Formulaic Language:
The Premise of Prediction 公式化语言：
预测的前提

Why abbreviate?为什么要缩写？

Source status: the local abbreviation base for this act is OHRE Appendix II. The article on proper names and Latin epigraphic abbreviation is a retrieval target for direct page-level confirmation of the worked IDR/ISM examples.资料状态：本幕在本地能核查的缩写依据是《牛津罗马铭文学手册》附录二。关于专名与拉丁铭文缩写的专文仍是待取回文献，将用来逐页核查下面的 IDR / ISM 例证。 Bruun & Edmondson 2015: 787–813, Appendix II; Strechie article retrieval target.

Two registers — Latin dense, Greek restricted两种传统：拉丁密集，希腊稀疏

The asymmetry, in numbers — but with a twist不均衡，用数字来看；但有一个转折

How to read the two findings together. McLean's claim is about abbreviation density per inscription (counting D M as one abbreviation, etc.). Our chart is about formula-phrase density per inscription (counting D(is) M(anibus) as one formula). Latin's funerary inscriptions are short and abbreviation-dense; Greek's decree inscriptions are long and formula-dense. Both traditions are formulaic, but they distribute the formulaic-ness differently — Latin into the abbreviation-letter, Greek into the full-phrase. 两个发现放在一起看：McLean 讲的是缩写密度（每条铭文的缩写计数）。本图显示的是套语密度（每条铭文的套语短语计数）。拉丁丧葬铭文短而缩写密集；希腊法令铭文长而套语密集。两种传统都是公式化的，不同的只是公式化的分布：拉丁集中在缩写字母上，希腊集中在整段短语上。

The reference works — already inadequate by design参考工具：本来就不完整

"This list of some common epigraphic abbreviations used in Latin inscriptions lays no claim to completeness. A work of a different magnitude would be required for such an ambition … Yet, a reader of epigraphic texts will discover that the ingenuity of the Roman stonecutter and/or his clients surpassed what modern compilers of wordlists have accomplished. Even with the help of an extensive list some puzzles will remain."“本表收录的拉丁铭文缩写绝不敢说完备……即使备好这张长表，谜题仍然存在。” Bruun & Edmondson 2015: 787, Appendix II.

The point. The reference work itself names the gap. The point is not that lists are bad; the point is that the abbreviation system is open-ended in a way no list can close. The editor restoring an inscription has the list and still has to make a call. 编者修复某条铭文时，即使有这张表，仍然要自己判断。这里的要点在于：缩写体系本身是开放的，没有任何一张表能把它全部收尽；参考工具齐不齐全反倒是次要的。

A — the alphabetical opener (51 entries)字母 A — 起首一栏（51 条）

Bruun & Edmondson 2015: 787–88, Appendix II.

D — the funerary opener (27 entries)字母 D — 葬铭起首（27 条）

Bruun & Edmondson 2015: 791, Appendix II.

Note. D M is famous; D D D is also a standard dedication-formula closer. D·E·R·I·C on its own would baffle an unprepared reader — it sits inside a senatorial-decree formula, which Act II will return to. 补注：D M 前面已多次见到；D D D 也是题献铭文常见的收尾。D·E·R·I·C 如果孤立出现就很难解读，因为它属于元老院决议套语的一环，留到第二幕详细讨论。

P — the biggest menu (38 entries)字母 P — 最长的菜单（38 条）

Branching factor. Six entries in the table above contain at least four distinct expansions for the same letter-pattern. The editor disambiguates by context. When context fails (broken stone, no surrounding text), restoration becomes uncertain. 分支因子：表中有六条各自至少有四种展开方式。编者靠上下文消歧；上下文不存在（石头残损、文字中断）时，修复就无法确定。

Bruun & Edmondson 2015: 803–05, Appendix II.

S — the verb-and-virtue letter (27 entries)字母 S — 动词与德目（27 条）

Bruun & Edmondson 2015: 808–10, Appendix II.

V — votive + military (23 entries)字母 V — 还愿与军务（23 条）

Note the funerary / votive divide. A reader who sees V S L M at the bottom of a stone reads votive ("paid his vow"). A reader who sees V A at the bottom reads funerary ("lived N years"). Same letter, two genre-bound expansions, picked by what came before. 还愿与丧葬的区别：石尾如果出现 V S L M，那是还愿铭的收尾（“还其愿”）；如果出现 V A，就属于丧葬（“享年若干”）。同一个字母有两种属体（即文类）的展开，取决于前文是什么语境。

Bruun & Edmondson 2015: 812–13, Appendix II.

L · M · N — three mid-alphabet samples (24 + 20 + 15 entries)L、M、N 三字母 — 中段样本

Bruun & Edmondson 2015: 798–800, Appendix II.

Worked example 1 — IDR III/2 9 (CIL III 7983), tria nomina + tribe实例一：IDR III/2 9（CIL III 7983），三名加部族

M · IVL · PAP · IVSTVS · DEC
COL · OB · HON · PONTIF
CAMPVM · CVMSVIS
ADITIBVS · CLVSIT
ET · STATVAM · POSIT

M(arcus) Iul(ius) Pap(iria tribu) Iustus dec(urio)
col(oniae) ob hon(orem) pontif(icatus)
campum cum suis
aditibus clusit
et statuam posit.

IDR III/2 9 worked example; direct article check remains a retrieval target. Local abbreviation base: Bruun & Edmondson 2015, Appendix II.

Worked example 2 — IDR II 49, prefect dedicates to Jupiter实例二：IDR II 49，营长向至高朱庇特题献

I · O · M
IVL · VALER · EMER
PRAEF · COH · I
BRIT · ⟨…⟩

I(ovi) O(ptimo) M(aximo)
Iul(ius) Valer(ius) Emer(itus)
praef(ectus) coh(ortis) I
Brit(tonum) ⟨…⟩

IDR II 49 worked example; direct article check remains a retrieval target. Local abbreviation base: Bruun & Edmondson 2015, Appendix II.

Worked example 3 — IDR I 3 (CIL XVI 163), imperial titles in compression实例三：IDR I 3（CIL XVI 163），元首尊号的压缩

IMP · CAES · DIVI
HADR · F · DIVI · TRAIAN
PARTH · NEP · DIVI · NERV
PRONEP · T · AEL · HADRIANVS
ANTONINVS · AVG · PIVS
P · M · TRIB · POT · ⟨N⟩
COS · ⟨N⟩ · P · P · ⟨…⟩

Imp(erator) Caes(ar) divi
Hadr(iani) f(ilius) divi Traian(i)
Parth(ici) nep(os) divi Nerv(ae)
pronep(os) T(itus) Ael(ius) Hadrianus
Antoninus Aug(ustus) Pius
p(ontifex) m(aximus) trib(unicia) pot(estate) ⟨N⟩
co(n)s(ul) ⟨N⟩ p(ater) p(atriae) ⟨…⟩

IDR I 3 worked example; direct article check remains a retrieval target. Local abbreviation base: Bruun & Edmondson 2015, Appendix II.

Worked example 4 — IDR II 145, parents bury their daughter实例四：IDR II 145，父母为女儿立碑

D · M
ET · IVLIAE · GRA-
TAE · FIL · PAREN-
TES · POSVER

D(is) M(anibus)
et Iuliae Gra-
tae fil(iae) paren-
tes posuer(unt).

IDR II 145 worked example; direct article check remains a retrieval target. Local abbreviation base: Bruun & Edmondson 2015, Appendix II.

Worked example 5 — ISM II 263, a Black-Sea funerary inscription实例五：ISM II 263，黑海沿岸的葬铭

D · M
T · VIBIVS · VARIVS
MIL · LEG · XI · CL
⟨signa cohortis…⟩
VIX · ANN · XXVIII
MILITAV · ANN · VIII
HEREDES · B · M · F

D(is) M(anibus)
T(itus) Vibius Varius
mil(es) leg(ionis) XI Cl(audiae)
⟨signa cohortis…⟩
vix(it) ann(is) XXVIII
militav(it) ann(is) VIII
heredes b(ene) m(erenti) f(ecerunt).

ISM II 263 worked example; direct article check remains a retrieval target. Local abbreviation base: Bruun & Edmondson 2015, Appendix II.

Worked example 6 — IDR II 1, deity abbreviated as an acronym实例六：IDR II 1，用缩字代替神名

I · O · M · DOLICH · ⟨…⟩
ET · S · I · M

I(ovi) O(ptimo) M(aximo) Dolich(eno) ⟨…⟩
et S(oli) I(nvicto) M(ithrae)

IDR II 1 worked example; direct article check remains a retrieval target. Local abbreviation base: Bruun & Edmondson 2015, Appendix II.

The Greek abbreviation tradition — three engines希腊缩写传统：三种机制

"Though the Greeks did not employ abbreviations nearly to the same extent as the Romans, abbreviations were used as a means of reducing labor and saving space on the stone's surface. An early system of abbreviation was developed in [Rhodes]. … Any system of abbreviation requires either that the meanings of abbreviations be self-evident, or that the readers be familiar with the conventions of abbreviation."“希腊人用缩写远不如罗马人密集，但也是为了省工省石。最早的缩写体系大约出自罗德岛……任何缩写体系，要么意思不言自明，要么读者必须熟悉它的惯例。” McLean 2002: 49, §2.05.

Greek abbreviation marks — the diacritical shorthand希腊缩号：标点式的省写记号

Why the marks matter. Where the engraver omitted the mark, a modern editor must infer abbreviation from context — the same problem as Latin, only with smaller margin for error because Greek does not pile shortened letters as densely. 缩号为什么要紧：石工如果漏刻，今天的编者就要靠上下文推断；这一点和拉丁语境相同；只是希腊缩写堆叠较少，能容的误差更小。

McLean 2002: 49–50, §2.06.3.

