Exploratory research atlas

Programming-Language Atlas

Normalized information distance, approximated with LZMA compression across 100 carefully sampled programming-language corpora.

100 corpora LZMA NCD 4 repository-disjoint folds 64 KiB per object

Interactive three-dimensional atlas

Drag to rotate. Scroll or pinch to zoom.
Your browser needs canvas support to display the interactive atlas.

Select a language to inspect its full-space density and nearest neighbours.

The default view is an ordinal display of the full normalized compression distance matrix. Screen-space whitespace is not direct evidence of an information gap.

Method

From NID to a computable map

Normalized information distance is a universal theoretical distance between objects. It is uncomputable, so this atlas uses normalized compression distance as a practical approximation.

NID

Kolmogorov complexity K measures the length of the shortest program describing an object. Up to logarithmic terms, the normalized distance is:

NID(x,y) = [K(xy) - min(K(x), K(y))] / max(K(x), K(y))

NCD

Because K cannot be computed, compressor length C is substituted. LZMA is order-sensitive, so this experiment conservatively takes the larger concatenation length:

NCD_C(x,y) = [max(C(xy), C(yx)) - min(C(x), C(y))] / max(C(x), C(y))

Corpus

Each language contributes four repository-disjoint, source-only objects of exactly 64 KiB. The displayed matrix averages all cross-fold NCD comparisons, reducing dependence on any single repository draw.

Map

The full NCD matrix is converted to a three-dimensional ordinal embedding. The fit seeks to preserve the rank ordering of distances. Thin lines are the minimum spanning tree of the raw matrix, and point size is median raw NCD to the five nearest languages.

Read carefully

This display preserves much, but not all, of the full NCD ordering. NCD remains compressor-dependent, and screen-space whitespace is not direct evidence of an information gap.

Current finding

A strong signal, conditional geometry

LZMA NCD strongly distinguishes repeated samples of the same language from samples of different languages: within-versus-between AUC is 0.9946. Delete-one-fold tests preserve pairwise ranks at a median 0.987 and five-neighbour overlap at 0.843.

That supports an exploratory map, but not a claim of canonical coordinates. LZMA also missed one synthetic Markov-relation check, while the PPMd alternative badly violated self-identity. The global NCD signal is stronger than any exact local reading.