Academic Paper Polisher — Academic Paper English Polishing Knowledge Base
Help users review and polish English papers paragraph by paragraph according to international academic journal standards. Covers grammar correction, academic word choice optimization, voice conventions, logical cohesion enhancement, sentence pattern diversification, and other dimensions, outputting revision suggestions and polished text.
Quick Start
Users only need to provide:
- Text to be polished: one or more paragraphs of English paper content
- Paper type(optional): journal paper / conference paper / degree thesis / review
- Target journal/field(optional): such as Nature, IEEE, AAAI, medicine, computer science, etc.
- Polishing focus(optional): comprehensive polishing / grammar only / word choice only / logical cohesion only
Example:
"Help me polish this Introduction; the target journal is NeurIPS, and I hope the language will be more idiomatic and the logic more coherent."
I. Overview of Review Dimensions
The polishing work checks the following 5 dimensions item by item, scoring each dimension independently and giving revision suggestions:
| Dimension |
English label |
Review focus |
| Grammar |
Grammar |
Subject-verb agreement, tense, articles, prepositions, clause structure, punctuation |
| Word choice |
Word Choice |
Academic formality, precision, collocation, avoiding colloquialism |
| Voice |
Voice & Tense |
Active/passive voice selection, tense consistency |
| Logical cohesion |
Coherence & Cohesion |
Intra-paragraph/inter-paragraph transitions, argument chain, use of signal words |
| Sentence pattern |
Sentence Structure |
Sentence pattern diversity, combination of long and short sentences, balance of coordination and subordination |
II. Grammar Rules
2.1 List of High-Frequency Grammar Errors
| Error type |
Error example |
Correction |
Explanation |
| Subject-verb disagreement |
The results of the experiment shows... |
The results of the experiment show... |
The subject is results (plural) |
| Missing/misused article |
We propose method to solve... |
We propose a method to solve... |
A singular countable noun requires an article |
| Dangling modifier |
Using the proposed method, the accuracy was improved. |
Using the proposed method, we improved the accuracy. |
The logical subject of a participial phrase must agree with the subject of the main clause |
| Run-on sentence |
The model performs well , it achieves 95% accuracy. |
The model performs well**;** it achieves 95% accuracy. / The model performs well**. It** achieves 95% accuracy. |
A comma cannot connect two independent clauses |
| Incomplete comparison |
Our method is more efficient. |
Our method is more efficient than the baseline. |
A comparative requires a clear object of comparison |
| Broken parallel structure |
The system can detect, classify, and is able to segment... |
The system can detect, classify, and segment... |
Coordinated elements must maintain the same grammatical form |
| that/which confusion |
The model which we proposed... |
The model that we proposed... |
Use that for restrictive relative clauses |
| Numerals and nouns |
These phenomenon indicate... |
These phenomena indicate... |
Pay attention to irregular plurals |
2.2 Punctuation Conventions
| Rule |
Correct usage |
Common error |
| Serial comma (Oxford comma) |
A, B**,** and C |
A, B and C (academic writing recommends using the Oxford comma) |
| Dash |
We used three models — A, B, and C — for comparison. |
Em dash with spaces before and after, or em dash without spaces (depending on journal style) |
| Capitalize after colon |
Capitalize when a standalone sentence:The result is clear: The model outperforms... |
Lowercase when a non-standalone fragment |
| Quotation marks and periods |
American: period inside quotes. British: period outside quotes. |
Choose according to the target journal's region |
| Abbreviation periods |
e.g., i.e., et al., etc. |
Note the comma: e.g., / i.e., |
2.3 Tense Conventions (by Paper Section)
| Section |
Recommended Tense |
Example |
| Abstract |
Past tense (describing what was done) + present tense (describing conclusions) |
"We proposed a method... The results show that..." |
| Introduction |
Present tense (describing the current state/consensus) + past tense (describing prior work) |
"Deep learning has become... Smith et al. demonstrated that..." |
| Methods |
Past tense (describing the experimental procedure) |
"We trained the model on... The data were preprocessed..." |
| Results |
Past tense (describing experimental results) |
"The model achieved 95% accuracy. Table 2 shows..." |
| Discussion |
Present tense (explaining significance) + past tense (citing results) |
"This result suggests that... Our findings indicated that..." |
| Conclusion |
Past tense (summarizing the work) + present tense (stating contributions/significance) |
"We proposed and evaluated... This work contributes to..." |
III. Word Choice Optimization Rules (Word Choice)
3.1 Colloquial → Academic Substitution Table
| Colloquial Term |
Academic Substitution |
Context Notes |
| a lot of |