Greek numerals — abbreviation as notation希腊数字：用缩号表示数

Acrophonic = abbreviation by initial. The Attic acrophonic system uses the first letter of the number-word as the numeral itself. The same logic underlies a vast amount of Greek epigraphic abbreviation: take the first letter (or first two), let the reader supply the rest from context. Acrophonic survives into the Athenian Tribute Lists (Act III) — the Η for 100 drachmas is the same Η as hekaton. 首音字母就是缩写：阿提卡的首音数字用词首字母代表这个数本身。希腊大量铭文缩写也是同一个道理：取词首一两个字母，其余由读者从语境补足。这套首音字母数字系统一直沿用到雅典贡金表（第三幕）：表示 100 德拉克玛的 Η，就是 ἑκατόν 的 Η。

McLean 2002: 51, §2.07 + §2.08.

The branching factor — when context fails分支因子：上下文失效时

What the menu does not show. Each "entry" is itself a head of a sub-tree: the entry S·D branches into "salutem dicit" (letter opener), "sub dolo", "sacrum dedit", and more — each picked by the inscription's genre. The real branching factor is the product of letter-menu × genre-menu × position-in-text. 表里没有显示的是：每个“条目”本身又是一棵子树。S·D 下面分出 “salutem dicit”、“sub dolo”、“sacrum dedit” 等好几个分支，取决于铭文的属体。实际的分支因子，是字母菜单 × 属体菜单 × 文中位置三者的乘积。

Bruun & Edmondson 2015: 787–813, counts from OHRE Appendix II.

The closing problem — what narrows the menu?小结：怎样收窄这个菜单？

The forward-pull. Latin and Greek abbreviation lists exist because abbreviations recur in fixed positions of fixed phrase-types. The phrase-types are formulae. Act II picks them up where Act I leaves them. 由此向前：拉丁与希腊的缩写表之所以编得出来，就是因为缩写有固定位置，而且嵌在固定的短语类型里。这种短语类型就是套语。第二幕接着讲。

Formula = phrase-unit + slots套语：短语单位与槽位

ὁ δεῖνα [divinity] ἀνέθηκε

The Formula Dossier's bank — 58 formulae across 21 case studies套语词典：21 个案例中的 58 条套语

The bank, by the numbers — top 15 formulae by corpus hits从语料数说起：命中数前 15

Three observations. (1) The Latin D M opener at 77,785 hits dwarfs everything; Latin's funerary habit is the corpus's gravitational centre. (2) Greek ἀγαθῇ τύχῃ and ἐπειδή sit second-tier — predictably high because they open thousands of honorific decrees. (3) The five canonical Athenian prescript formulae (ἔδοξεν τῇ βουλῇ, ἔδοξεν τῇ βουλῇ καὶ τῷ δήμῳ, ἐπὶ X ἄρχοντος, ἐπὶ τῆς … πρυτανείας, X ἐγραμμάτευεν) all sit in the 800–2,700 range — a tight cluster that reflects the small surviving Athenian-decree corpus relative to the imperial-period funerary one. 有三点可以观察：（1）拉丁 D M 开头有七万七千多次命中，远超其他所有套语；拉丁丧葬习惯是整个语料的引力中心。（2）希腊 ἀγαθῇ τύχῃ 与 ἐπειδή 排在第二级，因为它们开启了数千件荣誉法令，数量自然较多。（3）雅典前言的五条主要套语（ἔδοξεν τῇ βουλῇ、ἔδοξεν τῇ βουλῇ καὶ τῷ δήμῳ、ἐπὶ X ἄρχοντος、ἐπὶ τῆς … πρυτανείας、X ἐγραμμάτευεν）都在 800-2,700 区间，是紧密的一束，反映现存的雅典法令比帝政期丧葬铭文少。

Tally source: formula-dossier/dossier.html · DATA.summary.by_formula, derived from EDCS_text_cleaned_2022-09-12 (Latin) and I.PHI 2022 (Greek). All 54 formulae have corpus counts.

F.EDOXEN — "It seemed good to the council and the people"F.EDOXEN — “议事会与人民议决”

Henry 1977: 13–26, on the bouleutic / demotic distinction in Attic decree-prescripts.

F.EPI_ARCHONTOS / F.EPI_PRYTANEIAS — dating the decreeF.EPI_ARCHONTOS / F.EPI_PRYTANEIAS — 为法令纪年

Henry 1977: 27–49, on the prytany-secretary triangulation.

F.EPEIDE + F.DEDOCHTHAI — the whereas-clause and the be-it-resolvedF.EPEIDE 与 F.DEDOCHTHAI — 鉴于……议决……

McLean 2002: 214–25 and 232–43, on decrees, honorific inscriptions, commendation, crowns, and publication; Henry 1977 remains the prescript control for Act II.

F.EPAINESAI / F.STEFANOSAI / F.ANAGRAPSAI — commend, crown, publishF.EPAINESAI / F.STEFANOSAI / F.ANAGRAPSAI — 表彰、加冠、刻石

McLean 2002: 35–37 and 232–43, on restoration by formula, honorific clauses, and publication language; Henry 1977 remains the prescript control for Act II.

Latin SC opening — F.LAT_SENATUM_HABUIT family拉丁元老院决议的开头

Sherk 1969, SC procedural formulae; specialist senatorial-procedure bibliography remains a retrieval target.

F.LAT_DE_EA_RE_ITA_CENSUERUNT + F.LAT_CENSUERE — the deliberation hingeF.LAT_DE_EA_RE_ITA_CENSUERUNT 与 F.LAT_CENSUERE — 审议的关节

SC voting formulae keyed here to Sherk 1969; specialist Senate procedure volume remains a retrieval target.

Latin funerary — the four-formula stack拉丁丧葬：四套语堆叠

The stack. A typical 1st–2nd c. CE epitaph stacks four formulae: opening · subject's name + biographical numbers · dedicators + relation · closing. Five lines of text, four formula-units; the human payload is the name and the numbers. This stack is why Laudatio Turiae (Act III) reads as a non-formulaic outlier — it is a personalised oration where the formulae have given way to first-person narrative. 堆叠结构：典型的一至二世纪墓志层叠四组套语：起首 · 亡者姓名加生平数字 · 立碑者及其关系 · 收尾。全文五行，套语四组；属于个人的独有内容只有人名和数字。第三幕要讨论的图利娅颂辞之所以读起来“非套语”，是因为它是以第一人称叙述的颂辞，套语让位给了个人化的记述。

Bennett 2020: 63 (Table 3.2 — the EDCS search strings that make D·M and D·M·S mutually exclusive), 84 (60,728 of 104,007 = 58%), 85 (Table 4.2, by region).

F.AUTOKRATOR_SEBASTOS_LEGEI — the Greek imperial-edict openerF.AUTOKRATOR_SEBASTOS_LEGEI — 元首敕令的希腊语开头

The bilingual cross-check. A Greek inscription opening with this formula tells the editor she is reading an imperial response translated from Latin. The Latin original may exist (and often does, in archival rosters); the bilingual pairing then becomes a reciprocal restoration tool — each face supplies what the other lost. The Persicus edict (Act III · Case D) is the demonstration. 双语互证：希腊铭文用这条套语开头，编者就知道它是从拉丁文翻译过来的元首回复。拉丁原文可能还存在（常存于档案册中）；双语对照就成了相互修复的工具：一面缺的，另一面或许能补上。第三幕 D 案的 Persicus 政令就是明显的例子。

Sherk 1969: 217–34, imperial-edict translation conventions.

Formula × geography — the same slot, regional menus套语 × 地理：同一槽位，地区菜单

The point. A Black-Sea editor restoring a damaged publication-line does not look in Athens for parallels; the local variant στηλάμωνον is what fits. Regional knowledge is part of formula knowledge. 黑海的编者修补一条公布套语的残行，不应该到雅典找平行文本；当地的变体 στηλάμωνον 才合适。地域知识是套语知识的一部分。

McLean 2002: 35–37 + 240, regional publication-formula variants.

Formula × period — the same slot, diachronic drift套语 × 时段：同一槽位的历时变化

Horizontally scrollable: drag or trackpad-scroll the chart left/right to read the full timeline. The chart's natural width is 1,000 px; on narrow screens the wrapper supplies a scrollbar so labels stay readable. 本图可以左右滚动：用触控板或鼠标横向拖动就能看全。图的原始宽度是 1,000 像素；屏幕不够宽时，容器自带滚动条，保证字迹可读。

What the timeline shows. Greek formulae (teal) cluster in the 5th c. BCE → 1st c. AD band — the polis-decree era. Latin funerary formulae (gold) emerge sharply in the late Republic / early Empire and dominate the corpus only after Augustus. The two traditions barely overlap; the apparent "Greek dominance" in the early bars is genre, not geography — there are simply more Greek decrees from 500–100 BCE than there are surviving Latin documents. 寿命图显示：希腊套语（青色）集中在公元前 5 世纪至公元 1 世纪之间，也就是 polis 法令的时代。拉丁丧葬套语（金色）在共和末年至帝政初期迅速兴起，奥古斯都之后才主导语料。两种传统几乎不重叠；早期看到的“希腊为主”是属体造成的，与地理无关：公元前 500-100 年间存世的希腊法令本来就多于拉丁文献。

Lifespan data: DATA.formulae in formula-dossier/dossier.html, first_attested + last_attested fields, scholarly anchors per Henry 1977 (Greek) and Sherk 1969 (Latin). Hit counts: DATA.summary.by_formula.

Henry 1977: 13–82, prescript evolution; Rhodes & Lewis 1997, on the institutional history.

Formula × language — Greek ↔ Latin crossings套语 × 语言：希腊语与拉丁语互译

Persicus IK Ephesos 17–19 + Sherk 1969: 217–34, bilingual-edict translation conventions.

