Experiment 001A — Round 2 Preregistration v1.0

1. Research Question｜研究问题

English

This experiment tests whether English words beginning with the consonant M are disproportionately associated with the semantic domain HUMAN, compared with predefined control initial-letter groups. It tests a statistical tendency rather than shared etymology.

中文

本实验检验：英语中以辅音 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。本实验检验统计性语义倾向，而不是证明共同词源。

2. Hypotheses｜假设

English

Null hypothesis — H₀

Words beginning with M are not more likely to belong to HUMAN than words in the predefined controls.

Research hypothesis — H₁

Words beginning with M are more likely to belong to HUMAN than words in the predefined controls.

中文

零假设 H₀：M 开头词进入 HUMAN 语义场的比例不高于对照组。
研究假设 H₁：M 开头词进入 HUMAN 语义场的比例高于对照组。

3. Scope of the Claim｜假说范围

English

The experiment tests only whether M-initial English words show an elevated statistical association with HUMAN-related meanings. It does not test whether M universally means HUMAN, whether all M words have HUMAN meanings, whether M and HUMAN share a historical origin, whether individual examples prove the hypothesis, or whether the same relationship exists in other languages.

中文

本轮只检验英语 M 开头词是否在统计上更倾向于 HUMAN 相关语义。不检验 M 是否普遍表示“人”、所有 M 词是否都与 HUMAN 有关、M 与 HUMAN 是否同源、少数个案能否证明假说，或该关系是否存在于其他语言。

4. Primary Semantic Domain｜主要语义场

English

A word is coded HUMAN = 1 when its principal dictionary meaning directly denotes, categorizes, identifies, describes, or names a human being or a socially recognized class of human beings. Otherwise, HUMAN = 0.

中文

当一个词的主要词典义直接指称人、对人分类、表示人的身份、描述社会认可的人类类别或直接命名人类成员时，标记 HUMAN = 1；否则标记 HUMAN = 0。

5. Secondary Semantic Categories｜次级语义类别

English

Four labels are coded separately to prevent subjective expansion of HUMAN:

• PERSON: an individual human or type of person.
• PEOPLE: humans collectively, a population, group, community, or human category.
• IDENTITY: a human role, occupation, kinship relation, social status, or recognized category.
• HUMAN-ATTRIBUTE: a human-associated characteristic, state, behavior, or property that does not itself denote a human; it does not automatically count as HUMAN = 1.

中文

• PERSON：直接指称某个人或一类人。
• PEOPLE：人类集合、人口、群体、社群或人类类别。
• IDENTITY：人的角色、职业、亲属关系、社会地位或认可身份。
• HUMAN-ATTRIBUTE：与人有关但本身不指称人的特征、状态、行为或属性；不得自动计为 HUMAN = 1。

6. Inclusion Rule｜纳入标准

English

A word enters the main HUMAN category only when its dictionary-defined primary lexical meaning directly refers to humans. Qualifying types may include human beings, person categories, kinship roles, occupations, social roles, titles, identity categories, and human collectives. Judgment must follow the preregistered dictionary definition rather than researcher association.

中文

只有当词典规定的主要词义直接指向人时，该词才能进入主要 HUMAN 类别。可纳入类型包括人类成员、人类类别、亲属角色、职业、社会角色、头衔、身份类别和人类集合。判断必须依据预注册词典释义，不依据研究者联想。

7. Exclusion Rules｜排除标准

English

The following do not count as HUMAN solely by association: human-used objects; activities that do not denote their performer; body parts, physiological processes, or bodily substances; mental or emotional concepts; metaphorical personification of non-human entities; proper names, place names, company names, and trademarks; and technical abbreviations, acronyms, symbols, or formula-like items unless they function as ordinary words in the selected source.

For example, music is not HUMAN, while musician may be HUMAN.

