Notes

Lay meaning out as geometry — words as points, closeness as kinship — and the space between two concepts fills with intermediate words. Draw the straight line from one to the other, snap to the nearest word at each step, and you get a road. The catch: the endpoints don't choose it, the space does — relayout the same words and the same two anchors route through different go-betweens. With two vector spaces, a deviation tableau whose winner flips, and a nearest-neighbour walk. The musing: the meaning-making hides in the middle, not the ends.

Lay meaning out as geometry — words as points, closeness as kinship — and the space between two concepts fills with intermediate words. Draw the straight line from one to the other, snap to the nearest word at each step, and you get a road. The catch: the endpoints don't choose it, the space does — relayout the same words and the same two anchors route through different go-betweens. With two vector spaces, a deviation tableau whose winner flips, and a nearest-neighbour walk. The musing: the meaning-making hides in the middle, not the ends.

Brazilian Portuguese stress is nearly predictable from the right edge — penult by default, final when the last syllable is heavy. Lean too hard on that cue and a model erases the real minority it can't foresee: the proparoxytone (antepenultimate) tail. A 3-syllable window with weight sliders and a categoricity dial shows the tail collapsing toward zero as the cue hardens, with weight-and-edge tableaux that never reach the antepenult. The musing: certainty is a kind of forgetting.

Brazilian Portuguese stress is nearly predictable from the right edge — penult by default, final when the last syllable is heavy. Lean too hard on that cue and a model erases the real minority it can't foresee: the proparoxytone (antepenultimate) tail. A 3-syllable window with weight sliders and a categoricity dial shows the tail collapsing toward zero as the cue hardens, with weight-and-edge tableaux that never reach the antepenult. The musing: certainty is a kind of forgetting.

A vowel inventory isn't a list of segments but a field of contrasts: each vowel a territory carved out by its neighbours. Drag one across the space and only the nearby territories redraw — the global dispersion barely stirs. With a Voronoi mosaic, a live contrast graph, and a MinDist tableau where a vowel's best place flips with its neighbourhood. The musing: a phoneme is an address among other addresses, not an essence.

A vowel inventory isn't a list of segments but a field of contrasts: each vowel a territory carved out by its neighbours. Drag one across the space and only the nearby territories redraw — the global dispersion barely stirs. With a Voronoi mosaic, a live contrast graph, and a MinDist tableau where a vowel's best place flips with its neighbourhood. The musing: a phoneme is an address among other addresses, not an essence.

Surprisal is a clean idea — how unexpected a word was, −log p — and reading time seems to grow with it. But does it grow in a straight line or a bent one? Toggle the vertical ruler (milliseconds vs log) and turn the surprisal's exponent: the same cloud straightens or bends, and the 'best' exponent is different on each ruler. The musing: the nonlinearity you find may live in the measuring stick, not the mind.

Surprisal is a clean idea — how unexpected a word was, −log p — and reading time seems to grow with it. But does it grow in a straight line or a bent one? Toggle the vertical ruler (milliseconds vs log) and turn the surprisal's exponent: the same cloud straightens or bends, and the 'best' exponent is different on each ruler. The musing: the nonlinearity you find may live in the measuring stick, not the mind.

A 'shared phonetic space' isn't a place you can point to — it's where two categories stop being separable. Project each language's cloud onto the axis that best tells them apart, drag the centres together, and watch the overlap grow; then let bilinguals compress the gap and drift into the ambiguous middle. With a perception tableau that splits its bet exactly in the overlap. The musing: sharing a space is a matter of degree, not a line.

A 'shared phonetic space' isn't a place you can point to — it's where two categories stop being separable. Project each language's cloud onto the axis that best tells them apart, drag the centres together, and watch the overlap grow; then let bilinguals compress the gap and drift into the ambiguous middle. With a perception tableau that splits its bet exactly in the overlap. The musing: sharing a space is a matter of degree, not a line.