Formula × abbreviation — the work the formula does套语 × 缩写：套语做了什么

A note on what S alone is not. In a senatus consultum, the resolution-verb censuere / censuerunt is not abbreviated as S. It appears either spelled out (or as CENS, CENSV, CENSVER) — or, in the canonical hinge-formula D·E·R·I·C (de ea re ita censuerunt), the final letter that carries it is a C, not an S. The senatorial register is where S·C and S·P·Q·R belong; S alone in a senatorial context most naturally reads as senatus. 说明：单独的 S 字不是什么。元老院决议中，决议动词 censuere / censuerunt 不用单独的 S 表示。它的写法或是全字，或是 CENS、CENSV、CENSVER；或者像枢纽套语 D·E·R·I·C（即 de ea re ita censuerunt）那样，末字母承担的是 C，并非 S。属于元老院语域的是 S·C 与 S·P·Q·R；元老院文书中单独的 S，最自然的读法是 senatus。

The synthesis. Act I's abbreviation menus are open; Act II's formula menus narrow them. But formulae have menus too — F.EDOXEN_B vs F.EDOXEN_BD vs F.EDOXEN_D, F.LAT_PLACERE in four variants. Where the formula menu offers more than one slot-position for the inscription, the editor is back to choosing. Which leads to Act III. 合起来看：第一幕的缩写菜单是开放的；第二幕的套语菜单把它收窄。但套语菜单也不一定唯一：F.EDOXEN_B、F.EDOXEN_BD、F.EDOXEN_D 有区别，F.LAT_PLACERE 又有四个变体。如果某条铭文上套语菜单不能定于一种，编者就要重新做选择。第三幕接着讲。

Prediction by formula vs prediction by experience靠套语的预测，与靠经验的预测

The four cases below sit at: ATL near End A · Segesta one step toward B (the slot is formulaic but the choice within it is contested) · Laudatio mid-spectrum (formulaic virtues but unique-narrative outside them) · Persicus mid-spectrum (bilingual cross-check provides leverage, but the specific reforms cannot be predicted). 四个案例在这条光谱上的位置：贡金表靠近 A 端 · 塞格斯塔向 B 端偏一步（槽位有套语，但槽内的选择有争议）· 图利娅居中（德目段有套语，外侧的个人叙述没有）· Persicus 居中（双语互证有帮助，但具体改革内容无法预测）。

Case C · Athenian Tribute Lists — the fiscal-list genreC 案 · 雅典贡金表 — 财政清单的属体

The Athenian Tribute Lists, local handles for vols. II–IV plus related Meritt/Dow/West/Paarmann scholarship; ATL vol. I direct page checks remain a retrieval target. Project's SCPP edition: governance/athenian-tribute-list.html.

The four anchor years四锚年

The four years span 40 years and one war-driven discontinuity (the 425/4 reassessment doubled many cities' assessment to fund the Peloponnesian War). The formula stays stable across the discontinuity; the figures inside it change. This is the deck's first demonstration of a formula doing the work of inheritance: it carries continuity through historical rupture. 四个年份跨四十年，中间有一次战时断裂（公元前 425/4 的重估，为筹措伯罗奔尼撒战争军费，许多城邦数额加倍）。套语跨过断裂仍然稳定；里面的数字则在变。这是本课第一次看到套语承担传承功能：在历史断裂中维持连续。

A complete entry vs a damaged entry完好条目对残损条目

Καρπάθιοι     ΗΗΗ

[. . . . .]θιοι     [. .]Η

The editor's question: with the formula in hand, how much of the lost text should she restore? 编者要回答的问题是：手里已经有套语，缺失的文本应该还原多少？

ATL maximal — the Meritt / Wade-Gery / McGregor editionATL 最大派 — Meritt / Wade-Gery / McGregor 版本

[Καρπά]θιοι     [Η]ΗΗ

The visual. A reader who opens ATL I or II sees pages of [restored-name] [restored-figure] entries; the brackets become almost decorative. The editor's hand is dense but invisible to a reader who skims. 视觉效果：读者翻开 ATL I 或 II，看到整页的 [补名] [补数] 条目，方括号几乎成了装饰。编者的手笔密集而隐蔽，浏览的人不容易察觉。

Paarmann minimal — 2007Paarmann 最小派 — 2007

[. . . . .]θιοι     [. .]Η

The visual. A reader who opens Paarmann's edition sees pages of [. . .]θιοι [. .]Η entries; the unbracketed letters look sparse. The editor's hand is visible. 视觉效果：读者翻开 Paarmann 的版本，看到整页的 [. . .]θιοι [. .]Η 条目，不带括号的字母显得稀疏。编者的手笔清楚可见。

The disagreement is methodological, not philological分歧在方法，不在文本

"The two editions read the same letters on the same stone. They disagree not about what survives but about what a formula entitles an editor to write."“两个版本读的是同一块石头上的同一些字母。石头上留存什么并无分歧；分歧在于套语授权编者写下什么。” M-WK13.E6–E7 ATL-EXPANSION-PLAN.md, on the editorial-school distinction.

The deeper question. The Athenian Tribute Lists are foundational evidence for the Athenian Empire's fiscal history. If the editions disagree by ~37 percentage points in how much of a printed entry is editorial restoration (the ~46% vs ~9% of the specimen entry above), historians using them disagree by that much too — about who paid what, when. ATL is the case where prediction by formula almost works, and where the residual gap matters historically. 更深一层的问题：雅典贡金表是雅典帝国财政史的根本证据。两个版本在条目中有多少属于编者修复这一点上相差约三十七个百分点（上文样例条目约 46% 对约 9%）；使用它们的论述（谁在什么时候缴了多少）也会差这么多。贡金表就是靠套语预测几乎可行的案例，剩下的那部分差距，在历史意义上仍然要紧。

The 425/4 reassessment — the formula's stress test公元前 425/4 年的重估；套语的压力测试

In 425/4 BCE, the politician Thoudippos sponsored a decree (IG I³ 71) reassessing the empire's tribute upward — many cities doubled or tripled. The lists from 425/4 onwards record these new figures. 公元前 425/4 年，政客 Thoudippos 主导一项法令（IG I³ 71），把帝国贡金重估上调；许多城邦的数额加倍甚至涨到三倍。425/4 之后的表记下了这一新额。

The formula's role. The year-prescript stays identical pre- and post-425/4; the alphabet-block organisation stays identical; only the figures inside the quota-slot change. Editing the Year-39 (415/4) list, the editor restores the year-prescript and the alphabet-block frame mechanically — those are formula-determined. The figures she must read from the stone or, where lost, restore from the post-425/4 figure pattern (not from the pre-425/4 figures). 套语的角色：年份前言在 425/4 前后不变；字母组的组织也不变；只有数额槽位里的内容有变化。校订 Year 39（415/4）的表时，编者可以机械地补出年份前言和字母组的框架；这是套语决定的。数额则必须读石头；缺失的部分，应该按 425/4 之后的数额模式来补，不能按 425/4 之前的。

Lambert 2017, Two inscribed documents of the Athenian empire, on the Thoudippos decree.

Verdict — ATL is partially predictable by formula小结：ATL 靠套语可以部分预测

The pedagogical answer. The ATL is the deck's case where prediction by formula does most of the work. The remaining gap is where Paarmann's minimalism and ATL's maximalism part company — and where, increasingly, machine-learning tools will be asked to weigh in (Act IV). 到这里可以看出，贡金表是套语预测承担最大份额的案例。剩下的缺口，就是 Paarmann 的最小派与 ATL 的最大派分歧所在，也是机器学习工具逐渐被请来评判的地方（第四幕）。

How representative is the bracket? — the whole corpus at once方括号有多普遍？一次看遍全部语料

The pedagogical move. The auditor inverts the model's usual job: it does not fill brackets, it measures the ones already printed — reporting, per inscription, restored-vs-preserved-vs-declared-gap letters and ranking the heaviest brackets in the corpus. The ATL is not an outlier; it is the visible tip of an 11.5%-conjecture corpus. 教学上的用处：审计器把模型惯常做的事倒了过来：它不去填方括号，改为量已经印出来的方括号；逐条铭文报告臆补、留存、已声明缺位三类字母的数量，并把语料中方括号最重的铭文排序。贡金表并非异类：它是一个臆补比例达 11.5% 的语料的可见尖端。

EDCS letter-count / restoration auditor (epidoc/epipap/edcs-restoration-auditor) · audit over EDCS_merged_cleaned_attrs_2022-09-12 (537,262 records).

Case B · Segesta decree — the dating cruxB 案 · 塞格斯塔条约；纪年关键

An alliance between Athens and the non-Greek polis of Segesta (western Sicily). The text is a standard Athenian decree-prescript: enactment-formula (F.EDOXEN_BD), then archon-name and prytany. The lacuna sits in the archon-name slot. 雅典与西西里西部的非希腊 polis 塞格斯塔之间的盟约。文本是标准的雅典法令前言：颁令套语（F.EDOXEN_BD），后面接执政官名与 prytany。残缺的位置在执政官名的槽位。

ἔδοξεν τῆι βουλῆι καὶ τῶι δήμωι · [tribe-name]ος ἐπρυτάνευε
ἐπὶ [3-4 letter lacuna]ονος ἄρχοντος, ἐγραμμάτευε [N]ος
καὶ Σεγεσταῖοι …

Only the letters ]ΟΝ ΕΡΧΕ are securely read on the stone. The archon's name must end in -ων (genitive in -ονος); the two archon-names that fit and are in play are Ἁβρων and Ἀντιφῶν. 石头上只能确定 ]ΟΝ ΕΡΧΕ。执政官的名字应以 -ων 收尾（属格作 -ονος）；符合这个格式而又在讨论之列的执政官名有 Ἁβρων 和 Ἀντιφῶν 两个。

Raubitschek 1944 — [Ἁβρ]ονος → 458/7 BCERaubitschek 1944 — [Ἁβρ]ονος → 公元前 458/7

For forty-six years (1944–1990), the consensus reading was [Ἁβρ]ονος, on the basis of three-bar sigma palaeography: the stone uses the archaic three-bar sigma (Σ formed of three diagonal strokes) rather than the four-bar form, and the three-bar form was held to be confined to inscriptions before ~445 BCE. Habron was archon in 458/7. The alliance dated accordingly. 在整整四十六年间（1944–1990），通行的读法是 [Ἁβρ]ονος，依据是三划西格玛字形的论证：石上的西格玛是古式（三条对角笔划），并非四划式，而三划式被认为只用到公元前约 445 年之前。Habron 在公元前 458/7 年任执政官；盟约据此纪年。

Source-backed synthesis. Henry frames the Segesta dispute as a test case for the traditional dating canons of three-bar sigma and tailed rho: if Antiphon is right, the old palaeographic terminus has to yield to prosopographical and historical evidence.据 Henry 1992 综合。Henry 把塞格斯塔之争看作三划西格玛与带尾 rho 这两项传统断代准则的测试案例：如果 Antiphon 的读法成立，旧有的字形学上限就必须让位于人名学与史学证据。 Henry 1992: 137, "Through a Laser Beam Darkly."

