Paper Polisher: Academic Paper Refinement

The Specific Problem

In academic writing, papers often face two critical issues before submission: high plagiarism rates and noticeable AI-generated traces. Plagiarism detection systems like CNKI or Turnitin flagging duplicated content can lead to rejection or allegations of misconduct. Meanwhile, texts produced by AI writing tools (e.g., large language models) frequently exhibit patterned features such as exaggerated phrasing, vocabulary overuse, and formulaic structures, which reduce the paper's naturalness and credibility. Manual editing is time-consuming and prone to oversight, necessitating a systematic tool for language-level automation.

How the Skill Works

Paper Polisher operates through a three-stage pipeline, where the output of one stage feeds into the next, ensuring consistent processing. Core capabilities focus on linguistic rewriting and document formatting, without involving academic content creation.

Core Capabilities and Key Steps

• Stage 1: Plagiarism Reduction
  Rewrites text to lower detection rates while preserving academic meaning. Strategies, in priority order, include:
• Synonym/Near-synonym Replacement: For example, replacing “方法” with “技术” (method → technique) to maintain academic register.
• Sentence Restructuring: Breaking long sentences or merging short ones, such as converting subordinate clauses to pre-modifiers.
• Active-Passive Voice Conversion: Alternating between active and passive voice to add text diversity.
• Word Order Adjustment: Changing information sequence without altering logical relationships.

• Perspective Shifting: e.g., “X shows...” to “Research finds...”, or “According to X's analysis...”.
  Domain-specific terms, direct citation marks, formulas, data, and references are strictly retained.

• Stage 2: AI Trace Removal (Humanization)
  Eliminates AI-generated features to make language more natural and human-like. It checks and corrects the following patterns (based on Wikipedia: Signs of AI writing):

• Content Patterns: Exaggerated statements (e.g., stands as a testament to), empty popularity claims (e.g., has been cited in), superficial analysis phrases (-ing participle tails).
• Language Patterns: High-frequency AI words (e.g., additionally, crucial, delve), periphrastic verbs (e.g., using serves as instead of is), negation parallels, and tripartite structures.

• Style Patterns: Overuse of em dashes, AI collaboration artifacts (e.g., I hope this helps!), knowledge cutoff date statements.
  Injects "linguistic soul": ensures variation in sentence length and structure, allows expression of uncertainty, uses specific details over vague assertions, and prefers simple verbs like is or has over ornamental ones.

• Stage 3: Word Document Output
  Generates a structured refinement report using python-docx. The document includes a cover page, table of contents, original vs. revised comparison, modification summary, and references. Formatting specifications: A4 pages with 2.54cm margins, headers in bold Heiti font (18pt for level 1, etc.), body text in Song font 12pt with 1.5 line spacing, and tables in Light Grid Accent 1 style. Color coding is applied (red for original in plagiarism reduction, green for revised; orange for AI issues, blue for corrections).

Applicable Boundaries and Considerations

• Preserving original meaning is non-negotiable: Rewriting must not distort the text's intent, ensuring academic accuracy.
• Professional terms cannot be replaced: Domain-specific jargon must remain intact to maintain specialization.
• Does not write the paper: Only performs language-level plagiarism reduction and AI trace removal, without fabricating or modifying academic content.
• Does not remove citation marks: References must retain attribution to uphold academic integrity.
• Does not guarantee specific plagiarism rates: Effectiveness depends on the original text's duplication type and detection system variance; users must verify results independently.
• Segmented processing advice: Recommended to process paragraphs of 300-500 words each; start with a small segment to test style alignment before scaling up. Users should review the polished output to confirm comprehension and agreement with each sentence.