| name | academic-writing |
| description | Use when drafting sections or editing for clarity. Provides paragraph architecture (topic→evidence→interpretation→bridge), per-section tense discipline, hedging ladders, banned patterns, signal-word banks, and language-specific conventions (Turkish register, English articles). Covers paraphrase discipline with source fidelity. |
Academic Writing: Core Conventions
Paragraph Architecture (Universal Model)
Every paragraph should follow: Claim → Evidence → Interpretation → Bridge.
1. Claim (Topic Sentence)
- Function: States the main idea or assertion of the paragraph.
- Form: Usually the first sentence; sometimes the second (after a transition).
- Examples:
- "Deep learning has revolutionized image classification tasks."
- "Prior studies disagree on the effect of X."
- "Method A has three major limitations."
2. Evidence (Supporting Material)
- Function: Grounds the claim with data, citations, or examples.
- Forms:
- Citation: "Research shows that [@author2024key]."
- Data: "Our experiments yielded a 92% accuracy rate."
- Example: "For instance, in the MNIST dataset, convolutional networks outperformed…"
- Logic: "If we accept that X leads to Y, and we observe X, then…"
3. Interpretation (Explanation)
- Function: Explains why the evidence matters; connects it to the claim.
- Form: "This indicates that…" or "What this suggests is…" or "In other words…"
- Example: "This 5% improvement is significant because it suggests that the proposed loss function better captures…"
4. Bridge (Transition)
- Function: Connects this paragraph to the next; shows flow of thought.
- Forms:
- Forward-looking: "Having established X, we now turn to Y."
- Comparative: "Similarly, prior research on Y…"
- Causal: "Because X holds true, the implications for Y are…"
Full Example
Claim: "Deep learning models trained on insufficient data often overfit, reducing generalization to unseen examples." Evidence: "Chen et al. (2023) found that a CNN trained on only 500 images achieved 87% accuracy on training data but 61% on held-out test data ([@chen2023name])." Interpretation: "This 26-point gap indicates that the model learned noise or dataset-specific patterns rather than generalizable features." Bridge: "To address this, regularization techniques like dropout and data augmentation have become standard practice in the field."
Per-Section Tense Discipline
Introduction (Present Tense)
- Describe the field, state general knowledge, pose questions.
- Example: "Neural networks have been the dominant approach for image classification." (present perfect = established fact)
- Guideline: Use present tense to speak about the state of the field. Use past tense only for specific prior findings: "Smith et al. (2020) found that…"
Methods (Past Tense)
- Describe what you did (specific to your work).
- Example: "We recruited 50 participants and conducted semi-structured interviews lasting 60 minutes."
- Guideline: Everything you did is in the past (even if writing in present session). The procedure is complete before writing.
Results (Past Tense)
- Report what you found.
- Example: "Results showed that the proposed method achieved 94.2% accuracy, compared to 87% for the baseline."
- Exception: If presenting data in a table, the table itself is timeless (present tense OK for table headers: "Accuracy (%)").
Discussion (Present Tense + Hedging)
- Interpret findings, compare to literature, theorize.
- Example: "These findings suggest that the loss function improves generalization (present), as evidenced by our results (past). Prior work indicates similar patterns [@citation2024]."
- Guideline: Hedging is appropriate here (you're interpreting, not stating absolute truth).
Conclusion (Present Tense)
- Summarize and look forward.
