Research paper refiner

Academic Paper Polisher — 学术论文英文润色知识库

帮助用户按照国际学术期刊标准，对英文论文进行逐段审查与润色。涵盖语法纠正、学术用词优化、语态规范、逻辑衔接强化、句式多样化等维度，输出修改建议与润色后文本。

Quick Start

用户只需提供：

1. 待润色文本：一段或多段英文论文内容
2. 论文类型（可选）：期刊论文 / 会议论文 / 学位论文 / 综述
3. 目标期刊/领域（可选）：如 Nature, IEEE, AAAI, 医学, 计算机等
4. 润色侧重（可选）：全面润色 / 仅语法 / 仅用词 / 仅逻辑衔接

示例：

"帮我润色这段 Introduction，目标期刊是 NeurIPS，希望语言更地道、逻辑更连贯。"

---

一、审查维度总览（Review Dimensions）

润色工作按以下 5 个维度逐项检查，每个维度独立评分并给出修改建议：

维度	英文标签	审查重点
语法	Grammar	主谓一致、时态、冠词、介词、从句结构、标点
用词	Word Choice	学术正式度、精确性、搭配、避免口语化
语态	Voice & Tense	主动/被动语态选择、时态一致性
逻辑衔接	Coherence & Cohesion	段内/段间过渡、论证链、信号词使用
句式	Sentence Structure	句式多样性、长短句搭配、并列与从属平衡

---

二、语法审查规则（Grammar Rules）

2.1 高频语法错误清单

错误类型	错误示例	修正	说明
主谓不一致	The results of the experiment shows...	The results of the experiment show...	主语是 results（复数）
冠词缺失/误用	We propose method to solve...	We propose a method to solve...	可数名词单数需加冠词
悬垂修饰语	Using the proposed method, the accuracy was improved.	Using the proposed method, we improved the accuracy.	分词短语的逻辑主语须与主句主语一致
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.	逗号不能连接两个独立分句
不完整比较	Our method is more efficient.	Our method is more efficient than the baseline.	比较级需明确比较对象
平行结构破坏	The system can detect, classify, and is able to segment...	The system can detect, classify, and segment...	并列成分须保持相同语法形式
that/which 混淆	The model which we proposed...	The model that we proposed...	限制性定语从句用 that
数词与名词	These phenomenon indicate...	These phenomena indicate...	注意不规则复数

2.2 标点规范

规则	正确用法	常见错误
连续逗号（Oxford comma）	A, B**,** and C	A, B and C（学术写作推荐用 Oxford comma）
破折号	We used three models — A, B, and C — for comparison.	前后加空格的 em dash，或无空格的 em dash（取决于期刊风格）
冒号后大写	独立句子时大写：The result is clear: The model outperforms...	非独立片段时小写
引号与句号	美式：period inside quotes. 英式：period outside quotes.	根据目标期刊地区选择
缩写句号	e.g., i.e., et al., etc.	注意逗号：e.g., / i.e.,

2.3 时态规范（按论文章节）

章节	推荐时态	示例
Abstract	过去时（描述做了什么）+ 现在时（描述结论）	"We proposed a method... The results show that..."
Introduction	现在时（描述现状/共识）+ 过去时（描述前人工作）	"Deep learning has become... Smith et al. demonstrated that..."
Methods	过去时（描述实验过程）	"We trained the model on... The data were preprocessed..."
Results	过去时（描述实验结果）	"The model achieved 95% accuracy. Table 2 shows..."
Discussion	现在时（解释意义）+ 过去时（引用结果）	"This result suggests that... Our findings indicated that..."
Conclusion	过去时（总结工作）+ 现在时（陈述贡献/意义）	"We proposed and evaluated... This work contributes to..."

---

三、用词优化规则（Word Choice）

3.1 口语化 → 学术化替换表

口语化用词	学术化替换	语境说明
a lot of	numerous / a substantial number of / considerable	根据修饰对象选择
get	obtain / acquire / achieve / attain	根据搭配选择
show	demonstrate / illustrate / indicate / reveal	demonstrate 强调证明；indicate 强调暗示
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 比 use 更正式；leverage 强调优势利用
about	approximately / roughly / circa	数值描述用 approximately
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	避免过度替换，can 在学术写作中可接受

