terminology-normalizer

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Terminology Normalizer

术语归一化工具（Terminology Normalizer）

Purpose: make the draft read like one author wrote it by enforcing consistent naming (canonical terms + synonym policy), without changing citations or meaning.

用途：通过执行统一的命名规则（规范术语+同义词策略），让草稿读起来像是出自同一作者之手，全程不改动引用和原文含义。

Role cards (use explicitly)

角色卡（请明确使用）

Taxonomist (canonicalizer)

分类学家（规范制定者 Taxonomist）

Mission: decide one canonical term per concept and a light synonym policy.

Do:

Prefer taxonomy node names (
```
outline/taxonomy.yml
```
) as canonical labels when available.
Define a short synonym policy only where readers expect it (use sparingly).
Keep headings and tables aligned with canonical terms.

Avoid:

Renaming proper nouns (paper titles, benchmark names, model names).
Over-normalizing away meaningful distinctions (e.g., collapsing two different mechanisms into one word).

职责：为每个概念确定唯一的规范术语，制定精简的同义词使用规则。

需执行：

优先选用
```
outline/taxonomy.yml
```
中存在的分类节点名称作为规范术语标签
仅在读者有预期的场景下定义简短的同义词规则（尽量少用）
确保标题和表格内容与规范术语保持一致

需避免：

重命名专有名词（论文标题、基准测试名称、模型名称等）
过度归一化抹除有意义的差异（例如将两种不同的机制合并为同一个词）

Integrator (apply without drift)

整合员（无偏差执行者 Integrator）

Mission: apply replacements consistently without changing meaning or citations.

Do:

Keep replacements local and conservative; reread sentences that become ambiguous.
Preserve citation placement and subsection boundaries.

Avoid:

Introducing new claims while rewriting for terminology.
Moving citations across subsections.

职责：统一执行替换规则，全程不改变原文含义和引用内容。

需执行：

替换操作需保持局部性、保守性，对替换后出现歧义的句子要重新通读校验
保留引用的位置和子章节边界不变

需避免：

为适配术语重写内容时引入新的论述
跨子章节移动引用位置

Role prompt: Terminology Editor (one voice)

角色提示词：术语编辑（统一文风）

text

You are normalizing terminology in a technical survey draft.

Your job is to make the draft read like one author wrote it by enforcing consistent naming.

Constraints:
- do not add/remove citation keys
- do not move citations across ### subsections
- do not introduce new claims while renaming

Method:
- pick a canonical term per concept
- define allowed synonyms (optional, minimal)
- apply consistently across headings, prose, and tables

text

You are normalizing terminology in a technical survey draft.

Your job is to make the draft read like one author wrote it by enforcing consistent naming.

Constraints:
- do not add/remove citation keys
- do not move citations across ### subsections
- do not introduce new claims while renaming

Method:
- pick a canonical term per concept
- define allowed synonyms (optional, minimal)
- apply consistently across headings, prose, and tables

Inputs

输入文件

```
output/DRAFT.md
```
Optional (read-only context):
- ```
outline/outline.yml
```
  (heading consistency)
- ```
outline/taxonomy.yml
```
  (canonical labels)

```
output/DRAFT.md
```
可选（只读上下文）：
- ```
outline/outline.yml
```
  （用于标题一致性校验）
- ```
outline/taxonomy.yml
```
  （用于获取规范术语标签）

Outputs

输出文件

```
output/DRAFT.md
```
(in place)
Optional:
```
output/GLOSSARY.md
```
(short appendix/glossary table, if useful)

```
output/DRAFT.md
```
（原位修改）
可选：
```
output/GLOSSARY.md
```
（简短附录/术语表，如有必要可生成）

Workflow

工作流程

Use the role cards above.

Steps:

Build a glossary candidate list from the draft (10–30 key terms):

core objects (agent, tool, environment, protocol)
key components (planner/executor, memory, verifier)
evaluation terms (benchmark, metric, budget)

Choose canonical names and a synonym policy:

one concept = one canonical term
define allowed synonyms only when readers expect them (and use them sparingly)
if
```
outline/taxonomy.yml
```
exists: prefer taxonomy node names as canonical labels (avoid inventing new names)
if
```
outline/outline.yml
```
exists: keep section headings aligned with the same canonical terms

Apply replacements conservatively:

do not alter paper names, model names, benchmark names
keep terminology consistent across headings, prose, and table captions

Optional: write a small
```
output/GLOSSARY.md
```
:

term | canonical | allowed synonyms | notes

请使用上文的角色卡执行操作。

步骤：

从草稿中提取候选术语表（10-30个核心术语）：

核心对象（agent、tool、environment、protocol）
核心组件（planner/executor、memory、verifier）
评估术语（benchmark、metric、budget）

选定规范术语和同义词策略：

一个概念对应唯一的规范术语
仅在读者有预期的场景下定义允许使用的同义词（且尽量少用）
若存在
```
outline/taxonomy.yml
```
文件：优先选用分类节点名称作为规范术语标签（避免自行发明新名称）
若存在
```
outline/outline.yml
```
文件：确保章节标题与规范术语保持一致

保守执行替换操作：

不得修改论文名称、模型名称、基准测试名称
确保标题、正文、表格标题中的术语保持一致

可选：编写简短的
```
output/GLOSSARY.md
```
文件：

表头格式：

术语 | 规范名称 | 允许的同义词 | 备注

Mini examples (what to do / what to avoid)

小示例（正确/错误做法）

Bad (term drift):
```
tool API
```
,
```
tool interface
```
,
```
action schema
```
used interchangeably without a rule.
Better (canonical + light synonym policy): pick one canonical term (e.g.,
```
tool interface
```
) and allow one synonym only when first introduced (e.g.,
```
tool interface (API contract)
```
), then stick to canonical thereafter.
Bad (over-normalization): replacing distinct terms so a contrast disappears.
Better: keep distinct terms when they encode different mechanisms; normalize only spelling and naming consistency.

错误做法（术语漂移）：不加区分交替使用
```
tool API
```
、
```
tool interface
```
、
```
action schema
```
正确做法（规范术语+精简同义词策略）：选定一个规范术语（例如
```
tool interface
```
），仅在首次出现时允许使用一个同义词（例如
```
tool interface (API contract)
```
），后续全文统一使用规范术语
错误做法（过度归一化）：替换含义不同的术语，导致原文的对比关系消失
正确做法：当不同术语对应不同机制时保留差异，仅对拼写和命名的一致性做归一化处理

Guardrails (do not violate)

使用红线（严禁违反）

Do not add/remove citation keys.
Do not move citations across
```
###
```
subsections.
Do not introduce new claims while renaming.

不得新增或删除引用键
不得跨
```
###
```
子章节移动引用位置
重命名术语时不得引入新的论述

Troubleshooting

问题排查

Issue: normalization changes citation keys or moves citations

问题：归一化操作改动了引用键或移动了引用位置

Fix:

Revert; this skill must not add/remove keys or move citations across subsections.

修复方案：回退操作；本工具不得新增/删除引用键，也不得跨子章节移动引用位置。

Issue: synonyms policy is unclear

问题：同义词策略不清晰

Fix:

Define one canonical term per concept and list allowed synonyms; apply consistently across headings, tables, and prose.

修复方案：为每个概念定义唯一的规范术语，列出允许使用的同义词，在标题、表格和正文中统一执行。