Rhetorical Pattern Scanners#

These scanners detect rhetorical writing patterns that AI models overuse in German text. They only run when the language is set to deu.

A single occurrence of these patterns is normal in student writing. The scanners only trigger when patterns appear two or more times, and confidence increases with frequency.

Negative parallelism (negative-parallelism)#

Detects parallel negation constructions such as “Not only X, but also Y” and “Neither X nor Y”. The scanner matches the German equivalents of these patterns.

AI models overuse these constructions to appear rhetorically sophisticated.

Superficial analysis (superficial-analysis)#

Detects summary phrases that sound analytical but add no substance, such as “which illustrates X”, “this underscores”, “which shows the importance of X”. The scanner matches the German equivalents.

These phrases are filler — they restate the obvious rather than offering genuine analysis.

Promotional language (promotional-language)#

Detects advertisement-like vocabulary density, such as “groundbreaking”, “revolutionary”, “outstanding”, “unique”, “visionary”. The scanner matches the German equivalents.

Student essays rarely read like marketing copy. AI models, however, tend to inflate the significance of topics with promotional vocabulary.

Vague attribution (vague-attribution)#

Detects unsubstantiated attributions to unnamed authorities, such as “According to experts”, “Studies show”, “Scientists agree”. The scanner matches the German equivalents.

These phrases create an illusion of authority without citing actual sources. AI models use them frequently because they cannot reference real studies.

Hedging stack (hedging-stack)#

Detects German hedging adverbs (“möglicherweise”, “vermutlich”, “gewissermaßen”, “tendenziell”) that AI models stack to avoid committing to any claim. A single hedge is normal academic caution; several per 1,000 words signals evasive rather than careful writing.

Sentence-starter cascade (sentence-starter-cascade)#

Detects formal discourse markers that begin sentences (“Des Weiteren”, “Darüber hinaus”, “Folglich”, “Insgesamt”). Human writers use one or two of these per section; AI models chain them every second or third sentence.

Empty transition sentence (empty-transition-sentence)#

Detects German sentences that announce the next section without actually advancing the argument, such as “Im Folgenden wird X erläutert”, “Kommen wir nun zu Y”, “Ein weiterer Aspekt ist Z”. Professional writing usually integrates such transitions inline or omits them entirely.

Confidence scaling#

All rhetorical scanners use the same confidence model based on density per 1,000 words of text: