Rebuild speaking around intelligible interaction

2026-10-02 08:04:40 +08:00 · 2026-09-02 07:57:02 +08:00
parent 66083e3811
commit aeccc8ac94
31 changed files with 18532 additions and 17812 deletions
@@ -10,6 +10,7 @@ This register distinguishes repository-owned material from third-party sources.
 | Learning styles evidence review | Learning-principles and vocabulary chapters | [Pashler et al.](https://bjorklab.psych.ucla.edu/wp-content/uploads/sites/13/2016/07/Pashler_McDaniel_Rohrer_Bjork_2009_PSPI.pdf) | Linked research; short paraphrase, no paper stored | 2026-08-16 |
 | Vocabulary coverage research | Vocabulary chapters | [Hu & Nation (2000), Crossref record](https://api.crossref.org/works/10.64152%2F10125%2F66973), [Nation (2006)](https://doi.org/10.3138/cmlr.63.1.59), and [Schmitt, Jiang & Grabe (2011)](https://doi.org/10.1111/j.1540-4781.2011.01146.x) | Linked research; the Crossref endpoint is a stable bibliographic metadata record, while findings are paraphrased with limitations and no full text is redistributed | 2026-09-01 |
 | Focused and explicit L2 grammar instruction research | Bilingual Grammar chapters | [Norris & Ortega (2000), Crossref record](https://api.crossref.org/works/10.1111%2F0023-8333.00136) and [Spada & Tomita (2010), Crossref record](https://api.crossref.org/works/10.1111%2Fj.1467-9922.2010.00562.x) | Meta-analytic findings are paraphrased with construct, measure, feature, design, and generalisability limits; metadata links are provided and no paper is redistributed | 2026-09-02 |
+| Accent, intelligibility, and L2 pronunciation instruction research | Bilingual Speaking chapters | [Derwing & Munro (2005), Crossref record](https://api.crossref.org/works/10.2307%2F3588486), [Levis (2005), Crossref record](https://api.crossref.org/works/10.2307%2F3588485), and [Saito (2012), Crossref record](https://api.crossref.org/works/10.1002%2Ftesq.67) | Research framing is paraphrased conservatively to separate accent, communication outcomes, identity, listener role, and instruction; group findings are not written as individual guarantees and no paper is redistributed | 2026-09-02 |
 | ChatGPT Study Mode and Projects | AI chapters | [OpenAI Study Mode](https://openai.com/index/chatgpt-study-mode/) and [Projects Help](https://help.openai.com/en/articles/10169521-projects-in-chatgpt) | Product facts paraphrased and dated; availability may change | 2026-08-24 |
 | Gemini learning features | AI chapters | [Google Gemini Help](https://support.google.com/gemini/answer/16448384) | Product facts paraphrased with region/account limits | 2026-08-24 |
 | Claude Projects | AI chapters | [Anthropic Help](https://support.claude.com/en/articles/9517075-what-are-projects) | Product facts paraphrased with plan limits | 2026-08-24 |
@@ -8,6 +8,9 @@ All notable project-level changes are documented here. Content pages retain thei

 #### Changed

+- Rebuilt the bilingual Speaking chapter around unscripted baselines, reference variety without hierarchy, accentedness/intelligibility/comprehensibility distinctions, high-impact pronunciation, generation beyond shadowing, interaction repair, listener retelling, AI limits, and a fourteen-day transfer experiment.
+- Expanded the bilingual Speaking Evidence Card to capture three baseline conditions, recording consent, device and listener familiarity, actual listener recovery, comprehension effort, unfamiliar follow-ups, repair, recognition errors, and day-fourteen transfer.
+- Connected the new speaking method through the home pages, Reader's Guide, Toolkit Overview, glossary, Listening, AI-English, Job-search English, search-budget rules, research attribution, and regression coverage.
 - Added a bilingual Grammar chapter that starts from real-task baselines and form–meaning choices, prioritises errors by communicative impact, reconnects explicit explanation to output, and tests one structure through a fourteen-day delayed-transfer cycle.
 - Added a bilingual Grammar Evidence Card and connected it through Part I, the home pages, Reader's Guide, Toolkit Overview, glossary, search index, and continuous book navigation.
 - Made the existing bilingual biezou.com home-page entry an explicit third-party AI relay recommendation, while retaining current-site verification, privacy, pricing, terms, and availability boundaries.
@@ -45,7 +45,7 @@ AI 正在让答案变得前所未有地廉价。几秒钟里，我们可以得
  <a class="guide-path" href="./docs/templates/learning-state.md"><strong>终身学习系统</strong>从真实问题、当前基线和最小任务出发，让每轮学习都能被继续、复测与迁移。</a>
  <a class="guide-path" href="./docs/threads/part-3/1-ai-learning.md"><strong>使用 AI 学习一切</strong>让 AI 帮助提问、研究和反馈，同时把事实核验与最终判断留在人手中。</a>
  <a class="guide-path" href="./docs/threads/part-3/2-ai-development-and-resource-layer.md"><strong>AI 项目与资源层创业</strong>从需求、原型、代码和测试走向模型接入、治理、企业交付与商业验证。</a>
-  <a class="guide-path" href="./docs/threads/part-1/0-cefr.md"><strong>基础能力：英语</strong>用英语连接全球知识、技术文档、国际 AI 工具与跨文化协作。</a>
+  <a class="guide-path" href="./docs/threads/part-1/0-cefr.md"><strong>基础能力：英语</strong>用英语连接全球知识、技术文档与国际 AI 工具，以可理解度而不是口音相似度进入跨文化协作。</a>
  <a class="guide-path" href="./docs/threads/part-1/grammar.md"><strong>语法基础与真实表达</strong>不从背完整规则开始，而从时间、责任、条件与确定性是否被听清开始，用一项高影响结构完成十四天复测。</a>
  <a class="guide-path" href="./docs/threads/part-1/8-job-search-english.md"><strong>海外求职与远程协作</strong>把岗位描述拆成招聘沟通、项目说明、陌生追问和异步写作，用真实样本判断这一场面试还缺什么。</a>
  <a class="guide-path" href="./docs/threads/part-4/family-learning.md"><strong>家庭与中学生学习</strong>让孩子参与定义目标，让家长保护环境、隐私与安全，用十四天证据替代监控、代做和焦虑加码。</a>
@@ -63,7 +63,7 @@ AI 正在让答案变得前所未有地廉价。几秒钟里，我们可以得
 | 部分 | 核心问题 | 入口 |
 | --- | --- | --- |
 | 阅读指南与序章 | 我从哪里进入，又为什么要重新开始？ | [阅读指南](docs/threads/part-0/reader-guide.md) · [先不要急着改变人生](docs/threads/part-0/prologue.md) |
-| 第一部：打开输入 | 怎样建立英语和世界之间的通道，并把能力带进面试与远程协作？ | [本部导语](docs/threads/part-1/open-input.md) · [CEFR 自测](docs/threads/part-1/0-cefr.md) · [词汇、听力、阅读、口语、写作](docs/threads/part-1/2-vocabulary.md) · [求职英语](docs/threads/part-1/8-job-search-english.md) |
+| 第一部：打开输入 | 怎样建立英语和世界之间的通道，并把能力带进面试与远程协作？ | [本部导语](docs/threads/part-1/open-input.md) · [CEFR 自测](docs/threads/part-1/0-cefr.md) · [词汇](docs/threads/part-1/2-vocabulary.md) · [语法](docs/threads/part-1/grammar.md) · [听力](docs/threads/part-1/3-listening.md) · [阅读](docs/threads/part-1/4-reading.md) · [口语](docs/threads/part-1/5-speaking.md) · [写作](docs/threads/part-1/6-writing.md) · [求职英语](docs/threads/part-1/8-job-search-english.md) |
 | 第二部：把自己放回生活 | 能力、事业、关系、失败、选择与恢复怎样互相影响？ | [本部导语](docs/threads/part-2/return-to-life.md) · [我的故事](docs/threads/part-2/my-story.md) · [叙事与证据](docs/threads/part-2/narrative-and-evidence.md) · [恢复、选择、关系与创业](docs/threads/part-2/recovery.md) |
 | 第三部：借工具放大能力 | 怎样使用 AI，而不把判断力和注意力一起外包？ | [本部导语](docs/threads/part-3/amplify-ability.md) · [AI 学习](docs/threads/part-3/1-ai-learning.md) · [注意力、作品与证据](docs/threads/part-3/3-attention-and-judgment.md) · [项目实践](docs/threads/part-3/2-ai-development-and-resource-layer.md) |
 | 第四部：实践与恢复 | 学习如何回到身体、家庭与日常，并在需要时保护未成年学习者的主体性？ | [本部导语](docs/threads/part-4/practice-and-recovery.md) · [第一周实践](docs/threads/part-4/week-1.md) · [家庭学习](docs/threads/part-4/family-learning.md) · [生活系统](docs/threads/part-4/daily-system.md) · [节律](docs/threads/part-4/rhythm-and-compounding.md) |
@@ -62,9 +62,11 @@ function splitSearchSections(file: string, html: string) {
    "/templates/reader-field-note.md",
    "/templates/interview-evidence.md",
    "/templates/grammar-evidence.md",
+    "/templates/speaking-evidence.md",
    "/threads/part-4/family-learning.md",
    "/threads/part-1/8-job-search-english.md",
    "/threads/part-1/grammar.md",
+    "/threads/part-1/5-speaking.md",
  ].some((suffix) => normalizedFile.endsWith(suffix));
  const headingOnly = [
    "/docs/README.md",
@@ -45,7 +45,7 @@ AI 正在让答案变得前所未有地廉价。几秒钟里，我们可以得
  <a class="guide-path" href="./templates/learning-state"><strong>终身学习系统</strong>从真实问题、当前基线和最小任务出发，让每轮学习都能被继续、复测与迁移。</a>
  <a class="guide-path" href="./threads/part-3/1-ai-learning"><strong>使用 AI 学习一切</strong>让 AI 帮助提问、研究和反馈，同时把事实核验与最终判断留在人手中。</a>
  <a class="guide-path" href="./threads/part-3/2-ai-development-and-resource-layer"><strong>AI 项目与资源层创业</strong>从需求、原型、代码和测试走向模型接入、治理、企业交付与商业验证。</a>
-  <a class="guide-path" href="./threads/part-1/0-cefr"><strong>基础能力：英语</strong>用英语连接全球知识、技术文档、国际 AI 工具与跨文化协作。</a>
+  <a class="guide-path" href="./threads/part-1/0-cefr"><strong>基础能力：英语</strong>用英语连接全球知识、技术文档与国际 AI 工具，以可理解度而不是口音相似度进入跨文化协作。</a>
  <a class="guide-path" href="./threads/part-1/grammar"><strong>语法基础与真实表达</strong>不从背完整规则开始，而从时间、责任、条件与确定性是否被听清开始，用一项高影响结构完成十四天复测。</a>
  <a class="guide-path" href="./threads/part-1/8-job-search-english"><strong>海外求职与远程协作</strong>把岗位描述拆成招聘沟通、项目说明、陌生追问和异步写作，用真实样本判断这一场面试还缺什么。</a>
  <a class="guide-path" href="./threads/part-4/family-learning"><strong>家庭与中学生学习</strong>让孩子参与定义目标，让家长保护环境、隐私与安全，用十四天证据替代监控、代做和焦虑加码。</a>
@@ -63,7 +63,7 @@ AI 正在让答案变得前所未有地廉价。几秒钟里，我们可以得
 | 部分 | 核心问题 | 入口 |
 | --- | --- | --- |
 | 阅读指南与序章 | 我从哪里进入，又为什么要重新开始？ | [阅读指南](threads/part-0/reader-guide.md) · [先不要急着改变人生](threads/part-0/prologue.md) |
-| 第一部：打开输入 | 怎样建立英语和世界之间的通道，并把能力带进面试与远程协作？ | [本部导语](threads/part-1/open-input.md) · [CEFR 自测](threads/part-1/0-cefr.md) · [词汇、听力、阅读、口语、写作](threads/part-1/2-vocabulary.md) · [求职英语](threads/part-1/8-job-search-english.md) |
+| 第一部：打开输入 | 怎样建立英语和世界之间的通道，并把能力带进面试与远程协作？ | [本部导语](threads/part-1/open-input.md) · [CEFR 自测](threads/part-1/0-cefr.md) · [词汇](threads/part-1/2-vocabulary.md) · [语法](threads/part-1/grammar.md) · [听力](threads/part-1/3-listening.md) · [阅读](threads/part-1/4-reading.md) · [口语](threads/part-1/5-speaking.md) · [写作](threads/part-1/6-writing.md) · [求职英语](threads/part-1/8-job-search-english.md) |
 | 第二部：把自己放回生活 | 能力、事业、关系、失败、选择与恢复怎样互相影响？ | [本部导语](threads/part-2/return-to-life.md) · [我的故事](threads/part-2/my-story.md) · [叙事与证据](threads/part-2/narrative-and-evidence.md) · [恢复、选择、关系与创业](threads/part-2/recovery.md) |
 | 第三部：借工具放大能力 | 怎样使用 AI，而不把判断力和注意力一起外包？ | [本部导语](threads/part-3/amplify-ability.md) · [AI 学习](threads/part-3/1-ai-learning.md) · [注意力、作品与证据](threads/part-3/3-attention-and-judgment.md) · [项目实践](threads/part-3/2-ai-development-and-resource-layer.md) |
 | 第四部：实践与恢复 | 学习如何回到身体、家庭与日常，并在需要时保护未成年学习者的主体性？ | [本部导语](threads/part-4/practice-and-recovery.md) · [第一周实践](threads/part-4/week-1.md) · [家庭学习](threads/part-4/family-learning.md) · [生活系统](threads/part-4/daily-system.md) · [节律](threads/part-4/rhythm-and-compounding.md) |
@@ -45,7 +45,7 @@ It also keeps three kinds of claim separate:
  <a class="guide-path" href="./templates/learning-state"><strong>Lifelong Learning System</strong>Begin with a real problem, a current baseline, and a minimum task so each cycle can continue, be retested, and transfer.</a>
  <a class="guide-path" href="./threads/part-3/1-ai-learning"><strong>Learn Anything with AI</strong>Use AI for questions, research, and feedback while keeping fact-checking and final judgment in human hands.</a>
  <a class="guide-path" href="./threads/part-3/2-ai-development-and-resource-layer"><strong>AI Projects and Resource-layer Business</strong>Move from requirements, prototypes, code, and tests to model access, governance, enterprise delivery, and business validation.</a>
-  <a class="guide-path" href="./threads/part-1/0-cefr"><strong>Foundation: English</strong>Use English to reach global knowledge, technical documentation, international AI tools, and cross-cultural work.</a>
+  <a class="guide-path" href="./threads/part-1/0-cefr"><strong>Foundation: English</strong>Use English to reach global knowledge, technical documentation, and international AI tools, entering cross-cultural work through intelligibility rather than accent similarity.</a>
  <a class="guide-path" href="./threads/part-1/grammar"><strong>Grammar for Real Expression</strong>Begin with whether time, responsibility, condition, and certainty are understood, then retest one high-impact structure for fourteen days instead of memorising every rule first.</a>
  <a class="guide-path" href="./threads/part-1/8-job-search-english"><strong>Global Job Search and Remote Work</strong>Turn a job description into recruiter communication, project explanation, unfamiliar follow-ups, and asynchronous writing, then use real samples to locate the next gap.</a>
  <a class="guide-path" href="./threads/part-4/family-learning"><strong>Family and Middle-School Learning</strong>Let the learner help define the goal while adults protect environment, privacy, and safety; use fourteen days of evidence instead of surveillance or substituted work.</a>
@@ -63,7 +63,7 @@ When a term is unclear or you do not know which page to open next, use the [Glos
 | Part | Core question | Entry point |
 | --- | --- | --- |
 | Reader's Guide and Prologue | Where do I enter, and why begin again? | [Reader's Guide](threads/part-0/reader-guide.md) · [Do Not Rush to Change Your Life](threads/part-0/prologue.md) |
-| Part I: Open Input | How do I build a bridge between English and the world, then carry ability into interviews and remote collaboration? | [Part Introduction](threads/part-1/open-input.md) · [CEFR Self-check](threads/part-1/0-cefr.md) · [Vocabulary, Listening, Reading, Speaking, Writing](threads/part-1/2-vocabulary.md) · [Job-search English](threads/part-1/8-job-search-english.md) |
+| Part I: Open Input | How do I build a bridge between English and the world, then carry ability into interviews and remote collaboration? | [Part Introduction](threads/part-1/open-input.md) · [CEFR Self-check](threads/part-1/0-cefr.md) · [Vocabulary](threads/part-1/2-vocabulary.md) · [Grammar](threads/part-1/grammar.md) · [Listening](threads/part-1/3-listening.md) · [Reading](threads/part-1/4-reading.md) · [Speaking](threads/part-1/5-speaking.md) · [Writing](threads/part-1/6-writing.md) · [Job-search English](threads/part-1/8-job-search-english.md) |
 | Part II: Return to Life | How do ability, work, relationships, failure, choices, and recovery affect one another? | [Part Introduction](threads/part-2/return-to-life.md) · [My Story](threads/part-2/my-story.md) · [Narrative and Evidence](threads/part-2/narrative-and-evidence.md) · [Recovery, Decisions, Relationships, and Entrepreneurship](threads/part-2/recovery.md) |
 | Part III: Amplify Ability | How can I use AI without outsourcing judgment or attention? | [Part Introduction](threads/part-3/amplify-ability.md) · [AI Learning](threads/part-3/1-ai-learning.md) · [Attention, Artifacts, and Evidence](threads/part-3/3-attention-and-judgment.md) · [Project Practice](threads/part-3/2-ai-development-and-resource-layer.md) |
 | Part IV: Practice and Recovery | How does learning return to the body, family, and daily life while protecting a young learner's agency when relevant? | [Part Introduction](threads/part-4/practice-and-recovery.md) · [First Week Practice](threads/part-4/week-1.md) · [Family Learning](threads/part-4/family-learning.md) · [Daily System](threads/part-4/daily-system.md) · [Rhythm](threads/part-4/rhythm-and-compounding.md) |
@@ -42,6 +42,9 @@ Find the task you are facing, then follow “definition → evidence → next st
 | **Distributed practice** | Return at multiple points instead of cramming once | [Vocabulary](../threads/part-1/2-vocabulary.md) |
 | **Form–meaning choice** | Treat grammar as a choice that changes time, participants, condition, negation scope, logic, or certainty rather than only a rule label | [Grammar](../threads/part-1/grammar.md) |
 | **Grammar contrast card** | Preserve an original sentence, likely interpretation, intended meaning, form contrast, revision reason, and delayed retest for one real task | [Grammar Evidence Card](../templates/grammar-evidence.md) |
+| **Reference variety** | A relatively stable dictionary-audio, spelling, or pronunciation reference chosen to reduce early decisions; it does not rank other English varieties below it | [Speaking](../threads/part-1/5-speaking.md) |
+| **Intelligibility / comprehensibility** | The first records how much meaning the listener recovered; the second records effort such as guessing, replay, or confirmation. Both differ from accentedness | [Speaking](../threads/part-1/5-speaking.md) |
+| **Listener retelling** | A listener restates gist, detail, numbers, responsibility, and next step without seeing the script, showing what the sound actually carried | [Speaking Evidence Card](../templates/speaking-evidence.md) |
 | **Intensive listening/reading** | Use short material to check detail, structure, evidence, and language form | [Listening](../threads/part-1/3-listening.md) · [Reading](../threads/part-1/4-reading.md) |
 | **Parallel material** | A different source with similar difficulty and task for testing transfer | [Reading Evidence Card](../templates/reading-evidence.md) |
 | **Transfer** | Complete a related action after changing topic, listener, time, or task | [90-Day Cycle Map](../templates/90-day-cycle.md) |
@@ -63,6 +66,7 @@ Find the task you are facing, then follow “definition → evidence → next st
 | **Family Learning Agreement** | A private worksheet where learner and guardian record goals, concerns, support, AI and data boundaries separately, then review fourteen days of evidence together | [Family Learning Agreement](../templates/family-learning-agreement.md) |
 | **Job-search English Evidence Card** | One record connecting a real role, unscripted baseline, stories, unfamiliar follow-ups, asynchronous writing, delayed retest, and integrity boundaries | [Job-search English Evidence Card](../templates/interview-evidence.md) |
 | **Grammar Evidence Card** | Track one high-impact structure through a real-task baseline, meaning contrast, four-skill retrieval, feedback classification, and day-fourteen transfer | [Grammar Evidence Card](../templates/grammar-evidence.md) |
+| **Speaking Evidence Card** | A private record connecting reference variety, three unscripted baselines, listener retelling, comprehension effort, interaction repair, unfamiliar follow-ups, and delayed transfer | [Speaking Evidence Card](../templates/speaking-evidence.md) |
 | **Interaction repair** | When hearing, scope, or knowledge is incomplete, request repetition, confirm, ask for thinking time, and summarise agreement | [Speaking](../threads/part-1/5-speaking.md) · [Job-search English](../threads/part-1/8-job-search-english.md) |
 | **Artifact** | An output others can read, use, question, or improve; not necessarily a product | [Artifacts](../threads/part-3/4-artifacts-and-delivery.md) |
 | **Delivery** | Completing work under a real audience, user, or constraint and accepting the result | [90-Day Action Plan](../threads/part-5/90-day-plan.md) |
@@ -1,82 +1,172 @@
 ---
-title: Speaking Evidence Card
-description: Track speaking progress with recordings, transcripts, listener retellings, interaction repair, and delayed retakes instead of accent similarity alone.
-updated: 2026-08-31
+title: "Speaking Evidence Card: From Accent Anxiety to Verifiable Interaction"
+description: Preserve unscripted speech, unfamiliar follow-ups, listener retellings, reference variety, comprehension effort, repair, and delayed transfer without treating accent similarity or recognition scores as speaking ability.
+updated: 2026-09-02
 ---

