the lexical liar
Words against themselves
A contronym is a word that seems to mean its own opposite. Sanction permits and punishes. Cleave splits and clings. Screen shows and hides. The naive move is to write sanction = permit and punish and call the word a contradiction. But that flattens a context-indexed rule into a single dictionary slot. Restore the index and the trouble disappears.
A contronym is not a lexical contradiction. It is a context-sensitive semantic operator whose polarity context fixes.
The formal shape
Treat each word token as a function from context to polarity. There exist contexts that send it to P and contexts that send it to not-P, but there is no single context that sends it to both at once.
Many contronyms are not arbitrary opposites either. They are two directions of one reversible operation. Dust is change the dust-relation; context decides whether you add it to a cake or remove it from a shelf.
The Grim move
In Patrick Grim's work, totalized systems of truth or knowledge generate paradox: the Liar, the Knower, Gödel-style incompleteness all appear when you try to complete a self-referential whole. Contronyms are a small linguistic cousin. The paradox is manufactured by the demand for a total, context-free dictionary entry. Indexed meaning stays coherent; only the collapsed representation breaks. Collapse mode in this playground performs the erasure on purpose so you can watch the contradiction switch on.
What this playground is and is not
The sentence analyzer is a bag-of-cues keyword scorer with no syntax and no world model, and the polarity numbers are hand-assigned, not measured from a corpus. It is a transparent toy meant to make one claim legible: that lexical meaning is often an operator over context rather than a fixed object, and that the apparent self-contradiction of a contronym is an artifact of flattening that operator. The calibration table shows exactly where the toy scorer already disagrees with the hand labels.
Model changelog
- modelled each contronym as a semantic operator M(w, c) = f_w(c): a fixed deep operation whose polarity is set by context.
- ported eight contronyms (sanction, screen, dust, cleave, oversight, fast, seed, clip) with two attractors, three context frames, and three labelled examples each.
- polarity = 0.5 frame + 0.38 manual pull + 0.55 keyword-scored text, clamped to [-100, 100]; basin selected at +/-8 with an underdetermined middle.
- collapse-context mode erases the index and forces both senses into one slot, exhibiting the Grim-style paradox of totalization.
- semantic-attractor map: a point moving between two basins, drawn together into one slot when context collapses.
- calibration: the keyword scorer reads ten example sentences and is checked against the hand-labelled polarity on a 0..100 pull-toward-second-sense axis.