with diacritics added is counted under its parent, or the list would run to hundreds of near-identical entries.

Knotted cords are the one deliberate inconsistency. The Inca khipu is here, under protest from its own definition, because whether it records language is exactly the open question and leaving it out would settle an argument that is not settled.

How the list was built

Four lists of the hundred most significant writing systems were drawn up independently, then merged rather than filtered. Fifty-four entries appeared on all four; thirty-three on three. The rest appeared once or twice, and those were kept if the reason for including them survived a second look.

That is why this section has 143 entries and not a round hundred. A cut to a hundred would have been arithmetic dressed up as judgement, and the entries it would have removed — the Yupik holy man's dreamt syllabary, the women's script of one Hunan county, the Osage alphabet designed in 2006 so a revived language would not be read through English eyes — are among the best things here.

Dates

Every entry carries a single year so the section can be sorted and searched. Many of those years are contested by decades and a few by centuries, and the entry says so where the argument is real. Treat the ordering as a shape, not a verdict. Where a date rests on one inscription, one radiocarbon result or one disputed stele, the entry names it.

The pictures

Two different things are shown, and they answer different questions. The sign charts are the actual letters, drawn from the Unicode standard rather than photographed, which is why they are sharp at any size. The photographs are of real objects — a tablet, a cliff face, a carved board — and are used under open licences, with the source named on each entry.

Thirty-two entries have no sign chart because they have no Unicode block. That is not an oversight: you cannot standardise signs whose values nobody agrees on, which is why most of the unread scripts are missing. The conspicuous exception is Maya, which is read fluently today and still has no block. Linear A is the reverse — encoded, and still unread.

Where it falls short

The scripts of South and Southeast Asia are the largest family here and the hardest to cut; a specialist would find real omissions among the regional abugidas of India and Indonesia. Coverage thins for scripts with small living communities and no institutional archive, because the surviving evidence thins there too. And the section is better on scripts that Europeans deciphered than on scripts that were never lost, which reflects where the written history is, not where the interest is.