# 英语高级 · 第 3 周 Day 6|强度排序与错句复盘 > 系列:语种学习 / 英语 / 高级 · 每日学习|AI生成|2026-10-27|codebuddy **今天学什么** 今天解决"写的时候觉得清楚,别人读起来分不出"。判断力不能自证,只能靠盲测:把五版遮住档位,隔一天再排,排对了才算真的写出了差异。今天 60 分钟做两件事——盲排自己的五版并归因,再做 12 句改错。归因比分数重要:错位集中在相邻档,说明装置不典型;错位跨三档,说明命题内核早就变了。 ## 一、今日任务 | 环节 | 时长 | 具体动作 | 产出 | | --- | --- | --- | --- | | 盲排 | 20 min | 遮住档位与标题,把五段打乱,自己排一遍;再请一个人排一遍 | 两份排序 | | 归因 | 10 min | 逐处比对设计与实际,按三分类归因(装置/限定/证据) | 归因表 | | 改错 | 20 min | 12 句改错,每句写"错在哪一类",不写"我觉得" | 12 句改错 | | 记录 | 10 min | 错句本只增 3 条;写一条明日可执行的修正动作 | 错句本 + 1 条动作 | **盲排规则**:五段复制到新文档,删掉所有档位标记,随机排序;排的时候不许回看原稿,不许猜"我大概会怎么写"。自己排与请人排要分开做,先自己后别人,否则会被对方的顺序带跑。 **60 分钟怎么切**:这一天没有输入环节,全部用于诊断,所以时间要偏紧。盲排 20 分钟里,自己排 8 分钟、请人排 8 分钟、比对 4 分钟;归因 10 分钟只填表不解释;改错 20 分钟按"倒装—虚拟—省略—强度"顺序推进,每类 5 分钟;最后 10 分钟只做两件事:错句本记 3 条、写下明天要改的一个动作。超时就砍归因,不能砍改错——改错是唯一能直接迁移到下周的部分。 ## 二、核心词汇与搭配 | 词条 | 音标 | 词性 | 释义 | 高频搭配 | 语域 | | --- | --- | --- | --- | --- | --- | | calibration | /ˌkælɪˈbreɪʃn/(≈凯利布雷希恩) | n. | 校准,标定 | calibration of judgement | 学术 | | blind | /blaɪnd/(≈布莱恩德) | adj. | 盲测的,不知情的 | a blind ranking; blind review | 学术/通用 | | discriminate | /dɪˈskrɪmɪneɪt/(≈迪斯克里米内特) | v. | 区分,辨别 | discriminate between A and B | 学术 | | confound | /kənˈfaʊnd/(≈克恩福恩德) | v. | 混淆,使难以区分 | confound two effects | 学术 | | conflate | /kənˈfleɪt/(≈克恩夫莱特) | v. | 混为一谈 | conflate strength with evidence | 学术/媒体 | | granularity | /ˌɡrænjuˈlærəti/ | n. | 颗粒度,细化程度 | at this level of granularity | 学术/商务 | | slippage | /ˈslɪpɪdʒ/(≈斯利皮知) | n. | 滑移,偏离 | slippage between plan and delivery | 商务/政策 | | overclaim | /ˌəʊvəˈkleɪm/(≈欧弗克莱姆) | v. | 过度断言 | overclaim on the evidence | 学术/媒体 | | benchmark | /ˈbentʃmɑːk/(≈贝恩奇马克) | n. | 基准,参照点 | set a benchmark; against the benchmark | 商务/学术 | | variance | /ˈveəriəns/(≈怀里厄恩斯) | n. | 差异,偏差 | variance between raters | 学术/统计 | | # | 英文例句 | 中文 | | --- | --- | --- | | 1 | The calibration of your own judgement is the point of the blind test. | 盲测的意义在于校准你自己的判断。 | | 2 | In a blind ranking, the writer should not recognise his own versions. | 盲排时,作者不该认出哪段是自己写的。 | | 3 | Readers could not discriminate between the top two versions. | 读者分不出最高那两版。 | | 4 | Strong evidence and strong wording are easily confounded. | 强证据与强措辞很容易被混为一谈。 | | 5 | Do not conflate confidence with accuracy. | 不要把自信当成准确。 | | 6 | At this level of granularity, the difference is not meaningful. | 在这个颗粒度上,差别没有意义。 | | 7 | The slippage between what was designed and what was written is the finding. | 设计与写出的东西之间的偏离,本身就是发现。 | | 8 | The paragraph overclaims on a single case. | 这段凭一个案例就过度断言。 | | 9 | Hold the version against the benchmark set on Day 2. | 拿这一版对照 Day 2 定的基准。 | | 10 | Most of the variance came from two of the five versions. | 大部分偏差来自五版中的两版。 | > **语域提示**:calibration 在学术里指"判断与真值的吻合度",不是"调整参数";conflate 与 confound 常被混用,前者是"把两件事说成一件",后者是"两个因素纠缠导致无法分辨",改错时用词要准。 ## 三、语言点:盲排的三种归因 排序错了不是"感觉不准",而是三种可诊断的机械原因。诊断清楚,改一句就行;诊断不清,重写五遍还是错。 **规则** 1. **盲排必须双份**:自己一份、他人一份。自己排对而他人排错,说明你的设计依赖内部记忆;两份都错,说明装置本身不典型。 2. **合格线是一致 ≥4 档**,即五版中至少四版的相对顺序与设计相同。允许一处错位,因为 2 档与 3 档在语感上本就接近。 3. **错位位置决定原因**:错在相邻档(2↔3、4↔5),是装置不典型;错在跨档(1↔4),是命题内核或证据被改动。 4. **限定词泄漏是最隐蔽的一类**:低档句忘了加限定,高档句却加了 `arguably`,两档被拉平。检查方法是把五版的限定成分单独抄成一列,看是否单调递减。 5. **改错要写类名,不写感受**。写"倒装范围错"可以复用到下一句,写"读着别扭"什么也留不下。 **为什么 2 档与 3 档天然接近** 这两档的差别不在概率,而在"谁在负责"。`It may be that X` 是作者说"我只承认可能",`The evidence suggests that X` 是作者说"证据指向这里"。听众若只抓命题内容,两句话说的是同一件事,自然排不出来。所以盲排时允许这一对出错,也正因如此,设计时更要用标准锚句,不要自创 `it is possible to think that` 这类中间态——刻度越自创,越测不出。 **归因表(照此填写)** | 现象 | 归因 | 处理 | | --- | --- | --- | | 2 档被排成 3 档 | 装置不典型:用了 may be 而非标准锚句 | 换回 `It may be that…` | | 4 档被排成 5 档 | 限定词泄漏:高档句多了一个 arguably | 删掉多余的 hedge | | 1 档被排成 4 档 | 命题内核被改:低档句换了证据 | 恢复共用证据 | | 五版被排成同一档 | 第三格全部写成套话 | 重写第三格,各给不同退路 | **常见误用** - **用"我觉得不够强"当结论**:这不是归因,是重复症状。必须落到具体装置或具体词。 - **改错只抄正确答案**:抄一遍记不住,写"错在哪一类 + 下次怎么查"才记得住。 - **盲排前重读原稿**:记忆一启动,测试就失效,等于自己给自己泄题。 - **把 5 档写成 4 档还自我说服**:undoubtedly 与 it is likely 之间没有中间态,硬塞 `very likely` 会让刻度失真。 **什么时候用、什么时候不能用** 每次交付重要文本前用一次,尤其是摘要、提案、备忘录的结论段;日常练习不必每次都盲排。用得对,你会知道自己默认偏高还是偏低;用得滥,会把写作变成 calibration 游戏,反而写不出东西。 ## 四、句型与范例 | # | 句式 | 用途 | | --- | --- | --- | | 1 | Version 3 reads as version 2 because the modal does the work of a hedge. | 归因句,指向装置 | | 2 | The evidence is identical across all five; only the framing moves. | 说明内核未变 | | 3 | Two raters placed it one tier apart, which is within tolerance. | 报告容差 | | 4 | The slippage is systematic, not random: every low-tier version lost its qualifier. | 指出系统性偏差 | | 5 | This version overclaims relative to the benchmark. | 对照基准 | | 6 | The paragraph conflates strength of wording with strength of evidence. | 指出混淆 | | 7 | At this granularity, the distinction between 2 and 3 is not reliable. | 承认测不出 | | 8 | Rater variance was concentrated in the middle of the scale. | 描述分布 | | 9 | Restoring the shared evidence removed the cross-tier