Why the Roman Republic Had No PartiesForming a Faction Was Itself an Accusation

1. Self-designation and prohibition in the inscriptions

At present 182,853 Latin inscriptions can be searched by computer, and 984 of them contain collegium or a word of the same family. A collegium is a society or association: a guild formed by carpenters was called a collegium, and so was a mutual-aid burial society. In the same body of inscriptions, only 1 contains coniuratio or a word of its family. The word coniuratio literally means swearing together, and is usually translated as conspiracy or plot.

This 1 inscription is not one that some group carved for itself; it is a colonial law. In 44 BC Caesar ordered the founding of the colony of Urso, and article 106 of its law reads:

Quicumque colonus coloniae Genetivae erit, quae iussu Gai Caesaris dictatoris deducta est, ne quem in ea colonia coetum conventum coniurationem ……

Whoever shall be a colonist of the colonia Genetiva, which was founded by order of Gaius Caesar the dictator, shall not [hold] in that colony any gathering, assembly, sworn association ...¹

In this body of inscriptions coniuratio occurs only this once, and it occurs in a legal clause forbidding sworn association. No one used the word to name a group to which they belonged.²

Nearly all of the 984 inscriptions containing collegium were set up by the groups themselves: the titles of presidents, membership lists, the statutes of burial societies, and dedications to patrons. Among the inscriptions I sampled and registered, the first comes from Praeneste in the first century AD:

Publius Valerius Publi libertus Mahes, magister quinquennalis collegii fabrorum tignuariorum lustri XIII ……

Publius Valerius Mahes, freedman of Publius, quinquennial president of the guild of carpenters for the thirteenth lustrum ...³

The great difference between the two words' numbers in the inscriptions is not because one of them was rare everywhere else to begin with. I divided the corpus of classical Latin literature by date into three periods, Republican, Augustan, and early-to-middle Imperial, 5.14 million tokens in all; in it collegium occurs 195 times and coniuratio 225 times, the latter even slightly more often.⁴ Romans used both words when they wrote; but when they carved their own names and the names of their groups on stone, they used only the first.

What I want to know is why, for the Romans, saying of a group of people of roughly equal standing "you are in league together" was in itself an accusation. My concern is not which groups were later convicted, but by which definitions and with which expressions the Romans were already accusing a group of people when they put them under one name.

Why the Roman Republic had no parties is a long-standing question. I cannot answer the whole of it, and discuss only one part: if a group of Romans formed something like a party, what would their contemporaries have called them, and did those names carry praise or blame?

Below I call such groups horizontal political associations, meaning groups or bands that Romans of roughly equal standing formed for common interests; this differs from the vertical relationship between patron and client. If Romans wanted to present a horizontal political association as a legitimate group, they had only two ways of doing it. One was to present it as the fellowship of all mankind, that is, the relation by which all human beings join and deal with one another, which comes from Stoic doctrine; the other was to present it as a group authorized by the public authority, which comes from law. For a particular group of Romans gathered for a common interest, the Romans had no ready-made name by which to call them legitimately. Hence an accuser did not have to prove that these people had done anything wrong, only that they were in league: that they were merely a part of the whole citizen body and had no public authorization. The accused could not deny having joined with others; they could only dispute the name, arguing either that their side was the whole, or that the relationship should be called something else.

2. Which words to count, and on what basis

To learn how the Romans regarded horizontal political associations, one must first decide which words to count. These words cannot be chosen by a modern reader's feel for the language, so I registered them separately from three kinds of Roman texts. The names for groups in legal texts are collegium, sodalicium (society, association), corpus (corporate body; literally, body), societas (partnership, alliance), and universitas (the group as a whole). The names used for accusation in rhetorical works and histories are factio (faction), coniuratio (sworn association, conspiracy), conspiratio (unanimity, plot), coitio (private combination), and seditio (civil disturbance, mutiny). The words in letters and philosophical works that describe relations between people or the process of joining are amicitia (friendship), fides (good faith), consensio (agreement), coire (to come together), conspirare (to be of one mind), and consentire (to agree).⁵ There are forty-one words in all; two more, coetus (gathering) and conventus (assembly), were added later.⁶

Each word is counted once in each of four sources, given in four columns. The literature column counts the occurrences of the word, by lemma, in the three periods of the classical Latin corpus. A lemma groups all the inflected forms of a word under one base form; Latin nouns inflect for case and verbs for person, and without grouping them first one cannot count accurately. The LatinISE column comes from another corpus, annotated for genre and date. The inscriptions column uses the more than 180,000 inscriptions mentioned above and records the number of inscriptions in which the word appears, with the number of occurrences in parentheses. The Digest column counts the occurrences of the word in all fifty books of the Digest.

