German Language Scanners#

These scanners detect patterns specific to German text and only run when the language is set to deu (the default).

Umlaut digraphs (german)#

AI models sometimes replace German umlauts (ä, ö, ü) with ASCII digraphs (ae, oe, ue). This scanner detects such substitutions.

To avoid false positives, Quadrant IntegrityLens uses an allowlist of approximately 4,500 known German words that contain umlauts. A digraph is only flagged if the word (or a compound suffix) has a known umlaut form. This means English loanwords like “poem” or legitimate German words like “Abenteuer” (adventure) are not flagged.

Examples of flagged words:

• “ueber” → should be “über” (over/about)
• “Aerger” → should be “Ärger” (annoyance)
• “hoeren” → should be “hören” (to hear)

AI vocabulary (ai-vocabulary)#

This scanner measures the density of vocabulary and phrases that AI models overuse in German text. The patterns are organised in four tiers:

Tier A — Sentence-initial connectors#

Transitional phrases that AI uses excessively to link sentences, such as “Furthermore”, “Moreover”, “In this context”. The scanner matches the German equivalents of these phrases.

Tier B — Significance inflation#

Phrases that inflate the importance of ordinary observations, such as “plays a central role”, “is of great importance”, “makes an important contribution”. The scanner matches the German equivalents.

Tier C — Formulaic summaries#

Template-like summary phrases that add no content, such as “In summary, it can be said”, “It is important to note”, “In conclusion, it can be stated”. The scanner matches the German equivalents.

Tier D — Overused vocabulary#

Individual words that AI models favour disproportionately, such as “groundbreaking”, “multifaceted”, “interplay”, “holistic”, “decisive”. The scanner matches the German equivalents.

How confidence is determined#

The scanner counts AI vocabulary occurrences per ~1,000 words of text:

The scanner handles Swiss spelling variants (ss instead of ß) and ASCII digraph fallbacks automatically.

English-style heading capitalisation (heading-capitalisation-english-style)#

Detects Markdown headings that follow English Title Case rules — every content word capitalised — rather than proper German sentence case, where only the first word and nouns are capitalised. AI models trained on English-dominated corpora frequently carry this habit over into German output.

Flagged:

• Die Bedeutung Der Nachhaltigkeit Im Unternehmen
• Ein Überblick Über Die Methodik

Not flagged (correct German):

• Die Bedeutung der Nachhaltigkeit im Unternehmen
• Ein Überblick über die Methodik

The scanner only inspects headings with three or more words. Confidence scales with the raw number of violating headings: one is Low, two or three are Medium, four or more are High.