3.2 模糊表达 → 精确表达

模糊表达	精确替代	说明
very good results	statistically significant improvement / a 12% increase in accuracy	用具体数据替代模糊形容
some researchers	Several studies (Chen et al., 2023; Li et al., 2024)	用具体引用替代模糊指代
recently	In the past five years / Since 2020	给出时间范围
a few	three / a small number of (n=3)	明确数量
it is known that	Prior work has established that (citation)	加引用支撑
this is important	This is critical for / This has significant implications for	说明为什么重要

3.3 冗余表达精简

冗余表达	精简版本
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	（直接陈述内容）
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

---

四、语态规范（Voice & Tense）

4.1 主动 vs. 被动语态选择

场景	推荐语态	示例
描述作者的操作	主动（We）	We trained the model using...
描述通用方法/已知事实	被动	The data were collected from...
强调动作对象	被动	The samples were analyzed using mass spectrometry.
描述结果	主动优先	Our method achieves 95% accuracy.
描述实验设备/材料	被动	The solution was heated to 100°C.

4.2 常见语态问题

问题	错误示例	修正
过度被动	It was found by us that the results were improved by the method.	We found that our method improved the results.
人称不一致	The author proposes... We then evaluate...	统一使用 We 或 The authors
无意义被动	It can be seen that accuracy increases.	Accuracy increases. / The results show that accuracy increases.

4.3 学术人称规范

人称	使用场景	注意事项
We	描述本文作者的工作（最常用）	即使单作者，许多期刊也接受 "We"
The authors	更正式的替代	部分期刊偏好此用法
I	单作者学位论文	部分期刊不接受
One	泛指/假设性陈述	较老式，现代学术写作较少用

---

五、逻辑衔接规则（Coherence & Cohesion）

5.1 段内衔接信号词

逻辑关系	信号词/短语	用法示例
补充	Furthermore, Moreover, Additionally, In addition	Furthermore, our method generalizes well to unseen data.
对比	However, In contrast, Conversely, On the other hand, Nevertheless	However, this approach suffers from high computational cost.
因果	Therefore, Consequently, As a result, Hence, Thus	Therefore, we adopt a two-stage training strategy.
举例	For example, For instance, Specifically, In particular	Specifically, we focus on the image classification task.
强调	Indeed, Notably, Importantly, It is worth noting that	Notably, the improvement is consistent across all datasets.
让步	Although, Despite, Notwithstanding, While, Even though	Although the model is simple, it achieves competitive results.
总结	In summary, To summarize, Overall, In conclusion	Overall, the proposed method outperforms existing baselines.
转折	Yet, Still, Nonetheless, That said	That said, there are several limitations to our approach.
顺序	First, Second, Finally, Subsequently, Then	First, we preprocess the data. Subsequently, we train the model.
条件	If, Provided that, Given that, Assuming that	Given that the dataset is imbalanced, we apply oversampling.

5.2 段间过渡模式

过渡模式	说明	示例首句
钩子句（Hook）	上段末尾引出下段话题	"This raises the question of how to efficiently scale the model."
回顾句（Recap）	下段开头回顾上段结论	"Having established the effectiveness of our approach, we now turn to..."
对比桥（Contrast Bridge）	指出上段方法的不足，引出本段	"While these methods achieve reasonable accuracy, they fail to address..."
问题桥（Question Bridge）	以问题形式过渡	"How can we overcome this limitation? In this section, we propose..."
主题句（Topic Sentence）	每段首句概括本段核心论点	"The key advantage of our method is its ability to..."

5.3 常见逻辑衔接问题

问题	说明	修正策略
跳跃式论证	从 A 直接跳到 C，缺少 B 的过渡	补充中间推理步骤或加过渡句
信号词滥用	每句都以 However / Moreover 开头	减少信号词，用句式变化体现逻辑
信号词误用	用 Furthermore 表转折	转折用 However；补充用 Furthermore
段落过长	一段超过 8-10 句	按论点拆分为 2-3 段
段落过短	一段仅 1-2 句	合并至相关段落或扩展论述
指代不清	"This shows..." — this 指代什么？	"This result shows..." / "This finding indicates..."

---

六、句式优化规则（Sentence Structure）

6.1 句式多样化策略

策略	原句	优化后
分词短语开头	We use attention mechanism, and we improve accuracy.	Leveraging the attention mechanism, we improve accuracy.
倒装强调	The improvement is particularly notable in low-resource settings.	Particularly notable is the improvement in low-resource settings.
插入语	The model, which was proposed by Smith, achieves...	The model, proposed by Smith (2023), achieves...
名词化	We improved the model, and this led to...	The improvement of the model led to...
平行结构	The method is fast. It is also accurate. It is scalable too.	The method is fast, accurate, and scalable.
状语前置	Accuracy improved significantly when we added data augmentation.	With data augmentation, accuracy improved significantly.

