英语长句并不是许多单词的随机堆积。它通常只有一个有限动词承担主句判断，其余信息围绕这个判断扩展。阅读的第一步不是翻译，而是确认谓语要求哪些参与者，以及这些位置是否已经填满。
为什么认识句中每个单词，仍可能误解整句？因为词义不能告诉你谁支配谁。我们要区分主句、从句和短语，并从动词的论元结构出发寻找主干。
give 通常要求给予者、对象和接受者。be, seem, become 等系词后的补语不能随意删除。The committee rejected the proposal.
rejected 是有限动词，要求“谁拒绝”和“拒绝什么”。The committee 与 the proposal 分别填入两个论元位置。
The committee rejected the proposal after a brief discussion.
after a brief discussion 说明时间，删除后核心判断仍完整，因此是附加语。
The evidence suggests that the treatment may reduce anxiety.
主句谓语是 suggests；它的内容论元不是单个名词，而是整个 that 从句。不能把 reduce 误当主句谓语。
What the participants remembered remained unclear.
remained 是主句中心。What the participants remembered 整体充当主语，内部的 remembered 属于嵌入分句。
Researchers working under severe time constraints often overlook alternative explanations.
working 没有独立时态，只修饰 Researchers；主句有限动词是 overlook。
The assumption that markets always correct themselves has shaped public policy.
has shaped 承担主句判断。that markets... 说明 assumption 的内容，不与 assumption 构成主谓关系。
There appears to be no simple solution.
there 是形式位置，真正需要理解的存在对象是 no simple solution。主干不是“那里出现”，而是“似乎不存在简单方案”。
Whether the effect persists depends on how the outcome is measured.
主句谓语是 depends on；Whether... 是主语分句，how... 是介词 on 的补足语。
Scientific models do not reproduce every feature of the world. They select a limited set of relationships and represent those relationships in a tractable form. What a model excludes may matter as much as what it includes, because an omitted factor can determine where the model stops being reliable. A careful reader therefore asks not only what the model predicts but also which assumptions make that prediction possible.
第一句主干是 models do not reproduce features；第二句有共享主语的两个并列谓语 select 与 represent。第三句以两个 what 分句形成对比，第四句的 asks 支配两个并列的间接疑问分句。段落主线是“模型选择信息，因此必须检查选择的边界”。
训练语境：学术科研
找出主干：The results reported in the appendix do not support the stronger hypothesis.
主干是 The results do not support the stronger hypothesis。reported in the appendix 是过去分词短语，修饰 results；它没有独立时态，不能充当主句谓语。
训练语境：学术科研
说明 that 从句的功能：The claim that the policy reduced inequality remains controversial.
主句是 The claim remains controversial。that the policy reduced inequality 给出 claim 的具体内容，是同位内容从句；它不是 claim 的宾语，也不是主句。
训练语境：学术科研
为什么不能删除 valid：The reviewers considered the argument valid.
consider 在这里要求宾语 the argument 和宾语补语 valid，共同表达“认为论证有效”。删除 valid 后只剩“审稿人考虑了该论证”，动词意义和命题都改变了。
训练语境：学术科研
找出两个层级的谓语：What the survey reveals challenges what the theory predicts.
主句谓语是 challenges。主语分句内部谓语是 reveals，宾语分句内部谓语是 predicts；两个 what 分句分别占据主语和宾语位置。
训练语境：学术科研
判断 using national data 修饰谁：Using national data, the team estimated the effect.
句首现在分词短语的逻辑主语默认与主句主语一致，因此是 the team 使用全国数据。若实际使用数据的是别人，这个结构就会产生悬垂修饰。
训练语境：学术科研
解释主干：There is little evidence that the intervention caused the change.
这是存在结构，核心判断是“几乎没有证据”。that... 说明 evidence 的内容；there 不指地点，不能译作“在那里”。
训练语境：真实生活
找出必要成分：The delay made everyone anxious.
made 是使役结构，要求原因 The delay、受影响对象 everyone 和结果状态 anxious。anxious 是宾语补语，不是可随意删除的描述。
训练语境：真实生活
分析：What you said sounded reasonable at the time.
主句谓语是系词 sounded，补语是 reasonable。What you said 整体作主语，at the time 是时间附加语。
训练语境：真实生活
解释歧义：I saw the man with the telescope.
介词短语可能修饰 saw，表示“我用望远镜看见”；也可能修饰 the man，表示“我看见拿望远镜的人”。句法允许两种附着，必须借语境消歧。
训练语境：真实生活
找主句谓语：The package that arrived yesterday appears to be damaged.
主句有限动词是 appears，to be damaged 是它的补语。arrived 属于修饰 package 的关系从句，不能把昨天到达误作全句最终判断。
In debates about automated diagnosis, the most consequential disagreement is not whether an algorithm can identify patterns that escape a clinician's immediate attention. It is whether the institution that deploys the algorithm can explain which kinds of uncertainty the system has been trained to ignore. A model that performs well on historical data may still fail when the clinical population changes, because the relationships it has learned are often relationships among recorded variables rather than among the underlying biological processes. For that reason, a responsible evaluation asks not only how accurate the model is, but also what has to remain true for that accuracy to mean what its advocates claim.
本段是原创论文语域仿真材料，不复制论文正文；训练目标是让你读真实论文时能主动恢复主干、论元和附加信息。
the most consequential disagreement is not whether... 的主干是 disagreement is not whether...，about automated diagnosis 只是议题范围。It is whether...，It 回指上一句的 the most consequential disagreement。A model may fail；that performs well... 是定语从句，because... 给出失败原因。a responsible evaluation asks...，后面两个 what/how 从句是 asks 的内容论元。高阶语境：学术科研
请写出第三句的最小主干，并说明哪些成分不能删、哪些成分只是解释条件。
第三句最小主干是 A model may fail。但若为了理解论文论证，不能只停在这三个词上，还要保留 when the clinical population changes，因为它限制失败发生的情形；because... 是原因从句，用来解释为什么历史数据表现好仍可能失败。that performs well on historical data 修饰 model，不是主句谓语。
高阶语境：学术科研
第二句中的 which kinds of uncertainty the system has been trained to ignore 在句法上充当什么成分？为什么这不是一个独立问句？
它是 explain 的宾语内容，是嵌入式疑问分句。独立问句会使用疑问语序，如 which kinds ... has the system...；这里保持陈述语序 the system has been trained，说明它被嵌入到 can explain 后面，作为“解释的内容”。
高阶语境：真实生活
把最后一句改写成生活场景：比如评价一个导航软件。要求保留 not only..., but also... 的双重追问结构。
可以写成：A careful user asks not only how fast the navigation app finds a route, but also what has to remain true for that route to be reliable. 这里 not only 后问速度，but also 后问可靠性的前提，结构与原句一致。
遇到长句时先圈出所有动词，再问哪些动词有独立时态。随后检查主句动词要求的位置是否填满。若删去某成分后动词意义改变，它更可能是补语而不是附加语。