-# Speaking Evidence Card
+# Speaking Evidence Card: From Accent Anxiety to Verifiable Interaction

-Speaking ability has to be heard in a real task. Copy this card into a private note for one explanation, interaction, or repair task at a time. Do not record passwords, client data, unauthorised third-party information, or unnecessary personal details.
+Copy this card into a private note for one real situation at a time. Keep raw recordings private by default. Before recording another person or using customer, family, medical, or unreleased project information, obtain necessary permission or practise with fictional details.

-## 1. Define the task and boundary
+## 1. Define the Task, Listener, and Boundary

 ```markdown
-# Speaking Evidence — YYYY-MM-DD
+# Speaking Evidence - YYYY-MM-DD

-Real context:
-Task: explain / interact / repair / transfer
+Real situation:
+Task: explain / answer / repair / collaborate / transfer
 Listener and relationship:
+What the listener must know or do afterwards:
 Time limit:
-Completion standard:
-Content that must stay private:
+Allowed notes / retakes / captions / dictionary / AI:
+Recording and upload permission:
+Content that must remain private:
 ```

-Decide what may be recorded or shared first. For client, family, or colleague content, get necessary consent or use fictional details.
+Write completion as a listener result, such as "can retell the risk and next step", not only "sounds fluent".

-## 2. Save an unaided first take
+## 2. Record Reference Variety and Technical Conditions

 ```markdown
-Recording location:
-First-take length:
-One-sentence message I want to deliver:
-Two details the listener should remember:
-First pause or restart:
+Current reference variety or audio source:
+Why it fits:
+Differences I only need to recognise, not force myself to imitate:
+Other English varieties added this cycle:
+
+Device / microphone:
+Network / room noise:
+Recording or call platform:
+Listener familiarity with the topic and my accent:
 ```

-Do not ask AI to rewrite or script the first take. Its value is showing what you can do without scaffolding.
+When device, network, or listener familiarity changes, do not assign the entire score difference to ability.

-## 3. Classify problems by impact
+## 3. Preserve Three Unscripted Baselines

-| Segment/time | Problem | Does it affect understanding? | One repair for next time |
-| --- | --- | --- | --- |
-| | chunk / pronunciation / grammar / organisation / pause / interaction | yes / no | |
-
-Repair errors that change meaning, block the task, or recur. An accent difference is not automatically an error; do not sacrifice clarity and naturalness to imitate one speaker.
-
-## 4. Record interaction repair
+| Sample | Condition | Recording | First break | Task completed? |
+| --- | --- | --- | --- | --- |
+| 90-120 second monologue | Explain a real experience/problem without a script | | | |
+| Five-turn exchange | Include at least one unfamiliar follow-up | | | |
+| One interaction repair | Repetition, scope confirmation, rephrasing, or summary | | | |

 ```markdown
-What I did not hear or express clearly:
-Repair line I used:
-How the other person responded:
-How I confirmed shared understanding:
-Chunk to prepare next time:
+One-sentence main point:
+Two details the listener must hear:
+Critical number / time / responsibility / condition:
+What I most expect to be misunderstood:
 ```

-Repair actions include asking for repetition, confirming a keyword, rephrasing, requesting thinking time, and summarising agreement. They are part of communication, not proof of failure.
+Do not remove pauses, clean noise, or ask AI for a script first. The baseline shows what remains when scaffolding leaves.

-## 5. Retake and listener feedback
+## 4. Preserve What the Listener Actually Heard

-| Version | Conditions | Task 0–2 | Comprehensibility 0–2 | Organisation 0–2 | Interaction/repair 0–2 | Feedback |
-| --- | --- | ---: | ---: | ---: | ---: | --- |
-| First take | | | | | | |
-| Retake | | | | | | |
+Ask the listener to hear the recording before seeing a transcript:

-Ask the listener to answer from the recording alone:
+| Listener/condition | Retold gist | Retold details | Guess/replay point | Next response |
+| --- | --- | --- | --- | --- |
+| Listener A | | | | |
+| Listener B or new device | | | | |

-1. What was the main point?
-2. Which two details were clearest?
-3. Which sentence required guessing?
-4. What would you ask next?
+```markdown
+Noticeable accent that remained intelligible:
+Place where meaning was actually lost:
+Place that required effort without misunderstanding:
+Familiarity or bias that may shape the feedback:
+```

-## 6. Test delayed transfer
+One subjective impression is not a conclusion. Preserve what the person heard and could do, not only whether you resembled a prestige accent.

-| Date | Changed condition | Output | Result |
-| --- | --- | --- | --- |
-| Day 1 | Original situation, unscripted | First take and retake | |
-| Days 3–7 | Related situation or new listener | 60–120 second retelling/dialogue | |
-| Day 30 | New topic, time pressure, or real task | Five-minute explanation/meeting/interview | |
+## 5. Build a High-Impact Problem Map

-If fluency appears only with a familiar script, but disappears after the condition changes, make chunk retrieval and interaction repair the next focus instead of memorising more scripts.
+| Segment/time | Listener may hear | Intended meaning | Type | Task impact | This cycle's action |
+| --- | --- | --- | --- | --- | --- |
+| | | | sound / ending / stress / pause / chunk / grammar / organisation | meaning-changing / task-blocking / recurring / effortful / low-impact difference | |

-Related chapter: [Speaking: Make Meaning Arrive](../../threads/part-1/5-speaking.md)
+Choose only one to three problems. When a difference marks reference variety, region, or personal voice and does not raise comprehension cost, record it without forcing it away.
+
+## 6. Move from Chunks into Unfamiliar Follow-ups
+
+```markdown
+Five replaceable chunks:
+1.
+2.
+3.
+4.
+5.
+
+Topic variation:
+Position variation:
+Listener variation:
+Unfamiliar follow-up:
+Unscripted answer:
+```
+
+When fluency exists only inside a memorised script, reduce complete-sentence memorisation and increase keyword reconstruction plus follow-ups.
+
+## 7. Record Interaction Repair
+
+| Break | How I noticed | Repair action | Listener response | How shared understanding was confirmed |
+| --- | --- | --- | --- | --- |
+| Not heard / scope unclear / thinking needed / expression failed / agreement unstable | | | | |
+
+```markdown
+My repetition or confirmation request:
+My rephrasing:
+My final summary of agreement:
+If repair failed, the one change for next time:
+```
+
+Repair is not a penalty. Failing to notice a misunderstanding, or failing to restore the task after noticing it, is the training signal.
+
+## 8. Record Teacher, Listener, and AI Feedback
+
+```markdown
+Reviewer / tool / model version and date:
+Did it use raw audio or a transcript?
+Speech-recognition transcript errors:
+Problems it marked as high impact:
+Did it separate accent difference, misunderstanding, effort, grammar, register, and style?
+Dictionary audio, teacher, or real-listener check:
+What I accepted / rejected, and why:
+```
+
+Recognition scores change with microphone, noise, network, model, and accent. Treat them as clues only.
+
+## 9. Test Delayed Retention and Transfer
+
+| Time | What is removed | Changed condition | Output | Listener result |
+| --- | --- | --- | --- | --- |
+| Day 1 | No AI-written script | Original situation | First take and retake | |
+| Days 3-7 | Old script and model audio removed | New topic or listener | 90-120 seconds plus follow-up | |
+| Day 14 | Recent feedback removed | New device, accent, or time pressure | Collaboration/decision explanation | |
+| Day 30 | All practice prompts closed | Real meeting, interview, or conversation | Complete task | |
+
+```markdown
+What remained retrievable after seven days:
+What failed again after conditions changed:
+Did the listener guess or replay less often?
+Next variable:
+```
+
+## 10. Score and Decide
+
+Score 0-2: `0=not completed`, `1=prompt-dependent or unstable`, `2=completed under the current condition`.
+
+| Item | Day 1 | Day 7 | Day 14 | Evidence reason |
+| --- | ---: | ---: | ---: | --- |
+| Complete task and main line | | | | |
+| Listener recovers critical information | | | | |
+| Understanding does not require repeated guessing/replay | | | | |
+| Handle an unfamiliar follow-up | | | | |
+| Notice and repair an interaction break | | | | |
+| Complete after listener, device, or topic changes | | | | |
+
+```markdown
+Current evidence supports: continue same problem / add variation / move to another problem / pause
+Most important evidence:
+Conclusion I still cannot draw:
+Smallest next task:
+Next review date:
+```
+
+Related chapters: [Speaking: Make Meaning Arrive](../threads/part-1/5-speaking.md) | [Listening](../threads/part-1/3-listening.md) | [Job-search English](../threads/part-1/8-job-search-english.md) | [Evidence Chain Template](evidence-chain.md)
@@ -25,7 +25,8 @@ If you do not know where to begin, choose one sheet. An incomplete sheet used on
 | I recognise words but cannot use them | [Vocabulary Audit](vocabulary-audit.md) | [Vocabulary](../threads/part-1/2-vocabulary.md) | It cannot replace contextual performance with card counts |
 | I know the rule but still repeat errors in real expression | [Grammar Evidence Card](grammar-evidence.md) | [Grammar](../threads/part-1/grammar.md) | One corrected sentence, a style preference, or AI polish cannot prove stable grammar ability |
 | I cannot hear or explain what I read | [Listening Resource Audit](listening-audit.md) or [Reading Evidence Card](reading-evidence.md) | [Listening](../threads/part-1/3-listening.md) · [Reading](../threads/part-1/4-reading.md) | It cannot use minutes played or highlights as proof of understanding |
-| I cannot speak or write with confidence | [Speaking Evidence Card](speaking-evidence.md) or [Writing Evidence Card](writing-evidence.md) | [Speaking](../threads/part-1/5-speaking.md) · [Writing](../threads/part-1/6-writing.md) | It cannot speak for you before a real listener |
+| I sound fluent only with a script, fear accent difference, or freeze on follow-ups | [Speaking Evidence Card](speaking-evidence.md) | [Speaking](../threads/part-1/5-speaking.md) | Accent similarity, recognition scores, and one retake cannot prove stable interaction ability |
+| I cannot write, or can only show a tool-polished product | [Writing Evidence Card](writing-evidence.md) | [Writing](../threads/part-1/6-writing.md) | It cannot complete writing and revision transfer before a real reader |
 | I do not know how to involve AI | [AI Task Brief](ai-task-brief.md) | [AI Learning Log](ai-learning-log.md) | It cannot turn a model response into a fact or final judgment |
 | I have an idea but no deliverable | [Artifact Brief and Delivery Card](artifact-brief.md) | [AI Project Scorecard](ai-project-scorecard.md) | It cannot find users, carry responsibility, or prove a commercial result |
 | I want to analyse a person or project story | [AI Case Review](ai-case-review.md) | [Narrative and Evidence](../threads/part-2/narrative-and-evidence.md) | It cannot turn a public narrative into an independent audit |
@@ -24,7 +24,7 @@ Read the [Prologue: Do Not Rush to Change Your Life](prologue.md), then open the

 ### I want to improve one concrete skill

-Use [CEFR Goals and Self-check](../part-1/0-cefr.md) to name the context and baseline, then enter grammar, listening, reading, speaking, or writing. When familiar rules still change time, responsibility, condition, or certainty in real expression, read [Grammar: Let Structure Serve Meaning](../part-1/grammar.md), then track one high-impact structure with the [Grammar Evidence Card](../../templates/grammar-evidence.md). Choose one main line at a time and use the relevant [evidence cards](../../reference/glossary.md#tools-and-delivery) to preserve a first sample, feedback, and a retest. More study is not necessarily deeper study; completing the action in a new situation is where ability starts to show.
+Use [CEFR Goals and Self-check](../part-1/0-cefr.md) to name the context and baseline, then enter grammar, listening, reading, speaking, or writing. When familiar rules still change time, responsibility, condition, or certainty in real expression, read [Grammar: Let Structure Serve Meaning](../part-1/grammar.md), then track one high-impact structure with the [Grammar Evidence Card](../../templates/grammar-evidence.md). When fluency exists only inside a memorised script, accent difference creates fear, or an unfamiliar follow-up ends participation, enter [Speaking: Make Meaning Arrive](../part-1/5-speaking.md) and use the [Speaking Evidence Card](../../templates/speaking-evidence.md) to preserve reference variety, listener retelling, interaction repair, and follow-up transfer. Choose one main line at a time and use the relevant [evidence cards](../../reference/glossary.md#tools-and-delivery) to preserve a first sample, feedback, and a retest. More study is not necessarily deeper study; completing the action in a new situation is where ability starts to show.

 ### I am preparing for a global role or remote interview

@@ -1,7 +1,7 @@
 ---
 title: "Listening: From Sound Recognition to Real Understanding"
 description: Use CEFR can-do goals, gist, detail, segmentation, and retelling to build comparable 7-day, 30-day, and 12-week evidence.
-updated: 2026-09-01
+updated: 2026-09-02
 ---

 # Listening: From Sound Recognition to Real Understanding
@@ -97,6 +97,8 @@ For a 30-second to 3-minute clip, use five steps: listen for gist, mark unclear

 Intensive practice also helps you notice pronunciation patterns: linking, reductions, pauses, chunking. Find your weak spots and train them deliberately.

+Do not stop at line-by-line repetition after intensive listening. Continue through the five levels in [Speaking](5-speaking.md): read aloud, delay-shadow, close and retell, then answer a follow-up the material did not supply. Repeating the model shows finer observation; reorganising meaning under a new question begins to show speaking transfer.
+
 In high school I used a very “dumb” method: I listened over and over until I could recite and write down the whole text. I spent a lot of time on *New Concept English* Book 3. Back then, I couldn’t really see the improvement and even wondered if it was pointless—until I scored 115/120 on the English exam.

 There are many options, but *New Concept English* Book 3 and 4 are solid classics for this kind of training.
@@ -1,226 +1,263 @@
 ---
 title: "Speaking: Make Meaning Arrive"
-description: Use CEFR goals and recording evidence to train task completion, comprehensibility, fluency, interaction, and revision.
+description: Begin with unscripted baselines, a reference variety, and real listeners to train comprehensibility, pronunciation, chunks, interaction repair, and transfer without treating accent removal as the goal.
 updated: 2026-09-02
+sources_checked: 2026-09-02
 ---

 # Speaking: Make Meaning Arrive

-Source (Chinese edition): [Speaking](../../../threads/part-1/5-speaking.md)
+I once imagined speaking as a live examination. The sentence had to be arranged before I opened my mouth. The pronunciation should resemble the recording. Pauses should disappear, and errors should remain unheard. The harder I tried to speak well, the higher the threshold became.

-Speaking is not about performing one “correct” accent. It is about making meaning arrive clearly and naturally within a real interaction. Pronunciation, vocabulary, and grammar matter, but they serve comprehension, interaction, and task completion.
+A real conversation is not a performance completed alone. Another person asks an unfamiliar question, misunderstands, adds information, changes direction, or enters with a different accent, pace, device, and history. Speaking ability therefore does not live only in pronunciation. It includes finding the main line, noticing that the other person has not followed, trying another wording, admitting what is not yet known, and continuing the shared task after repair.

-## Quick Overview
+This chapter does not promise accent removal or present American, British, or any other single voice as the destination of English. It offers a smaller path: choose a reference variety related to your audience, preserve unpolished recordings, ask real listeners to retell what reached them, repair only one to three high-impact problems, then test the result with unfamiliar questions, different listeners, and time pressure.

- build intelligible pronunciation with reliable audio and physical practice, not Chinese approximations;
- move familiar expressions into your mouth through reading aloud, retelling, role-play, and real conversations;
- record an unaided first take and repair only the one to three issues that most affect understanding;
- judge progress by whether a listener can restate your point, not by accent conformity.
+## Chapter at a Glance

-## Define the Speaking Task First
+- Define an explanation, interaction, repair, or collaboration task before choosing an exercise.
+- Preserve unscripted monologue, question-and-answer, and interaction-repair baselines instead of substituting a memorised script.
+- Use one reference variety for consistency while gradually learning to understand varied real-world Englishes.
+- Separate accentedness, intelligibility, and comprehensibility instead of calling every noticeable difference an error.
+- Prioritise pronunciation and rhythm problems that change words, time, numbers, responsibility, or information focus.
+- Use shadowing to observe and imitate; use retelling, follow-ups, and collaboration to generate meaning.
+- Treat AI and speech recognition as clues; listener retelling and next action are stronger evidence.
+- Track one real situation and a few high-impact problems for fourteen days.

-Speaking practice is not only a monologue. Decide whether you need to explain, interact, repair a misunderstanding, or transfer a skill to a new situation, then choose the exercise.
+## 1. Define the Speaking Task

-| Task | Conditions | Evidence of completion |
+"Improve speaking" is too large to train directly. Rewrite it as an action that will happen:
+
+| Task | Practice condition | Evidence of completion |
 | --- | --- | --- |
-| Explain | 60–120 seconds, unscripted, describing an experience, view, or process | A listener can write your gist and two key details |
-| Interact | At least five turns with questions and follow-ups | Keep the conversation notes and mark one question you used to move it forward |
-| Repair | Start from one real uncertain point and practise clarifying, restating, and confirming | Record how the misunderstanding was found, repaired, and confirmed |
-| Transfer | Repeat with a new person, topic, time limit, or audience | Compare task completion, not accent similarity |
+| Explain | 60-120 seconds, unscripted, describing an experience, view, or process | A listener can retell the gist and two key details |
+| Answer | At least five turns, including one unfamiliar follow-up | The response does not miss the question entirely and advances the conversation once |
+| Repair | Preserve a moment of unclear hearing, unclear scope, or failed expression | Repetition, confirmation, or rephrasing restores shared understanding |
+| Collaborate | Meeting, interview, customer conversation, or joint decision | The other person knows the conclusion, disagreement, owner, and next step |
+| Transfer | Change topic, listener, device, pace, or time limit | A related task remains possible without the old script |

-Copy the [Speaking Evidence Card](../../templates/speaking-evidence.md) to keep the recording, transcript, listener feedback, and next variable together.
+Before starting, write who is listening, why they are listening, and what they need to know or do afterwards. Speaking is not making all English impressive. It is completing this task inside this relationship under the current conditions.

---
+Copy the [Speaking Evidence Card](../../templates/speaking-evidence.md) to keep recordings, listener retellings, repair actions, and delayed retests together.

-## IPA / Phonics
+## 2. Preserve Three Unpolished Baselines

-If you want to speak better, start with pronunciation. IPA is a map, not the destination. Any Chinese approximation is only a cue for mouth shape or airflow; it cannot replace dictionary audio, feedback, or comparing your own recording.
+A person may speak smoothly in a monologue and lose direction during follow-ups. Another may pronounce clearly but fail to confirm what was actually asked. Preserve at least three samples:

-There is no shortcut, but the sound inventory is limited. Hear a sound, inspect the symbol, practise it, and return to it inside real words and sentences.
+1. **Monologue baseline**: speak for 90-120 seconds about a real experience or problem without a script.
+2. **Interaction baseline**: complete five question-and-answer turns with one unknown question.
+3. **Repair baseline**: record one request for repetition, scope confirmation, rephrasing, or summary of agreement.

-Recommended playlist:
+Keep the raw audio. Do not remove pauses, reduce noise, or ask AI to rewrite before saving it. Record the device, network, notes, retake permission, and listener familiarity with the topic. Performances under different conditions are not directly comparable.

- [Teach Reading with Phonics - American English Pronunciation!](https://www.youtube.com/playlist?list=PL9BB1D7256440E08B)
+Ask the listener to answer from audio alone:

-### Vowels (quick reference)
+```markdown
+The main point I heard:
+The two clearest details:
+Where I had to guess or replay:
+What I would ask or do next:
+```

- cop /ɑ/ — like “ah”
- the /ə/ — “uh” (unstressed)
- cup /ʌ/ — like “uh” (stressed)
- boot /u/ — “oo”
- book /ʊ/ — shorter “oo”
- beat /i/ — “ee”
- bit /ɪ/ — “ih”
- make /eɪ/ — “ay”
- head /e/ — “eh”
- had /æ/ — “a” as in “cat”
- law /ɔ/ — “aw”
- now /aʊ/ — “ow”
- bite /aɪ/ — “eye”
- boy /ɔɪ/ — “oy”
- go /oʊ/ — “oh”
+A transcript can reveal vocabulary and structure. It cannot replace stress, pause, stance, or interaction timing. Listen before reading.

-### Consonants (quick reference)
+## 3. Choose a Reference Variety without Creating a Hierarchy

- web /w/
- yes /j/
- father /f/
- very /v/ (like /f/ but with voice)
- red /r/ (before a vowel)
- car /r/ (after a vowel)
- light /l/ (before a vowel)
- well /l/ (after a vowel)
- night /n/ (before a vowel)
- pen /n/ (after a vowel)
- mom /m/
- sing /ŋ/
- roads /dz/
- let’s /ts/
- boss /s/
- rose /z/ (voiced /s/)
- thanks /θ/
- them /ð/
- just /dʒ/
- check /tʃ/
- she /ʃ/
- Asia /ʒ/
- try /tr/
- dry /dr/
- pet /p/
- bed /b/
- too /t/
- do /d/
- kiss /k/
- go /g/
- how /h/
+English varies across regions, communities, professions, and individuals. American and British English are common reference points, not the whole language and not ranks of quality.

-Channel recommendation:
+Choose a reference variety through four questions:

- [EnglishAnyone](https://www.youtube.com/channel/UCrJHj7MDQhmQ9iFuACdoWCg) (great for speaking practice)
+1. **Real audience**: which colleagues, customers, teachers, exams, or communities do you mainly encounter?
+2. **Stable material**: can dictionary audio, courses, and feedback remain reasonably consistent over time?
+3. **Task cost**: do spelling, vocabulary, or pronunciation differences actually affect the current delivery?
+4. **Identity**: are you willing to use this voice over time, or are you trying to hide yourself inside imitation?

---
+A reference variety reduces early decisions: follow one reliable audio source for a word and keep spelling consistent inside a formal document. It does not require excluding other varieties. Real collaborators may bring English shaped by India, Singapore, Nigeria, China, the United States, Britain, or many other language histories. Listening practice must gradually add different accents, speeds, and interaction styles.

-## Speak Out Loud
+Write two boundaries:

-Speaking quietly and speaking out loud are very different.
+```markdown
+Current reference variety:
+Why it fits this task:
+Other varieties I need to understand gradually:
+Differences I only need to recognise, not force myself to imitate:
+```

-Using *New Concept English* Book 3/4 as an example:
+Consistency serves learning. Tolerance of difference serves reality. "Choose one" must not become "only one is correct."

-1. listen carefully first
-2. read out loud
-3. at the beginning, don’t worry about linking or pausing—just focus on clear, accurate sounds
-4. once you can read smoothly, start mimicking the speaker’s rhythm and intonation
+## 4. Separate Accentedness, Intelligibility, and Comprehensibility

---
+Three ideas need separate names:

-## Real Conversations: When Words Arrive Slowly
+| Idea | Meaning in this chapter | How to observe it |
+| --- | --- | --- |
+| Accentedness | How different the speech sounds from a listener's familiar reference | It shows that speech sounds different, not that the task failed |
+| Intelligibility | How much wording, meaning, and relationship the listener actually recovered | Ask for the gist, details, numbers, responsibility, and next step |
+| Comprehensibility | How much effort the listener needed to understand | Did they replay, guess, pause, or confirm repeatedly? |

-When you first speak English with another person, the idea in your head may be richer than the sentence you can produce. That is a common phase. Allow simple sentences, then practise moving the interaction forward instead of treating one pause as a verdict on your ability.
+An accent can be noticeable and easy to understand. Speech can also resemble a prestige model while unclear focus, numbers, or logic blocks the task. Listeners bring experience, familiarity, and bias, so one negative reaction must not be assigned entirely to the speaker.

-Many people speak like this at first:
+Use more than one condition: at least two listeners, a different device, or a new topic. Preserve what each listener actually heard instead of only an impression of native-likeness.