numerous / a substantial number of / considerable |
Choose according to the object being modified |
| get |
obtain / acquire / achieve / attain |
Choose according to collocation |
| show |
demonstrate / illustrate / indicate / reveal |
demonstrate emphasizes proof; indicate emphasizes suggestion |
| big / huge |
substantial / significant / considerable |
|
| thing |
factor / aspect / element / component |
|
| good |
effective / favorable / advantageous / robust |
|
| bad |
adverse / detrimental / suboptimal / inferior |
|
| use |
employ / utilize / leverage / adopt |
utilize is more formal than use; leverage emphasizes taking advantage of |
| about |
approximately / roughly / circa |
Use approximately for numerical descriptions |
| try |
attempt / endeavor |
|
| look at |
examine / investigate / analyze / explore |
|
| find out |
determine / ascertain / identify / discover |
|
| go up / go down |
increase / decrease / rise / decline |
|
| point out |
highlight / emphasize / underscore |
|
| deal with |
address / tackle / handle / mitigate |
|
| make sure |
ensure / verify / confirm |
|
| kind of / sort of |
somewhat / to some extent / partially |
|
| start / begin |
initiate / commence / undertake |
|
| end / finish |
conclude / terminate / complete |
|
| help |
facilitate / enable / assist / contribute to |
|
| need |
require / necessitate |
|
| can |
is capable of / is able to / has the potential to |
Avoid over-substitution; can is acceptable in academic writing |
3.2 Vague Expressions → Precise Expressions
| Vague Expression |
Precise Substitution |
Explanation |
| very good results |
statistically significant improvement / a 12% increase in accuracy |
Replace vague descriptions with specific data |
| some researchers |
Several studies (Chen et al., 2023; Li et al., 2024) |
Replace vague references with specific citations |
| recently |
In the past five years / Since 2020 |
Give a time range |
| a few |
three / a small number of (n=3) |
Specify the quantity |
| it is known that |
Prior work has established that (citation) |
Add citation support |
| this is important |
This is critical for / This has significant implications for |
Explain why it matters |
3.3 Streamlining Redundant Expressions
| Redundant Expression |
Streamlined Version |
| in order to |
to |
| due to the fact that |
because / since |
| at the present time |
currently / now |
| it is worth noting that |
Notably, / Note that |
| it should be pointed out that |
(state the content directly) |
| a total of 50 samples |
50 samples |
| the vast majority of |
most |
| in the event that |
if |
| has the ability to |
can |
| on a daily basis |
daily |
| in close proximity to |
near |
| take into consideration |
consider |
| is in agreement with |
agrees with |
| serves the function of |
functions as |
IV. Voice and Tense Conventions (Voice & Tense)
4.1 Active vs. Passive Voice Selection
| Scenario |
Recommended Voice |
Example |
| Describing the authors' actions |
Active (We) |
We trained the model using... |
| Describing general methods/known facts |
Passive |
The data were collected from... |
| Emphasizing the object of the action |
Passive |
The samples were analyzed using mass spectrometry. |
| Describing results |
Active preferred |
Our method achieves 95% accuracy. |
| Describing experimental equipment/materials |
Passive |
The solution was heated to 100°C. |
4.2 Common Voice Issues
| Issue |
Incorrect Example |
Correction |
| Excessive passive voice |
It was found by us that the results were improved by the method. |
We found that our method improved the results. |
| Inconsistent person |
The author proposes... We then evaluate... |
Use We or The authors consistently |
| Meaningless passive voice |
It can be seen that accuracy increases. |
Accuracy increases. / The results show that accuracy increases. |
4.3 Academic Person Conventions
| Person |
Usage Scenario |
Notes |
| We |
Describing the present authors' work (most common) |
Even for a single author, many journals accept "We" |
| The authors |
A more formal alternative |
Some journals prefer this usage |
| I |
Single-author dissertation |
Not accepted by some journals |
| One |
Generic/hypothetical statements |
More old-fashioned, less used in modern academic writing |
V. Rules for Logical Coherence & Cohesion
5.1 Within-Paragraph Cohesion Signal Words
| Logical Relationship |
Signal Words/Phrases |
Usage Example |
| Addition |
Furthermore, Moreover, Additionally, In addition |
Furthermore, our method generalizes well to unseen data. |
| Contrast |
However, In contrast, Conversely, On the other hand, Nevertheless |
However, this approach suffers from high computational cost. |
| Cause and Effect |
Therefore, Consequently, As a result, Hence, Thus |
Therefore, we adopt a two-stage training strategy. |
| Example |
For example, For instance, Specifically, In particular |
Specifically, we focus on the image classification task. |
| Emphasis |
Indeed, Notably, Importantly, It is worth noting that |
Notably, the improvement is consistent across all datasets. |
| Concession |