中文

仅因联想关系，以下类型不计为 HUMAN：人使用的物体；不指称执行者的活动；身体部位、生理过程或体液；心理或情绪概念；对非人实体的拟人化；人名、地名、公司名与商标；以及未在所选词频来源中作为普通词使用的技术缩写、首字母缩写、符号或公式式词项。例如 music 不计为 HUMAN，而 musician 可以计入。

8. Polysemy Rule｜多义词规则

English

Polysemous words are classified by the primary or highest-frequency contemporary dictionary sense in the preregistered reference dictionary. A rare or historical human sense cannot place a word into HUMAN. Rare senses may remain in a notes field.

中文

多义词依据预注册参考词典中的当代第一义或最高频义分类。若第一义不指人，只有罕见义或古义表示人，则主分析中 HUMAN = 0；罕见义可保留在备注栏。

9. Sampling Principle｜抽样原则

English

The Round 2 sample must not be manually selected to support or reject the hypothesis. Words are sampled from a predefined modern English frequency source. The procedure is determined before semantic annotation, and the M and control samples are frequency-matched as closely as practical.

中文

Round 2 样本不得为支持或否定假说而人工挑选。词项从预先确定的现代英语词频来源抽取；抽样程序必须在语义标注前确定，M 组与对照组应尽可能进行频率匹配。

10. M Group｜实验组

English

The experimental group contains eligible English lexical items beginning orthographically with M / m. Orthographic M is primary because the sampling source is lexical and frequency based. Spelling–pronunciation divergence may be analyzed separately later.

中文

实验组由拼写上以 M / m 开头的合格英语词项构成。由于抽样来源基于词项和词频，主实验采用正字法首字母 M；拼写与发音不一致的情况可在以后单独分析。

11. Control Groups｜对照组

English

Control initial-letter groups are specified before semantic results are examined. The design must not select only letters expected to perform poorly. At least one pooled control is used; individual control letters are also reported when sample size permits. The primary comparison is M vs. Controls, not M versus one conveniently chosen letter.

中文

对照首字母组必须在查看语义结果前确定，不得只选择预期表现较差的字母。至少使用一个合并对照组；样本量允许时，也分别报告各对照字母。主要比较为 M 与 Controls，而不是 M 与某个方便挑选的单一字母。

12. Frequency Matching｜频率匹配

English

Because common and rare words may differ systematically, M and control samples are frequency-matched. Where possible, words are assigned to comparable high-, medium-, and lower-frequency bands. Exact bands are fixed before annotation.

中文

高频词与低频词可能存在系统差异，因此 M 组和对照组要进行频率匹配。条件允许时，词项划入可比较的高频、中频和较低频区间；精确分段必须在标注前冻结。

13. Evidence-Family Deduplication｜证据家族去重

English

Closely related forms are not fully independent evidence. man, men, manly, manhood may represent one morphological family rather than four observations; the same applies to mother, motherhood, motherly. Every sampled item receives an Evidence Family ID, such as EF-MAN-001.

中文

紧密相关的词形不得视为完全独立的证据。例如 man, men, manly, manhood 可能只构成一个词汇或形态证据家族，而非四个独立观察；mother, motherhood, motherly 同理。每个样本词项都获得一个 Evidence Family ID。

14. Primary and Deduplicated Analyses｜主分析与去重分析

English

Analysis A — Token/Lexeme-Level: all eligible sampled lexical items are analyzed individually.

Analysis B — Evidence-Family-Level: related forms are collapsed into evidence families. The effect is more convincing if it survives deduplication.

中文

分析 A——词项层：所有合格样本词项分别分析。
分析 B——证据家族层：紧密相关词形合并为证据家族。如果观察到的效应在去重后仍然存在，假说才更具说服力。

15. Counterexamples｜反例

English

Every eligible M word remains in the dataset regardless of semantic category. Non-HUMAN M words are not noise and cannot be deleted. If sampled, examples such as machine, metal, mountain, music, minute remain. The final dataset preserves supporting cases, counterexamples, neutral cases, and excluded cases with reasons.