-1) think in Chinese  
-2) translate into English in their head  
-3) search for the right word  
-4) get stuck and freeze
+## 5. Repair High-Impact Pronunciation Relationships First

-After enough listening and speaking practice, your brain starts building “reflex phrases”. In certain situations you’ll respond instantly without translating.
+IPA is a map, not the destination. It can locate tongue position, airflow, voicing, and mouth shape, but it cannot replace reliable audio, your recording, and a real listener.

-That automaticity comes from repeated listening and use of high-frequency phrases, not from memorising isolated expressions.
+### Sounds and Word Meaning

-A minimum interaction turn can be:
+Practise only contrasts that currently change a word or produce recurring misunderstanding. Whether `/ɪ/` and `/i:/`, `/r/` and `/l/`, or voiced and voiceless consonants matter depends on your words, audience, and error record. Do not distribute equal time across every sound.

-1. **Confirm**: restate the key words you heard.
-2. **Respond**: give the core meaning you can express now.
-3. **Repair**: ask for repetition or try another wording when needed.
-4. **Advance**: ask one question so the conversation keeps moving.
+Use five steps: distinguish two words by ear; observe articulation; say them slowly; return them to a short sentence; choose again inside an unfamiliar sentence. If the isolated word is clear but disappears inside speech, the problem has moved from one sound to rhythm or retrieval.

-Useful repair lines include `Could you say that again?`, `Do you mean ...?`, `Let me put it another way.`, and `I need a second to think.` They are tools for continuing an interaction, not empty filler.
+### Endings, Stress, and Boundaries

---
+In real tasks, endings and stress may carry time, number, word class, or information focus:

-## Sing English Songs
+```text
+work / worked
+fifteen / fifty
+REcord / reCORD
+We need the BLUE file. / We NEED the blue file.
+```

-If you’re already in love with melodies, why not sing?
+Do not ask only whether every sound is accurate. Ask whether the listener heard the past, number, keyword, and contrast. A small difference in sound colour usually matters less than a missing critical ending or misplaced sentence focus.

-Lately I’ve been into Troye Sivan. Singing can also be a low-pressure pronunciation exercise:
+### Chunks, Pauses, and Stance

- organize the lyrics
- write down unknown words and learn them first
- sing along (yes, I actually bought a microphone)
+Divide a long sentence into breathable meaning units: context, claim, reason, condition, and next step. Pause between relationships instead of cutting a phrase at random.
+
+```text
+Based on the current test, / I recommend a smaller release, / because rollback is still available.
+```
+
+Rhythm is not a performance of authenticity. It helps the listener predict structure. Make the main line clear before adding linking, reduction, and subtle intonation.
+
+## 6. Shadowing Is Not the Destination: Move from Imitation to Generation
+
+Shadowing can reveal sounds, rhythm, and pauses. When you stay behind the recording, however, the content, word order, and next line were all decided by someone else. It cannot by itself prove that you can generate meaning in a real conversation.
+
+Move one piece of material through five levels:
+
+1. **Listen and mark**: write the gist, stress, pauses, and unclear moments.
+2. **Read aloud**: make the text clear without copying every detail.
+3. **Delayed shadow**: remain slightly behind the audio and observe rhythm without racing it.
+4. **Close and retell**: preserve meaning in your own word order.
+5. **Answer an unfamiliar follow-up**: respond to a question the material did not supply.
+
+If level four collapses, do not add endless shadowing repetitions. Shorten the material, reduce the information, and rebuild from three remembered keywords.
+
+## 7. Build Flexible Scaffolds with Chunks
+
+Speaking needs structures that can be retrieved quickly, but a complete memorised script locks fluency inside one condition. Use replaceable chunks:
+
+```text
+The main issue is ...
+What changed was ...
+I am not certain about ..., but the current evidence suggests ...
+Could we first clarify ...?
+My recommendation is ..., because ...
+```
+
+Vary each chunk by topic, position, and audience. Then add one surprise: the listener disagrees, the data is incomplete, or only thirty seconds remain.
+
+A chunk is scaffolding, not a mask. The goal is not to repeat one sentence everywhere. It is to build a main line and let the other person's response change what comes next.
+
+## 8. Interaction Repair Is Ability, Not Remediation
+
+Misunderstanding is ordinary in conversation. The danger is losing shared understanding and continuing because neither person wants to expose the break.
+
+Practise five repair types:
+
+| Situation | Repair action | Example |
+| --- | --- | --- |
+| Not heard | Ask for repetition or slower delivery | `Could you say the last part again?` |
+| Scope unclear | Confirm the question | `Are you asking about the cause or the next step?` |
+| Thinking needed | Request limited time | `Let me think for a moment.` |
+| Expression failed | Rephrase | `Let me put that another way.` |
+| Agreement unstable | Summarise and confirm | `So we will test it today and decide tomorrow. Is that right?` |
+
+Do not only memorise repair lines. Ask a partner to speed up, change the question, misunderstand a number, or challenge the evidence. Observe whether you notice the break and restore the task.
+
+For global job search or remote work, continue to [Job-search English](8-job-search-english.md) and keep project explanation, unfamiliar follow-ups, admission of unknowns, and asynchronous summary inside one scenario.
+
+## 9. Divide Work among Real Listeners, Teachers, and AI
+
+| Role | Useful work | What it cannot prove alone |
+| --- | --- | --- |
+| Self | Preserve raw audio, mark pauses, compare versions | Familiarity with your own voice can fill information the listener missed |
+| Real listener | Retell the gist, locate guessing, and continue the task | One listener may be shaped by familiarity and bias |
+| Teacher/coach | Observe articulation, rhythm, teachable problems, and sequence | Feedback still has to return to your task and transfer sample |
+| AI/speech recognition | Offer transcript candidates, generate follow-ups, locate possibly unclear segments | Results change with accent, microphone, noise, model, and network |
+
+When asking AI for feedback, submit the task and your first take before requesting analysis. Ask it to separate possible misrecognition, confident misrecognition, organisation, grammar, register, and style. Preserve the raw audio and tool version. Do not turn a recognition score into pronunciation truth.
+
+For customer, colleague, student, family, medical, or unreleased project material, obtain recording and upload permission first. Otherwise use fictional details or local processing. The value of real interaction cannot depend on crossing privacy boundaries.
+
+## 10. Put Practice Back into Life
+
+I really did buy a microphone, and I have sung along repeatedly to songs I loved. Singing cannot replace questions, repair, and real collaboration, but it lets sound leave internal judgment and enter the body. Low-pressure practice has a place: read a passage you care about, leave a voice message, sing a song, or describe something that actually happened today.

 ![speaking practice setup](../../../assets/speaking-practice.svg)

-Keep the interest and make the goal concrete. Singing is not an exam: enjoy the sound first, then notice whether sentences become easier to hear and say.
+Connect relaxed practice to a real task. Take one chunk from a lyric and use it in your own sentence tomorrow. Let a voice message lead to a follow-up. Let reading aloud lead to retelling after the page closes. Interest brings you back; evidence tells you whether you moved.

---
+A twelve-minute loop:

-## Prepare Common Questions for Yourself
+1. Two-minute unscripted first take.
+2. Three minutes to replay and mark one high-impact problem.
+3. Three minutes of sound, chunk, or repair practice.
+4. Two-minute retake.
+5. Two minutes for an unfamiliar follow-up and the next variable.

-To build “English intuition”, prepare a set of common topics and practice them repeatedly until you can speak smoothly and reorder sentences freely:
+## 11. A Fourteen-Day Speaking Experiment

- introduce yourself
- introduce your hometown
- how life has been recently
- a few things that make you happiest
- a small but unforgettable memory
- your opinion on a product (e.g., iPad Pro)
+| Day | Action | Evidence |
+| --- | --- | --- |
+| 1 | Define one real situation; record a monologue and five-turn exchange | Raw audio, conditions, listener retelling |
+| 2 | Choose a reference variety and reliable audio | Choice reason and differences not requiring imitation |
+| 3 | Find one high-impact pronunciation or rhythm problem | Error segment and task impact |
+| 4 | Practise distinction, slow production, a short sentence, and an unfamiliar sentence | Recordings under four conditions |
+| 5 | Read, delay-shadow, and retell the same material from memory | Imitation-generation difference |
+| 6 | Build five replaceable chunks | Three topic variations |
+| 7 | Remove the old script and complete the first delayed retest | New-topic 90-120 second recording |
+| 8 | Ask a partner to create one hearing or scope break | Repair process and confirmation result |
+| 9 | Change listener or device | Listener retelling under changed conditions |
+| 10 | Listen to an unfamiliar English variety and retell | Gist, details, and uncertainty |
+| 11 | Repair only the problem that still recurs | Third version and rejected low-impact advice |
+| 12 | Let AI or a partner ask three unfamiliar follow-ups | Unscripted answers and transcript errors |
+| 13 | Explain a collaboration or decision under time pressure | Conclusion, owner, and next step |
+| 14 | Close prompts, complete a new situation, and choose the next cycle | Evidence for keep, adjust, or move on |

---
+Fourteen days is not a fluency deadline. It answers a smaller question: without the familiar script, after the listener changes or a follow-up arrives, is this expression still usable?

-## Use YouTube Channels by Task
+## 12. Evidence That Speaking Is Becoming Ability

-The channels below are entry points for pronunciation and expression practice, not a subscription list. Channels can rename, stop updating, or change captions; check title, captions, access date, and task before starting. Replace or unsubscribe when a source produces no output for two weeks.
+Stronger evidence includes:

- [EnglishAnyone](https://www.youtube.com/channel/UCrJHj7MDQhmQ9iFuACdoWCg)
- [Speak English With Vanessa](https://www.youtube.com/channel/UCxJGMJbjokfnr2-s4_RXPxQ)
- [Doing English with Julian Northbrook](https://www.youtube.com/channel/UC1ZSvn6VIWqM68J_2USwH3Q)
- [A.J. Hoge](https://www.youtube.com/c/AJHogeEffortlessEnglish)
- [AccurateEnglish](https://www.youtube.com/channel/UCMTcsanYhBtOb096XegDZQA)
+- A listener can retell the gist and key details after one pass.
+- Numbers, time, responsibility, conditions, and next steps are misunderstood less often.
+- A pause or error does not end participation.
+- You can confirm a question, request repetition, rephrase, and summarise agreement.
+- Seven days later, a related task remains possible without the old script.
+- Topic, listener, device, or accent can change without destroying the task.
+- You can separate necessary repair, reference-variety difference, and identity preference.
+- You can explain why you accepted or rejected feedback from AI, a teacher, or a listener.

-Less is more. Pick a few and practise a lot. When suggesting a channel, include its task, last-checked date, and public link.
+Fluency is not packing more words into each minute or editing out every pause. It is attention moving away from "Do I sound enough like somebody else?" and back toward "Are we still understanding the same thing?"

---
+## Sources and Boundaries

-## Add Real Interaction Safely
+- [Derwing & Munro (2005), Second Language Accent and Pronunciation Teaching](https://api.crossref.org/works/10.2307%2F3588486): the review separates accent from communication outcomes, foregrounds mutual intelligibility, and notes the social consequences of accent.
+- [Levis (2005), Changing Contexts and Shifting Paradigms in Pronunciation Teaching](https://api.crossref.org/works/10.2307%2F3588485): the article discusses a shift from accent reduction toward intelligibility, identity, World Englishes, and negotiation between speaker and listener.
+- [Saito (2012), Effects of Instruction on L2 Pronunciation Development](https://api.crossref.org/works/10.1002%2Ftesq.67): a synthesis of fifteen quasi-experimental intervention studies; this chapter does not turn group averages into a personal outcome guarantee.
+- Pronunciation, vocabulary, register, and acceptability change across region, community, task, and listener. Important delivery should be checked against the current audience, reliable dictionary audio, professional feedback, and real-task retesting.

-Real interaction does not have to begin with a complete stranger. Start with a classmate, colleague, teacher, or moderated platform partner; voice messages and role-play can add uncertainty gradually.
-
-Do not share identity documents, address, financial, medical, or client information. If someone pressures you, crosses a boundary, or makes you unsafe, end the interaction and use the platform’s block/report tools.
-
-Further reading: [Why we’re afraid to talk to strangers](https://github.com/byoungd/me/blob/main/blog/3.why-we-are-afraid-to-communicate-with-strangers.md)
-
---
-
-## CEFR Task Goals
-
- **A1–A2**: complete short introductions, shopping, directions, and appointment interactions.
- **B1**: explain experiences, plans, and reasons coherently and handle common follow-ups.
- **B2**: participate fairly naturally, explain specialist matters, and maintain a position.
- **C1–C2**: organise flexibly and negotiate fine meaning in complex or pressured settings.
-
-For job search or remote work, connect speaking practice to [Job-search English](8-job-search-english.md). Project explanation is only the first layer; practise unfamiliar follow-ups, thinking time, scope confirmation, admission of unknowns, and summaries of agreement. Use the [Job-search English Evidence Card](../../templates/interview-evidence.md) to keep interaction and asynchronous writing inside one role-specific record.
-
-## Inspectable Practice and Feedback
-
-Choose a real scenario and record an unscripted two-minute first take. Transcribe it and mark only pauses, chunks, grammar, or pronunciation that affect understanding. Repair one to three issues and record again after a delay. Ask a listener to restate the message from the audio alone.
-
-When one structure repeatedly changes time, responsibility, condition, or strength of stance, do not only correct the transcript. Move it into the [Grammar Evidence Card](../../templates/grammar-evidence.md), compare intended meaning with likely interpretation, then test retrieval inside an unfamiliar follow-up.
-
-Score 0–2 for task completion, comprehensibility, organisation, interaction, and accuracy/range. A different accent is not an error; prioritise misunderstandings and persistent blocks.
-
-Put the unscripted first take, revised take, listener retelling, and unfamiliar-question transfer in the [Evidence Chain Template](../../templates/evidence-chain.md) to separate smoother speech from stable ability.
-
-## Seven Days, Thirty Days, Twelve Weeks
-
- **Seven days**: four 60–120 second outputs in one scenario, each with first and second takes.
- **Thirty days**: one monologue, one role-play, and one real or peer conversation weekly; track recurring pauses and repair strategies.
- **Twelve weeks**: add unfamiliar questions, time pressure, and real audiences; finish with a five-to-ten-minute presentation, interview, or meeting.
+Related entry points: [Listening](3-listening.md) | [Grammar](grammar.md) | [Learning English with AI](7-ai.md) | [Speaking Evidence Card](../../templates/speaking-evidence.md) | [Evidence Chain Template](../../templates/evidence-chain.md)

 ## Closing: Let Meaning Arrive

 Speaking is not removing an accent or reaching a life without hesitation. It is letting another person understand what you mean, why you mean it, and how they might respond, even when time is short and the sentence is imperfect.

-You may pause, choose another phrase, or admit that a word has not arrived. Fluency is not the absence of error. It is remaining inside the conversation after an error, repairing what failed, and listening for the place where the other person did not understand.
+You may pause, try another wording, or admit that a word has not arrived. Fluency is not permanent correctness. It is remaining in the conversation after an error, willing to repair and willing to hear where the other person did not understand.

-When you speak the first imperfect sentence and seriously wait for the answer, language stops being a performance completed alone. It becomes a relationship: meaning leaves you, finds an echo in another person's understanding, and returns carrying a new question.
+When you say the first imperfect sentence and wait seriously for the answer, language stops being a performance completed alone. It becomes a relationship: meaning leaves you, gains an echo inside another person's understanding, and returns carrying a new question.
@@ -111,6 +111,8 @@ A 12-minute session:

 Speech recognition is a clue, not proof of natural pronunciation. The stronger test is whether a trustworthy listener understands once, whether repetition is needed, and whether you can continue in a new question.