error. | 说明修正有效 | | 10 | The blind test failed on one pair; the rest held. | 结论句 | **示范:一段 100 词的复盘结论** > Two raters placed the five versions in the same order except for one pair: the designed 2 was read as a 3 by both. The evidence is identical across all five, so the fault is in the device rather than the content. Version 2 used `may cut` without the `it may be that` frame, which readers took as a hedged prediction rather than a bare possibility. Restoring the frame removed the error. The remaining variance sits between 4 and 5, within tolerance. 中文:两位评分者排出的顺序一致,只有一对例外:设计的第 2 档被两人都读成第 3 档。五版证据完全相同,所以问题在装置而非内容。第 2 档只用了 may cut,没有 `it may be that` 的框架,读者把它当成"带保留的预测"而不是"纯粹的可能性"。恢复框架后误差消失。剩下的偏差在 4 与 5 之间,属容差范围。 这段的价值在于:它没有说"我排错了",而是定位到"哪一个词导致哪一档被读高",并给出了可复用的修正。 ## 五、练习 **A 词汇与元语言(8 题)** 1. 用 calibration 写一句,说明盲排的目的。 2. 改正:`The two effects are easy to conflate each other.` 3. 用 discriminate 写一句,说明读者分不出 4 档与 5 档。 4. 用 overclaim 写一句,指出某段凭单一案例下结论。 5. 改正:`The variance between raters are small.` 6. 用 granularity 写一句,说明当前刻度太粗。 7. 用 benchmark 写一句,说明用 Day 2 的表作参照。 8. 用 slippage 写一句,说明设计与交付之间的偏离。 **B 改错(6 组,12 句)**:每句写出错类名 + 改正。 倒装 1. Not only the firm lost the contract, but it also lost its licence. 2. Rarely the board has been so divided. 3. Only when the figures were revised the scale of the shortfall became clear. 4. Under no circumstances the fund should be used for this purpose. 虚拟 5. If the regulator would intervene, the merger would collapse. 6. Were it not the delay, the project would have finished on time. 7. Had the board acted sooner, the loss was avoided. 省略与非谓语 8. The scheme, if implementing at all, will take a decade. 9. Corrections should be made where is necessary. 强度 10. Undoubtedly, it may be that the policy has worked. 11. One could argue that undoubtedly the evidence proves the case. 12. The data suggests that the trend may possibly be reversing, perhaps. **C 输出(2 题)** 1. 盲排记录:两份排序 + 归因表(现象/归因/处理三列)。 2. 12 句改错稿 + 错句本新增不超过 3 条。 ## 六、参考答案 **A** 1. `Blind ranking is a calibration exercise, not a test of taste.` 强调目的是校准判断。 2. `The two effects are easily conflated.` 或 `easy to confuse with each other`。conflate 是及物动词,不接 each other。 3. `Readers could not discriminate between versions 4 and 5.` discriminate between 是固定搭配。 4. `The paragraph overclaims on the basis of a single case.` overclaim on 后接证据。 5. `The variance between raters is small.