paper-copy-editing

name

paper-copy-editing

description

Academic English copy-editing for research papers. Use when the user asks to "copy-edit my paper", "improve the writing", "fix the English", "polish the language", "make it sound more academic", "tighten the prose", or "check grammar and style". Also use when the user is a non-native English speaker polishing a draft, when preparing a camera-ready version, or when reviewer feedback mentions "writing quality" or "language issues". This goes beyond grammar checkers by understanding academic writing conventions, hedging, and discipline-specific style.

Paper Copy-Editing

Improve the language quality of an academic paper while preserving the author's voice and technical meaning. Copy-editing is not rewriting — the ideas and structure stay the same; the prose becomes clearer, tighter, and more precise.

Principles

Preserve meaning exactly. A copy-edit must never change what the paper claims. If a sentence is ambiguous, flag it and ask the author rather than guessing.

Respect the author's voice. Academic writing has personal style. Don't flatten everything into the same generic tone. If the author uses "we" consistently, don't switch to passive voice. If they prefer short sentences, don't merge them.

Edit in passes. Don't try to fix everything at once. Each pass focuses on one category. This prevents edits from interfering with each other and makes the changes easier for the author to review.

Editing Passes

Pass 0: Section Title Capitalization

Check this first because it's a global, mechanical fix.

Paragraph-level titles (e.g., \paragraph{}, \subsubsection{}) should always use sentence case: capitalize only the first word and proper nouns.

Section and subsection titles (\section{}, \subsection{}) need a consistent convention — either title case ("Experimental Results on ImageNet") or sentence case ("Experimental results on ImageNet"). Before asking the user:

1. Scan the existing draft for a dominant pattern. If 80%+ of section titles already use one convention, adopt that convention and fix the outliers.
2. If there is no clear pattern (new draft, or mixed), ask the user: "Do you prefer sentence case or title case for section titles? Sentence case capitalizes only the first word and proper nouns. Title case capitalizes major words."
3. Apply the chosen convention consistently to all \section{} and \subsection{} titles throughout the paper.

Pass 1: Clarity and Precision

The most important pass. Every sentence should say exactly what it means.

Vague quantifiers — Replace with specifics when possible.

• "significantly outperforms" → "outperforms by 3.2 F1 points" (if the number exists)
• "much larger" → "4× larger" or "larger (2.1M vs 0.5M parameters)"
• "recently" → "since 2022" or name the specific work

Hedge appropriately — Academic writing requires careful hedging, but too much weakens the message.

• Over-hedged: "It might be possible that our method could potentially improve..."
• Under-hedged: "Our method solves the problem of..." (unless it provably does)
• Right: "Our method improves performance on X by Y" (factual, bounded claim)

Dangling modifiers and unclear antecedents:

• "Using our method, the accuracy improved." → "Using our method, we improved accuracy."
• "This shows that..." → What does "this" refer to? Make it explicit.

Nominalizations — Turning verbs into nouns makes prose heavier.

• "We performed an evaluation of..." → "We evaluated..."
• "The utilization of attention mechanisms..." → "Using attention mechanisms..."
• "He made a proposal to use Rust." → "He proposed to use Rust."

"There is/are" constructions — These bury the real subject. Rewrite with a concrete subject:

• "There are many approaches to this problem." → "Many approaches address this problem."
• "There is a trade-off between speed and accuracy." → "Speed and accuracy trade off."

Pass 2: Tense and Voice

Tense conventions in ML/CS papers:

• Present tense for established facts, general truths, and descriptions of your method as it exists: "Our model consists of three components."
• Past tense for specific experiments you ran and their results: "We trained on ImageNet and achieved 92.3% accuracy."
• Present tense for citing others' work in general: "Smith et al. [2023] propose a framework that handles..."
• Past tense for specific past findings of others: "Smith et al. [2023] reported a 5% improvement."

Check for tense consistency within each section. The most common error is mixing present and past when describing the same experiment.

Active vs. passive voice:

• Avoid passive voice unless strongly justifiable. Passive voice is ambiguous because it hides the subject:
  • Bad: "LLM was applied to fuzzing." (Who applied it? The authors or prior work?)
  • Good: "We applied LLM to fuzzing."
• Active is generally preferred: "We propose..." not "It is proposed that..."
• Passive is appropriate when the agent is truly irrelevant: "The model was trained for 100 epochs" (who trained it is obvious from context)
• Don't mechanically convert all passive to active — sometimes passive reads better

Pass 3: Conciseness

Academic papers have page limits. Every word should earn its place.