+Choose a reference variety related to the real audience, then add other English varieties gradually. Do not let a model turn "sounds different" automatically into "wrong". Use the [Speaking Evidence Card](../../templates/speaking-evidence.md) to preserve device, listener familiarity, recognition errors, listener retelling, and interaction repair, separating accentedness, actual misunderstanding, and comprehension effort.
+
 ## 6. Reading: Train the Evidence Chain

 Before AI, mark claims, evidence, definitions, examples, and unknowns. Then ask AI to compare your marks and quote the original locations. Do not accept “the article basically says”.
@@ -142,6 +142,8 @@ Change one condition per practice: wording, duration, listener, entry point, or

 Remote interviews carry delay, accent variation, device limits, and time pressure. The more complex the answer, the more it needs a line the listener can hold.

+Accent difference alone is not evidence of interview failure. Use the [Speaking Evidence Card](../../templates/speaking-evidence.md) to preserve device, listener retelling, replay points, and interaction repair. When the listener recovers the answer, evidence, and next step, do not spend role-preparation time imitating one prestige accent.
+
 A useful order is:

 1. **One-sentence answer**: answer the question directly.
@@ -8,18 +8,18 @@
      "language": "zh-CN",
      "chapters": 62,
      "images": 20,
-      "bytes": 2639115,
-      "sha256": "7eca7be3994e2d1e3f899d2dd215b2880d65759890e25361d34ca7bf6771c83b",
-      "sourceSha256": "07571d4ec27c2ab16f75865bcaf21bb5754cdcdce3a025d4882030db1915c5e4"
+      "bytes": 2644791,
+      "sha256": "ff5e4a1360babe1c1b012d070328cd415c1904abeb160f7da8c59ab9235377f5",
+      "sourceSha256": "8994664f85be8f5b98cc388615e5f19c6e3252c8dbdbe0ea401afaa3eb2817ed"
    },
    "en-US": {
      "file": "life-level-up-guide-en.epub",
      "language": "en-US",
      "chapters": 62,
      "images": 20,
-      "bytes": 2607500,
-      "sha256": "0018ef795ffec6aa208e0c477312404532b3aa49d5cd348190e4f4e8926066c1",
-      "sourceSha256": "9bc26578e3d2373cfa2e4305d708d88341ec78775f461ac5d42d90c138aedaa1"
+      "bytes": 2613125,
+      "sha256": "8057b8675901b62da8e7090883b62be584a8684014e897acb18c6a954ce4cc2b",
+      "sourceSha256": "38fb0cbb3b948962757da2b4218892892cdb41372221fd069f9828a3e2ecbdfb"
    }
  }
 }
@@ -7,14 +7,14 @@
    "zh-CN": {
      "file": "life-level-up-guide-zh.pdf",
      "language": "zh-CN",
-      "pages": 294,
+      "pages": 299,
      "chapters": 62,
-      "bytes": 4521989,
-      "sha256": "2edcd1d9637efe640a9ad173a835b5339f318dd06888f612bb4217ad861dfd2a",
-      "sourceEpubSha256": "7eca7be3994e2d1e3f899d2dd215b2880d65759890e25361d34ca7bf6771c83b",
-      "semanticSha256": "c8e8ba0cada21e39572c647a64d3d71e31c686452ce3d195d5a559522179af2a",
-      "outlineEntries": 712,
-      "linkAnnotations": 703,
+      "bytes": 4546114,
+      "sha256": "9abe42b78bcf3e3ff5c014cbfc343d7452d2759a2c5430240051a15b158841cd",
+      "sourceEpubSha256": "ff5e4a1360babe1c1b012d070328cd415c1904abeb160f7da8c59ab9235377f5",
+      "semanticSha256": "34190a29ae9ed9476ebe494624ad16dff127bafe3dd4c74e64abeede4e52cb99",
+      "outlineEntries": 719,
+      "linkAnnotations": 712,
      "images": 21,
      "fonts": [
        {
@@ -42,14 +42,14 @@
    "en-US": {
      "file": "life-level-up-guide-en.pdf",
      "language": "en-US",
-      "pages": 329,
+      "pages": 337,
      "chapters": 62,
-      "bytes": 3720134,
-      "sha256": "534f4d4bc5808a04394ce82634bd810f96159829163a8240519e2f356901d41b",
-      "sourceEpubSha256": "0018ef795ffec6aa208e0c477312404532b3aa49d5cd348190e4f4e8926066c1",
-      "semanticSha256": "f143faa8bd6f85bac103ed7c9c0842b86a3f1c4084d73b1032c7fa9861fd5caa",
-      "outlineEntries": 712,
-      "linkAnnotations": 715,
+      "bytes": 3742305,
+      "sha256": "228aa729ba548972176e715c3ebdd379194de6925bc95ec488236454df72612c",
+      "sourceEpubSha256": "8057b8675901b62da8e7090883b62be584a8684014e897acb18c6a954ce4cc2b",
+      "semanticSha256": "f68798dba0c18b6fc916fcfd12f9dbf4c56bdcdd25ecec553cbf0178b97a78a0",
+      "outlineEntries": 719,
+      "linkAnnotations": 724,
      "images": 21,
      "fonts": [
        {
@@ -42,6 +42,9 @@ updated: 2026-09-02
 | **间隔练习** | 在多个时间点回来，而不是一次集中刷完 | [词汇篇](../threads/part-1/2-vocabulary.md) |
 | **形式—意义选择** | 不把语法只看作规则名称，而是比较不同结构怎样改变时间、参与者、条件、否定范围、逻辑或确定性 | [语法篇](../threads/part-1/grammar.md) |
 | **语法对比卡** | 围绕一个真实任务保存原句、可能理解、真实意图、形式对比、修改理由与延迟复测 | [语法证据卡](../templates/grammar-evidence.md) |
+| **参考变体** | 为减少初期决策而选择的一套相对稳定的词典音频、拼写或发音参照；它不代表其他英语变体低一等 | [口语篇](../threads/part-1/5-speaking.md) |
+| **可理解度 / 理解难度** | 前者记录听众实际恢复了多少意思，后者记录理解需要多少猜测、重听或确认；两者都不同于口音显著度 | [口语篇](../threads/part-1/5-speaking.md) |
+| **听众复述** | 听众不看稿，用自己的话说出主旨、细节、数字、责任与下一步，以检查声音实际传达了什么 | [口语证据卡](../templates/speaking-evidence.md) |
 | **精听/精读** | 用短材料核对细节、结构、证据和语言形式 | [听力篇](../threads/part-1/3-listening.md) · [阅读篇](../threads/part-1/4-reading.md) |
 | **平行材料** | 难度和任务相近、但内容不同的材料，用来测试迁移 | [阅读证据卡](../templates/reading-evidence.md) |
 | **迁移** | 换主题、听众、时间或任务后仍能完成相近动作 | [九十日行动总表](../templates/90-day-cycle.md) |
@@ -63,6 +66,7 @@ updated: 2026-09-02
 | **家庭学习共同协议** | 学习者与家长分别记录目标、担心、支持、AI 与数据边界，并用十四天证据共同复查的私密工作表 | [家庭学习共同协议](../templates/family-learning-agreement.md) |
 | **求职英语证据卡** | 将真实岗位、无稿基线、故事、陌生追问、异步写作、延迟复测和诚信边界放在同一份记录中 | [求职英语证据卡](../templates/interview-evidence.md) |
 | **语法证据卡** | 一次只追踪一个高影响结构，连接真实任务基线、意义对比、四项调用、反馈分类与第 14 天迁移 | [语法证据卡](../templates/grammar-evidence.md) |
+| **口语证据卡** | 连接参考变体、三种无稿基线、听众复述、理解成本、互动修复、陌生追问与延迟迁移的私密记录 | [口语证据卡](../templates/speaking-evidence.md) |
 | **互动修复** | 没听清、范围不明或暂时未知时，请求重复、确认、争取思考时间并总结共识 | [口语篇](../threads/part-1/5-speaking.md) · [求职英语篇](../threads/part-1/8-job-search-english.md) |
 | **作品** | 能被别人阅读、使用、质疑或改进的输出，不限于产品 | [作品篇](../threads/part-3/4-artifacts-and-delivery.md) |
 | **交付** | 在真实受众、用户或约束下完成并接受结果 | [九十日行动篇](../threads/part-5/90-day-plan.md) |
@@ -1,82 +1,172 @@
 ---
-title: 口语证据卡
-description: 用录音、转写、听众复述、互动修复和延迟重录记录口语进步，不把口音相似度当成唯一标准。
-updated: 2026-08-31
+title: 口语证据卡：从口音焦虑到可验证互动
+description: 保存无稿独白、陌生追问、听众复述、参考变体、理解成本、互动修复和延迟迁移，不把口音相似度或语音识别分数当成口语能力。
+updated: 2026-09-02
 ---

-# 口语证据卡
+# 口语证据卡：从口音焦虑到可验证互动

-口语能力要在真实任务里被听见。把这张卡复制到私有笔记中，每次记录一个说明、互动或修复任务。不要写入密码、客户资料、未获授权的第三方信息或不必要的隐私。
+把这张卡复制到私有笔记中，一次只记录一个真实场景。原始录音优先保持私密；录到他人声音、客户信息、家庭内容、医疗信息或未公开项目时，先取得必要同意，或改用虚构信息练习。

-## 1. 定义任务与边界
+## 1. 定义任务、听众与边界

 ```markdown
 # Speaking Evidence — YYYY-MM-DD

 真实场景：
-任务：说明 / 互动 / 修复 / 迁移
+任务：说明 / 问答 / 修复 / 协作 / 迁移
 听众与关系：
+结束后对方需要知道或做什么：
 时间限制：
-完成标准：
-本次不公开的内容：
+允许提纲 / 重录 / 字幕 / 词典 / AI：
+录音与上传许可：
+必须保持私密的内容：
 ```

-先决定哪些内容可以被录音和分享。涉及客户、家人或同事时，取得必要同意，或改用虚构信息练习。
+完成标准要写成听众结果，例如“能复述风险和下一步”，不要只写“说得流利”。

-## 2. 保存无稿首版
+## 2. 记录参考变体与技术条件
+
+```markdown
+当前参考变体或音频来源：
+选择理由：
+只需识别、不必强迫模仿的差异：
+本轮新增的其他英语变体：
+
+设备 / 麦克风：
+网络 / 房间噪声：
+录音或通话平台：
+听众是否熟悉主题与我的口音：
+```
+
+设备、网络、听众熟悉度改变时，不要把分数差异全部解释成能力变化。
+
+## 3. 保存三种无稿基线
+
+| 样本 | 条件 | 录音位置 | 首次断点 | 是否完成任务 |
+| --- | --- | --- | --- | --- |
+| 90-120 秒独白 | 无稿说明真实经历/问题 | | | |
+| 五轮问答 | 至少一个陌生追问 | | | |
+| 一次互动修复 | 请求重复、确认范围、重述或总结 | | | |

 ```markdown
-录音位置：
-首版时长：
 我想传达的一句话主旨：
-听众需要记住的两个细节：
-首次卡顿或改口的位置：
+听众必须听见的两个细节：
+关键数字 / 时间 / 责任 / 条件：
+我最担心被误解的地方：
 ```

-不要先让 AI 重写或生成稿件。首版的价值是显示你在没有支架时能完成什么。
+不要先剪掉停顿、降噪或让 AI 生成稿件。首版要显示支架撤掉后能完成什么。

-## 3. 按影响分类问题
+## 4. 保存听众实际听见的内容

-| 片段/时间 | 问题 | 是否影响理解 | 下一次只改什么 |
-| --- | --- | --- | --- |
-| | 词块 / 发音 / 语法 / 组织 / 停顿 / 互动 | 是 / 否 | |
+请听众先听录音，不看转写：

-优先修复会改变意思、阻断任务或反复出现的问题。口音不同不等于错误；不要为了像某个说话者而牺牲清楚和自然。
-
-## 4. 记录互动修复
+| 听众/条件 | 复述的主旨 | 复述的细节 | 猜测/重听位置 | 下一步反应 |
+| --- | --- | --- | --- | --- |
+| 听众 A | | | | |
+| 听众 B 或新设备 | | | | |

 ```markdown
-我没有听清/说清的地方：
-我使用的修复句：
-对方如何回应：
-我如何确认双方理解一致：
-下一次要提前准备的词块：
+口音显著但仍然理解的地方：
+真正造成意思丢失的地方：
+理解需要额外努力但未误解的地方：
+听众反馈中可能存在的熟悉度或偏见：
 ```

-可用的修复动作包括请对方重复、确认关键词、换一种说法、请求思考时间和总结共识。它们是沟通能力的一部分，不是失败的证明。
+一次主观评价不是结论。优先保存对方听见了什么、需要做什么，而不是“像不像某种标准口音”。

-## 5. 重录与听众反馈
+## 5. 建立高影响问题地图

-| 版本 | 条件 | 任务完成 0–2 | 可理解度 0–2 | 组织 0–2 | 互动/修复 0–2 | 反馈 |
-| --- | --- | ---: | ---: | ---: | ---: | --- |
-| 首版 | | | | | | |
-| 重录 | | | | | | |
+| 片段/时间 | 听众可能听成 | 我的原意 | 类型 | 任务影响 | 本轮动作 |
+| --- | --- | --- | --- | --- | --- |
+| | | | 音 / 词尾 / 重音 / 停顿 / 词块 / 语法 / 组织 | 意思改变 / 任务阻断 / 高频重复 / 理解费力 / 低影响差异 | |

-请听众只根据录音回答：
+每轮只选一到三个问题。若差异只表明参考变体、地区或个人声音不同，且没有提高理解成本，就记录而不必强行修掉。

-1. 你听到的主旨是什么？
-2. 哪两个细节最清楚？
-3. 哪一句需要猜测？
-4. 你会追问什么？
+## 6. 从词块进入陌生追问

-## 6. 延迟迁移
+```markdown
+本轮五个可替换词块：
+1.
+2.
+3.
+4.
+5.

