The script¶

This is the grammar glyph implements: version 8 of the grid script, where a text is a knowledge graph and spacing is structure. In the story it is our way of representing their language, not their language itself: they speak by resonance, in air, and the grid is how the translating machine writes that speech down. Nobody has seen them use any script.

The five rules¶

1. Spacing is structure. Every mark is a whole 3×3 part. The gaps between parts say how they group: 1 empty cell between the parts of a word, 2 between the words of a triplet, 4 between triplets. Lines are 3 empty rows apart.
2. Words are pairs. A word is a kind narrowed by a which, KIND.WHICH. A one-part word is just its kind, 3 cells wide. Only numbers have more parts.
3. Every triplet is one edge. A text is a run of triplets, (node, relation, node), read left to right. The same word in two triplets is the same node, and a lone ONE in a node slot means that: the triplet before.
4. Stacking means another voice. Only a new party adds a band to a triplet: a 3-row strip under the band above, one empty row between. Each party has one symbol.
5. A part answers the part above it. Read every part in a lower band against the nearest drawn part above it, in the same position of the same slot:

Two triplets on one line, "We saw your world. We will say that.":

+++..+.+......+++.+++....+++..+...........
+.+...+..+....+++.+......+.+..++....+...+.
+++..+.+......+++.+++....+++..+++.........

SELF, 2 cells, LIGHT.BEFORE (1 cell inside), 2 cells, BODY.OTHER; then 4 cells and the next triplet.

Parts¶

SELF      OTHER     STAR      LIGHT
###       ###       .#.       #.#
#.#       #..       ###       .#.
###       ###       .#.       #.#

BODY      AIR       WATER     TIME
###       #.#       ###       #.#
###       ...       ...       #.#
###       #.#       ###       #.#

VOICE     PATH      NOT       ONE
#..       ...       #..       ...
##.       ###       .#.       .#.
###       ...       ..#       ...

BEFORE    AFTER     OPEN      COUNT
...       ...       .#.       ...
#..       ..#       #.#       ...
...       ...       .#.       ###

Relations¶

A relation is one part (the action) plus a which that marks time or polarity:

Numbers¶

COUNT followed by base-512 digits, most significant first. Each digit is a part's worth of nine bits, 256 down to 1, left to right and top to bottom, and the digits are one cell apart like the parts of any word. COUNT alone is zero.

COUNT.137:

.....+.
......+
+++...+

glyph number N writes any number this way, and COUNT.2219 in a page is accepted and written as COUNT.4.171.

COUNT is the only kind whose positions are read as bits, and it never takes a part, so a number can never be drawn like a word. (Under v6, numbers were ONE plus bits, and 495 was drawn exactly like ONE.SELF, one of us.)

A 0 digit is a blank position. Inside a line that is fine, because the next gap shows where the number ends, but a number ending in 0 digits as the last word of a line can't be told from a shorter one. glyph render warns when that happens.

"That" and questions¶

• ONE alone in a node slot means that: the triplet before.
• OPEN in any position asks what goes here? The asker leaves an empty band for the answer, and whatever is drawn under OPEN replaces it. OPEN as a relation's which asks yes or no.
• How many? is OPEN.COUNT (what number?), answered with a COUNT word.

When the gaps fit two readings¶

ONE, BEFORE and AFTER are each one dot, a cell apart, so a dot next to a gap can sometimes be read either way. BODY.AFTER | OTHER (will have, you) and BODY | ONE.OTHER (have, one of you) are the same drawing. This is kept on purpose (L, 2026-10-08): like any real decipherment, the machine reads such a line the way that makes most sense against its lexicon, a relation with a tense marker and words it already knows, and it can be wrong.

How the machine reads a page¶

1. Split the drawing into lines at the three-row gaps, and count the bands: that is the number of voices.
2. Split each line into triplets, words and parts from the 1, 2 and 4 cell gaps. Anything that won't split into whole parts is a picture, not text. ONE, BEFORE and AFTER and number digits have empty edges of their own, so when more than one split fits, the machine keeps the one that reads best against the vocabulary.
3. Read the first band of each triplet as an edge. Same word, same node. A lone ONE points at the triplet before.
4. Apply rule 5 to every part in a lower band, and attach the result to that node or edge with the voice that said it.

glyph decode does steps 1 and 2; glyph graph does 3 and 4.