Academic Paper Polish

Academic Paper Polish

Help the user write and polish English for academic research papers, with a focus on CS/ML fields. The repository at the working directory contains curated vocabulary, phrase banks, and section templates drawn from real published papers.

Workflow

When the user provides text to polish or asks for writing help:

1. Identify the section — determine which part of the paper (Abstract, Introduction, Related Work, Method, Experiment, Conclusion, Rebuttal) or ask if unclear
2. Diagnose the issues — look for: vague vocabulary, weak transitions, informal tone, lack of precision, repetitive sentence starters, missing evidence claims
3. Apply improvements using vocabulary and phrase banks in the reference files
4. Provide 2–3 alternatives so the user can choose the best fit; explain the nuance between options

Section-Specific Guidance

Read references/section-phrases.md for ready-to-use sentence templates organized by paper section.

Abstract (3-part structure)

• Opening: describe the problem/gap existing methods face — "Existing approaches face a dilemma...", "Despite the rapid advancement of..."
• Middle: describe your method and key innovations — "In this work, we present...", "A key part of our approach is..."
• Closing: show results — "Extensive experiments demonstrate that our framework significantly outperforms SOTA methods on..."

Introduction (4-paragraph structure)

1. Background & motivation — "An explosively growing line of research...", "has enjoyed tremendous growth..."
2. Limitations of existing work — "The main drawback...", "prior attempts hamper...", "an inevitable problem..."
3. Your solution — "To address this, we propose...", "Motivated by this, we introduce..."
4. Contributions — "In summary, our contributions are:" followed by bullet points

Related Work

• Group by sub-category, not chronologically
• "The most related works can be divided into... The first focuses on... The second..."
• Handle concurrent work: "The concurrently developed X. The central distinction between these and ours is that..."
• "Our method builds upon this line of research"

Method

• Introduce the full framework first, then individual components
• "Our system consists of three major components: (1)... (2)... (3)..."
• For math notation: "where X denotes...", "we formulate this as..."
• "Specifically, our model first adapts... Then, we... Finally,..."

Experiment

• Standard structure: Dataset → Baselines → Metrics → Quantitative Results → Qualitative Results → Ablations
• Use w/ and w/o for ablation notation (with / without)
• Figure captions: bold the key phrase, e.g. "Qualitative comparison of our method against baselines."
• "We conduct extensive experiments to evaluate our model."

Conclusion

• Sentence 1: "We present [method] for [task]."
• Sentence 2: "The key idea underpinning the proposed method is to..."
• Sentence 3: "Extensive experiments demonstrate that our method achieves state-of-the-art performance."
• Sentence 4: Future impact — "We hope this work brings us one step closer to..."

Vocabulary & Phrases

Read references/vocabulary.md for curated vocabulary organized by use case:

• Positive evaluation (abstract/good concepts): unprecedented, phenomenal, intriguing, seminal, indispensable...
• Positive evaluation (concrete methods): ameliorate, synergistic, complementary, maturity...
• Negative evaluation (critiquing baselines): hamper, hinder, deficiency, deteriorate, inferior, vulnerable...
• Difficulty vocabulary: elusive, tackle, address, a remedy to...
• Transitional connectors: notably, in contrast, as a remedy, in tandem, meanwhile...
• Quantity expressions: a myriad of, plethora of, a wide range of...
• Useful phrases: off-the-shelf, on par with, in stark contrast to, take an orthogonal direction...
• Domain-specific terms: leverage, streamline, harness, empirically, agnostic, hallucinate (amodal)...

Abbreviations & Latin Terms

Latin terms (italicize): per se, de facto, vice versa, post hoc, ad-hoc

AI-Assisted Polishing

Read references/ai-polish.md for ready-to-paste prompts for ChatGPT/Claude polishing, including:

• The Nature (2024) recommended template
• System presets for consistent academic style
• Dimension-specific polishing: precision, conciseness, coherence, academic tone, formality

Key Writing Principles

• Academic tone: prefer "generate" over "produce", "analyze" over "look at", "demonstrate" over "show"
• Precision: avoid vague claims; add specifics ("improves by 3.2 PSNR", "reduces training time by 40%")
• Variety: avoid repeating the same sentence opener; rotate transitions (notably / in contrast / more specifically / furthermore)
• Evidence-backed claims: support assertions with "experiments demonstrate...", "as shown in Fig. X..."
• Active vs. passive: use active voice for your contributions ("We propose..."), passive for results ("The model is trained on...")
• Avoid these weak phrases: "it can be seen that", "in order to", "due to the fact that" → replace with direct alternatives