-| 日期 | 改变的条件 | 输出 | 结果 |
-| --- | --- | --- | --- |
-| 第 1 天 | 原场景、无稿 | 首版与重录 | |
-| 第 3–7 天 | 相近场景或新听众 | 60–120 秒复述/对话 | |
-| 第 30 天 | 新主题、时间压力或真实任务 | 5 分钟说明/会议/面试 | |
+主题变式：
+立场变式：
+听众变式：
+陌生追问：
+无稿回答位置：
+```

-如果只在熟悉稿件中变流利，换条件后仍卡住，就把下一周期重点放在词块调用和互动修复，而不是继续背稿。
+若只在背熟稿件里流畅，换条件后无法继续，就减少完整句背诵，增加关键词重构与追问。

-相关章节：[口语篇：让意思清楚到达](../threads/part-1/5-speaking.md)
+## 7. 记录互动修复
+
+| 断点 | 我怎样发现 | 修复动作 | 对方回应 | 如何确认恢复共同理解 |
+| --- | --- | --- | --- | --- |
+| 没听清 / 范围不明 / 需要思考 / 表达失败 / 共识不稳 | | | | |
+
+```markdown
+我主动请求重复或确认的句子：
+我换用的说法：
+我最后总结的共识：
+若修复失败，下一次只改什么：
+```
+
+修复不是扣分项。没有发现误解、或发现后无法恢复任务，才是需要训练的地方。
+
+## 8. 教师、听众与 AI 反馈记录
+
+```markdown
+反馈者 / 工具 / 模型版本与日期：
+它使用的是原始音频还是转写：
+语音识别转写错误：
+它判断为高影响的问题：
+它是否区分口音差异、理解错误、理解费力、语法、语域与风格：
+我怎样用词典音频、教师或真实听众复核：
+我采纳 / 拒绝什么，为什么：
+```
+
+语音识别分数会受麦克风、噪声、网络、模型和口音影响，只能作为线索。
+
+## 9. 做延迟保持与迁移
+
+| 时间 | 撤掉什么 | 改变的条件 | 输出 | 听众结果 |
+| --- | --- | --- | --- | --- |
+| 第 1 天 | 不使用 AI 稿件 | 原场景 | 首版与重录 | |
+| 第 3-7 天 | 撤掉原稿与示范音频 | 新主题或新听众 | 90-120 秒 + 追问 | |
+| 第 14 天 | 撤掉最近反馈 | 新设备、口音或时间压力 | 协作/决策说明 | |
+| 第 30 天 | 关闭全部练习提示 | 真实会议、面试或对话 | 完整任务 | |
+
+```markdown
+七天后仍然能调用的部分：
+换条件后再次失败的部分：
+听众是否更少猜测或重听：
+下一轮变量：
+```
+
+## 10. 评分与去留
+
+每项 0-2：`0=未完成`，`1=依赖提示或不稳定`，`2=当前条件下完成`。
+
+| 项目 | 第 1 天 | 第 7 天 | 第 14 天 | 证据理由 |
+| --- | ---: | ---: | ---: | --- |
+| 任务与主线完成 | | | | |
+| 听众实际恢复关键信息 | | | | |
+| 理解不需要持续猜测或重听 | | | | |
+| 能处理陌生追问 | | | | |
+| 能发现并修复互动断点 | | | | |
+| 换听众、设备或主题后仍能完成 | | | | |
+
+```markdown
+当前证据支持：继续同一问题 / 增加变式 / 转向另一问题 / 暂停
+最重要的依据：
+仍不能推出的结论：
+下一项最小任务：
+下一次复查日期：
+```
+
+相关章节：[口语篇：让意思清楚到达](../threads/part-1/5-speaking.md) ｜ [听力篇](../threads/part-1/3-listening.md) ｜ [求职英语篇](../threads/part-1/8-job-search-english.md) ｜ [证据链模板](evidence-chain.md)
@@ -25,7 +25,8 @@ updated: 2026-09-02
 | 单词认识却不会使用 | [词汇审计](vocabulary-audit.md) | [词汇篇](../threads/part-1/2-vocabulary.md) | 不能用卡片数量替代语境中的表现 |
 | 规则似乎懂了，真实表达仍反复出错 | [语法证据卡](grammar-evidence.md) | [语法篇](../threads/part-1/grammar.md) | 不能把一次改对、风格偏好或 AI 润色当成稳定语法能力 |
 | 听不清或读完说不清 | [听力资源审计卡](listening-audit.md) 或 [阅读证据卡](reading-evidence.md) | [听力篇](../threads/part-1/3-listening.md) · [阅读篇](../threads/part-1/4-reading.md) | 不能把播放时长或划线数量当成理解证据 |
-| 不敢开口或写不出来 | [口语证据卡](speaking-evidence.md) 或 [写作证据卡](writing-evidence.md) | [口语篇](../threads/part-1/5-speaking.md) · [写作篇](../threads/part-1/6-writing.md) | 不能替你完成真实听众面前的表达 |
+| 只在背稿时流畅、害怕口音或遇到追问就卡住 | [口语证据卡](speaking-evidence.md) | [口语篇](../threads/part-1/5-speaking.md) | 不能把口音相似度、语音识别分数或一次重录当成稳定互动能力 |
+| 写不出来或只能展示工具润色后的成品 | [写作证据卡](writing-evidence.md) | [写作篇](../threads/part-1/6-writing.md) | 不能替你完成真实读者面前的写作与修订迁移 |
 | 不知道怎样让 AI 参与 | [AI 任务简报](ai-task-brief.md) | [AI 学习记录](ai-learning-log.md) | 不能把模型回答自动变成事实或最终判断 |
 | 有想法却没有可交付作品 | [作品简报与交付卡](artifact-brief.md) | [AI 项目评分卡](ai-project-scorecard.md) | 不能替你找到用户、承担责任或证明商业结果 |
 | 想分析一个人物或项目故事 | [AI 经历案例复盘](ai-case-review.md) | [叙事与证据篇](../threads/part-2/narrative-and-evidence.md) | 不能把公开叙事当成独立审计 |
@@ -24,7 +24,7 @@ next:

 ### 我想提高一项具体能力

-从[CEFR 目标与自测](../part-1/0-cefr.md)确认场景和基线，再进入英语的语法、听、读、说、写章节。若规则似乎懂了，但表达仍反复改变时间、责任、条件或确定性，先读[语法篇：让结构服务于意思](../part-1/grammar.md)，再用[语法证据卡](../../templates/grammar-evidence.md)追踪一项高影响结构。每次只选一条主线，并用对应的[证据卡](../../reference/glossary.md#工具与交付)保存首版、反馈和复测。学得更多不等于学得更深；能够在新场景中完成，才开始接近能力。
+从[CEFR 目标与自测](../part-1/0-cefr.md)确认场景和基线，再进入英语的语法、听、读、说、写章节。若规则似乎懂了，但表达仍反复改变时间、责任、条件或确定性，先读[语法篇：让结构服务于意思](../part-1/grammar.md)，再用[语法证据卡](../../templates/grammar-evidence.md)追踪一项高影响结构。若只在背熟稿件里流畅、害怕口音不同或遇到追问就退出，进入[口语篇：让意思清楚到达](../part-1/5-speaking.md)，用[口语证据卡](../../templates/speaking-evidence.md)保存参考变体、听众复述、互动修复和陌生问题迁移。每次只选一条主线，并用对应的[证据卡](../../reference/glossary.md#工具与交付)保存首版、反馈和复测。学得更多不等于学得更深；能够在新场景中完成，才开始接近能力。

 ### 我正在准备海外岗位或远程面试

@@ -1,7 +1,7 @@
 ---
 title: 听力篇：从声音辨认到真实理解
 description: 按 CEFR can-do 目标训练主旨、细节、语块切分与复述，用可比较样本建立 7 天、30 天和 12 周听力计划。
-updated: 2026-09-01
+updated: 2026-09-02
 ---

 # 听力篇：从声音辨认到真实理解
@@ -82,6 +82,8 @@ updated: 2026-09-01

 第一遍记不住、跟不上、很多单词听不清，都很正常。重要的是每一轮都知道自己修正了什么。
 精听还可以帮助自己学习单词发音、连读、断句等。分析自己在听力方面的弱项，针对性的去锻炼。
+
+精听完成后，不要停在逐句跟读。沿着[口语篇](5-speaking.md)的五级路径继续：朗读、延迟跟读、合上复述，再回答材料没有提供的陌生追问。能重复原音说明观察更细，能在新问题中重组意思才开始接近口语迁移。
 我高中时采用过很笨、也很有效的一种方法：反复听《新概念英语》第三册，直到能够大致复述和拼写整篇文章。那时我并没有立刻看出变化，甚至怀疑自己在做无用功；后来回头看，真正留下来的不是某一篇课文，而是对句子节奏和常见表达的熟悉。
 听力材料可选的非常多，在这里我先简单的推荐《新概念英语》第三册和第四册。

@@ -1,156 +1,258 @@
 ---
 title: 口语篇：让意思清楚到达
-description: 按 CEFR 目标训练任务完成、可理解度、流畅、互动与修订，用录音证据建立口语进阶计划。
+description: 从无稿基线、参考变体与真实听众出发，训练可理解度、发音、词块、互动修复和陌生问题迁移，不把消灭口音当成口语目标。
 updated: 2026-09-02
+sources_checked: 2026-09-02
 ---

 # 口语篇：让意思清楚到达

-口语不是把自己训练成某一种口音，而是在有限时间里，让意思清楚、自然地到达另一个人那里。发音、词汇和语法都重要，但它们最终要服务于理解、互动和完成任务。
+我曾经把口语想成一种临场考试：开口之前要先把句子在心里排好，发音要像录音里的人，停顿最好消失，错误不能被听见。于是，越想说好，开口前需要跨过的门槛越高。
+
+真正的对话不是独自完成的表演。另一个人会追问、误解、补充、改变话题，也会带着不同口音、语速、设备和背景进入现场。口语能力因此不只住在发音里。它还包括：能否先说出主线，能否听见对方没有跟上，能否换一种说法，能否承认暂时不知道，并在一次修复之后继续共同完成事情。
+
+本章不承诺让口音消失，也不把美式、英式或任何单一声音写成英语的终点。它提供一条更可执行的路径：选择一个与你的受众相关的参考变体，保存未经修饰的录音，让真实听众复述所听见的意思；每轮只修一到三个高影响问题，再用陌生追问、不同听众和时间压力检查能力是否留下。

 ## 本章速览

- 用可靠音频和口腔动作建立可辨认的发音，而不是依赖中文谐音；
- 用朗读、复述、角色对话和真实交流把熟悉表达推到嘴边；
- 录下无稿第一版，只修最影响理解的 1–3 个问题；
- 记录听众能否复述你的主旨，用结果而不是口音评价进步。
+- 先定义说明、互动、修复或协作任务，再决定练什么；
+- 保存独白、问答与互动修复三种无稿基线，不用背稿代替口语；
+- 选择一种参考变体保持学习材料相对一致，同时逐步适应多种真实英语；
+- 区分口音显著度、可理解度和理解难度，不把“听起来不同”直接判成错误；
+- 优先修会改变词义、时间、数字、责任或句子重点的发音与节奏问题；
+- 跟读负责观察和模仿，复述、追问与协作才负责生成；
+- AI 和语音识别可以提供线索，真实听众的复述与下一步行动才是更强证据；
+- 十四天只追踪一个真实场景与少量高影响问题。

-## 先定义口语任务
+## 1. 先把口语任务说清楚

-口语训练不只是一段独白。先决定你要说明、互动、修复误解，还是把同一能力迁移到新情境，再选择练习形式。
+“提高口语”太大，无法直接训练。把它改写成一个会发生的动作：

 | 任务 | 练习条件 | 完成证据 |
 | --- | --- | --- |
-| 说明 | 60–120 秒，无稿解释经历、观点或流程 | 听众能写出你的主旨和两个关键细节 |
-| 互动 | 至少五轮问答，允许停顿和追问 | 保存对话要点，并标出你主动推进的一次提问 |
-| 修复 | 从一次真实的不确定点开始，练习澄清、重述和确认 | 记录误解如何被发现、修复和确认 |
-| 迁移 | 换人物、主题、时间压力或听众重做任务 | 比较两次任务完成度，而不是只比较口音 |
+| 说明 | 60-120 秒，无稿讲经历、观点或流程 | 听众能复述主旨和两个关键细节 |
+| 问答 | 至少五轮，包含一个陌生追问 | 回答没有完全偏题，并主动推进一次对话 |
+| 修复 | 故意保留一个没听清、范围不明或表达失败的时刻 | 双方通过重复、确认或重述恢复共同理解 |
+| 协作 | 会议、面试、客户沟通或共同决策 | 对方知道结论、分歧、责任人与下一步 |
+| 迁移 | 换主题、听众、设备、语速或时间限制 | 不依赖原稿仍能完成相近任务 |

-可以复制[口语证据卡](../../templates/speaking-evidence.md)，把录音、转写、听众反馈和下一次变量放在同一处。
+开始前写下：谁在听，为什么听，结束后对方需要知道或做什么。口语不是把所有英语都说得漂亮，而是在当前条件下完成这一个关系中的任务。

-## 音标
+可以复制[口语证据卡](../../templates/speaking-evidence.md)，把录音、听众复述、修复动作和延迟复测放在同一处。

-音标是地图，不是终点。中文近似最多只能提醒你注意口形和气流，不能替代词典音频、老师反馈或自己的录音对比。不同英语变体的音素和符号也可能略有差异，先选定一种发音资源，持续跟随即可。
+## 2. 保存三种未经修饰的基线

-先听、再看符号、再大声练习。不要为了把每个符号背得漂亮，牺牲了真实句子的节奏和可理解度。
-推荐大家跟着这个Playlist进行学习：[Teach Reading with Phonics - American English Pronunciation!](https://www.youtube.com/playlist?list=PL9BB1D7256440E08B)
+同一个人可能独白流畅，却在追问中失去方向；也可能发音清楚，却不会确认对方真正问了什么。至少保存三种样本：

-### 元音
+1. **独白基线**：无稿说 90-120 秒，说明一个真实经历或问题；
+2. **互动基线**：完成五轮问答，其中一题事先未知；
+3. **修复基线**：记录一次请求重复、确认范围、换一种说法或总结共识。

-先用下面的元音最小对立练习听辨和口形：
+每份样本保留原始录音，不先降噪、剪掉停顿或用 AI 重写。记录设备、网络、是否看提纲、是否允许重录、听众是否熟悉主题。条件不同，表现不能直接比较。

-| 对立 | 例词 | 检查点 |
+让听众只根据录音回答：
+
+```markdown
+你听到的主旨是什么：
+哪两个细节最清楚：
+哪里需要猜测或重听：
+你接下来会问什么或做什么：
+```
+
+转写稿能暴露词汇和结构，不能替代声音中的重音、停顿、语气和互动时机。先听，再看文字。
+
+## 3. 选择参考变体，不制造高低等级
+
+英语存在地区、社群、职业与个人差异。美式和英式只是常见参考入口，不是全部英语，也不是高低等级。
+
+选择参考变体时看四件事：
+
+1. **真实受众**：你主要面对哪类同事、客户、教师、考试或社区？
+2. **稳定材料**：词典音频、课程和反馈者能否长期保持相对一致？
+3. **任务成本**：拼写、词汇或发音差异是否真的影响当前交付？
+4. **个人身份**：你是否愿意长期使用这种声音，而不是把自己藏进模仿里？
+
+参考变体的作用是减少初期决策：同一个词先跟随一套可靠音频，同一份正式文本保持拼写一致。它不要求你排斥其他变体。真实协作中，听众可能来自印度、新加坡、尼日利亚、中国、美国、英国或更多语言背景；因此还要逐步增加不同口音、语速和互动风格的输入。
+
+可以给自己写两条边界：
+
+```markdown
+当前参考变体：
+选择它因为：
+我需要逐步听懂的其他变体：
+哪些差异只需识别，不必强迫自己模仿：
+```
+
+一致性服务于学习，差异耐受服务于现实。不要让“选一种”变成“只有一种正确”。
+
+## 4. 把口音、可理解度与理解难度分开
+
+三个概念需要分开：
+
+| 概念 | 本章怎样使用 | 怎样观察 |
 | --- | --- | --- |
-| /ɪ/ — /iː/ | `bit` — `beat` | 舌位、时长和句中重音是否都能听出差别 |
-| /æ/ — /e/ | `bad` — `bed` | 下颌开合与舌位是否稳定 |
-| /ʌ/ — /ɑː/ | `cut` — `cart` | 元音位置和目标变体是否一致 |
+| 口音显著度 | 听者感到声音与其熟悉参照有多大差异 | 只能说明“听起来不同”，不能单独证明任务失败 |
+| 可理解度 | 听者实际恢复了多少词语、主旨和关系 | 让听众复述主旨、细节、数字、责任和下一步 |
+| 理解难度 | 听者为了理解需要付出多少努力 | 是否需要重听、猜测、暂停或多次确认 |