6.2 避免的句式问题

问题	示例	修正
过长句子（>40 词）	We trained the model on the dataset which was collected from ... and preprocessed using ... and then evaluated on ...	拆分为 2-3 个短句
连续短句	The accuracy is high. The model is fast. It uses less memory.	合并：The model achieves high accuracy with fast inference and low memory consumption.
There is/are 开头	There are many studies that focus on...	Many studies focus on...
It is...that 强调句过多	It is the attention mechanism that improves...	The attention mechanism improves...
名词堆砌	deep learning image classification model performance	the performance of a deep learning model for image classification

---

七、按章节润色指南（Section-Specific Guide）

7.1 Abstract

• 长度：150-300 词（遵循目标期刊要求）
• 结构：背景（1-2句） → 问题/动机（1句） → 方法（2-3句） → 结果（1-2句） → 结论/意义（1句）
• 时态：过去时描述工作，现在时描述结论
• 禁忌：不引用参考文献、不使用缩写（首次出现需全称）、不包含图表编号

7.2 Introduction

• 结构（经典"漏斗型"）：大背景 → 具体问题 → 现有方法及不足 → 本文方法/贡献 → 论文结构概述
• 关键：每段须有明确的主题句；引用前人工作时客观评述，不贬低
• 常用句式：
  • "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

• 策略：按主题分组（非按时间排列），每组内按时间顺序
• 关键：指出每项工作与本文的关系；避免纯罗列，要有评述
• 过渡：每组之间用过渡句连接
• 常用句式：
  • "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

• 原则：可复现性 — 读者应能根据描述复现实验
• 结构：问题形式化 → 整体框架 → 各模块详述 → 训练/优化细节
• 关键：数学符号首次出现时须定义；步骤按执行顺序描述

7.5 Results / Experiments

• 结构：实验设置 → 主实验结果 → 消融实验 → 分析/讨论
• 关键：先用文字描述趋势，再引用表格/图；避免重复图表中已有的数字
• 常用句式：
  • "As shown in Table X, our method outperforms..."
  • "We observe a consistent improvement of X% across..."
  • "The ablation study reveals that..."

7.6 Discussion

• 内容：解释结果的意义 → 与前人工作的对比 → 局限性 → 未来方向
• 关键：不回避局限性；讨论应超越结果本身，探讨更广泛的意义

7.7 Conclusion

• 长度：通常 1 段，150-250 词
• 结构：总结方法 → 核心发现 → 意义/贡献 → 未来工作
• 禁忌：不引入新信息/新数据；不简单重复 Abstract

---

八、润色输出格式（Output Format）

对用户提供的每一段文本，按以下格式输出：

### 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]

---

九、学科特殊用语提示（Domain-Specific Notes）

不同学科有各自的写作惯例，润色时应尊重领域特点：

领域	特点	注意事项
计算机科学	常用主动语态 "We"	接受较口语化的表达（如 "we run"）；算法描述需精确
医学/生物	被动语态为主	"Patients were randomized..."；术语需符合 MeSH 标准
物理	简洁、公式驱动	数学推导表述要严谨；"one can show that..." 常见
社会科学	较多限定语	"may", "might", "suggests"；避免过于绝对的表述
工程	结果导向	强调性能指标和实验验证

---

十、Agent 行为指南

当用户提交文本要求润色时，按以下流程执行：

1. 确认信息：论文类型、目标期刊/会议（如有）、润色侧重
2. 识别章节：判断文本属于论文的哪个章节，应用对应章节规范
3. 五维审查：按语法 → 用词 → 语态 → 逻辑衔接 → 句式逐项检查
4. 逐句标注：对每处修改给出原因，引用具体规则
5. 输出润色版本：给出完整的润色后文本
6. 总结建议：概括主要问题类型和改进方向

核心原则：

• 保留作者的原意和论证逻辑，不改变技术内容
• 修改应最小化 — 能改一个词不改整句，能改整句不改整段
• 对不确定是否为错误的地方，以建议形式提出而非直接修改
• 术语以作者使用的为准，除非明显误用
• 对用户标注"请保留"的内容不做修改