中文

所有合格 M 词都必须留在数据集中，与其语义类别无关。非 HUMAN 的 M 词不是噪声，不能删除。若 machine, metal, mountain, music, minute 等词进入样本，必须保留。最终数据同时保存支持案例、反例、中性案例和附理由的排除案例。

16. Blind Annotation｜盲标

English

Semantic annotators are not told that the experiment tests M. The interface omits the “M hypothesis,” expected direction, Pilot results, and lists of supposed supporting words. Annotators receive only the lexical item, standardized dictionary definition, and semantic coding instructions.

中文

不得告知语义标注者实验正在检验字母 M。标注界面不得出现“M 假说”、预期方向、Pilot 结果或所谓支持词列表。标注者只接收词项、标准化词典释义和语义编码说明。

17. Two Independent Annotators｜双人独立标注

English

Two annotators independently code every item and do not discuss items during initial coding. Each records HUMAN, PERSON, PEOPLE, IDENTITY, HUMAN-ATTRIBUTE, and UNCERTAIN as 0 or 1.

中文

每个词项由两名标注者独立编码，初次编码期间不得讨论具体词项。每人分别记录 HUMAN、PERSON、PEOPLE、IDENTITY、HUMAN-ATTRIBUTE 和 UNCERTAIN 的 0/1 标签。

18. Disagreement Resolution｜分歧处理

English

Initial disagreement is preserved, not overwritten. After independent coding, the disagreement rate is reported, inter-annotator agreement is calculated where appropriate, disputed items undergo predefined adjudication, and both original labels plus the final adjudicated label remain in the dataset.

中文

初始标注分歧必须保留，不能直接覆盖。独立标注后，报告分歧率，适当时计算标注者间一致性，争议词项按预定程序裁决，并在数据中保留两份原始标签与最终裁决标签。

19. Primary Outcome Measure｜主要结果变量

English

The principal quantity is the difference or ratio between these probabilities.

中文

主要结果比较 P(HUMAN | M) 与 P(HUMAN | Controls)；核心关注量是两个概率之间的差值或比值。

20. Effect Size｜效应量

English

The experiment reports effect size, not only significance. At minimum, it includes an interpretable measure such as risk difference and/or odds ratio, with a 95% confidence interval.

中文

实验必须报告效应量，而不仅是统计显著性。至少报告一种可解释的指标，如风险差和/或优势比，并附 95% 置信区间。

21. Statistical Significance｜统计显著性

English

A suitable categorical comparison test is selected according to sample characteristics. The p-value is reported but is not the sole criterion. Interpretation prioritizes effect direction, magnitude, confidence interval, robustness after evidence-family deduplication, and statistical significance—in that order.

中文

根据样本特征采用适当的类别比较检验。报告 p 值，但不把它作为唯一标准。解释依次优先考虑效应方向、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。

22. Reporting Standard｜结果报告标准

English

The final report includes N, P(HUMAN|M), P(HUMAN|Controls), effect size, 95% CI, and p-value, plus the sampling source and procedure, exclusions, annotation agreement, evidence-family-adjusted result, and complete counterexample list.

中文

最终报告至少包括 N、P(HUMAN|M)、P(HUMAN|Controls)、效应量、95% CI 和 p 值；另须报告抽样来源与程序、排除项、标注一致性、证据家族校正结果和完整反例清单。

23. Decision Rule｜结论规则

English

The hypothesis is not judged solely by p < .05. Final assessment considers direction, magnitude, uncertainty, and robustness.

• Supported: a meaningful positive effect is consistent and survives robustness checks.
• Tentatively Supported: a positive effect appears, but uncertainty or robustness limits remain.
• Inconclusive: evidence is insufficiently precise in either direction.
• Not Supported: the predicted elevation is not observed.
• Contradicted: informative evidence consistently runs against the prediction.