The inscriptions and the Digest were not lemmatized and could only be matched on the first few letters of a word form, so the numbers in these two columns cannot be converted into ratios and compared with the first two columns. The literary corpus counts lemmatized lemmas, while the inscriptions count the various word forms that begin with certain letters; the two are not the same unit of count. Hence this table can only compare how large the numbers of the various words are within one column; one column's number cannot be divided by another's.

Of the forty-one words, the six that bear directly on this question are as follows:⁷

Taking each class of words together gives the same result. The seven accusatory words together appear in 48 inscriptions of this body, 44 of them with factio. I read these 44 one by one: 35 refer to circus racing factions, 4 to theatrical troupes, 4 have a political sense, and 1 cannot be determined. Once the racing and theatrical uses are removed, accusatory words remain in only 8 of the 180,000 inscriptions. The six group names together appear in 1,399 inscriptions. In the literary corpus these same seven accusatory words total 993 occurrences over the three periods and these same six group names 1,117, the former being 0.89 times the latter; in the inscriptions the ratio is 0.034, a twenty-six-fold difference.⁸

The 4 inscriptions in which factio has a political sense are concentrated in a very late period. In AD 196 to 197, one from Pannonia records a campaign undertaken at the emperor's command to suppress the Gallic faction; in 313, one from Africa, dedicated to Constantine, calls him the extinguisher of the tyrant's faction; in 315 to 316, one at Rome, the Arch of Constantine, records his campaign against the tyrant and his whole faction; in 324, one at Rome calls him the extinguisher of the tyrants' factions. No inscription of the Republican period or of the early Principate uses the word in this way. Only after the late second century did public monuments begin to call political enemies a factio, whereas Cicero had long used the word in this way in literary works.⁹

The same word can mean entirely different things in inscriptions and in literature. In Cicero, sodalicium always appears together with the lex Licinia de sodaliciis and refers to a band formed for electoral corruption; it is a criminal charge. I read the 23 inscriptions containing the word one by one, and none treats it as a charge: all are names that wool-combers, fullers, marble-workers, carpenters, youth associations, and various cult groups gave to their own groups.¹⁰ universitas is the reverse case: it does not occur once in the three periods of the classical Latin literary corpus, occurs 46 times in the whole Digest, and appears in 2 inscriptions, one of which is the self-designation of a community of Jews resident at Ostia, between AD 151 and 200, on buying land with jointly contributed money.¹¹ Of the two words Roman law used to name groups, one does not appear in literature at all, and the other (corpus) in literature mainly means "body", so that its use in the sense of "group" cannot be seen from frequency counts alone.¹²

One association's statutes carry clauses of permission and of prohibition on the same stone. These are the statutes of a burial society at Lanuvium in AD 136, which begin by copying a clause of a decree of the senate:

kaput ex senatus consulto populi Romani: quibus permissum est convenire collegiumque habere liceat, qui stipem menstruam conferre volent in it collegium, coeant neque sub specie eius collegi nisi semel in mense coeant stipem conferendi causa unde defuncti sepeliantur

Clause from the decree of the senate of the Roman people: those who have been permitted to assemble may have an association; those who wish to pay a monthly fee into this association may come together, but they may not, under the pretext of this association, come together more than once a month, and the fees paid are for burying deceased members.

In a later section of the statutes, the society uses an accusatory word to restrain its own members:

si quisquis seditionis causa de loco in alium locum transierit, ei multa esto HS IIII nummum

Whoever moves from one seat to another for the sake of making a disturbance shall be fined four sesterces.¹³

The people who carved these statutes knew that whether their group could exist depended on official permission; they also knew which word was used to rebuke people.