The downstream cascade. Many other 5th-c. Athenian decrees were dated relative to the Segesta decree, via the three-bar sigma criterion. The 458/7 date for Segesta thus anchored a chronology covering hundreds of inscriptions across the Athenian imperial archive. 连锁效应：其他许多公元前五世纪的雅典法令按三划西格玛标准纪年，以塞格斯塔条约为参照。塞格斯塔 458/7 的纪年因此系着数百条铭文的年代框架（也就是雅典帝国的档案）。

Chambers, Gallucci & Spanos 1990 — [Ἀντιφ]ονος → 418/7 BCEChambers、Gallucci 与 Spanos 1990 — [Ἀντιφ]ονος → 公元前 418/7

In 1990, three scholars applied laser-enhanced photography to the same stone in the Acropolis Museum. The technique illuminated traces of letters at the edge of the lacuna that earlier autopsies had missed. The traces fit Ἀντιφῶν better than Ἁβρων. Antiphon was archon in 418/7 BCE. The alliance moves forty years younger. 1990 年，三位学者用激光增强摄影检视卫城博物馆的原石。这种技术让早先观察漏掉的残缺边缘字痕显现出来。显现的笔痕与 Ἀντιφῶν 比与 Ἁβρων 更吻合。Antiphon 在公元前 418/7 年任执政官。盟约的纪年向后移了四十年。

Source-backed synthesis. Chambers, Gallucci, and Spanos combine measurement, computer-enhanced photography, and laser imaging to argue that the visible traces fit Antiphon rather than Habron; they conclude that the inscription belongs to 418/7 BCE.据 Chambers、Gallucci 与 Spanos 1990 综合。三位作者合用测量、电脑增强摄影与激光成像，主张可见的笔痕吻合 Antiphon 而非 Habron；他们的结论是这方铭文应归入 418/7 BCE。 Chambers, Gallucci & Spanos 1990: 38–63, ZPE 83.

The cascade reverses. If the alliance is 418/7, then many other inscriptions dated by the three-bar sigma criterion need re-dating. The criterion itself comes under pressure: if Segesta's three-bar sigma is from 418/7, then perhaps three-bar sigma was used later than 445 after all. A single restoration unwinds part of a chronological framework. 连锁反转：如果盟约属于 418/7，那么按三划西格玛纪年的其他铭文都要重新纪年。这条标准本身也受到压力：如果塞格斯塔的三划西格玛属于 418/7，那么三划西格玛可能一直用到 445 之后。一处补字，就松动了一部分年代框架。

What changes downstream — two empires下游的变化：两种帝国

Three letters. The historical interpretation hinges on three letters in a single archon-name slot. The formula tells us where the archon-name sits; the formula does not pick between Habron and Antiphon. 三个字母的事：历史诠释悬在执政官槽位里的三个字母上。套语指出执政官名所在的位置；但在 Habron 与 Antiphon 之间，套语没有办法做出选择。

Badian 1989 — "History from Square Brackets"Badian 1989 — “出自方括号的历史”

"history should not be written with any confidence from what is inside square brackets in published inscriptions."“不应该凭已刊铭文方括号里的字，自信地书写历史。” Badian 1989: 59–70, "History from 'Square Brackets'", ZPE 79.

Badian's article is the methodological warning the Segesta case needs: bracketed restorations may be reasonable, but the historical argument has to name the lacuna, the constraint, and the viable alternatives instead of treating the supplement as transparent evidence. Badian 这篇文章给出了塞格斯塔案需要的方法论警告：方括号里的修复也许能成立，但史学论证必须明确列出缺位、约束和可行的其他方案，不能把补出的字当作透明的证据。

Verdict — Segesta is contested, not predictable小结：塞格斯塔有争议，不可预测

The pedagogical answer. Segesta sits one step toward End B from ATL. The formula is intact; one slot is contested; the contest carries 40 years of empire chronology. The case demonstrates that formulaic predictability does not equal historical certainty — even when the formula does its work perfectly, the slot it identifies can still be the heaviest bracket in the corpus. 教学结语：塞格斯塔比贡金表更靠近 B 端一步。套语完好；一个槽位有争议；这一个争议带着四十年的帝国年代。本案说明：套语可以预测，不等于历史意义确定；即使套语完全起了作用，它指出的那个槽位，仍然可能是整个语料里分量最重的方括号。

Case A · Laudatio Turiae — laudatio funebris as typeA 案 · 图利娅颂辞；丧葬颂辞的体裁

A monumental Latin funerary oration: a husband's praise of his wife, surviving on a fragmentary marble tablet recovered from secondary use as a catacomb loculus-lid. The longest preserved private Latin funerary inscription. Mantzilas 2017's comparative study of 5 Roman laudationes mulierum (Murdia, Hadrian's mother-in-law, Iulia Procilla, etc.) finds an 8-slot bank shared across all five — making the Laudatio a type-specimen of the genre, not the outlier it is sometimes presented as. 一篇拉丁丧葬颂辞的纪念性文本：丈夫称颂亡妻的话，保存在一块残损的大理石板上，石板原来做墓窟的 loculus，大概经过二次利用。这是现存最长的私人拉丁丧葬铭文。Mantzilas 2017 比较罗马五篇妇女颂辞（Murdia、Hadrian 的岳母、Iulia Procilla 等），发现五篇共享一个八槽位的套语库；这使图利娅颂辞成为该体裁的典型，而非有时所说的“孤例”。

The 8-slot bank: subject-formula · address-opener · vita-narrative · virtue-catalogue · distinguishing-event praise · counterfactual closer · lament · ritual closer. Each slot is recognisable; the contents vary. 八槽位库：主体套语 · 称呼起首 · 生平叙述 · 德目目录 · 殊事赞颂 · 反事实结束 · 哀悼 · 仪式收尾。每个槽位都能辨认，内容则各不相同。

Mantzilas 2017, Laudationes Mulierum, 5-case comparative table; SCPP edition governance/laudatio-turiae.html.

The virtue-catalogue slot — formulaic, restorable德目目录槽位；套语式，可以补

Lines 30–34 of the left column. Wistrand 1976's text, with editorial restoration in brackets. 左列 30–34 行。文本取 Wistrand 1976，编者修复的部分用方括号标出。

(30) Domestica bona pudici[t]iae, opsequi, comitatis, facilitatis,
(31) lanificii stud[i, religionis] sine superstitione, o[rnatus non
(32) conspici]endo, cultus modici cur [memorem? Cur dicam de tuorum cari-]
(33) tate, familiae pietate, [c]um aeque matrem meam ac tuos parente[s
(34) col]ueris eandemque quietem [ill]i quam tuis curaveris …

Translation. Why should I record [your] domestic virtues — your modesty, compliance, affability, accessibility, wool-working, religion without superstition, restrained dress, modest appearance? Why should I speak of your love for [your] family, your devotion to kin, [since] you loved my mother as you did your own parents, and gave her the same peace as theirs … 译文。我何必列举 [你] 家中的种种美德：贞洁、顺从、和悦、亲和、纺织的劳作、不流于迷信的虔诚、不张扬的装束、有节制的仪容？何必说你对 [你] 家族的爱、对亲属的忠诚？[因为] 你爱我的母亲像爱你自己的父母一样，让她得到同样的安顿。

The 8-virtue list — pudicitia, obsequium, comitas, facilitas, lanificium, religio sine superstitione, ornatus non conspiciendus, cultus modicus — appears in the same order in Murdia's laudatio (CIL VI 10230) and four other Latin laudationes mulierum. The list is a closed-vocabulary slot; the editor restoring missing words consults the 8-virtue bank. 八项德目：pudicitia、obsequium、comitas、facilitas、lanificium、religio sine superstitione、ornatus non conspiciendus、cultus modicus；以同样的顺序见于 Murdia 颂辞（CIL VI 10230）和其他四篇拉丁妇女颂辞。这个目录是封闭词汇的槽位；编者补缺字时，参考的就是这八项德目库。

Wistrand 1976: 21; Mantzilas 2017, 5-case bank.

Mommsen 1863 — restoring the right-column lacunaeMommsen 1863；修补右列的残缺

The right column (II 0–17) survived in two heavily-damaged Villa Albani fragments. Mommsen, in his 1863 Berlin Academy edition (later CIL VI 1527), restored lines II 0–9 entirely from formula and parallel — he had to: the stone provided only line-endings of some letters per line. The restorations were canonical for 86 years; Dessau accepted them in ILS 8393 (1906); Durry reproduced them in his 1950 Budé. 右列（II 0–17）保存在两块残损较重的 Villa Albani 残片上。Mommsen 在 1863 年的柏林科学院版（后收入 CIL VI 1527）凭套语和平行文本，把 II 0–9 行全部补出；他不得不这样做：石头上每行只剩行尾的几个字母。这个修复维持了八十六年；Dessau 在 ILS 8393（1906）接受了它；Durry 在 1950 年的 Budé 版又照印了它。

The wager. Each bracketed phrase is Mommsen's prediction. He had the formula-frame (Wistrand 1976: 11 reconstructs the rhetorical pattern: a series of rhetorical-question slots, each carrying a virtuous-act + the husband's gratitude). The frame is solid; the specific words are the editor's call. 下方注：每一个方括号里的短语都是 Mommsen 的预测。他手里有套语提供的框架（Wistrand 1976: 11 重构了这一修辞模式：一连串修辞问句的槽位，每个承接一项德行 + 相应的感恩）。框架是稳固的；具体用词，则是编者的判断。

Gordon 1950 — a rediscovered fragmentGordon 1950；一块重新出现的残片

"In January 1949, while looking over the odds and ends in the 'Antiquario' of the Museo Nazionale Romano in Rome, in search of dated inscriptions, the writer discovered what looked at once like a piece of the Laudatio Turiae … the museum's records have nothing to indicate either its provenience or the date of its finding or of arrival in the museum."“一九四九年一月，笔者在罗马国立罗马博物馆（Museo Nazionale Romano）的”古物保存室“里翻检零散藏品，本意是寻找有纪年的铭文；忽然看到一件东西，第一眼就像是图利娅颂辞的一块残片……博物馆档案里没有任何记录能说明它的来源、发现时间或入馆日期。” Gordon 1950: 223, "A New Fragment of the Laudatio Turiae", AJA 54.