- Example: "This work demonstrates that X is possible. Future research should investigate whether Y applies to…"
Hedging Ladder (Claim Strength vs. Evidence Strength)
Match hedging language to your evidence strength:
| Strength | Verb/Phrase | Example |
|---|
| 5 (Certain) | demonstrate, show, prove, establish | "Results demonstrate that X causes Y." (use only with strong evidence: RCT, multiple studies) |
| 4 (Strong) | indicate, suggest, provide evidence that | "Findings indicate that X likely causes Y." (good evidence: consistent pattern, multiple datasets) |
| 3 (Moderate) | suggest, imply, hint at, appear to | "Results suggest that X may influence Y." (decent evidence: one solid study, needs replication) |
| 2 (Weak) | may suggest, might, could | "These findings may suggest that X could influence Y." (limited evidence: single study, small N) |
| 1 (Uncertain) | may possibly, might tend to | "One might speculate that X possibly influences Y." (very limited, avoid in academic writing) |
Examples
| Claim | Strength | Hedged Phrasing |
|---|
| Your hypothesis is supported | 4–5 | "Results demonstrate the hypothesis" (RCT, large sample, clear effect) |
| You found a pattern but it's not universal | 3–4 | "This suggests a pattern; further work is needed to confirm" |
| One prior study found something | 2–3 | "One study indicates that X; however, replication is needed" |
| You're speculating | 1–2 | Avoid or use: "Future work could explore whether…" |
Banned Patterns (Eliminate These)
1. Vague "This" (without noun anchor)
| Bad | Good |
|---|
| "The model was trained for 100 epochs. This improved accuracy." | "The model was trained for 100 epochs. This training regime improved accuracy." Or: "As a result, accuracy improved." |
| "We used data augmentation. This is important." | "We used data augmentation. This technique is important because…" |
Fix: Replace "This" with "This [noun]" or rewrite to avoid ambiguity.
2. Chained Nominalizations (noun → noun → noun pileup)
| Bad (Hard to parse) | Good (Clear structure) |
|---|
| "The implementation of the adoption of the framework requires understanding of the theory underlying…" | "To implement the framework, researchers must understand the underlying theory…" |
| "Improvement in model accuracy through regularization application shows regularization method effectiveness." | "Regularization improves model accuracy, demonstrating regularization's effectiveness." |
Fix: Use verbs instead of nouns where possible. "Implementation requires" not "Implementation of X requires."
3. Empty Intensifiers (add no meaning)
| Bad | Good |
|---|
| "This is a very important finding." | "This finding is significant because [explain why]." |
| "The results are quite remarkable." | "The results show [specific outcome that is remarkable]." |
| "The method is extremely novel." | "The method differs from prior work by [specific difference]." |
Fix: Show the importance via specificity, not adjectives. Let data/analysis speak.
4. Weak Passivity (hiding the agent)
| Bad | Good |
|---|
| "It was found that X improves Y." | "Our experiments show that X improves Y." |
| "It is believed that…" | "Researchers believe that…" or "Current theory suggests that…" |
| "The model was trained." (by whom? when?) | "We trained the model for 100 epochs using SGD with learning rate 0.001." |
Fix: Use active voice when possible. Passive is OK for methods if agent is clear from context.
5. Hedging Gone Wrong (too much or too little)
| Bad (Over-hedged) | Bad (Overconfident) | Good |
|---|
| "It might possibly be suggested that perhaps X could conceivably influence Y." | "X definitely causes Y." (without evidence) | "Evidence suggests that X influences Y." |
| "One could speculate that…" (avoid) | "Our single experiment proves X." | "Our results indicate X; further validation is needed." |
Fix: Match hedging to evidence. Use the ladder above.
Signal-Word Bank (Transitions & Connectivity)
Contrast / Opposition
- "However," "In contrast," "Conversely," "Yet," "Although," "Despite this," "On the other hand."
- Example: "Traditional methods are faster; however, they sacrifice accuracy."
Concession (accepting a point before arguing yours)
- "Granted," "Admittedly," "One could argue that," "To be fair," "It is true that…"
- Example: "Granted, prior work has shown X; our results suggest Y is also important."
Causality / Reason
- "Because," "Since," "As a result," "Therefore," "Consequently," "Thus," "This explains why."
- Example: "Because the dataset was imbalanced, we used stratified sampling."
Addition / Reinforcement
- "Furthermore," "Moreover," "In addition," "Also," "Similarly," "Likewise."
- Example: "The method is accurate. Moreover, it is computationally efficient."
Emphasis / Importance
- "Notably," "Significantly," "Importantly," "It is worth noting that," "Crucially."
- Example: "Notably, this is the first study to combine X and Y."
Temporal / Sequential
- "First," "Subsequently," "Meanwhile," "Finally," "Earlier," "Following this."