3. How praise and blame were judged sentence by sentence

Counting how often words occur shows only which words the Romans used often, not how they regarded the things those words referred to. To know their attitudes, one must judge sentence by sentence.

In the lemmatized literary corpus I extracted every complete sentence containing any of fifteen core words, 2,463 sentences in all: 822 from the Republican period, 723 from the Augustan, and 918 from the early-to-middle Imperial. Sentences are divided at full stops, question marks, and exclamation marks, and no sentence crosses a chapter boundary.¹⁴

Each sentence is judged on three items. The first is which kind of text the sentence belongs to, divided by genre into nine kinds: oratory, historiography, biography, letters, imperial rescripts, poetry together with the novel and drama, philosophy and rhetoric, technical and scholarly works, and legal texts. The second is which kind of combination the sentence speaks of, divided into alliance against the state, political faction and electoral combination, organized group, external alliance, and private relationship; there is also a category "none of these", which includes the mating of animals, grammatical examples, and cases where people gather in one place without forming a group. The third is the attitude the sentence takes toward the group it names, divided into condemnation, neutral, approval, legal prohibition, and legal permission, with two further cells, "irrelevant" and "cannot judge".¹⁵

The third item judges the sentence's attitude toward the group it names, not the author's attitude toward the whole affair. If a sentence merely records some alliance, such as who its leader was, in which year it was exposed, or who came forward to inform, the sentence is judged neutral even if the whole work condemns that alliance; to be judged condemnation, the sentence itself must contain a pejorative word, a criminal charge, or a penalty, or the narrator must say outright that the thing is bad. To avoid circular reasoning there is a further rule: the fact that the word being judged itself carries the sense of a charge does not count as "the sentence contains a charge". The word coniuratio itself carries a charge; if the word were taken as evidence of itself, every sentence containing it would be judged condemnation, and the count would only reproduce the classification of the word list.¹⁶

This judging was done independently twice by others, and whoever did the second pass could not see the answers of the first. The two passes disagreed in 549 cells in all. A cell here means the result for one sentence on one item of judgment; each item of judgment is called an axis, and a sentence on which the passes disagree on two axes counts as two cells. I first drew sixty cells and adjudicated them myself, checking "whether at least one of the two answers is correct": in 3 cells both were wrong, 0.0500, below the threshold of one in ten set in advance, and so, as originally planned, I did not re-judge every disagreement cell by cell.¹⁷ Later I nevertheless adjudicated all 549 cells. Only after finishing did I realize that this check must be computed separately for the two axes, or one cannot see which axis has gone wrong: on the axis "which kind of combination", both passes were wrong in 29 of 111 cells, 0.2613; on the attitude axis, in 18 of 277 cells, 0.0650, and after removing 12 cells in which, because of the cross-axis rule, the correct answer was never among the two candidate answers, in 6 of the remaining 265 cells, 0.0226. When I designed the check, I put both axes into one denominator and also counted in another remaining batch of 94 cells: the three batches together make 482 cells, with both passes wrong in 48, 0.0996, just under one in ten. This merged ratio hid the difference between the two axes: from it alone one cannot see that the error rate on one axis was already two and a half times the threshold.¹⁸

The error rate on the axis "which kind of combination" exceeded the threshold not because the judges could not read the original, but because I had altered the judging criteria myself. After the two passes were complete, I twice revised the classification table, adding two subheadings to the category "none of these": first, people gathering in one place without forming a group; second, the word being used in this sentence in a sense outside the "association" group of meanings. Before these additions, when the judges met sentences such as Livy's account of the women gathering at Veturia's house, or Curtius' account of a crowd gathering for a funeral, the table had no category for them, and they could only pick one of the association categories or enter "cannot judge". After the additions, the correct answer for these sentences was no longer among the two candidates the two passes had given. So each time the classification table is revised, one should record at the same time which kind of already-judged sentences the revision affects, and whether the conclusion of the earlier check still holds for that kind.¹⁹