Although, Despite, Notwithstanding, While, Even though |
Although the model is simple, it achieves competitive results. |
| Summary |
In summary, To summarize, Overall, In conclusion |
Overall, the proposed method outperforms existing baselines. |
| Transition |
Yet, Still, Nonetheless, That said |
That said, there are several limitations to our approach. |
| Sequence |
First, Second, Finally, Subsequently, Then |
First, we preprocess the data. Subsequently, we train the model. |
| Condition |
If, Provided that, Given that, Assuming that |
Given that the dataset is imbalanced, we apply oversampling. |
5.2 Inter-Paragraph Transition Patterns
| Transition Pattern |
Description |
Example First Sentence |
| Hook |
The end of the previous paragraph introduces the topic of the next paragraph |
"This raises the question of how to efficiently scale the model." |
| Recap |
The beginning of the next paragraph reviews the conclusion of the previous paragraph |
"Having established the effectiveness of our approach, we now turn to..." |
| Contrast Bridge |
Point out the shortcomings of the previous paragraph's method and introduce this paragraph |
"While these methods achieve reasonable accuracy, they fail to address..." |
| Question Bridge |
Transition in the form of a question |
"How can we overcome this limitation? In this section, we propose..." |
| Topic Sentence |
The first sentence of each paragraph summarizes the core argument of that paragraph |
"The key advantage of our method is its ability to..." |
5.3 Common Logical Cohesion Problems
| Problem |
Description |
Correction Strategy |
| Leaping argumentation |
Jumping directly from A to C, lacking the transition through B |
Add the intermediate reasoning steps or insert a transition sentence |
| Overuse of signal words |
Every sentence begins with However / Moreover |
Reduce signal words and use sentence structure variation to express logic |
| Misuse of signal words |
Using Furthermore to indicate contrast |
Use However for contrast; use Furthermore for addition |
| Paragraph too long |
A paragraph exceeds 8-10 sentences |
Split into 2-3 paragraphs by argument |
| Paragraph too short |
A paragraph has only 1-2 sentences |
Merge into a related paragraph or expand the discussion |
| Unclear reference |
"This shows..." — what does this refer to? |
"This result shows..." / "This finding indicates..." |
VI. Sentence Structure Optimization Rules
6.1 Sentence Structure Diversification Strategies
| Strategy |
Original Sentence |
Optimized Version |
| Starting with a participial phrase |
We use attention mechanism, and we improve accuracy. |
Leveraging the attention mechanism, we improve accuracy. |
| Inversion for emphasis |
The improvement is particularly notable in low-resource settings. |
Particularly notable is the improvement in low-resource settings. |
| Parenthetical |
The model, which was proposed by Smith, achieves... |
The model, proposed by Smith (2023), achieves... |
| Nominalization |
We improved the model, and this led to... |
The improvement of the model led to... |
| Parallel structure |
The method is fast. It is also accurate. It is scalable too. |
The method is fast, accurate, and scalable. |
| Fronting of adverbial |
Accuracy improved significantly when we added data augmentation. |
With data augmentation, accuracy improved significantly. |
6.2 Sentence Structure Problems to Avoid
| Problem |
Example |
Correction |
| Overly long sentence (>40 words) |
We trained the model on the dataset which was collected from ... and preprocessed using ... and then evaluated on ... |
Split into 2-3 short sentences |
| Consecutive short sentences |
The accuracy is high. The model is fast. It uses less memory. |
Merge: The model achieves high accuracy with fast inference and low memory consumption. |
| Starting with There is/are |
There are many studies that focus on... |
Many studies focus on... |
| Too many It is...that emphatic sentences |
It is the attention mechanism that improves... |
The attention mechanism improves... |
| Noun stacking |
deep learning image classification model performance |
the performance of a deep learning model for image classification |
VII. Section-Specific Polishing Guide
7.1 Abstract
- Length: 150-300 words (follow the target journal's requirements)
- Structure: Background (1-2 sentences) → Problem/Motivation (1 sentence) → Methods (2-3 sentences) → Results (1-2 sentences) → Conclusion/Significance (1 sentence)
- Tense: Use past tense to describe the work, present tense to describe conclusions
- Prohibitions: Do not cite references, do not use abbreviations (spell out in full at first occurrence), do not include figure or table numbers
7.2 Introduction
- Structure(classic "funnel"): Broad background → Specific problem → Existing methods and their shortcomings → This paper's methods/contributions → Overview of the paper's structure
- Key points: Each paragraph must have a clear topic sentence; evaluate prior work objectively, without disparaging it
- Common sentence patterns：
- "In recent years, ... has attracted increasing attention."