The rediscovered fragment supplied the line-endings of right-column lines II 0–9 — the same lacunae Mommsen had filled in 1863. For the first time in 86 years, his restorations could be checked against the actual stone. 这块重新出现的残片补出了右列 0–9 行的行尾；就是 Mommsen 在 1863 年补出的那段残缺。八十六年来第一次，他的修复可以对照实际的石头了。

The contemporaneity. Wistrand 1976: 10 notes the cruel timing: Durry's full 1950 Budé edition went to press just months before Gordon's AJA paper. Durry's text is "Mommsen-orthodox" because at the moment of publication no other text was available. 时间上的巧合：Wistrand 1976: 10 说这个时机很残酷：Durry 1950 年的 Budé 全本送印才几个月，Gordon 的 AJA 文章就刊出了。Durry 的文本仍然守着“Mommsen 正统”，因为付印的时候，新残片还没有发表。

Mommsen vs the stone — the eight-line box-scoreMommsen 对石头；八行对照得分

"Such great scholars and epigraphists as Mommsen and Vollmer proved to have been unable to reconstitute the original Latin text in more than one out of eight instances. On the other hand it must be acknowledged that they did not miss the sense of any line."“像 Mommsen 和 Vollmer 这样的大学者、大铭刻学家，八处里也只有一处复原出了原来的拉丁文。但也应当承认，在各行的文意上，他们都没有偏差。” Wistrand 1976: 10, reporting the post-Gordon critique.

Did the meaning actually change? — side-by-side含义是否真的变了？并列对照

The pattern. Of 8 lines: 4 preserve sense, 3 are partial (agency or detail shifts), only 1 (II 6) reverses meaning. The husband's letter-perfect words were largely unrecoverable; his argument was almost entirely recoverable from formula and rhetoric alone. This is what Wistrand's "they did not miss the sense of any line" measures. An editor with the genre-pattern in hand can recover what the husband is saying even when she gets the words wrong. 规律：八行中，四行保住含义，三行部分偏移（主语或细节），仅一行（II 6）含义反转。丈夫确切的字句几乎不可复原；他的论说则可以靠套语和修辞几乎全部复原。这就是 Wistrand 所说“在各行的文意上都没有偏差”的实际测量。编者只要把握体裁的套语，即使字词不准，也仍能读出丈夫在说什么。

What this means for the prediction question. Formula-driven restoration is a strong predictor of meaning, a weaker predictor of words, and a very weak predictor of the specific personal narrative. The Laudatio reads in the husband's voice not because the words are recoverable, but because the rhetorical-question frame is. The voice survives formulaic restoration; the man's exact phrasing did not. 这对预测问题的含义：套语驱动的修复，预测意义的能力较强，预测字词的能力较弱，对个人化叙事的预测力则极弱。颂辞能读出“丈夫的声音”，靠的是修辞问句的框架可以复原，而字句本身并不能复原。声音留下来了；那个人确切的措辞没有留下。

Della Corte's spatial critique — the 15-line problemDella Corte 的空间批评；十五行问题

After Gordon's fragment fixed the line-width of column II at 18–24 characters, Della Corte's post-Gordon critique went line-by-line through Mommsen's restoration of the rest of the right column and found 15 lines where Mommsen's restoration was either too short or too long for the now-known slot width. Each off-width restoration is a candidate for revision — not necessarily wrong, but no longer formula-justified at its current length. Gordon 的残片已经确定右列每行 18–24 个字符的宽度，Della Corte 之后的批评于是逐行检查 Mommsen 在右列其余部分的修复，发现有十五行的修复不是过短就是过长，与现在已知的槽位宽度不合。每一个过宽或过窄的修复都成了可以讨论的候选：它们未必都错，但在现有长度上已经没有套语上的根据。

"Mommsen had filled lines 1–2 and 4–9 [of the right column] as follows … and his restorations were accepted as standard (by CIL and Dessau, for example). He had got one line (7) quite right, had correctly restored tu[ae quam] in 2, inminen[tia] in 5, and per[iculo] in 9, and had correctly guessed sororem in 8. He had also sensed what was missing in lines 2, 8, and partly in 5, but he was off the track in lines 1, 4, and 6."“Mommsen 已经补出右列 1–2 与 4–9 行……他的修复曾被 CIL 和 Dessau 等接受为标准。其中一行（7）完全正确；2 行的 tu[ae quam]、5 行的 inminen[tia]、9 行的 per[iculo] 都补对了；8 行的 sororem 也猜对了。2、8 与 5 的一部分，他也察觉到缺了什么；只有 1、4、6 三行，他偏离了正轨。” Gordon 1950: 226.

Verdict — Laudatio sits mid-spectrum小结：图利娅处在中段

The pedagogical answer. Laudatio's formulaic frame is fully restorable; its individual words within the frame are not. The editor's experience reads the husband's voice and proposes plausible language — and the experience is sometimes spectacularly wrong, as the Mommsen-vs-Gordon box-score shows. The case demonstrates that within a single inscription, different slots have different predictability profiles. 笔者按：图利娅颂辞的套语框架可以补；框架之内的具体字词不可以。编者凭经验读出丈夫的声音，提出可能成立的措辞；这种经验有时很不准，Mommsen 对照 Gordon 的得分就是明证。这个案例告诉我们：同一方铭文之内，不同槽位的可预测性各不相同。

Case D · Persicus on the Artemision — proconsular reform edictD 案 · Persicus 关于阿尔忒弥斯神庙的政令；总督的改革颁令

Engelmann, H., Knibbe, D. et al. 1979–81. Die Inschriften von Ephesos (IK 11). Project SCPP edition governance/fabius-persicus-artemision.html carries the Greek text in 14 sections with full English translation.

The preamble — Claudius authorises, Persicus reflects开篇 — 克劳狄许可，Persicus 引论

Τιβερίου Κλαυδίου Καίσαρος Σεβαστοῦ
   [Γερμανικοῦ προτρέ]ψαντος αὐτοῦ
   ἐπίκριμα συμφέρον τῇ [τε Ἐφεσίων πόλει]
   καὶ ὅλῃ τῇ ἐπαρχείᾳ· ὃ ἐν Ἐφέσῳ
   προέθηκ[εν καὶ πρὸ εʹ Καλ(ανδῶν)] …
ἐν ταύτῃ διὰ παντὸς τῇ ὑπολήψει
   ὑπάρχων [πρὸ πάντων] τοὺς τῶν
   ἐπαρχιῶν προεστῶτας ἄρχοντα[ς]

[Paullus Fabius Persicus … proconsul of
Asia, sent,] at the direction of Tiberius
Claudius Caesar Augustus Germanicus, his
pronouncement (ἐπίκριμα), advantageous to
both the city of the Ephesians and the
whole province, which he published at
Ephesos on the 5th day before the Kalends …
Though I have always held the view that, above
all, the officials set over the provinces …

Two slots in one passage. The ἐπίκριμα-authorisation clause names the source of the proconsul's authority (the emperor's earlier ruling); the reflective opener positions Persicus rhetorically as the dutiful magistrate. The Greek ὑπόληψις ("view, opinion") corresponds to Latin opinio or iudicium; the Greek ὑπάρχων ("being, holding") matches Latin cum sim / existens. The translator was working in a standard proconsular register. 一段之内的两个槽位：ἐπίκριμα-许可句指明总督权力的来源（皇帝的先行裁决）；引论用修辞把 Persicus 放在尽职官员的位置上。希腊语 ὑπόληψις（“观点”）对应拉丁语 opinio 或 iudicium；希腊语 ὑπάρχων（“处于、抱持”）对应拉丁语 cum sim / existens。译者遵循的是标准的总督语域。

Greek text from project SCPP edition, ll. 1–7 (IK 11 no. 17 init.). English translation: project edition §0–§1.

The diagnosis — temples in ruin, Artemis's wealth squandered诊断病情：庙宇倾圮，阿尔忒弥斯的资财被虚耗

διόπερ φορτικὴν μὲν ἐπίγνωσιν
   ἀνεδεξάμην, ἀναγκαίαν δὲ τῇ
   λαμπροτάτηι Ἐφεσίων πόλει …
[πολλαὶ γὰρ θεῖαι οἰκίαι ἢ διὰ πυρὸς
   διεφθαρμέναι ἢ διὰ] … τὰς ἱερωσύνας
   ὥσπερ ἐν ἀπαρτείᾳ ἀποδίδονται …

Wherefore I have undertaken an investigation
that is burdensome, but necessary for the
most brilliant city of the Ephesians …
For many temples have either been destroyed
by fire or have fallen, in their collapse,
into unsightly ruin; and they sell off the
priesthoods as if at an auction; they call up
bidders of every sort to buy them, and then
they do not appoint as priest the worthiest …

Three translation choices to flag. (1) φορτικός "burdensome" is the standard Greek-Latin loan-equivalent for Latin onerosus / gravis — proconsular self-deprecation register. (2) ἐπίγνωσις renders Latin cognitio, the technical term for a magistrate's judicial enquiry. (3) ἀπαρτεία "auction" is rare; the more common Greek word for "auction" is πρᾶσις; the translator's choice suggests Latin licitatio (the Roman public auction procedure) was uppermost in his ear. The Greek reader hears, faintly, the Latin underlying register. 三处译笔值得注意：（1） φορτικός（“繁重”）是拉丁语 onerosus / gravis 的标准希腊对译；这是总督自我谦抑的语域。（2） ἐπίγνωσις 译拉丁语 cognitio（官员司法调查的专名）。（3） ἀπαρτεία（“竞价”）比较少见；希腊语更常用 πρᾶσις；译者这样选，似乎是拉丁语 licitatio（罗马公开拍卖的程序）在他耳边回响。希腊读者隐约能听到拉丁底本的语域。

Greek text: project SCPP edition, IK 11 no. 17 ll. 18–38. Latin parallels: Sherk 1969 (RDGE) on the chancery-Latin underlying Greek-face proconsular edicts.