After the 549 cells were adjudicated, the table contained 2,451 sentences, each with one result on attitude: condemnation 822, neutral 801, approval 329, irrelevant 476, legal prohibition 19, legal permission 1, cannot judge 3.²⁰

The figure of 822 condemnations is too high and can be taken only as an upper bound. There are 798 sentences that both passes judged condemnation. With a fixed random seed I drew a hundred of them and reread each without any search-hit markings, asking only whether the sentence itself contained a pejorative word, a charge, or a penalty: 35 did not, 0.35. Even if the eight I was unsure about on rereading (the questions were all "does being put to death count as a penalty" and "does extreme violence count as a pejorative word") were all changed to "yes", the proportion would still be 0.27.²¹ This shows that the condemnation column was not judged strictly on "this sentence only"; a considerable share of the sentences were judged by readers who had read the surrounding text, or from their own knowledge of the affair.

I therefore put every occurrence of these words back into the original text and reread it, with the two sentences before and after attached for reference, but in judging attitude I looked only at the sentence itself, not at the sentences around it. In this way I reread 2,049 items, covering fourteen words. From the third word on, the previous judgments were masked before rereading: the material I received contained only the number, period, author, work, reference, the sentence itself, and the two sentences before and after it, and the words "condemnation", "approval", "neutral", "legal prohibition", and "legal permission" each appeared 0 times in the material. So for these words the agreement rate between the two judgments is not an inflated upper bound.²²

Comparing the results of the rereading with the previous judgments cell by cell, the agreement rates of the various words lie between 0.5257 and 0.8841.²³ The numbers of cells judged condemnation and judged approval both fell. The earlier judgments of coniuratio and factio lay mostly at the condemnation end, and the rereading shows only a fall in condemnations; collegium originally had 42 cells judged approval and 15 judged condemnation, and both fell after rereading. When all 652 items for societas had been read, condemnation fell from 114 cells to 71 and approval from 186 to 143, each down by 43, exactly the same.²⁴ So the approval figures in the earlier table were likewise too high and can be taken only as upper bounds. The "this sentence only" method also misjudges three kinds of sentence: first, ironic sentences, such as "so sacred in your heart is the bond of partnership", which is literally favourable but in fact mocking; second, verse running over the line, as in the two lines of Virgil where the half-sentence that characterizes the affair is in the next line, whereas when Seneca quotes the same two lines in prose, both lie within one sentence; third, the corpus treats the full stop in abbreviations such as rei p. as the end of a sentence, and so cuts out incomplete fragments. These three problems cannot be removed by changing the rules for dividing sentences.²⁵

Before computing ratios, two further steps are needed. The judging material assigns each sentence one headword, and when two target words occur in one sentence, the sentence counts under only one of them. In addition, some occurrences of these fourteen words do not refer to any kind of association; for example, coeo in agricultural writing, natural history, and medical writing refers to wounds closing, soil and crops joining, and animals mating, and these sentences are not counted in the denominator. Of the 2,049 items, 75 were excluded under this rule. The word most affected is coeo: it occurs 476 times in the corpus, and only 109 items enter the denominators of the ratios below. The denominator of every ratio that follows is the number of items after exclusion.²⁶

4. Condemnation has nothing to do with part of speech

My initial supposition was this. In Latin the same root often has both a verb and a noun, as with conspirare and conspiratio, coire and coitio, sociare and societas; condemnation might be aimed mainly at nouns that call a group of people a single entity, or at nouns that name the action of "combining". Once the material was read, the condemnation rates of the words (the proportion of a word's counted items that were judged condemnation) did not support this supposition.

Setting nouns and verbs of the same root side by side, the three pairs do not agree. In two pairs the condemnation rates are nearly equal: coniuratio 0.5429 (210 items) against coniuro 0.5091 (55 items), and societas 0.1081 (592 items) against coeo 0.1009 (109 items). In the third, conspiratio 0.7059 (34 items) against conspiro 0.5000 (26 items), the noun is higher than the verb, as the supposition expected, but each word has fewer than forty items. Taking only the items in Cicero, these six words in order are coniuro 1.0000 (10 items), coniuratio 0.7465 (71 items), conspiratio 0.5000 (10 items), coeo 0.2000 (20 items), societas 0.1181 (237 items), conspiro 0.0000 (7 items): coniuro is higher than coniuratio and coeo higher than societas, so in these two pairs the verb is higher than the noun, contrary to the supposition; only in the conspiratio pair is the noun still higher. Which side is higher changes with the pair and with the author. These values show that whether a word is a verb or a noun has no bearing on whether it is condemned.²⁷