-### 辅音与最小对立
+一个口音可以明显，却仍然容易理解；一句话也可能听起来接近某种标准，却因为重点、数字或逻辑关系不清而阻断任务。听者也会带着经验、偏见和熟悉度进入判断，所以不要把一次负面反应全部归因于说话者。

-中文谐音最多只能提醒你注意口形，不能提供可迁移的目标音。先选一种英语变体，使用词典音频和自己的录音，练习下面几组最小对立；不同变体的音素和符号可能略有差异。
+更可靠的判断来自多个条件：至少两位听众、不同设备或一次新主题；同时保存他们听见了什么，而不只保存“像不像母语者”的评价。

-| 对立 | 例词 | 检查点 |
+## 5. 发音训练先修高影响关系
+
+音标是地图，不是终点。它可以帮助定位口形、舌位、气流与声带状态，却不能替代可靠音频、自己的录音和真实听众。
+
+### 音与词义
+
+只练当前样本里会改变词义或造成反复误解的对立。例如 `/ɪ/` 与 `/iː/`、`/r/` 与 `/l/`、清辅音与浊辅音是否重要，要由你的词语、受众和错误记录决定，不必平均刷完所有音。
+
+使用五步：听辨两个词；观察口腔动作；慢速说；放回短句；在陌生句中再次选择。若孤立词能说对、句子里仍消失，问题已从单音转向节奏或调用。
+
+### 词尾、重音与边界
+
+真实任务中，词尾和重音可能携带时间、数量、词性或信息重点：
+
+```text
+work / worked
+fifteen / fifty
+REcord / reCORD
+We need the BLUE file. / We NEED the blue file.
+```
+
+不要只问“每个音准不准”。问听众是否听见过去、数字、关键词与对比。一个轻微音色差异的优先级，通常低于丢失关键词尾或把句子重点放错。
+
+### 句块、停顿与语气
+
+把长句按意思分成可呼吸的块：背景、主张、理由、条件、下一步。停顿放在关系之间，而不是随机切断词组。
+
+```text
+Based on the current test, / I recommend a smaller release, / because rollback is still available.
+```
+
+节奏不是为了表演“地道”，而是帮助听者预测结构。先让主线清楚，再增加连读、弱读和语气细节。
+
+## 6. 跟读不是终点：从模仿走向生成
+
+跟读能帮助观察声音、节奏和停顿，但完全跟在原音后面时，内容、词序和下一句都已经由别人决定。它不能单独证明你会在真实对话中生成意思。
+
+把同一段材料走完五级：
+
+1. **听与标记**：写下主旨、重音、停顿和没听清的位置；
+2. **朗读**：看文本说清楚，不急着复制所有细节；
+3. **延迟跟读**：与原音保持短暂距离，观察节奏而不抢速度；
+4. **合上复述**：保留意思，用自己的词序说；
+5. **陌生追问**：回答材料没有预先提供的问题。
+
+若第四级一开始就崩塌，不要回去无限增加跟读次数。缩短材料、减少信息点，练习从记住三个关键词开始重新组织。
+
+## 7. 用词块建立可变的支架
+
+口语需要一些能快速调用的结构，但背完整稿件会把流畅锁在固定条件里。更有用的是可替换词块：
+
+```text
+The main issue is ...
+What changed was ...
+I am not certain about ..., but the current evidence suggests ...
+Could we first clarify ...?
+My recommendation is ..., because ...
+```
+
+每个词块至少做三种变化：换主题、换立场、换听众。再加入一个意外条件，例如“对方不同意”“数据不完整”“只剩三十秒”。
+
+词块是脚手架，不是面具。真正的目标不是随时说出同一句，而是能借它建立主线，再根据对方的反应改变下一句。
+
+## 8. 互动修复不是补救，是能力
+
+对话中出现误解并不异常。异常的是双方已经失去共同理解，却因为害怕暴露问题继续往下说。
+
+可以练五类修复：
+
+| 情况 | 修复动作 | 示例 |
 | --- | --- | --- |
-| /θ/ — /s/ | `think` — `sink` | 舌尖、摩擦气流和声带状态 |
-| /v/ — /w/ | `vest` — `west` | 嘴唇形状与是否持续摩擦 |
-| /r/ — /l/ | `right` — `light` | 舌尖是否接触上齿龈 |
+| 没听清 | 请求重复或降低速度 | `Could you say the last part again?` |
+| 范围不明 | 确认问题 | `Are you asking about the cause or the next step?` |
+| 需要思考 | 争取有限时间 | `Let me think for a moment.` |
+| 表达失败 | 换一种说法 | `Let me put that another way.` |
+| 共识不稳 | 总结并请确认 | `So we will test it today and decide tomorrow. Is that right?` |

-先听、再看符号、再大声练习，最后把音放回真实句子的节奏、重音和停顿。不要为了把符号背得漂亮，牺牲可理解度。
+不要只背修复句。让同伴故意说快、改变问题、误解一个数字或追问证据，观察你能否发现断点并恢复任务。

-推荐的发音入口：[Teach Reading with Phonics - American English Pronunciation!](https://www.youtube.com/playlist?list=PL9BB1D7256440E08B)
+如果目标是海外求职或远程协作，进入[求职英语篇](8-job-search-english.md)，把项目说明、陌生追问、承认未知和异步总结放进同一场景。

-## 大声说英语
+## 9. 真实听众、教师与 AI 怎样分工

-大声说和默读是两种不同的训练。以《新概念英语》第三册和第四册为例，可以先听一遍，再朗读，起初只追求清楚和可辨认，不急着模仿所有连读和语气。读得顺畅后，再挑一小段模仿节奏、重音和停顿。
+| 角色 | 适合做什么 | 不能单独证明什么 |
+| --- | --- | --- |
+| 自己 | 保存原始录音、标记停顿、比较版本 | 熟悉自己的声音，容易补全听者没听见的信息 |
+| 真实听众 | 复述主旨、指出需要猜测之处、继续行动 | 一位听众可能受熟悉度与偏见影响 |
+| 教师/教练 | 观察口腔动作、节奏、可教问题与训练顺序 | 反馈仍需回到你的任务和迁移样本 |
+| AI/语音识别 | 转写候选、生成追问、定位可能不清片段 | 识别结果会受口音、麦克风、噪声、模型与网络影响 |

-一个 10 分钟闭环是：朗读 2 分钟，合上材料复述 3 分钟，回听录音 2 分钟，挑一个问题重录 3 分钟。每轮只改 1–2 项，才能听见自己的变化。
+向 AI 请求反馈时，先提交任务和自己的第一版，再要求它区分：可能听错、明确听错、表达组织问题、语法问题、语域选择和风格偏好。保留原始音频与工具版本，不把识别分数写成发音真相。

-## 交流
+涉及客户、同事、学生、家庭、医疗或未公开项目时，先取得录音与上传许可；否则使用虚构信息或本地处理。真实交流的价值不能建立在越过隐私边界之上。

-在刚开始使用英语与他人交流时，脑中的意思比嘴里的句子丰富，是很常见的阶段。先允许自己用简单句，再练习把对话往前推进，不要把一次卡顿解释成能力结论。
+## 10. 把练习放回生活

-很多人在开口说英语时，会先想好中文，再逐句翻译，找不到某个词就卡住。大量听力、复述和场景练习会让一些高频表达变得更自动，但这种自动化来自反复使用，不来自背诵几句孤立的口头禅。
-
-一个最小互动回合可以是：
-
-1. **确认**：复述你听到的关键词；
-2. **回应**：先给出当前能表达的核心意思；
-3. **修复**：听不清时请对方重复，或换一种说法；
-4. **推进**：补一个问题，让对方知道你仍在参与。
-
-可用的修复句包括：`Could you say that again?`、`Do you mean ...?`、`Let me put it another way.` 和 `I need a second to think.` 这些句子不是口头禅，而是让互动继续的工具。
-
-## 唱英文歌
-
-我想很多人都会被优美的旋律迷倒，何不尝试哼上一曲。
-最近我很喜欢戳爷的歌，也把唱歌当作一种轻松的语音练习。可以这样开始：
-
- 整理歌词，把歌词里的生词抄下来先学习
- 跟着播放器开唱，我的确买了一个麦克风！
+我确实买过一个麦克风，也曾跟着喜欢的歌反复唱。唱歌不能替代问答、修复和真实协作，但它让声音离开脑内审查，进入身体。低压力练习有它的位置：朗读一段喜欢的文字，给朋友留一条语音，唱一首歌，或用英语讲述今天真正发生的小事。

 ![口语练习示意图](../../assets/speaking-practice.svg)

-兴趣明确、目标具体，练习就更容易持续。唱歌不是考试，先享受声音，再观察自己是否更能听见和说出句子。
+关键是让轻松练习与真实任务相连：从歌词里取一个词块，第二天用在自己的句子里；从语音留言进入一次追问；从朗读进入合上材料后的复述。兴趣负责让你回来，证据负责告诉你是否前进。

-## 为自己准备一些常见的问题
+一个 12 分钟闭环可以是：

-为了培养自己的英语直觉，我推荐大家给自己准备一些常用的问题，反复练习，直到说的比较流利，能自由的改变句子的顺序为止。
+1. 2 分钟无稿首版；
+2. 3 分钟回听并标记一个高影响问题；
+3. 3 分钟做音、词块或修复练习；
+4. 2 分钟重录；
+5. 2 分钟回答陌生追问并写下下次变量。

- 介绍自己
- 介绍家乡
- 最近过得怎么样
- 令自己最开心的几件事
- 回忆一件记忆犹新的小事
- 谈谈自己对某个产品的看法，如（iPad Pro)
+## 11. 十四天口语实验

-## 按任务使用 YouTube 频道
+| 天数 | 动作 | 留下的证据 |
+| --- | --- | --- |
+| 1 | 定义一个真实场景，录独白与五轮问答 | 原始录音、条件、听众复述 |
+| 2 | 选择参考变体与可靠音频 | 选择理由与不必模仿的差异 |
+| 3 | 找一个高影响发音或节奏问题 | 错误片段与任务影响 |
+| 4 | 完成听辨、慢说、短句、陌生句练习 | 四种条件录音 |
+| 5 | 对同一材料朗读、延迟跟读、合上复述 | 模仿与生成差异 |
+| 6 | 建立五个可替换词块 | 三种主题变式 |
+| 7 | 撤掉原稿，完成第一次延迟复测 | 新主题 90-120 秒录音 |
+| 8 | 让同伴制造一次没听清或范围不明 | 修复过程与确认结果 |
+| 9 | 更换听众或设备 | 条件变化后的听众复述 |
+| 10 | 听一种不熟悉的英语变体并复述 | 主旨、细节与不确定点 |
+| 11 | 只修仍然反复的一项问题 | 第三版与拒绝的低影响建议 |
+| 12 | 让 AI/同伴提出三个陌生追问 | 无稿回答与转写误差 |
+| 13 | 在时间压力下完成协作或决策说明 | 结论、责任人和下一步 |
+| 14 | 关闭提示，完成新场景并决定下一轮 | 保留/调整/转向的依据 |

-下面只是发音和表达练习的入口，不是订阅清单。频道会改名、停更或改变字幕；开始前确认标题、字幕、访问日期和练习目标，连续两周没有产出就更换或取消订阅。
+十四天不是口语速成期限。它只用来回答一个较小的问题：在没有熟悉稿件、换了听众或出现追问之后，这项表达是否仍然可用？

- [EnglishAnyone](https://www.youtube.com/channel/UCrJHj7MDQhmQ9iFuACdoWCg)
+## 12. 怎样判断口语正在变成能力

- [Speak English With Vanessa](https://www.youtube.com/channel/UCxJGMJbjokfnr2-s4_RXPxQ)
+更强的证据包括：

- [Doing English with Julian Northbrook](https://www.youtube.com/channel/UC1ZSvn6VIWqM68J_2USwH3Q)
+- 听众一次能复述主旨和关键细节；
+- 数字、时间、责任、条件与下一步更少被误解；
+- 出现停顿或错误后能继续，而不是退出对话；
+- 能主动确认问题、请求重复、换一种说法并总结共识；
+- 七天后撤掉原稿仍能组织相近任务；
+- 换主题、听众、设备或口音后仍能完成；
+- 能区分必须修复的问题、参考变体差异与身份偏好；
+- 对 AI、教师或听众的反馈能解释采纳与拒绝理由。

- [A.J. Hoge](https://www.youtube.com/c/AJHogeEffortlessEnglish)
+流畅不是单位时间里塞进更多词，也不是把停顿全部剪掉。它是注意力逐渐从“我听起来够不够像别人”转回“我们是否仍在理解同一件事”。

- [AccurateEnglish](https://www.youtube.com/channel/UCMTcsanYhBtOb096XegDZQA)
+## 来源与边界

-> 学习频道不在多，在于多加练习；如果推荐新频道，请注明适合的任务、最近核验日期和公开链接。
+- [Derwing & Munro (2005), Second Language Accent and Pronunciation Teaching](https://api.crossref.org/works/10.2307%2F3588486)：研究综述区分口音与沟通结果，并把相互可理解性放在重要位置，同时提醒口音具有社会影响。
+- [Levis (2005), Changing Contexts and Shifting Paradigms in Pronunciation Teaching](https://api.crossref.org/works/10.2307%2F3588485)：文章讨论从减少口音转向可理解度、身份、世界英语与说听双方协商的范式变化。
+- [Saito (2012), Effects of Instruction on L2 Pronunciation Development](https://api.crossref.org/works/10.1002%2Ftesq.67)：对 15 项准实验干预研究进行综合；本章不把群体平均结果写成个人效果保证。
+- 发音、词汇、语域和可接受性会随地区、社群、任务和听众变化。重要交付应结合当前受众、可靠词典音频、专业反馈与真实任务复测。

