Computational Style Diagnostics: Passive Voice POS Tagging & Lexical Density (Ure / TTR)

Readability formulas only measure word and sentence length—they cannot detect passive voice, weak `-ly` adverb clutter, or vocabulary repetition. Conversely, naive regex style checkers flag `'She was tired'`, `'The shop is closed'`, and `'friendly advice'` as errors because they look only at `-ed` and `-ly` suffixes! CheckReadability runs an in-browser Compromise Part-of-Speech (POS) disambiguation pass alongside curated exception lexicons and Ure's Lexical Density equation to separate true grammatical bloat from clean idiomatic English.
Key Compliance & Engineering Takeaways
- Verbal Passive vs. Adjectival State: In 'The window was broken by vandals', 'broken' is a dynamic eventive passive; in 'He was tired after dinner' or 'Our team is excited', the `-ed` word is a predicate adjective describing a state—never passive voice!
- The `-ly` Suffix Fallacy: Over 60 common English nouns and adjectives end in `-ly` ('family', 'supply', 'early', 'only', 'friendly', 'costly', 'timely', 'scholarly', 'elderly', 'orderly')—POS tagging prevents false adverb highlights.
- Why Lexical Density Beats Raw TTR on Long Articles: Under Heaps' Law, Type-Token Ratio (`Unique / Total`) inevitably drops from ~75% at 100 words to ~32% at 3,000 words because function words ('the', 'of', 'to') repeat. Ure's Lexical Density (`Content / Total`) remains invariant to document length!
- Information Packing Threshold: When Ure's Lexical Density climbs past `60%`, nouns are stacked onto nouns ('enterprise cloud security posture remediation configuration') without prepositions or verbs—signaling immediate cognitive overload.
Mathematical & Algorithmic Specification
Exact peer-reviewed equations & linguistic rules executed locally inside CheckReadability's browser engine.
Ure's Lexical Density (Ure, 1971)
Verified SpecLD% = ( ContentWords / TotalWords ) × 100Variables: ContentWords = lexical nouns, main verbs, adjectives & manner adverbs (excluding closed-class grammatical stopwords)
Psychometric & Engine Note: Spoken English typically sits at 35–42% lexical density; clear web/editorial prose sits at 48–56%; dense academic/legal abstracts exceed 60–68%.
Type-Token Ratio — TTR (Vocabulary Diversity)
Verified SpecTTR% = ( UniqueWordTypes / TotalWordTokens ) × 100Variables: UniqueWordTypes = distinct normalized lowercase word forms (`|V| / N`)
Psychometric & Engine Note: Naturally decreases with passage length (`N`) under Heaps' Law (`|V| ≈ K × N^β`), so compare TTR across similar-length drafts.
POS-Disambiguated Verbal Passive Detector
Verified SpecMatch = Aux[be/get] + Adv? + Participle[regular -ed/-en ∨ irregular] ∧ ¬(StativeAdjective ∧ ¬byAgent)Variables: Filters copular adjectival states ('was tired', 'are closed', 'is interested') unless followed by an explicit 'by <agent>' PP
Psychometric & Engine Note: Eliminates the #1 complaint writers have with naive regex passive highlighters.
Linguistic Disambiguation: Regex-Only Checkers vs. CheckReadability POS Engine
| English Construction | CheckReadability (POS + Lexicon) | Naive Regex Style Checkers | Grammatical Explanation (CGEL) |
|---|---|---|---|
| 'The clinical study was approved by the board.' | Flagged as Passive Voice | Flagged as Passive Voice | True eventive verbal passive with 'by' agent phrase |
| 'Patients were tired and worried before surgery.' | Clean (Ignored — State Adj.) | False Positive ('were tired') | Copular verb + participial adjective describing emotional/physical state |
| 'The clinic is closed on Sundays.' vs. 'was closed by inspectors' | Context-aware ('by' check) | Flags both blindly | Distinguishes stative property from dynamic agentive action |
| 'She gave friendly, timely advice to the family.' | 0 Adverbs Flagged (Pass) | 3 False Adverbs ('-ly') | Adjectives ('friendly', 'timely') & noun ('family') excluded via POS |
Worked Linguistic & Formula Sensitivity Examples
Measured directly using CheckReadability's 12-formula consensus & POS engine.
Case Study: Unpacking >65% Lexical Density 'Noun Stacks' into Readable Prose
“Automated Kubernetes cluster node resource utilization anomaly detection threshold configuration documentation.”
“This guide shows how to set alert thresholds that spot unusual CPU and memory spikes across your Kubernetes nodes.”
Why this passes: Adding finite verbs ('shows', 'set', 'spot') and prepositions ('across') drops lexical density into the natural 48–55% sweet spot so readers parse relationships on the first pass.
1. Adjectival vs. Verbal Passives in English Grammar
In Huddleston and Pullum's Cambridge Grammar of the English Language (CGEL), English `be + past participle` sequences divide into two distinct syntactic categories: verbal passives and complex-intransitive adjectival passives.
True verbal passives correspond to an active transitive clause with an implicit or explicit agent ('The report was written [by Sarah]' → 'Sarah wrote the report'). Adjectival passives describe a static condition ('She was delighted', 'The API is deprecated', 'We are committed to privacy') and accept degree modifiers like 'very' ('very tired', whereas *'very written' is ungrammatical). Filtering stative participial adjectives unless followed by an explicit `by`-agent eliminates annoying false alarms.
2. Jean Ure & M. A. K. Halliday's Lexical Density Benchmarks
Linguist Jean Ure (1971) and systemic-functional grammarian M. A. K. Halliday (1985) demonstrated that the single sharpest structural difference between spoken conversation and dense written academic prose isn't sentence length—it is Lexical Density.
Spoken dialogue conveys ideas through dynamic verbs and clauses ('When you heat the gas up, it expands quickly' — 44% lexical density). Academic and bureaucratic prose nominalizes those verbs into dense noun phrases ('Rapid thermal gas expansion occurs upon heating' — 71% lexical density). Checking `/lexical-density-checker` alongside readability grade levels pinpoints 'noun-stuffed' prose that even short-sentence tricks cannot hide.
- 35% – 44% Lexical Density: Conversational transcripts, podcasts, speeches, and fiction dialogue.
- 46% – 55% Lexical Density: Ideal sweet spot for clear journalism, SaaS copy, patient handouts, and documentation.
- 58% – 70%+ Lexical Density: Heavy noun stacking, nominalizations, or missing function words—unpack noun chains into verbs.
Run These Offline Readability & Compliance Tools Now
100% browser-local execution with built-in sample documents — 0 bytes uploaded
Vocabulary Diversity (TTR), Lexical Density & Keyword Repeater
Measures Type-Token Ratio (TTR), content-to-function word ratio & overused 2-word phrases.
Adverb (-ly), Hedge & Weasel Word Highlighter
Spots -ly adverbs, weak intensifiers ('very', 'really') & hedge words diluting your prose.
Live Readability Studio
Live 12-formula consensus + inline highlights for hard sentences, passive voice & jargon.
Frequently Asked Questions
Is passive voice always bad?
No! Keeping passive voice at or below 10% of sentences is the editorial gold standard because passive voice is genuinely useful when the actor is unknown ('My bike was stolen'), irrelevant ('The samples were centrifuged at 3,000 RPM'), or when maintaining topic continuity at the start of a sentence.
Why did my Type-Token Ratio (TTR) drop when my article got longer?
Because English has a finite set of ~150 grammatical function words ('the', 'and', 'to', 'in'), longer texts inevitably repeat function words (Heaps' Law). Use Ure's Lexical Density % alongside Repeated Bigrams in `/lexical-density-checker` for length-independent diagnostics.
Does CheckReadability's POS tagger upload my text anywhere?
Never—Compromise (~75 KB) is bundled directly inside client-side JavaScript and executes in under 3 milliseconds in local browser RAM.
More Formula & Linguistic Engineering Specifications