Redundant phrases:

• "in order to" → "to"
• "due to the fact that" → "because"
• "a total of 500 samples" → "500 samples"
• "it is worth noting that" → delete, then just state the thing
• "as can be seen in Figure 3" → "Figure 3 shows" or "(Figure 3)"

Filler words that contribute no meaning — delete on sight:

• "kind of", "sort of", "basically", "actually", "really", "quite"

Redundant adjectives and adverbs:

• "completely novel" → "novel" (novelty isn't partial)
• "very unique" → "unique"
• "currently existing" → "existing" or "current"
• "first introduce" → "introduce"

Throat-clearing — Sentences that delay the point:

• "It is important to note that X plays a crucial role." → "X is important because..."
• "In this paper, we present a method. The method is..." → "We present a method that..."

Pass 4: Terminology Consistency

A paper should use the same term for the same concept throughout.

• Build a term list as you read. If "feature extraction" appears in Section 2 and "feature encoding" in Section 4 for the same operation, pick one.
• Check that acronyms are defined on first use and used consistently after.
• Verify that math notation matches prose: if the loss is called $\mathcal{L}$ in equations, don't call it "L" or "the loss function $L$" in text.

Pass 5: Academic Register

The writing should sound like a published paper, not a blog post or a textbook.

Too informal:

• "We basically just fine-tuned BERT" → "We fine-tuned BERT"
• "This trick works really well" → "This approach is effective"
• "We tried a bunch of things" → "We explored several alternatives"

Too formal / archaic:

• "Herein we delineate..." → "We describe..."
• "The aforementioned method" → "This method" or name it
• "It behooves us to note" → just state the point

Avoid first-person overuse:

• Not every sentence needs "we". Alternate with passive and noun-subject constructions.
• "We use X. We then apply Y. We compute Z." → "We use X. Y is then applied to produce Z."

Pass 6: Grammar

Specific grammar rules that reviewers (and native speakers) notice immediately.

"Which" vs "that" — Use "that" in a restrictive clause (essential to meaning) and "which" in a nonrestrictive clause (parenthetical, set off by commas):

• Wrong: "Rust that is safe is popular." (implies some Rust is not safe)
• Right: "Rust, which is safe, is popular." (adds a fact about Rust)
• Right: "The method that uses attention outperforms the baseline." (distinguishes from other methods)

Comma splices and conjunctive adverbs — You cannot join two independent clauses with just a comma. Use a conjunction, a semicolon, or separate sentences:

• Wrong: "C is dangerous, Rust is safe."
• Right: "C is dangerous, but Rust is safe." (coordinating conjunction)
• Wrong: "C is dangerous, however Rust is safe."
• Right: "C is dangerous; however, Rust is safe." (semicolon + conjunctive adverb)

This applies to "however", "therefore", "moreover", "furthermore", "thus", "consequently", and "nevertheless" — they are adverbs, not conjunctions.

"Fewer" vs "less" — "Fewer" for countable nouns, "less" for uncountable:

• Wrong: "ten items or less" → Right: "ten items or fewer"
• Wrong: "fewer feedback" → Right: "less feedback"
• Wrong: "less parameters" → Right: "fewer parameters"

Articles (a, an, the) — A singular countable noun needs an article:

• Wrong: "We trained model on dataset."
• Right: "We trained a model on the dataset."
• "The" requires a unique referent (unique by fact or in context):
  • Right: "the first Transformer model" (unique by fact)
  • Right: "Our system has an encoder and a decoder. The encoder processes..." (unique in context)
  • Wrong: "Our system has two encoders. The encoder is..." (which one?)

"Compare with" vs "compare to" — "Compare with" for examining differences; "compare to" for asserting similarity:

• "Our results compare favorably with the baseline." (examining differences)
• "Some have compared Transformers to a Swiss Army knife." (asserting similarity)

How to Present Edits

Show changes at the paragraph level with clear before/after:

**Before:** We propose a method that is able to significantly improve the performance
of existing models by utilizing a novel attention mechanism.

**After:** We propose a method that improves existing models by 3.2 F1 points using
a cross-document attention mechanism.

**Changes:** replaced vague "significantly improve" with specific gain; replaced
"is able to" with direct verb; replaced "novel attention mechanism" with specific name;
removed "utilizing" (nominalization).

Group edits by section so the author can review one section at a time.

References

• HKU SEC writing advice
• Verb tense in scientific manuscripts (UNLV)
• Which vs that (Grammarly)

What This Skill Does NOT Do

• Does not restructure sections or rearrange arguments (use research-paper-writing)
• Does not check factual claims or experimental validity
• Does not fix LaTeX formatting (use latex-paper-formatting)
• Does not check references (use reference-checking)
• Does not catch typos or numbering errors (use paper-proofreading for the final pass)