🗣️ Druckify

Measured from transcribed lectures. Drucker speaks in short clauses, addresses you, asks questions, and opens sentences with And.

Most frequent trigrams in the lectures

Distinctiveness chart (Evan’s chart — trigram log-ratio against a general-English baseline)

This is what makes the phrasing his rather than merely frequent — the Druckify dials target this list first.

How it works

1. Corpus mining — every number here comes from lecture transcripts (words, sentences, sentence length, “you” rate, question rate).
2. Deterministic rewriting — contractions, plain-English substitutions, sentence splitting, sentence openers (And so, And then, Now, You see), discourse markers, and targeted trigram rules.
3. Distinctiveness targeting — the over-used-trigram chart (log-ratio, Evan’s chart) is the priority list: will have to, one doesn’t, in this country, let me say, there is no, it isn’t, plus the don’t-family.
4. The engine never adds facts. The only opt-in exception is extra rhetorical questions.
5. Tone safety: profanity and insults are stripped from the input, so they can never appear in the output. If the text is hostile, only a language model can properly reframe it — use druckify-lm for that.

Honest limitations

• This is a style demonstration, not a simulation of the person; the output is not his writing.
• He spoke mid-century American English; forcing it onto modern jargon (KPI, roadmap, stakeholder) can read as pastiche — which is useful if you want the contrast.
• No language model is used: quality scales with how well your input tolerates spoken clauses.

Provenance & disclosure

• Rules derived from archive lecture transcripts (Claremont Colleges Digital Library collection), used as aggregate statistics.
• Over-used-trigram rankings: Evan’s chart (log-ratio against a general-English baseline).
• Generated text is AI-assisted pastiche; disclose it wherever the output is published.