` variance 不可数,配 is。 6. `At this level of granularity, the scale cannot separate 2 from 3.` 7. `Hold each version against the benchmark set on Day 2.` hold against 是"对照检验"。 8. `The slippage between design and delivery was systematic.` **B** 1. 助动词缺失。`Not only did the firm lose the contract, but it also lost its licence.` 句中无助动词,必须借 did。 2. 倒装未发生。`Rarely has the board been so divided.` 频率副词前置引起倒装。 3. 倒装范围错。`Only when the figures were revised did the scale of the shortfall become clear.` 从句不倒,主句倒。 4. 倒装未发生。`Under no circumstances should the fund be used for this purpose.` 否定介词短语前置,情态动词前置。 5. 虚拟混用。`Were the regulator to intervene, the merger would collapse.` 条件从句不能用 would。 6. 省略不当。`Were it not for the delay, the project would have finished on time.` for 不能省。 7. 主句时态错。`Had the board acted sooner, the loss would have been avoided.` 与过去事实相反,主句用 would have done。 8. 非谓语错。`The scheme, if implemented at all, will take a decade.` 被动关系用过去分词。 9. 省略过度。`Corrections should be made where necessary.` 删到可复原为止,不能删出残缺。 10. 强度叠用。`Undoubtedly, the policy has worked.` 或 `It may be that the policy has worked.` 强化语与情态不能同现。 11. 强度叠用 + 越档。`The evidence suggests that the case is strong.` 归属与 booster 并存会让读者无法定位作者立场。 12. 三重 hedge。`The data suggest that the trend may be reversing.` 留一层即可,另注意 data 在正式文体中常作复数。 **改错的优先顺序** 12 句里如果时间不够,先改倒装(4 句)与强度(3 句)。倒装错在结构,一错整句读不通,考试与工作中最致命;强度错在语用,改起来最快,往往删一个词就完成。虚拟的错多半是 would 位置,规则单一,练三次就稳定;省略的错最难自查,因为删对了读起来"本来就这样",删错了才别扭,所以放到最后,靠朗读判断。这个顺序不是重要性排序,而是"投入产出比"排序:同样 5 分钟,改倒装能救 4 句,改省略只能救 1 句。 **C 评分维度** - 两份排序齐全(3 分);一致 ≥4 档(5 分);归因表每行落到具体装置/词(5 分);12 句改错错类名正确 ≥10 句(5 分);错句本 ≤3 条且每条附"下次怎么查"(2 分)。 **参考:一句合格与一句不合格的归因** 合格:`Version 2 was read as 3 because may cut without the it-may-be-that frame signals a hedged prediction; restoring the frame fixed it.` 不合格:`Version 2 was read as 3; I think the wording was not strong enough.` 后者没有指出任何装置,下次还会错。 ## 七、今日输出与自查 - [ ] 文件 `Day06_盲排与改错.md`,含两份排序、归因表、12 句改错。 - [ ] 排序一致 ≥4 档;不足 4 档,说明要回到 Day 5 重写第三格,而不是调副词。 - [ ] 归因表每一行都指向具体装置或具体词,没有"语气不够"这类描述。 - [ ] 12 句改错全部写出错类名(助动词缺失/倒装范围/虚拟混用/省略过度/强度叠用)。 - [ ] 错句本新增 ≤3 条,每条写"下次怎么查",例如"看到 Only when 先找主句"。 - [ ] 把五版的限定成分单独抄成一列,检查是否随档位单调变化;出现"低档没限定、高档有限定"就重排。 ## 八、明日预告 Day 7 是本周高压日:抽一个题目准备 10 分钟,做 5 分钟陈述录音,回听写 3 个具体问题。陈述必须含立场、2 个证据、1 处让步,hedge 不超过 5 处。今天改错里那些"强度叠用"的句子,明天在口语里会以另一种形式复发:口语不会写出 `undoubtedly…may`,而是先说"毫无疑问",停两秒,再补一句"当然也可能"。提前留意这个组合,听到自己这样说就立刻二选一。