- Example: "First, we collected data. Subsequently, we analyzed it."
Summary / Conclusion
- "In summary," "In conclusion," "To summarize," "Overall," "In short," "To recap."
- Example: "In summary, our findings suggest three key implications."
Example / Illustration
- "For instance," "For example," "Specifically," "To illustrate," "Consider…"
- Example: "Our method applies to various domains. For instance, in medical imaging, it achieved 98% accuracy."
Paragraph Cohesion Checklist
Before submitting a draft:
Language-Specific Guidance
Turkish Academic Writing
See turkish-academic.md for:
- Passive voice norm (yapılmıştır, bulunmuştur)
- Term unity (one concept = one Turkish term throughout)
- TDK number/unit conventions
- Common expression errors
English Academic Writing
See english-academic.md for:
- Article usage (the #1 error of Turkish authors)
- Collocation bank for academic verbs
- Nominalization-overload repair patterns
Paraphrase Discipline
See paraphrase.md for:
- Source-faithful paraphrase technique
- "Change the structure too" rule
- Self-check questions
- When to quote verbatim (with page-number discipline)
Claim Strength: Never Say More Than Your Source
A citation can be real, resolvable and correctly formatted while the sentence
built on it still misrepresents the cited work. /verify-citations answers
"does this reference exist?"; it cannot answer "does it say that?".
The ordered scale (rungs 0–5, TR + EN), the one-directional rule (writing
weaker than your source is always safe; writing stronger is the defect), the
attributive-participle trap ("reduced capacity" is an outcome, not a claim), and
the evidence→rung table live in claim-strength.md.
Deterministic half: python scripts/claim_check.py compares each cited
sentence's rung against the evidence cards for the sources it cites, and flags
empirical assertions carrying neither a citation nor an own-work marker.
WARN-only — a rung difference is often legitimate; the check exists so you make
that call deliberately.
Quick Reference: Tense Table
| Section | Tense | Example |
|---|
| Introduction | Present (general knowledge); past (prior studies) | "Research shows that X is important. Smith et al. found that…" |
| Methods | Past (what we did) | "We recruited 50 participants. Data were analyzed using…" |
| Results | Past (what we found) | "Results showed a 5% improvement. The method achieved 92% accuracy." |
| Discussion | Present (interpretation, theory); past (our results) | "Our findings suggest that…. We found that…" |
| Conclusion | Present (broader implications, general statements) | "This work demonstrates that X is possible. Future work should investigate Y." |
One-Sentence Summaries Per Reference File
- turkish-academic.md: Use passive voice as the norm, maintain term unity (EN term in parentheses on first mention), apply TDK conventions for numbers/units.
- english-academic.md: Master article usage (a/an vs. the), use strong academic collocations (not "have an issue" but "face a challenge"), avoid nominalization overload.
- paraphrase.md: Paraphrase with structure change, cite the source, check that meaning is preserved without copying phrasing, quote verbatim only for key passages (with page numbers).
- claim-strength.md: Keep every claim on the rung its source actually supports; weakening is safe, strengthening is the defect; watch for adjectives silently promoted into verbs.
Practice Exercises
Exercise 1: Fix the vague "this"
"The model achieved 95% accuracy. This was a major breakthrough."
Answer: "The model achieved 95% accuracy. This performance was a major breakthrough (an 8-point improvement over prior work)."
Exercise 2: Fix the nominalizations
"The implementation of the algorithm through the use of data augmentation resulted in accuracy improvement."
Answer: "Implementing the algorithm with data augmentation improved accuracy."
Exercise 3: Match hedging to evidence
Evidence: One study with n=20 found X. Your phrasing: "X is proven."
Answer: "Preliminary evidence suggests that X may occur; larger studies are needed to confirm."
Exercise 4: Add signal words
"The method is fast. It is accurate. It is cheap. These are advantages."
Answer: "The method is fast. Moreover, it is accurate. Furthermore, it is cheap. Together, these characteristics make it attractive for deployment."
References & Resources
- See reference files for language-specific guidance:
- APA 7 Style Guide: aspastyle.apa.org
- Merriam-Webster for academic collocations and usage.