The reform-list — Greek and reconstructed Latin in full sentences改革段：希腊原文与可复原拉丁文的全句对照

After the diagnosis the edict turns to its substantive reforms. Each is introduced by ὁμοίως ("Likewise"), translating the standard Latin chancery-bullet item. Three of the reform-clauses are reproduced below in full: the Greek text from the SCPP edition (IK 11 no. 18), a published English translation, and the Latin reconstruction implied by the Greek (parallels from Sherk 1969 / Cooley 2025). 诊断之后，政令转入具体改革。每条以 ὁμοίως（“同前”）开头，对应拉丁文书的标准分条词 item。下列三条以全句呈现：希腊原文（采自 SCPP 校本，IK 11 no. 18）、已刊英译、以及由希腊文反推的拉丁对应（Sherk 1969 / Cooley 2025）。

ὁμοίως ὅσοι ὄντες ἐλεύθεροι
ἔργα δημοσίων δούλων ἐκτελοῦσιν
καὶ τὰ κοινὰ χρήματα ἐπιβαροῦσιν
[ἀπολυέσθωσαν], ἀντικατασταθήσονται
δὲ δημόσιοι δοῦλοι ἀντ' αὐτῶν.

EN: Likewise, all those who, being free
men, are performing the service of
public slaves and burdening the common
funds [shall be discharged]; public slaves
shall be appointed in their place.

LAT (recon.): item, quicumque cum
liberi sint, opera servorum publicorum
fungantur et rationes communes onerent,
[dimittantur], et eorum loco servi
publici substituantur.

ὁμοίως τοὺς ἱερονείκας,
ὅσοι τῆς Ἀρτέμιδος ἱεροὶ εἰς τὸν
τοῦ ὀψωνίου λόγον λέγονται, οὐκ
ἀρέσκει μοι ἀντὶ μισθοῦ διηνεκοῦς
ἀλλὰ τειμῆς λαμβάνειν.

EN: Likewise, as for the sacred-games
victors who are said to be sacred to
Artemis on account of their stipend —
it is not my decision that they receive
it as a permanent salary; rather, as
an honorary payment.

LAT (recon.): item, hieronicas, quotquot
Dianae sacri propter stipendii rationem
dicuntur, non placet mihi accipere
[id] perpetua mercede, sed honoris
gratia.

ὁμοίως τοὺς ὑμνῳδούς, εἰς οὓς οὐκ
ὀλίγον τῆς πόλεως ἀνάλωμα δαπανᾶται,
ἀρέσκει μοι τοὺς ἐφήβους ἀντ' αὐτῶν
ὑμνεῖν, τοῦ δὲ Περγαμηνῶν ἔθνους
ἐξαιρέτου ὄντος ⋯

EN: Likewise, as for the hymnodes —
upon whom no small part of the city's
revenues is spent — it is my decision
that the ephebes should sing in their
place; the people of Pergamon being
exempted ⋯

LAT (recon.): item, hymnodos in quos
non parva pars vectigalium urbis
insumitur, placet mihi ephebos pro
ipsis hymnos canere, gente Pergamenorum
excepta ⋯

What the full-sentence comparison shows. Three translation strategies coexist within a few lines: (1) calque — the Greek absolute participle in §6 ὄντες ἐλεύθεροι follows the Latin cum liberi sint structurally; (2) register loan — ἱερονείκας (§8) keeps the technical Latin term hieronica in Greek transliteration, recognised across the Asian provincial chancery; (3) idiomatic recasting — ἀρέσκει μοι ("it is pleasing to me") renders Latin placet mihi, the standard Roman magistrate's "I rule" formula — the Greek translator chose the same person and mood, preserving Persicus's first-person voice across the language barrier. The bilingual cross-check therefore restores the proconsul's syntax and voice on top of the lexicon. 全句对照显示：短短几行里并存三种译法。第一种是仿造：§6 的希腊独立分词 ὄντες ἐλεύθεροι 在结构上仿照拉丁 cum liberi sint。第二种是语域借用：§8 的 ἱερονείκας 直接保留拉丁专名 hieronica 的希腊化转写，在亚细亚行省文书里通行。第三种是语义重铸：ἀρέσκει μοι（“于我所悦”）译拉丁 placet mihi，即罗马官员“我裁决”的标准语式，希腊译者保留了同样的人称和语气，让 Persicus 的第一人称声音跨语言而不中断。所以，双语互证不只复原词汇，也复原总督的句法和语气。

Greek text: project SCPP edition, IK 11 no. 18 ll. 65–110, with brackets [...] marking editorial restoration. Latin reconstructions per Sherk 1969 (RDGE) proconsular-edict register; Aeneas 2025 §6 (Cooley) on Latin-Greek translation conventions in IK Ephesos 17–19.

Verdict — the bilingual cross-check helps with form, not substance小结：双语互证有助于形式，无助于实质

Where Persicus sits on the spectrum. The proconsular protocol and the item-anaphora are as formulaic as anything in Acts I–II; an editor missing those slots could restore them mechanically from Sherk's parallels. The specific reforms — what Persicus actually decreed about the Artemision's bequests, hieronicae, and hymnodes — are the document's payload. No formula generates this content; no Greek-Latin pair recovers it. The Greek face supplies the words; the meaning requires the editor to know the temple, the period, the proconsul. Persicus sits mid-spectrum with Laudatio: formulaic frame, experience-dependent content. Persicus 在光谱上的位置：总督的官衔、礼节段和 item 式的反复，公式化程度与第一、二幕所列的没有差别；编者如果失去这些槽位，可以从 Sherk 的平行文本机械补出。具体改革（Persicus 究竟对神庙的遗赠、获胜运动员的俸给、唱诗员做了什么规定）才是这份文献独有的内容。没有套语能生成这些内容；没有希拉对照能恢复它的意义。希腊面提供字句，意义则要靠编者了解这座庙、这个时代、这位总督。Persicus 与图利娅一样处在光谱中段：框架由套语提供，内容依赖经验。

The four cases on the predictability spectrum四个案例在可预测性光谱上

The forward-pull. All four cases sit somewhere between mechanical formula-prediction and pure editorial experience. Could a machine learning system, trained on the corpus, do better — at any of them? Act IV takes up the question. 接下来的问题：四个案例都处在机械的套语预测与纯粹的编者经验之间。一个在语料上训练出来的机器学习系统，能不能在其中任何一案做得更好？第四幕接着问这个问题。

Two papers, one frame两篇论文，同一个框架

"the frequent use of abbreviations, a hallmark of the Latin epigraphic habit, further complicates our interpretation of these inscribed artefacts."“缩写的频繁使用是拉丁铭文习惯的标志，也让这些刻石的解读更加复杂。” Assael et al. 2025: 141, Aeneas introduction.

Ithaca — architecture and corpusIthaca；架构与语料

Assael et al. 2022: Methods + Extended Data Fig. 1; architecture diagram simplified from the paper's Fig. 1.

Ithaca — three benchmark numbersIthaca；三项基准数字

"While Ithaca alone achieves 62% accuracy when restoring damaged texts, the use of Ithaca by historians improved their accuracy from 25% to 72%, confirming the synergistic effect of this research tool."“Ithaca 单独工作的修复准确率为 62%；历史学家借助 Ithaca，准确率从 25% 升到 72%，显示了这一研究工具的协同效应。” Assael et al. 2022: 280, abstract.

The synergy claim. The headline figure is not Ithaca's 62%; it is the human-machine combined 72%. Neither alone matches it. The implication: the right framing is not "AI replaces editor" but "AI gives editor a working partner". 协同的论点：真正的标题数字是人机合作的 72%，而非 Ithaca 单独的 62%。两者单独都达不到。这意味着合适的框架是“AI 给编者一个能一起工作的搭档”，而非“AI 取代编者”。

Case study — the Chalcis decree (IG I³ 40)案例；哈尔基斯法令（IG I³ 40）

An oath of allegiance sworn by the city of Chalcis to Athens, traditionally dated to 446/5 BCE on the three-bar-sigma criterion (the same criterion that gave us Segesta's 458/7). Recent palaeographic and historical scholarship has challenged the date, proposing 424/3 BCE on the basis of contextual evidence. 哈尔基斯城向雅典宣誓效忠的誓文。传统上按三划西格玛标准定为公元前 446/5（就是给塞格斯塔 458/7 的那条标准）。近年的古文字与历史研究对这个纪年提出挑战，依上下文证据主张公元前 424/3。

"the (in)famous Chalcis decree (IG I³ 40 …), which records an oath of allegiance sworn by the city of Chalcis to Athens and traditionally dated to 446/5 BC, is attributed by Ithaca to 420 BC, therefore concurring with the lower dating hypothesis of 424/3 BC proposed by more recent scholarship. … Ithaca's predictions are on average only 5 years off the more recent re-evaluations."“（臭）名昭著的哈尔基斯法令（IG I³ 40 …），记录哈尔基斯城向雅典宣誓的效忠誓，传统定为公元前 446/5，如今 Ithaca 把它归入公元前 420 年左右，与近年学界提出的 424/3 较低纪年方案相合。…… Ithaca 的预测平均只与近年的重新评估相差五年。” Assael et al. 2022: 282, Chalcis-decree case study.