SMOG vs. Flesch-Kincaid: Why 100% vs. 75% Comprehension Shifts Scores by ~2 Grades
Mathematical and psychometric breakdown of why Harry McLaughlin's SMOG Index consistently reports 1.5 to 2.5 grade levels higher than Flesch-Kincaid Grade Level on identical text.

Why Readability Calculators Disagree: Silent-E, Affixes & Syllable Counting Algorithms
Deep dive into English orthographic-to-phonemic syllable counting: how silent '-e', '-ed'/'-es' inflections, vowel hiatus ('create', 'chaos'), syllabic '-le', and exception dictionaries govern Flesch, Fog, and SMOG accuracy.

Character Formulas (ARI, Coleman-Liau, LIX/RIX) vs. Syllable Formulas Benchmarked
Mathematical and linguistic comparison of character-length formulas (ARI, Coleman-Liau, LIX, RIX) against syllable-based formulas (Flesch-Kincaid, SMOG, Gunning Fog).

Scoring UI Microcopy, Forms & Bullet Lists Without Full Sentences (FORCAST)
Mathematical specification of Caylor et al.'s (1973) FORCAST formula: why sentence-based formulas fail on UI strings, survey items, and slide decks, and how monosyllable ratios solve UX readability.

How CheckReadability Computes Consensus Grade Without Scale Distortion
Open mathematical specification of CheckReadability's 7-formula Consensus Grade Level, CEFR A1–C2 mapping, and why non-grade index scales (FRE, Dale-Chall, LIX) are isolated.

Fry Readability Graph vs. Lexicon Formulas (New Dale-Chall & Spache) Explained
Technical specification of Edward Fry's 100-word visual coordinate graph (including invalid zones), Chall & Dale's 3,000-word +3.6365 step function, and George Spache's unique sight-word formula.