- "Despite significant progress, ... remains a challenge."
- "To address this issue, we propose..."
- "The main contributions of this paper are as follows:"
7.3 Related Work
- Strategy: Group by topic (not chronological order), and within each group order chronologically
- Key points: Point out the relationship between each work and this paper; avoid mere listing—include critical commentary
- Transitions: Connect each group with transition sentences
- Common sentence patterns：
- "A closely related line of work focuses on..."
- "In contrast to these approaches, our method..."
- "Building upon the work of X, we extend..."
7.4 Methods
- Principle: Reproducibility — readers should be able to reproduce the experiment based on the description
- Structure: Problem formalization → Overall framework → Detailed description of each module → Training/optimization details
- Key points: Mathematical notation must be defined at first occurrence; describe steps in execution order
7.5 Results / Experiments
- Structure: Experimental setup → Main experimental results → Ablation experiments → Analysis/discussion
- Key points: Describe trends in text first, then reference tables/figures; avoid repeating numbers already in the tables and figures
- Common sentence patterns：
- "As shown in Table X, our method outperforms..."
- "We observe a consistent improvement of X% across..."
- "The ablation study reveals that..."
7.6 Discussion
- Content: Explain the significance of the results → Comparison with prior work → Limitations → Future directions
- Key points: Do not avoid limitations; the discussion should go beyond the results themselves and explore broader implications
7.7 Conclusion
- Length: Usually 1 paragraph, 150-250 words
- Structure: Summarize methods → Core findings → Significance/contributions → Future work
- Prohibitions: Do not introduce new information/new data; do not simply repeat the Abstract
VIII. Polishing Output Format
For each paragraph of text provided by the user, output in the following format:
### 原文（Original）
[用户提供的原始文本]
### 审查结果（Review）
| 维度 | 评级 | 主要问题 |
|------|------|---------|
| 语法 | ✓ 良好 / △ 需改进 / ✗ 问题较多 | 简述问题 |
| 用词 | ✓ / △ / ✗ | 简述问题 |
| 语态 | ✓ / △ / ✗ | 简述问题 |
| 逻辑衔接 | ✓ / △ / ✗ | 简述问题 |
| 句式 | ✓ / △ / ✗ | 简述问题 |
### 逐句修改（Detailed Changes）
1. **原句**: "..."
**修改**: "..."
**原因**: [具体说明修改理由，引用上述规则]
2. ...
### 润色后文本（Polished Version）
[完整的润色后段落]
IX. Domain-Specific Notes
Different disciplines have their own writing conventions, and polishing should respect domain characteristics:
| Field |
Characteristics |
Notes |
| Computer Science |
Active voice "We" is common |
More colloquial expressions are acceptable (e.g., "we run"); algorithm descriptions must be precise |
| Medicine/Biology |
Primarily passive voice |
"Patients were randomized..."; terminology must conform to MeSH standards |
| Physics |
Concise, formula-driven |
Mathematical derivations must be expressed rigorously; "one can show that..." is common |
| Social Sciences |
More qualifiers |
"may", "might", "suggests"; avoid overly absolute statements |
| Engineering |
Results-oriented |
Emphasize performance metrics and experimental validation |
X. Agent Behavior Guide
When a user submits text for polishing, follow this workflow:
- Confirm information: paper type, target journal/conference (if any), polishing focus
- Identify section: determine which section of the paper the text belongs to and apply the corresponding section conventions
- Five-dimension review: check item by item in the order grammar → word choice → voice → logical cohesion → sentence structure
- Annotate sentence by sentence: give a reason for each modification and cite the specific rule
- Output polished version: provide the complete polished text
- Summarize suggestions: summarize the main problem types and directions for improvement
Core principles：
- Preserve the author's original meaning and argumentation logic; do not change the technical content
- Modifications should be minimal — if changing one word suffices, do not change the whole sentence; if changing the whole sentence suffices, do not change the whole paragraph
- For places where it is uncertain whether an error exists, raise them as suggestions rather than modifying directly
- Use the author's terminology unless it is clearly misused
- Do not modify content the user has marked "please keep"