What Ithaca does here. Independently of the three-bar-sigma palaeographic argument, and independently of the historical-context argument, Ithaca's chronology head predicts a date close to the lower-dating hypothesis. The model's prediction is convergent evidence — neither replacing the human arguments nor confirming them mechanically, but landing on the same chronology by a different path. Ithaca 在这里做了什么：它不依赖三划西格玛的古文字论证，也不依赖史学的上下文论证，年代头独立预测出一个接近较低纪年方案的年份。这个预测是汇合的证据；它不取代人的论证，也不机械地证实人的论证；它沿另一条路到达同一个纪年。

Where Ithaca would help our cases — Segesta and ATLIthaca 能帮本课什么；塞格斯塔与贡金表

Where Ithaca cannot reach — Latin, abbreviations, unique contentIthaca 做不到的地方；拉丁、缩写、独有内容

Three limits of Ithaca, all addressed (in part) by Aeneas in the next half of Act IV. Ithaca 的三个限制，都由 Aeneas 在本幕后半部分回应。

Aeneas — Latin, contextual, generativeAeneas；拉丁、上下文、生成式

The 2025 follow-up from the same Assael / Sommerschield group, with the British classicist Alison Cooley joining the team. Three substantive advances over Ithaca: 2025 年 Assael / Sommerschield 团队的后续工作，英国古典学家 Alison Cooley 加入。相对 Ithaca 有三项实质进步：

Assael et al. 2025: 141–43, Aeneas introduction and architecture.

The dual of restoration — the model as auditor, not only generator修复的对偶：模型作审计者，不只是生成者

Week 14 asks the model to fill gaps. The sharper, dual question turns Aeneas' unknown-length head around: feed EDCS's declared [N] gap-lengths to the model and flag where its predicted length disagrees with the editor's count. The markup itself encodes the editor's confidence — and the corpus shows editors almost never confess ignorance of the count: 第十四周请模型填缺。更锋利的对偶问题，是把 Aeneas 的变长修复头反过来用：把 EDCS 声明的 [N] 缺位长度喂给模型，标出模型预测的长度与编者数字相左的地方。这个标记本身就写进了编者的信心；而语料显示，编者几乎从不承认自己不知道字数：

Why this matters. EDCS commits to a letter-count 1.18 million times but admits an unknown length only 43 times — so almost every count is a falsifiable claim a model can test. This turns the corpus from a training set into an object of critique: the same markup that trains the model also gives the model something to check. 为什么重要：EDCS 已经承诺具体字数 118 万次，承认长度未知只有 43 次；所以几乎每一个计数都是模型可以证伪的论断。这把语料从训练集变成了批判对象：训练模型用的同一批标记，也给了模型可以核验的东西。

Convention counts from the EDCS auditor (epidoc/epipap/edcs-restoration-auditor); Aeneas unknown-length head, Assael et al. 2025 §architecture.

Aeneas — Table 1 benchmark numbersAeneas；表一基准数字

The headline. Historian solo achieves a Character Error Rate (CER) of 39%; Aeneas alone achieves 23.1%; historian working with Aeneas parallels + prediction achieves 21.4% — slightly better than Aeneas solo. The synergy claim is replicated and strengthened. For dating: historian solo at 31.3 years off, Aeneas solo at 12.8 years, combined at 14.1 years — Aeneas alone is the best on this metric, suggesting human input slightly hurts dating accuracy where the model has strong corpus-priors. 要点：历史学家单独的字符错误率（CER）39%；Aeneas 单独 23.1%；历史学家用 Aeneas 平行 + 预测合作 21.4%，略优于 Aeneas 单独。协同的论点被复制并加强。年代上：历史学家单独差 31.3 年，Aeneas 单独 12.8 年，合作 14.1 年；Aeneas 单独反而是该项最佳，说明在模型已有强语料先验时，人的介入反而稍损纪年准确度。

Assael et al. 2025: 143, Table 1.

Aeneas on the abbreviations problem (Act I)Aeneas 回应缩写问题（第一幕）

Aeneas was designed in awareness of the Latin abbreviation problem. The architecture handles abbreviations in three ways: Aeneas 的设计本来就考虑了拉丁缩写问题。它的架构用三种方式处理：

• Training on already-expanded text. The LED corpus standardises abbreviations to their expanded forms wherever the editorial tradition has already done so; the model learns the formula-environment in expanded form, then can recognise compressed forms by translation. 训练时用已展开的文本。LED 语料在编辑传统已展开缩写的地方统一用展开式；模型在展开式里学套语的环境，再通过对应关系识别缩写式。
• Contextual parallel-retrieval. When restoring an abbreviation, Aeneas retrieves parallel inscriptions where the same abbreviation appears in the same formula-slot. The parallels show how previous editors expanded the abbreviation, narrowing the candidate menu. 上下文平行检索。修复一个缩写时，Aeneas 检索同一缩写出现在同一套语槽位里的平行铭文。这些平行文本显示此前的编者怎样展开这个缩写，从而收窄候选菜单。
• Variable-length generation. Because abbreviations expand to texts of unknown length, the variable-length restoration head is what makes abbreviation-expansion tractable. Ithaca's fixed-length approach could not generate a 22-character expansion of a 4-letter abbreviation; Aeneas can. 变长生成。因为缩写展开后的长度未知，只有变长修复头才能处理展开。Ithaca 的定长方法没法把四个字母的缩写生成为二十二个字符的展开式；Aeneas 可以。

The architectural payoff. The combination — contextual parallels + variable-length generation — is what Act I asked for. The OHRE Appendix II's 432 entries are no longer the editor's only narrowing-tool; Aeneas's corpus-wide retrieval substitutes for an editor who has memorised every parallel inscription. Aeneas does not replace the editor; it gives her a memory she could not otherwise have. 架构上的回报：组合；上下文平行加变长生成，就是第一幕需要的。牛津附录二的 432 条目，不再是编者唯一的收窄工具；Aeneas 的语料级检索可以替代编者记不住的那些平行铭文。Aeneas 不取代编者，只是给她一份原本得不到的记忆。

Aeneas on the formula problem (Act II)Aeneas 回应套语问题（第二幕）

Aeneas's parallel-retrieval is a formula-identification engine in disguise. When the model returns the top-N parallels for a target inscription, those parallels are inscriptions sharing the target's formula-signature. The historian reading the parallels reads (in effect) a corpus-grounded answer to "which formula does this inscription deploy?" Aeneas 的平行检索，实际上是一台伪装的套语辨识引擎。模型对一条目标铭文返回前 N 条平行时，这些平行就是与目标共享套语标识的铭文。历史学家读这些平行，等于得到一个语料级的回答：“这条铭文用了哪些套语？”

"historians agreed that parallels generated by Aeneas enhanced their contextual understanding and interpretation of the target inscriptions. More specifically, when provided with Aeneas' parallels, historians reported that these could serve as a starting point for historical inquiry 75% of the time. This increased to 90% when Aeneas' predictions for the three epigraphic tasks were also included."“历史学家一致认为，Aeneas 生成的平行文本增进了他们对目标铭文的上下文理解与诠释。具体来说，拿到 Aeneas 的平行文本时，历史学家报告可以作为历史研究起点的占 75%。再加上 Aeneas 三项铭文任务的预测，比例升到 90%。” Assael et al. 2025: 143–44.

What this means for Act II. The project's 58-formula bank is a human-curated answer to the question "which formulae recur across the Greco-Roman corpus?" Aeneas offers a complementary answer: not a curated bank, but a per-inscription parallel-pool. The two can be triangulated — the bank tells the editor what formula-categories exist; Aeneas tells the editor which inscriptions in the corpus exhibit that formula. Together they collapse some of the editor's burden of memorisation. 这对第二幕有什么意义：本项目的 58 套语库，是对“希罗语料中复现哪些套语？”这一问题的人工编次回答。Aeneas 提供互补的回答：它给出的是逐铭文的平行池，而非编次好的库。二者可以互证；库告诉编者有哪些套语类别，Aeneas 告诉编者语料中哪些铭文呈现该套语。合起来用，可以减轻编者记忆的一部分负担。

Aeneas on the prediction problem (Act III, all 4 cases)Aeneas 回应预测问题（第三幕的四个案例）

The pattern. Aeneas closes the formula-and-parallel gap; it does not close the experience-and-substance gap. The editor's role is preserved precisely at the slots Act III's verdict-tables flagged as experience-dependent. 规律：Aeneas 弥合套语与平行的缺口，弥合不了经验与实质的缺口。编者的角色，就留在第三幕结论表里那些“经验依赖”的槽位上。

Case study — Res Gestae Divi Augusti案例；奥古斯都神功录（Res Gestae）

The 2025 paper closes with a case study on the Res Gestae Divi Augusti (RGDA), the most famous Latin inscription of the early Empire — Augustus's autobiographical account of his deeds, transcribed onto walls at Ankara, Apollonia, and Antioch. Aeneas's contribution: not restoration (the text is well-preserved) but attribution analysis. 2025 年的论文以奥古斯都神功录（Res Gestae Divi Augusti, RGDA）作收尾案例；这是早期帝政最负盛名的拉丁铭文，奥古斯都自述功业的文本，刻在安卡拉、阿波罗尼亚、安提阿的墙上。Aeneas 的贡献在于归属分析；文本本身保存较好，用不着修复。

"Aeneas' hypotheses for attribution of the RGDA, aggregated across each chapter, [show that] each chapter was processed individually. The resulting distributions [suggest that] sites like Trento, Baetica and Ercolano are secondary to their shared function as provincial votive inscriptions [of a kind structurally similar to RGDA chapters]."“Aeneas 关于 RGDA 归属的假设按章汇集；各章独立处理 …… 所得分布显示，Trento、Baetica、Ercolano 等地的候选属于次要，它们与 RGDA 各章共享的功能是：行省所立的愿铭（与 RGDA 的章节结构相似）。” Assael et al. 2025: 144–45, RGDA chapter-wise attribution.