中文

• 支持：有意义的正向效应稳定出现，并通过相关稳健性检验。
• 初步支持：出现正向效应，但仍有不确定性或稳健性局限。
• 无法判断：数据不足以为任一方向提供足够精确的证据。
• 未获支持：预注册实验未显示预期的 HUMAN 比例提升。
• 被反证：信息充分的证据稳定地指向与预测相反的方向。

24. Critical Falsification Condition｜关键证伪条件

English

If Round 2 finds M is not greater than Controls, and the result remains non-supportive after frequency matching and evidence-family correction, the M → HUMAN hypothesis is downgraded in the Hypothesis Lab. It cannot be rescued by selecting favorable examples.

中文

如果 Round 2 发现 M 不高于 Controls，且结果在适当的频率匹配和证据家族校正后仍不支持假说，则在 Hypothesis Lab 中降级 M → HUMAN 假说；不得通过挑选个别有利案例加以挽救。

25. No Post-Hoc Rescue Rule｜禁止事后挽救规则

English

After outcomes are revealed, the primary experiment cannot be redefined by changing HUMAN definitions, deleting inconvenient counterexamples, replacing unexpectedly strong controls, inventing exclusions, counting rare senses only when favorable, or splitting and merging lexical families asymmetrically. Later work must be labeled Exploratory / Post-hoc Analysis and cannot replace the preregistered result.

中文

结果揭示后，不得通过修改 HUMAN 定义、删除不便的反例、更换意外表现较强的对照、增加新排除标准、只在有利时计算罕见义，或不对称地拆分与合并词汇家族来重定义主实验。任何后续分析必须标为探索性／事后分析，不能替代预注册结果。

26. Negative Results Policy｜负结果政策

English

Unilanguage retains and publishes negative results. A failed hypothesis is scientifically informative.

We publish hypotheses—including hypotheses our own experiments fail to support.

中文

我们公开记录假说，也公开记录实验未能支持的假说。

27. Interpretation Boundary｜解释边界

English

Even a supported result establishes only a lexical distributional association in sampled English vocabulary. It does not demonstrate universal sound symbolism, common historical or Proto-Indo-European origin, cross-language universality, or a causal relation between phoneme and meaning. Those require separate experiments.

中文

即使结果得到统计支持，也只建立所抽英语词汇中的词汇分布关联。它本身不能证明普遍语音象征、共同历史或原始印欧语起源、跨语言普遍性，或音位与意义之间的因果关系；这些需要独立实验。

28. Replication Rule｜复制规则

English

Only a stable positive Round 2 effect advances to broader replication: Round 3A with another English frequency source, Round 3B across languages, or Round 3C using phonological rather than orthographic M. These are outside Experiment 001A Round 2.

中文

只有 Round 2 产生稳定正向效应，假说才进入更广泛复制：Round 3A 使用另一英语词频来源，Round 3B 进行跨语言复制，Round 3C 检验语音而非正字法 M。这些均不属于 Experiment 001A Round 2。

29. Relationship to Experiment 001B｜与实验 001B 的关系

English

Experiment 001B — M ↔ N cannot reinterpret 001A. Experiment 001A must receive its own formal conclusion before 001B begins.

中文

Experiment 001B — M ↔ N 不得用于重新解释 001A。001A 必须先获得独立的正式结论，之后才开始 001B。

30. Preregistration Principle｜预注册原则

English

Once this document is frozen and the Round 2 sample is generated, primary rules cannot change after outcomes are inspected. Any necessary departure is documented as a Deviation from Preregistration, stating what changed, why, when, and whether it occurred before or after outcome inspection.

中文

本文件冻结并生成 Round 2 样本后，不得在查看结果后修改主要规则。任何必要变更都必须记录为偏离预注册，说明改了什么、为何修改、何时修改，以及修改发生在查看结果之前还是之后。

31. Research Philosophy｜研究原则

English

Experiment 001A is not designed to prove an idea proposed by Unilanguage. It tests whether that idea survives a predefined empirical procedure.

A hypothesis becomes more valuable when it can clearly fail.

中文

Experiment 001A 不是为了证明 Unilanguage 提出的想法，而是检验该想法能否经受预先确定的实证程序。

一个假说只有在能够被明确否定时，才真正开始成为可检验的研究假说。