How high a word's condemnation rate is depends on whether the Romans used it to rebuke some kind of alliance. The table below gives condemnation rates judged only from the sentence itself; the two right-hand columns give the same rate computed separately by author.²⁸

The figures by author matter as much as the overall figures. The same word seditio has a condemnation rate of 0.6066 (61 items) in Cicero, 0.2407 (162 items) in Livy, 0.2043 (93 items) in Tacitus, and 0.0000 (15 items) in Curtius; the highest and the lowest differ more than threefold.²⁹ This is not because the word changed its meaning over three hundred years, but because different genres use the same word to do different things. Speeches, collections of moral exempla, and political tracts need to characterize the affair within the sentence, and so the sentence contains pejorative words; annalistic history records events year by year and leaves the characterization to the reader, and so the sentence contains none. The contrast between Virgil and Florus is the clearest. Virgil writes "as when in a great people a seditio often arises, and the base rabble rage in their minds", the most famous description of civil disturbance in Latin poetry; but judged by this sentence alone the result is neutral, because "rage" and "base" describe those who cause the disturbance, not the disturbance itself. Florus writes "yet even amid these seditiones you would not without reason revere this sovereign people", the clearest positive view of civil disturbance in the imperial period; but judged by this sentence alone the result is likewise neutral, because the favourable word also describes the people.³⁰ This rule works the same way for praise and for blame: it measures "whether the sentence itself characterizes the word", not "whether the author liked the affair".

Computed by author, the order of the words does not change. Within one author, the condemnation rates fall clearly into two tiers: the terms of opprobrium the Romans used to rebuke some kind of alliance, and the words that could be used of any combination. In Cicero, coniuratio, factio, and seditio are all above 0.60, and societas and collegium are both below 0.12; in Livy, the first three are all above 0.24 and the last two below 0.08.³¹ With a different author each word's condemnation rate changes, but the order of the words relative to one another does not.

How many times the two classes differ depends on which words are compared. Taking only the items in Cicero and comparing seven words, the lowest-rated term of opprobrium, seditio (0.6066), is 3.03 times the highest-rated neutral word, coeo (0.2000); widening the range to the eleven words read, the lowest term of opprobrium becomes sodalitas (0.4286), while the highest neutral word is still coeo (0.2000), a difference of only 2.14 times.³² Without separating by author, taking all items together, the lowest term of opprobrium, factio (0.4581), is 4.24 times the highest neutral word, societas (0.1081); these two words are 0.7273 and 0.1181 in Cicero, and 0.3636 and 0.0743 in Livy.³³ The largest difference is between coniuratio (0.5429) and collegium (0.0455), 11.94 times; if the 6 cells in which collegium was judged legal prohibition are also counted as negative, its proportion is 14/176 = 0.0795, and the difference is 6.82 times. coniuratio is 5.02 times societas. These three words are 0.7465, 0.0781, and 0.1181 in Cicero, and 0.5778, 0.0000, and 0.0743 in Livy.³⁴

One group of cognates cannot be judged by the pooled figure alone, because their meaning changed between the two eras. conspiro has a condemnation rate of 0.5000 (26 items), which looks as if it belongs in the tier of terms of opprobrium, but the figure is made up of two sets of items that do not overlap: by the three periods, Republican 0/9, Augustan 4/4, early-to-middle Imperial 9/13; by author, 0.0000 (7 items) in Cicero and 0.7000 (10 items) in Suetonius. conspiratio changes in the same direction: Republican 5/10, Augustan 2/3, early-to-middle Imperial 17/21; 0.5000 (10 items) in Cicero, 0.8000 (5 items) in Suetonius.³⁵ In the Republican period the conspirare group of words was not a term of opprobrium for rebuking some kind of alliance; only in the imperial period did it become one. Cicero's own usage shows this best. In the four speeches Against Catiline, coniuratio occurs 14 times, always referring to Catiline and his associates; conspiratio occurs only 1 time, in the fourth speech, section 22: neque ulla profecto tanta vis reperietur quae coniunctionem vestram equitumque Romanorum et tantam conspirationem bonorum omnium confringere et labefactare possit, "no force will ever be found great enough to break and shake your union with the Roman knights and so great a conspiratio of all good men". The two words are formed in the same way, one "swearing together", the other "breathing together"; in 63 BC Cicero used the former for his opponents and the latter for his own side.³⁶

