The piece that blew the count open
With One Hand, Erase the Dagger had supplied only eighteen rows to the old lexicon. A wider sweep catches 101 strange single forms in that work alone, before counting phrase-level constructions such as MAL ARIA BELL WETHER, WORDS OF WOE ORB, A Jocular Englishing or the longer corruption chains. The source itself names “unadapted neologisms” and describes the method less as forced novelty than as embracing what “fails pleasingly”.
The piece practically announces the method itself. See the dedicated lexical-event page.
What turned up
- Neologism is only one class. The corpus also uses protologisms, nonce formations, corruption chains, rare inherited vocabulary, technical transplants, foreign-language fragments and semantic refunctioning.
- The hyphen is a morphology engine. 483 distinct forms contain two or more hyphens or comparable chained structure. Some are ordinary; many compress whole clauses into temporary lexical objects.
- Case is morphological. 119 forms use internal capitalization or case damage strongly enough to register as a separate event.
- Compounding is pervasive. 1,698 caught forms can be split into recognizable English roots. Some are standard compounds and therefore false positives; others are things such as thirstwench, fucktory, voidyelp, sweatgloss or remembranceless.
- Lexical density is not evenly distributed. A small group of works act like word-factories; others remain lexically plain while doing their strangeness through syntax, structure or image.
- Music continues the process. Current LP/DS and LTAP lyric witnesses add forms including audemartian, leechmilk, undeclareable, looky-looing and h-h-h-h-head, alongside proper-name and genre vocabulary.
Ways in
What the machine actually does
This is not a big-vocabulary trick. Roots get welded; suffixes counterfeit disciplines; spelling follows the mouth; borrowed words are dragged into alien jobs; useful errors stay useful; names buckle under rhythm; whole sentences get forced into the role of adjectives.
The lexical practice therefore sits at the intersection of Quantum Entendre, corruption chains, word formation, cross-modal thinking, musical language and useful wrongness / multilingualism.
Most of it happens once
Of the 5,111 filtered outlier forms, 4,558 occur in only one published work (89.2%), and 4,335 occur only once in the entire 212-work text snapshot (84.8%). Only 183 forms appear in three or more works. These figures still include names, slang, specialist vocabulary and other non-coinages, so they are not an originality score. They do show something more useful: most lexical strangeness is locally generated or locally recruited rather than maintained as a small stable house vocabulary.
This helps distinguish two processes. A few durable terms — mundus vult decipi, Xenochroma, Quantum Entendre, project names and recurring corrupted names — accumulate corpus-wide identity. Around them sits a much larger disposable vocabulary: words made, damaged, borrowed or welded for one immediate pressure and then abandoned. The lexicon therefore behaves less like a dictionary than like a field of local inventions with a small recurrent spine.
The pressure moves
The same conservative filter produces roughly 22–24 outlier forms per thousand word-like tokens in 2021–23, falls to about 18 in 2024 and 17 in 2025, then rises again to about 24 in the 2026 material. This should not be mistaken for a simple curve of “experimentalism”: later works can move their strangeness into syntax, structure, image, sound or interface instead. It does show that the lexical machine is episodic rather than steadily escalating.
| Year | Works | Tokens | Filtered outliers | Outliers / 1,000 tokens |
|---|---|---|---|---|
| 2021 | 15 | 20,181 | 453 | 22.4 |
| 2022 | 66 | 82,583 | 1,830 | 22.2 |
| 2023 | 27 | 37,315 | 899 | 24.1 |
| 2024 | 29 | 41,591 | 757 | 18.2 |
| 2025 | 30 | 42,716 | 729 | 17.1 |
| 2026 | 45 | 56,792 | 1,350 | 23.8 |
When a word eats the sentence
The scan also exposes an extreme edge that a normal word list tends to discard as noise. The corpus contains single token-like constructions hundreds of characters long, serial vocalizations, uninterrupted repetitions, and clause-sized hyphen chains. The important formal fact is not whether a tokenizer calls these “words.” It is that the writing repeatedly tests how much syntax can be forced inside a lexical shell. At that limit, morphology and sentence structure cease to be cleanly separable.