Feature: Tricolon Detector — Rule of Three Pattern Recognition

[craigtrim]

Overview

Detect tricolon (rule of three) patterns in text at multiple syntactic levels. This is one of the core AI stylistic tells identified in #69 — LLMs use tricolon as a default organizational skeleton with anomalously high frequency compared to human writing.

Background

Tricolon is the classical rhetorical device of three parallel syntactic units:

• Phrase level: "life, liberty, and the pursuit of happiness"
• Clause level: "It's not what you say, it's how you say it, or when you say it"
• Sentence level: "He came. He saw. He conquered."

The AI tell is not presence but density — where a human uses it once deliberately, an LLM reaches for it as the automatic shape of any three-part thought.

Detection Strategy

Surface Form (Oxford Tricolon)

Regex detection of comma-list-of-three patterns:

X, Y, and Z
X, Y, or Z

Clausal Form (Parallel Clause Tricolon)

Sentences containing exactly two commas that divide the sentence into three roughly equal-length segments. Pragmatic proxy for parallel clause detection without a full syntactic parser.

Sentential Form (Three-Sentence Sequences)

Consecutive sequences of short simple sentences of similar length within a paragraph — the "He came. He saw. He conquered." form.

Metrics

• tricolon_density: tricolon instances per 1000 words
• tricolon_type_distribution: breakdown across phrase / clause / sentence forms
• tricolon_paragraph_rate: proportion of paragraphs containing at least one tricolon

The signal is deviation from a human baseline corpus — anomalously high density is the AI fingerprint.

Dependencies

• Requires paragraph-level segmentation (see companion issue)
• Feeds into compound AI-tell co-occurrence scorer (see companion issue)

Related

• Feature: AI Stylistic Tell Detection — Tricolon, Stacked Sentences, and AI Vocabulary Fingerprints #69 — AI Stylistic Tell Detection (parent issue)

[craigtrim]

Paragraph Segmentation Dependency Is Resolved

The paragraph segmentation dependency noted in this issue is already provided by fast-sentence-segment. segment() returns list[list[str]] — paragraphs as outer lists, sentences as inner. See #71 for full analysis and implementation plan.

Once split_paragraphs() is added to _utils.py (a thin wrapper over the existing segment() call — no new dependency), the paragraph boundary constraint on sentential tricolon detection is lifted.

What This Unlocks Here

Sentential tricolon detection (three structurally parallel sentences in sequence) becomes fully tractable:

• Per-paragraph sentence list is available directly from split_paragraphs()
• Consecutive structure comparison within a paragraph window is straightforward

Blocking Order

1. split_paragraphs() added to _utils.py — Feature: Paragraph-Level Segmentation #71 (implementable immediately, closes Feature: Paragraph-Level Segmentation #71 and Feature: Paragraph-Level Segmentation #72)
2. This issue (Feature: Tricolon Detector — Rule of Three Pattern Recognition #70) — unblocked once Feature: Paragraph-Level Segmentation #71 is done

Related

• Feature: Paragraph-Level Segmentation #71 — adds split_paragraphs() to _utils.py (immediate prerequisite)
• Feature: Paragraph-Level Segmentation #72 — duplicate paragraph segmentation issue
• Feature: AI Stylistic Tell Detection — Tricolon, Stacked Sentences, and AI Vocabulary Fingerprints #69 — parent feature (AI Stylistic Tell Detection)
• Feature: AI-Tell Co-occurrence Scorer #75 — co-occurrence scorer (downstream consumer of this and other tells)
• craigtrim/fast-sentence-segment — source library providing native paragraph output
• NewlinesToPeriods silently corrupts text mid-word when hard-wrapped line breaks fall inside a word boundary fast-sentence-segment#42 — NewlinesToPeriods corruption on hard-wrapped text (affects paragraph boundary quality for plain-text sources like Project Gutenberg)