The approval rate likewise varies with genre, and must also be read with the generic composition in view. The approval rates of seven words are: societas 0.2416, sodalis 0.1796, collegium 0.1364, coeo 0.1009, factio 0.0194, seditio 0.0098, coniuratio 0.0000. The two most condemned words, factio and coniuratio, have only 3 cells judged approval between them (3/365 = 0.008); the three words that refer to persons or to approved groups all have approval rates above 0.13.³⁷ The generic composition of these approval rates also differs. The approval rate of sodalis is 0.2404 (104 items) in the author's own words and 0.0794 (63 items) in words the author quotes from others, a threefold difference; this runs opposite to the three other words that have approval cells (societas 0.1962 against 0.2832, coeo 0.0625 against 0.1522, collegium 0.1274 against 0.2000, the side quoting others being higher in each case). The difference is caused not by who the speaker is but by genre: on the side quoting others, sodalis comes almost entirely from comedy (Plautus, 44 items) and declamation, mostly as a form of address in the vocative, with few evaluative words; on the author's own side, it is first-person poetry, Ovid 21 items, Martial 21, Horace 8, Catullus 6. By author, Ovid is 0.4762, Horace 0.3750, Martial 0.2381, against Plautus 0.0455 and Cicero 0.0645, the highest and the lowest differing tenfold.³⁸ The approval rate of conspiro, 0.2692 (26 items), must also be read by period: of its 7 approval cells, 7 items come from Cicero, and all 7 of these items in Cicero are judged approval; after removing one fragment cut out by sentence division, it is 6 cells, 0.2308.³⁹

Between verbs of process and nouns of entity there is indeed one difference, but it lies not in the condemnation column but in the "legal prohibition" column. The proportion of coeo judged legal prohibition is 0.0917 (109 items), the highest among these words; collegium is 0.0341, societas 0.0034, and the two terms of opprobrium have not a single cell.⁴⁰ These ten cells come from only four passages: the statute quoted in Cicero's Pro Cluentio, "whoever has combined to bring about someone's conviction in a public trial" (the statute is quoted five times between sections 144 and 157, which together with one sentence in which Cicero discusses it makes six cells); the senate's decree of 186 BC prohibiting the Bacchanalia; the ancestral custom stated by the consul Postumius before the popular assembly (except in three cases, when the standard is raised for an electoral assembly, when a tribune announces a meeting of the plebeian council, and when a magistrate summons the people to a public meeting, citizens may not gather at will); and tribunes using their veto to stop the patricians from meeting to choose an interrex. The five cells of "legal permission" come from another kind of text: Gaius, Institutes 3.148 to 3.154b, the part on how a partnership is formed and how shares are determined, and the third category in a treaty recorded by Livy.⁴¹ For the same verb, the statutes of criminal and public law forbid this kind of coming together, while private law and treaty law prescribe the form in which it is constituted.