What this demonstrates. Aeneas reveals structural affinities the human-curated bibliography may not have surfaced. RGDA was not historically read in dialogue with provincial votive inscriptions; Aeneas's chapter-wise retrieval finds those affinities at the level of phrasing and function, opening a new comparative-reading possibility. The model finds parallels the historian would not have looked for — and the historian decides whether the parallels are meaningful or coincidental. 这说明：Aeneas 能揭示人工编次书目未必浮现的结构亲缘。RGDA 历来没有与行省愿铭对读过；Aeneas 的章节级检索在用词和功能层面发现了这层亲缘，开启一种新的比较阅读的可能。模型找出历史学家不会去找的平行；历史学家再判断这个平行有意义还是纯属巧合。

The two engines on Acts I–III两个引擎在第一到第三幕的应用

The synergy thesis — model + editor > either alone协同的论点；模型加编者优于任何一方单独工作

Both papers make a structurally identical claim. Ithaca 2022's headline figure: historian solo 25%, Ithaca solo 62%, historian + Ithaca 72%. Aeneas 2025's headline figure: historian solo CER 39%, Aeneas solo CER 23%, historian + Aeneas + prediction CER 21%. 两篇论文做的是结构相同的论证；笔者认为，它们实际上是同一个假说的两次实验。Ithaca 2022 的标题数字：历史学家单独 25%、Ithaca 单独 62%、合作 72%。Aeneas 2025 的标题数字：历史学家单独 CER 39%、Aeneas 单独 CER 23%、合作 21%。

"Aeneas demonstrates flexibility by handling both fixed and unknown restoration lengths. … Aeneas completely changed my historical focus."“Aeneas 显示出它的灵活性，能处理定长与未知长度的修复…… Aeneas 完全改变了我的史学焦点。” Assael et al. 2025: 144, qualitative feedback from participating historians.

What the synergy means in practice. The model widens the editor's parallel-pool from "what I have read" to "the entire corpus". The model also gives the editor confidence scores she would not otherwise have — the parallels' similarity scores, the restoration head's probability distribution. Confidence scores let the editor work on the contested slots with explicit numerical uncertainty, instead of with prose-only hedges. That is a methodological gain, not just a labour-saving one. 笔者认为协同实际指的是：模型把编者的平行池从“我读过的”扩大到“整个语料”。模型还给编者原本得不到的概率分：平行文本的相似度、修复头的概率分布。概率分让编者在有争议的槽位上用明确的数字化不确定性来工作，而不只用散文里的模糊措辞。这是方法上的进步，不只是省工。

The bracket reader's responsibility读方括号的人的责任

"My text will be based on that of Mommsen (and Durry). But there will be quite a few exceptions, I shall adopt a different text when I consider that there are good reasons for doing so. Such divergences from Mommsen's (and Durry's) text will be noted in the apparatus criticus and the author of the reading substituted will be named. In these cases I will also report suggestions made by other scholars. I think it is only fair that the reader should be acquainted with the alternative solutions which have been proposed."“我的版本将以 Mommsen（与 Durry）的版本为基础；但会有若干不同，在有充分理由时我会采纳别的读法。这些与 Mommsen（与 Durry）的差异都会在校勘栏里标明，并署上所采读法的提出者的名字。在这些地方，我也会报导其他学者提出的方案。我认为，让读者知道已经提出过的其他可行方案，才算公允。” Wistrand 1976: 12, on the editor's apparatus discipline.

Wistrand's discipline is the model's discipline. Whether the bracket was filled by Mommsen in 1863, by Wistrand in 1976, by Ithaca in 2022, or by Aeneas in 2025, the reader's claim on the editor / model is the same: name the alternative restorations, name the basis for the preferred one, name the residual uncertainty. The bracket is the editor's mark; the reader judges. The model does not change that economy. It changes what the editor knows when she places the bracket — and it makes the residual measurable: the EDCS letter-count auditor runs Wistrand's discipline across all 537,262 inscriptions at once, not just the four cases on this stage. Wistrand 的规矩，也就是模型的规矩。无论方括号是 Mommsen 1863 填的、Wistrand 1976 填的、Ithaca 2022 填的、还是 Aeneas 2025 填的，读者对编者或模型的要求都一样：列出其他可行的修复，列出所选读法的依据，列出剩余的不确定性。方括号是编者的标志；读者来评判。模型没有改变这场交换；它改变的是编者下笔时知道多少，而且让剩余的不确定可以计量：EDCS 字数审计器把 Wistrand 的规矩一次施加到全部 537,262 条铭文上，而不只是台上的四个案例。

Bridge to Week 14承接第十四周

Week 14 takes Ithaca as a working tool and asks the students to use it on the four Wk13 cases. The questions will be: 第十四周把 Ithaca 当作可以上手的工具，请学生把它用在本周的四个案例上。要问的是：

• On ATL. Does Ithaca's top-20 restoration of damaged year-entries cluster (favouring ATL maximal) or diffuse (favouring Paarmann minimal)? 论贡金表。Ithaca 对残缺年表条目的前 20 个修复，是聚集的（偏向 ATL 最大派），还是分散的（偏向 Paarmann 最小派）？
• On Segesta. What date does Ithaca's chronology head propose for IG I³ 11 independently of either Habron or Antiphon being supplied? 论塞格斯塔。在不提供 Habron 或 Antiphon 的情况下，Ithaca 的年代头为 IG I³ 11 给出什么纪年？
• On Laudatio. Ithaca cannot read the Laudatio (Latin); Week 14 substitutes Aeneas. Does Aeneas's parallel-retrieval surface other laudationes mulierum the project has not catalogued? 论图利娅颂辞。Ithaca 读不了图利娅（拉丁文）；第十四周用 Aeneas 代替。Aeneas 的平行检索能不能浮现本项目没有编录的其他妇女颂辞？
• On Persicus. Aeneas's bilingual cross-retrieval will return parallel proconsular edicts in both languages. Does the retrieval-set overlap with Sherk 1969's hand-curated list? 论 Persicus。Aeneas 的跨语言检索会返回双语的 proconsul 颁令平行。检索结果与 Sherk 1969 的人工编次表重合多少？

Reading & sources — Latin abbreviation + Greek tradition阅读与出处；拉丁缩写 · 希腊传统

Latin abbreviation references

Bruun, C. & Edmondson, J. (eds.) 2015. The Oxford Handbook of Roman Epigraphy. Oxford. — esp. Appendix II "Epigraphic Abbreviations", pp. 787–813.

Retrieval target: Strechie's article on proper nouns and Latin epigraphic abbreviation, for direct confirmation of the worked IDR / ISM examples.

Retrieval target: Lassère, Manuel d'épigraphie romaine, second edition, for comparison with the OHRE abbreviation list.

Retrieval target: Cagnat, Cours d'épigraphie latine, fourth edition, for comparison with the OHRE abbreviation list.

Latin funerary formulae

Bennett, L. 2020. Global Patterns of Commemoration in Roman Epitaphs: A Quantitative Spatial Analysis of the Most Common Formulae. PhD thesis, University of Kent (SECL), March 2020; Kent Academic Repository kar.kent.ac.uk/83390. — 104,007 Latin epitaphs drawn from EDCS. D·M and D·M·S are counted as mutually exclusive categories (Table 3.2, p. 63); overall shares p. 84; regional breakdown Table 4.2, p. 85; the sample is not restricted by date (p. 74).

Greek tradition references

McLean, B. H. 2002. An Introduction to Greek Epigraphy of the Hellenistic and Roman Periods (323 BC – AD 337). Ann Arbor. — esp. §2.05 "The Development of Abbreviations", §2.06.3 "Abbreviation Marks", §2.07–08 numerals, §5 Roman names in Greek inscriptions.

Henry, A. S. 1977. The Prescripts of Athenian Decrees. Mnemosyne Suppl. 49. Leiden. — the canonical study of bouleutic / demotic / prytany formulae.

Reading & sources — case-study sources + ML papers阅读与出处；案例的出处 · 机器学习论文

Case-study sources (Act III)

ATL. Meritt / Wade-Gery / McGregor, The Athenian Tribute Lists, four vols. Princeton — the maximal edition; local handles currently cover vols. II–IV and related Meritt/Dow/West scholarship, while vol. I remains a retrieval target. Paarmann, B. 2007. Aparchai and Phoroi. Diss. Fribourg. — the minimal edition.

Segesta. Meiggs, R. & Lewis, D. 1969. A Selection of Greek Historical Inscriptions, no. 37. Oxford. Chambers, M., Gallucci, R. & Spanos, P. 1990. "Athens' Alliance with Egesta in the Year of Antiphon." ZPE 83: 38–63. Henry, A. S. 1992. "Through a Laser Beam Darkly: Space-age Technology and the Egesta Decree (IG I³ 11)." ZPE 91: 137–146. Badian, E. 1989. "History from 'Square Brackets'." ZPE 79: 59–70.

Laudatio Turiae. Mommsen, T. 1863. CIL VI 1527. Wistrand, E. 1976. The So-called Laudatio Turiae: Introduction, Text, Translation, Commentary. Studia Graeca et Latina Gothoburgensia 34. Göteborg. Gordon, A. E. 1950. "A New Fragment of the Laudatio Turiae." AJA 54.3: 223–226. Della Corte's letter-count critique is used here through Wistrand/Gordon and remains a direct-article retrieval target. Mantzilas, D. 2017. Laudationes Mulierum.

Persicus. IK Ephesos 17–19 (Engelmann et al. 1979–81). Sherk, R. K. 1969. Roman Documents from the Greek East. Baltimore. Knibbe, D. 1979. On the Ephesian Artemis estates. Rogers, G. M. 2014. The Mysteries of Artemis of Ephesos.

Machine-learning papers

Assael, Y., Sommerschield, T., et al. 2022. "Restoring and attributing ancient texts using deep neural networks." Nature 603: 280–283. — Ithaca.

Assael, Y., Sommerschield, T., Cooley, A., et al. 2025. "Contextualizing ancient texts with generative neural networks." Nature (published 23 July 2025). — Aeneas.

Connect — every Wk13 surface in one place网状导览；Wk13 所有页面

End of deck · ←/→ to navigate · Cmd/Ctrl-P to print · ⓘ register-mode: introduction · voice profile: chinese-profile-CYW.md幻灯片完 · ←/→ 导航 · Cmd/Ctrl-P 打印 · ⓘ 语域：introduction · 文风守则：chinese-profile-CYW.md