-## 在安全边界内增加真实交流
-
-真实交流不必从完全陌生的人开始。可以先找同学、同事、老师或有明确身份的平台伙伴，也可以用语音留言和角色对话逐步增加不确定性。
-
-不要公开身份证件、住址、财务、医疗或客户信息；如果对方施压、越过边界或让你感到不安全，立即结束交流并使用平台举报/屏蔽功能。
-
-延伸阅读：[我们为何害怕与陌生人交流](https://github.com/byoungd/me/blob/main/blog/3.why-we-are-afraid-to-communicate-with-strangers.md)
-
-## CEFR 任务目标
-
- **A1–A2**：完成介绍、购物、问路、预约等短互动。
- **B1**：连贯说明经历、计划与理由，并处理常见追问。
- **B2**：较自然地参与讨论、解释专业问题并维护立场。
- **C1–C2**：在复杂或高压力场景中灵活组织表达、协商细微含义。
-
-如果目标是求职或远程协作，把口语训练接入[求职英语篇](8-job-search-english.md)：项目说明只是第一层，还要练陌生追问、争取思考时间、确认问题范围、承认未知和总结共识。使用[求职英语证据卡](../../templates/interview-evidence.md)把这些互动动作与异步写作放在同一份岗位证据里。
-
-## 可检查训练与反馈
-
-选一个真实场景，录两分钟无稿第一版；转写并标记影响理解的停顿、词块、语法和发音；只修 1–3 项；间隔后重录。邀请听众只根据录音复述你的主旨，以验证意思是否到达。
-
-若同一个结构反复改变时间、责任、条件或语气强弱，不要只在转写稿上改答案。把这一项转入[语法证据卡](../../templates/grammar-evidence.md)，比较原意、可能理解和修改理由，再回到陌生追问中测试能否即时调用。
-
-按 0–2 评分：任务完成、可理解度、组织、互动、准确与范围。口音不同不等于错误；优先处理会引起误解或持续阻断表达的问题。
-
-将无稿首版、修订版、听众复述和陌生问题迁移放进[证据链模板](../../templates/evidence-chain.md)，区分表达变顺与能力变稳。
-
-## 7 天、30 天、12 周
-
- **7 天**：围绕一个场景录 4 次 60–120 秒输出，保存首版和重录。
- **30 天**：每周一次独白、一次角色对话、一次真实或同伴交流；跟踪高频停顿和修复策略。
- **12 周**：逐步增加陌生问题、时间压力和真实听众；最终完成 5–10 分钟演讲、面试或会议任务。
+相关入口：[听力篇](3-listening.md) ｜ [语法篇](grammar.md) ｜ [用 AI 学英语](7-ai.md) ｜ [口语证据卡](../../templates/speaking-evidence.md) ｜ [证据链模板](../../templates/evidence-chain.md)

 ## 结语：让意思抵达

@@ -116,6 +116,8 @@ AI 可以生成填空、改错和替换练习，但释义、搭配和例句要

 语音识别结果只是线索，不等于自然发音。最可靠的验收是可信的听者是否能一次理解、是否需要重复，以及你能否在新问题中继续表达。

+先选择与真实受众相关的参考变体，再逐步加入其他英语变体；不要让模型把“听起来不同”自动改写成“错误”。使用[口语证据卡](../../templates/speaking-evidence.md)保存设备、听众熟悉度、识别错误、听众复述与互动修复，区分口音显著度、实际误解和理解成本。
+
 ## 六、阅读：训练证据链，而不是收集摘要

 先独立标记作者的主张、证据、定义、例子和未知项，再让 AI 比较你的标记。要求它给出原文位置，不接受“文章大意是”。
@@ -142,6 +142,8 @@ STAR（Situation, Task, Action, Result）可以帮助组织，却不能把复杂

 远程面试常受延迟、口音、设备和时间限制影响。回答越复杂，越需要先给对方一条可以抓住的主线。

+口音差异本身不是面试失败证据。用[口语证据卡](../../templates/speaking-evidence.md)保存设备、听众复述、需要重听的位置和互动修复；若对方能抓住结论、证据与下一步，就不要为了模仿单一口音牺牲岗位准备时间。
+
 一个实用顺序是：

 1. **一句结论**：直接回答问题；
@@ -276,6 +276,34 @@ test("grammar turns rules into meaning decisions and delayed evidence", () => {
  }
 });

+test("speaking prioritises intelligibility, variation, repair, and transfer", () => {
+  const cases = [
+    {
+      chapter: "threads/part-1/5-speaking.md",
+      card: "templates/speaking-evidence.md",
+      chapterTerms: ["选择参考变体，不制造高低等级", "把口音、可理解度与理解难度分开", "跟读不是终点", "十四天口语实验"],
+      cardTerms: ["设备 / 麦克风", "保存听众实际听见的内容", "语音识别分数", "第 14 天"],
+    },
+    {
+      chapter: "en/threads/part-1/5-speaking.md",
+      card: "en/templates/speaking-evidence.md",
+      chapterTerms: ["Choose a Reference Variety without Creating a Hierarchy", "Separate Accentedness, Intelligibility, and Comprehensibility", "Shadowing Is Not the Destination", "A Fourteen-Day Speaking Experiment"],
+      cardTerms: ["Device / microphone", "Preserve What the Listener Actually Heard", "Recognition scores", "Day 14"],
+    },
+  ];
+
+  for (const { chapter, card, chapterTerms, cardTerms } of cases) {
+    const chapterText = readFileSync(resolve(process.cwd(), "docs", chapter), "utf8");
+    const cardText = readFileSync(resolve(process.cwd(), "docs", card), "utf8");
+    for (const term of chapterTerms) expect(chapterText, chapter).toContain(term);
+    for (const term of cardTerms) expect(cardText, card).toContain(term);
+    expect(chapterText).toContain("api.crossref.org/works/10.2307%2F3588486");
+    expect(chapterText).toContain("speaking-evidence.md");
+    expect(cardText).toContain("part-1/5-speaking.md");
+    expect(chapterText).not.toContain("cop /ɑ/");
+  }
+});
+
 test("the entrepreneurship chapter advances through scenes instead of stacked binary contrasts", () => {
  const text = readFileSync(
    resolve(process.cwd(), "docs/threads/part-2/entrepreneurship.md"),
@@ -658,15 +686,15 @@ test("page-level search keeps nested chapter text discoverable", async ({ page }
  await page.goto("./");
  await page.getByRole("button", { name: "搜索", exact: true }).click();
  const zhSearchBox = page.locator(".VPLocalSearchBox");
-  await zhSearchBox.locator("input").fill("辅音与最小对立");
-  await expect(zhSearchBox.getByRole("link", { name: /口语篇/ }).first()).toBeVisible();
+  await zhSearchBox.locator("input").fill("过度依赖字幕");
+  await expect(zhSearchBox.getByRole("link", { name: /听力篇/ }).first()).toBeVisible();

  await page.keyboard.press("Escape");
  await page.goto("./en/");
  await page.getByRole("button", { name: "Search", exact: true }).click();
  const enSearchBox = page.locator(".VPLocalSearchBox");
-  await enSearchBox.locator("input").fill("Consonants");
-  await expect(enSearchBox.getByRole("link", { name: /Speaking/ }).first()).toBeVisible();
+  await enSearchBox.locator("input").fill("Over-relying on subtitles");
+  await expect(enSearchBox.getByRole("link", { name: /Listening/ }).first()).toBeVisible();
 });

 test("heading-only search keeps long-form chapters and tools discoverable without indexing their full prose", async ({ page }) => {
@@ -687,6 +715,10 @@ test("heading-only search keeps long-form chapters and tools discoverable withou
  await expect(zhSearchBox.getByRole("link", { name: /语法篇：让结构服务于意思/ }).first()).toBeVisible();
  await zhSearchBox.locator("input").fill("语法证据卡：从规则识别到真实表达");
  await expect(zhSearchBox.getByRole("link", { name: /语法证据卡：从规则识别到真实表达/ }).first()).toBeVisible();
+  await zhSearchBox.locator("input").fill("口语篇：让意思清楚到达");
+  await expect(zhSearchBox.getByRole("link", { name: /口语篇：让意思清楚到达/ }).first()).toBeVisible();
+  await zhSearchBox.locator("input").fill("口语证据卡：从口音焦虑到可验证互动");
+  await expect(zhSearchBox.getByRole("link", { name: /口语证据卡：从口音焦虑到可验证互动/ }).first()).toBeVisible();

  await page.keyboard.press("Escape");
  await page.goto("./en/");
@@ -706,6 +738,10 @@ test("heading-only search keeps long-form chapters and tools discoverable withou
  await expect(enSearchBox.getByRole("link", { name: /Grammar: Let Structure Serve Meaning/ }).first()).toBeVisible();
  await enSearchBox.locator("input").fill("Grammar Evidence Card: From Rule Recognition to Real Expression");
  await expect(enSearchBox.getByRole("link", { name: /Grammar Evidence Card: From Rule Recognition to Real Expression/ }).first()).toBeVisible();
+  await enSearchBox.locator("input").fill("Speaking: Make Meaning Arrive");
+  await expect(enSearchBox.getByRole("link", { name: /Speaking: Make Meaning Arrive/ }).first()).toBeVisible();
+  await enSearchBox.locator("input").fill("Speaking Evidence Card: From Accent Anxiety to Verifiable Interaction");
+  await expect(enSearchBox.getByRole("link", { name: /Speaking Evidence Card: From Accent Anxiety to Verifiable Interaction/ }).first()).toBeVisible();
 });

 test("Part I literary closings remain discoverable after bibliography pruning", async ({ page }) => {
@@ -862,6 +898,7 @@ test("home pages link to the reader guide", async ({ page }) => {
    "href",
    "./threads/part-1/grammar",
  );
+  await expect(page.locator('main a[href="./threads/part-1/5-speaking"]').first()).toBeVisible();

  await page.goto("./en/");
  await expect(page.getByRole("link", { name: "Reader's Guide", exact: true }).first()).toBeVisible();
@@ -877,6 +914,7 @@ test("home pages link to the reader guide", async ({ page }) => {
    "href",
    "./threads/part-1/grammar",
  );
+  await expect(page.locator('main a[href="./threads/part-1/5-speaking"]').first()).toBeVisible();
 });

 test("home pages expose biezou as a bounded external AI reference", async ({ page }) => {
@@ -951,6 +989,7 @@ test("toolkit overview routes readers by problem", async ({ page }) => {
  await expect(zhMain.getByRole("link", { name: "家庭学习共同协议", exact: true })).toBeVisible();
  await expect(zhMain.getByRole("link", { name: "求职英语证据卡", exact: true }).first()).toBeVisible();
  await expect(zhMain.getByRole("link", { name: "语法证据卡", exact: true }).first()).toBeVisible();
+  await expect(zhMain.getByRole("link", { name: "口语证据卡", exact: true }).first()).toBeVisible();

  await page.goto("./en/templates/toolkit");
  const enMain = page.locator("main");
@@ -961,6 +1000,35 @@ test("toolkit overview routes readers by problem", async ({ page }) => {
  await expect(enMain.getByRole("link", { name: "Family Learning Agreement", exact: true })).toBeVisible();
  await expect(enMain.getByRole("link", { name: "Job-search English Evidence Card", exact: true }).first()).toBeVisible();
  await expect(enMain.getByRole("link", { name: "Grammar Evidence Card", exact: true }).first()).toBeVisible();
+  await expect(enMain.getByRole("link", { name: "Speaking Evidence Card", exact: true }).first()).toBeVisible();
+});
+
+test("speaking pages turn accent anxiety into listener evidence and repair", async ({ page }) => {
+  await page.goto("./threads/part-1/5-speaking");
+  const zhMain = page.locator("main");
+  await expect(zhMain.getByRole("heading", { level: 2, name: /选择参考变体，不制造高低等级/ })).toBeVisible();
+  await expect(zhMain.getByRole("heading", { level: 2, name: /把口音、可理解度与理解难度分开/ })).toBeVisible();
+  await expect(zhMain.getByRole("heading", { level: 2, name: /互动修复不是补救，是能力/ })).toBeVisible();
+  await expect(zhMain.getByRole("link", { name: "口语证据卡", exact: true }).first()).toBeVisible();
+
+  await page.goto("./templates/speaking-evidence");
+  const zhCard = page.locator("main");
+  await expect(zhCard.getByRole("heading", { level: 2, name: /保存三种无稿基线/ })).toBeVisible();
+  await expect(zhCard.getByRole("heading", { level: 2, name: /保存听众实际听见的内容/ })).toBeVisible();
+  await expect(zhCard).toContainText("语音识别分数会受麦克风、噪声、网络、模型和口音影响");
+
+  await page.goto("./en/threads/part-1/5-speaking");
+  const enMain = page.locator("main");
+  await expect(enMain.getByRole("heading", { level: 2, name: /Choose a Reference Variety without Creating a Hierarchy/ })).toBeVisible();
+  await expect(enMain.getByRole("heading", { level: 2, name: /Separate Accentedness, Intelligibility, and Comprehensibility/ })).toBeVisible();
+  await expect(enMain.getByRole("heading", { level: 2, name: /Interaction Repair Is Ability, Not Remediation/ })).toBeVisible();
+  await expect(enMain.getByRole("link", { name: "Speaking Evidence Card", exact: true }).first()).toBeVisible();
+
+  await page.goto("./en/templates/speaking-evidence");
+  const enCard = page.locator("main");
+  await expect(enCard.getByRole("heading", { level: 2, name: /Preserve Three Unscripted Baselines/ })).toBeVisible();
+  await expect(enCard.getByRole("heading", { level: 2, name: /Preserve What the Listener Actually Heard/ })).toBeVisible();
+  await expect(enCard).toContainText("Recognition scores change with microphone, noise, network, model, and accent");
 });

 test("job-search English maps one real role into interview and remote-work evidence", async ({ page }) => {
@@ -972,6 +1040,7 @@ test("job-search English maps one real role into interview and remote-work evide
  await expect(zhMain).toContainText("远程岗位还在测试写作");
  await expect(zhMain).toContainText("虚构项目、职责、数据、客户或结果");
  await expect(zhMain.getByRole("link", { name: "求职英语证据卡", exact: true }).first()).toBeVisible();
+  await expect(zhMain.getByRole("link", { name: "口语证据卡", exact: true }).first()).toBeVisible();

  await page.goto("./en/threads/part-1/8-job-search-english");
  const enMain = page.locator("main");
@@ -981,6 +1050,7 @@ test("job-search English maps one real role into interview and remote-work evide
  await expect(enMain).toContainText("Remote Roles Also Test Writing");
  await expect(enMain).toContainText("Invent projects, responsibilities, data, customers, or results");
  await expect(enMain.getByRole("link", { name: "Job-search English Evidence Card", exact: true }).first()).toBeVisible();
+  await expect(enMain.getByRole("link", { name: "Speaking Evidence Card", exact: true }).first()).toBeVisible();
 });

 test("job-search evidence cards preserve unfamiliar follow-ups, async writing, and integrity", async ({ page }) => {
@@ -1099,6 +1169,8 @@ test("reader guide routes return visits to the right tools", async ({ page }) =>
  await expect(zhMain.getByRole("link", { name: "家庭学习共同协议", exact: true })).toBeVisible();
  await expect(zhMain.getByRole("link", { name: "语法篇：让结构服务于意思", exact: true })).toBeVisible();
  await expect(zhMain.getByRole("link", { name: "语法证据卡", exact: true })).toBeVisible();
+  await expect(zhMain.getByRole("link", { name: "口语篇：让意思清楚到达", exact: true })).toBeVisible();
+  await expect(zhMain.getByRole("link", { name: "口语证据卡", exact: true })).toBeVisible();

  await page.goto("./en/threads/part-0/reader-guide");
  const enMain = page.locator("main");
@@ -1111,6 +1183,8 @@ test("reader guide routes return visits to the right tools", async ({ page }) =>
  await expect(enMain.getByRole("link", { name: "Family Learning Agreement", exact: true })).toBeVisible();
  await expect(enMain.getByRole("link", { name: "Grammar: Let Structure Serve Meaning", exact: true })).toBeVisible();
  await expect(enMain.getByRole("link", { name: "Grammar Evidence Card", exact: true })).toBeVisible();
+  await expect(enMain.getByRole("link", { name: "Speaking: Make Meaning Arrive", exact: true })).toBeVisible();
+  await expect(enMain.getByRole("link", { name: "Speaking Evidence Card", exact: true })).toBeVisible();
 });

 test("echoes chapter separates harm, responsibility, and the next choice", async ({ page }) => {