Even without the attitude judgments, a difference can be seen from the role the word plays in the sentence. Of the 211 items read for coniuratio, in 43 the sentence equates it with a criminal charge, and in another 16 it is used as a technical term in a legal proposition, so that 59 in all occur in indictments, statutes, trials, or decrees of the senate; of the 164 items for factio only 2 are of this kind, a twenty-three-fold difference.⁴² Of the 177 items for collegium, 92 use it as a technical term in a legal proposition, and 0 as a charge.⁴³ The items in which the speaker uses the word for a group to which the speaker belongs show the difference more clearly still: collegium has 10 cells, the speakers being Cicero (of priestly colleges and the college of augurs), Tacitus (himself a member of the quindecimviral college), and the protagonist of Apuleius' novel, who has joined the priesthood of Isis; factio has 4 cells, all from the mouths of robbers in Apuleius; coniuratio has not one. Romans said "our collegium" to mean their own priestly college; no one said "our coniuratio".⁴⁴ The only word used in large numbers for one's own side is sodalis (companion, fellow member of a society): of the 167 items read, 55 are the speaker calling upon or identifying his own companions, about three in ten, while the highest proportion among the other words is that of coeo, 8 of 110 items; and there is not one cell in which sodalis is used to reproach a named opponent, the only zero among the seven words. No one was prosecuted on the ground that "you are someone's companion".⁴⁵

So even without relying on the attitude judgments, conclusions can be drawn about these two words: coniuratio was a criminal charge, with its own court, rewards, and graded reductions of penalty; factio was not a charge, no statute made it a crime, and it was only a word for rebuking opponents in political struggle.⁴⁶

That no charge bore the name seditio has a different cause from that of the previous two words. Of the 409 items read, in 0 does the sentence equate it with any charge.⁴⁷ Rome did have a prosecutable charge for "inciting disturbance", but it was not called seditio; it was called maiestas (treason). There are two independent pieces of evidence. Tacitus, describing Tiberius' revival of the law of treason, states the law's old content: "if anyone had diminished the majesty of the Roman people by betraying an army, or by stirring up the plebs with seditiones, or by bad administration of the state" (Annals 1.72). Cicero's rhetorical handbook sets out the formulations of both sides in a treason case, and the definition on the defendant's side is "majesty lies in the dignity of the empire and name of the Roman people, and it was diminished by one who brought matters to seditio through the violence of the multitude" (Partitiones oratoriae 105). The two passages are a century and a half apart and differ in genre, one a rhetorical handbook, the other a history. In both, seditio is only a description of an element of the offence, not the name of the offence itself.⁴⁸ The Romans themselves noticed this gap and were ashamed of it: Appius Claudius said in the senate that among the enemy anyone who started a seditio was put to death on the spot, whereas Roman military law had a specific offence for deserting the standards and leaving camp, but none specifically for seditio.⁴⁹

5. The contest over names in accusation and defence

The Bacchanalian decree of 186 BC: the rites could be licensed, association could not

In 186 BC the senate sent the allied districts a decree on the Bacchic cult. The decree was inscribed on a bronze tablet, found in 1640 at present-day Tiriolo, and now kept in Vienna.⁵⁰ The decree prohibits nine things clause by clause, and four of these clauses set out a procedure for permission: those who wished to keep a shrine, to attend the rites, to hold rites inside or outside the city, or to perform rites with more than five people together, had to apply to the urban praetor at Rome, and the matter was then decided with at least a hundred senators present; only with permission could they do it. The other five have no procedure for permission: there may be no priest, no master, no common fund, no appointment of officers, no swearing to one another, joint oaths, mutual guarantees, or mutual agreements, and no giving of pledges of good faith to one another. Violators were to be put to death.⁵¹

The difference between the two groups of clauses does not lie in the severity of the punishment. What could be licensed was the rite itself: where to worship, who would go to worship, how many would worship together. What could not be licensed were offices, money, and the relations by which members bound one another, that is, the conditions that organize a body of worshippers into a group. The senate did not forbid people to worship Bacchus; it forbade worshippers to form a group with people in charge, common property, and oaths sworn to one another, and it did not allow them to apply for a licence to do so.

The decree contains, from beginning to end, not a single noun naming an association. coniuratio, collegium, sodalitas, coetus, factio, societas, corpus: none of these words appears in its thirty lines, and not even the verb of coming together, coire, appears. When it speaks of gatherings, the decree sets numbers: no more than five persons, of whom no more than two men and no more than three women.⁵² What the decree prohibits by name are the specific things and acts that make people into a group, not some organization with a name. When Livy paraphrases the decree, he does use that verb: no one initiated into the Bacchic rites may wish to have come together or assembled for the sake of the rites (ne quis, qui Bacchis initiatus esset, coisse aut convenisse sacrorum causa velit); the persons the two consuls were ordered to investigate were "those who have come together and sworn" (qui coierint coniuraverintve, 39.14.8).⁵³ To present the Bacchic worshippers as a "conspiracy" with boundaries and leaders who could be named is the historian's narration; the decree itself does not put it that way.

