Unilanguage Experimental Research · UNI-EXP-001A-R2
Can a proposed cross-language pattern survive a predefined empirical test?
M → HUMAN 语义关联：一个跨语言观察，能否经受预先确定的实证检验？
Exploratory AI Result
Exploratory AI Result — Inconclusive. Both frozen passes point in the predicted direction, but every primary and evidence-family 95% confidence interval includes zero and every directional Fisher test has p > 0.05. This result does not replace or decide the preregistered human blind analysis.
探索性 AI 结果——无法判断。两份冻结标注均指向预测方向，但所有主要分析和证据家族分析的 95% 置信区间都包含 0，且所有单侧 Fisher 检验均为 p > 0.05。本结果不替代、也不裁定预注册真人盲标分析。
Research question
This experiment tests whether English words beginning orthographically with M are disproportionately associated with the semantic domain HUMAN, compared with predefined control initial-letter groups. It tests a lexical distributional tendency—not shared etymology, universal sound symbolism, or causation.
本实验检验：英语中以字母 M 开头的词，是否相对于预先确定的对照首字母组，更集中地分布于 HUMAN（人类）语义场。本实验只检验词汇分布倾向，不证明共同词源、普遍语音象征或因果关系。
Pilot study
The Pilot made the M → HUMAN proposal explicit and showed that it could be converted into a testable comparison. It also exposed major sources of researcher freedom: hand-picked examples, unstable semantic boundaries, related-word duplication, and control selection.
Pilot 将 M → HUMAN 观察转化为可比较的假说，同时暴露了人工选词、语义边界不稳定、同族词重复计数和对照选择等研究者自由度。
The Pilot is exploratory evidence only. It cannot decide the hypothesis. Round 2 therefore freezes definitions, sampling, annotation, counterexample handling, and reporting rules before outcomes are inspected.
Pilot 只提供探索性证据，不能裁决假说。Round 2 因而在查看结果前冻结定义、抽样、标注、反例处理和报告规则。
Round 2
The 31-section preregistration defines the claim, categories, sampling principle, controls, frequency matching, evidence-family deduplication, blind coding, analyses, reporting standard, and decision rule.
31 节预注册文件已经固定假说范围、语义类别、抽样原则、对照组、频率匹配、证据家族去重、盲标、分析、报告和结论规则。
View Full 31-Section Preregistration | 查看完整31项预注册规则Immutable public record · 不可覆盖的公开记录： Version v1.0 · 2026-08-19 · Jinkai Liu · UNI-EXP-001A-R2 · Frozen / Preregistered. Future framework versions do not replace this record.
Formal hypotheses
M-initial words are not more likely to belong to HUMAN than words in the predefined controls.
M 开头的词进入 HUMAN 语义场的比例不高于预先设定的对照组。
M-initial words are more likely to belong to HUMAN than words in the predefined controls.
M 开头的词进入 HUMAN 语义场的比例高于预先设定的对照组。
Method
Eligible modern English lexical items are sampled from a predefined frequency source. The M group and pooled controls are matched as closely as practical across fixed frequency bands.
从预先确定的现代英语词频来源抽样，并在固定频段内尽可能匹配 M 组与合并对照组。
Related forms receive an Evidence Family ID. Results are reported both by eligible lexeme and after related forms are collapsed.
相关词形获得 Evidence Family ID；结果同时报告词项层分析与证据家族去重分析。
HUMAN coding uses the primary or highest-frequency contemporary dictionary sense. Rare or historical senses cannot rescue a favorable case.
HUMAN 判断采用当代词典的第一义或最高频义；罕见义和古义不能用于事后挽救支持案例。
Bias control
标注者只看到词项、标准词典释义、固定分类说明和语义标签。
标注界面不出现 M 假说名称、预期方向、Pilot 结果或所谓支持词表。
Two annotators code each item independently. Initial disagreements, agreement statistics, adjudication, and both original labels are preserved.
每个词项由两名标注者独立编码；原始分歧、一致性统计、裁决结果及两份原标签全部保留。
Counterexamples policy
Every eligible M-initial word remains in the dataset regardless of category. Words such as machine, metal, mountain, music, and minute, if sampled, cannot be removed because they weaken the proposal.
所有符合资格的 M 开头词都必须留在数据中。若 machine、metal、mountain、music、minute 等词进入样本，不能因其削弱假说而被删除。
Supporting cases · Counterexamples · Neutral cases · Excluded cases with reasons
最终数据同时保存：支持案例、反例、中性案例，以及附排除理由的排除案例。
Reporting standard
The interpretation prioritizes effect direction, magnitude, confidence interval, robustness after evidence-family deduplication, and then statistical significance.
解释顺序优先考虑效应方向、效应大小、置信区间、证据家族去重后的稳健性，最后才是统计显著性。
Round 2 outcomes
Both frozen AI passes show a modest positive M-minus-Control HUMAN difference, while both 95% confidence intervals include zero. The evidence-family robustness analysis retains the same direction but remains imprecise. Under the requested three-level vocabulary, the result is Inconclusive. This is an Exploratory AI Result; Preregistered Human Annotation: Pending.
两份冻结 AI 标注均显示 M 组 HUMAN 比例略高于对照组，但两个 95% 置信区间都包含 0；证据家族去重分析保持同一方向，但精度仍不足。按本次要求的三级结论词汇，本结果为无法判断。本结果仅为探索性 AI 结果；预注册真人标注：待进行。
Decision record
Exploratory AI Result — Inconclusive: the estimated HUMAN effect is positive and consistent across AI-A, AI-B, and fixed-family robustness analyses, but all uncertainty intervals include zero and all directional Fisher tests exceed 0.05. The archived v1.0 five-level wording “Tentatively Supported” is preserved as a historical record; this three-level classification is the current public-safe conclusion. No preregistered human status has been assigned.
探索性 AI 结果——无法判断：AI-A、AI-B 与冻结证据家族稳健性分析均给出正向 HUMAN 效应，但所有不确定性区间都包含 0，且所有单侧 Fisher 检验均超过 0.05。v1.0 采用五级词汇的“初步支持”作为历史记录原样保留；当前公开安全的三级结论为“无法判断”。预注册真人结论尚未分配。
We publish hypotheses—including hypotheses our own experiments fail to support.
我们公开记录假说，也公开记录实验未能支持的假说。