High-Frequency AI Writing Terms in Chinese Theses - Reference List

This document lists high-frequency words that most readily reveal AI writing traces in Chinese degree theses and gives a recommended maximum occurrence count per document. The companion tone-thresholds.yaml is the authoritative configuration actually read by deai_check.py; this file is design documentation only.

How Thresholds Take Effect

• deai_check.py reads tone-thresholds.yaml at startup.
• For each term under term_thresholds:, exceeding its threshold in visible body text from one document (after parser extraction removes citations, formulas, and comments) triggers one [Script] LOW trace.
• Chinese is not tokenized; counts use substring matching.
• Change thresholds in the yaml file; code does not read this document.

Maintenance Cadence (This List Is a Snapshot, Not a Final State)

This list records current high-frequency AI writing terms, not permanent truths. As terms such as “赋能” and “彰显” are widely called out, deliberate authors filter them and their frequency declines, while new AI-preferred words continue to emerge. Review the list every six months against excess-vocabulary research and add/remove terms as needed; do not treat it as frozen.

• Last review: 2026-06
• Sources: Kobak et al., Sci. Adv. 2025; Geng & Trotta 2025

High-Frequency AI Chinese Connectors

These are not forbidden words; limited use is necessary. The threshold marks the point beyond which a reviewer is likely to perceive a formulaic pattern.

Term	Threshold	Note
首先	4	Marker of a templated argumentative opening
其次	4	Often stacked in a pair with “首先”
然而	5	Overused transition, often without genuine contrast
此外	4	Additive connector often used for stacking
因此	6	Inference/conclusion marker; a higher count is acceptable
另外	3	Duplicates the function of “此外”; consolidate them
进而	3	Progression word preferred by AI
而且	4	Duplicates the function of “并且”
显然	3	The more natural the context, the less this word is needed
通常	4	Vague frequency; replace when quantification is possible
一般	5	Weaker than “通常”
尤其	3	Emphasis word frequently used by AI

Content-Template Terms

Term	Threshold	Note
显著	5	Usually lacks support from a p-value or effect size
全面	3	A single study can rarely be “comprehensive”
深入	3	Marketing language
大量	3	Vague quantifier
众多	3	Vague quantifier
重要	5	Explain what makes it “important”
关键	5	Same as above
核心	4	A paper should not have too many “core” items
基本	4	Ambiguous
主要	5	Ambiguous
最为	3	Superlative rhetoric
极为	3	Superlative rhetoric
尤为	3	Superlative rhetoric

Repeated Paragraph Openings (Burstiness)

Trigger when three consecutive paragraphs begin with the same first four Chinese characters. Typical cases:

• “本节首先...” / “本节首先...” / “本节首先...”
• “为了进一步...” / “为了进一步...” / “为了进一步...”
• “在本章中...” / “在本章中...” / “在本章中...”

Fix: rewrite at least one paragraph with a different syntactic form, such as a fronted adverbial, contrastive transition, or question.

Paragraph-Opening Throat Clearing

Trigger when the first nonempty visible line of a paragraph matches one of these patterns:

• 综上所述 / 总而言之 / 总的来说 / 由此可见
• 值得指出的是 / 值得注意的是 / 需要指出的是 / 需要说明的是
• 不难发现 / 不难看出 / 众所周知 / 毋庸讳言
• Begins with “首先,” “其次,” “然而,” or “此外,”
• 一方面 / 另一方面

Each match records one [Script] LOW trace.

Punctuation Patterns

• When the document-wide count of “——” exceeds max_em_dashes_per_doc, record one aggregated trace at the first occurrence.
• Record one trace for each Chinese “！” or English “!” in body chapters between the abstract and conclusion. Formulas, code, and comments are stripped and excluded.

Content Outside This Table

• Syntax/grammar (covered by analyze_logic.py / editor self-check)
• Citation density (covered by check_references.py)
• Chapter structure (covered by check_format.py / map_structure.py)
• University naming rules (covered by templates/)
• Protected terms (covered by forbidden-terms.md)