There is another point where Livy disagrees with the bronze. When he summarizes the decree at 39.18.9, he turns three clauses that are absolute prohibitions on the bronze into conditions attached to a licence: once permitted, the rites could be performed, provided that no more than five took part in the sacrifice, that there be no common fund, and no master of the rites or priest (dum ne plus quinque sacrificio interessent, neu qua pecunia communis neu quis magister sacrorum aut sacerdos esset). What is prohibited is the same on both sides, but the legal character differs: on the bronze it is an absolute prohibition, in Livy a condition attached to a licence.⁵⁴

How Cicero made "being only a part" into a charge

The sentence that made "being only a part of the whole" into a charge comes from Cicero, De re publica 3.44:

Vides igitur ne illam quidem quae tota sit in factionis potestate posse vere dici rem publicam

You see, then, that not even that which lies wholly in the power of a factio can truly be called a state.⁵⁵

This sentence makes factio and res publica two mutually exclusive things: if a community is wholly controlled by a faction, it no longer qualifies to be called a state.

De officiis 1.85 states the same reasoning as a general principle:

Qui autem parti civium consulunt partem neglegunt rem perniciosissimam in civitatem inducunt seditionem atque discordiam ex quo evenit ut alii populares alii studiosi optimi cuiusque videantur pauci universorum

Those who look after one part of the citizens and neglect another bring into the city the most ruinous thing, namely disturbance and discord; from this it comes about that some seem to be populares, others devoted to every best man, and few to the whole.⁵⁶

The key lies in the last two words, pauci universorum, meaning that few look after the whole. Here Cicero places the names of both parties of his day, populares (the people's side) and optimates (the side of the best men), together under "looking after only a part": in his view both commit the same error, looking after a part and not the whole.

With this definition, an accuser need not prove what the other side has specifically done. The tribune Gaius Memmius once attacked, before the popular assembly, the nobles who had made peace with Jugurtha, and Sallust composed this speech for him in his history. Two consecutive sentences in it make the point clearly:

metum ab scelere suo ad ignaviam vestram transtulere, quos omnis eadem cupere, eadem odisse, eadem metuere in unum coegit. Sed haec inter bonos amicitia, inter malos factio est

They have transferred fear from their own crime to your cowardice; and what has driven them all together is wanting the same things, hating the same things, fearing the same things. But this among good men is friendship, among bad men faction.⁵⁷

Memmius first states what this combination consists of: wanting the same things, hating the same things, fearing the same things. He uses three verbs, not a noun. He gives the combination no neutral name; he can only give two names that carry praise or blame, and he says outright that which one to use depends on whether those who combine are good men or bad. And in this speech there is no procedure for deciding who is good and who is bad; the one who decides is the speaker himself, and Memmius is at this very moment calling the other side a factio. In the whole of the Jugurthine War this word occurs only six times, three of them in this chapter, and five of the six are applied to the nobles' side.⁵⁸

In court the point is clearer still. When Cicero defended Plancius, the prosecution charged Plancius under the lex Licinia de sodaliciis with electoral corruption through associations; Cicero did not deny the facts:

quos tu si sodalis vocas, officiosam amicitiam nomine inquinas criminoso

If you call these people sodales, you are staining a dutiful friendship with a criminal name.⁵⁹

He admits that his client had influence, admits that those who helped him had influence, admits that they exerted themselves together; he disputes only one thing, what this relationship should be called. He demands that the prosecution prove specific acts: dividing the members of a tribe into groups of ten, registering them on lists, acting as intermediary guarantors, making promises on another's behalf, distributing money, bribing. He spends three whole sections arguing that "proof must reach this level". That he argues this shows precisely that the prosecution could make out its case without proof reaching this level.⁶⁰

In section 37 of the same speech he uses the same word in the opposite direction. Explaining why this law allowed the prosecution to choose jurors from the defendant's tribe, he says that bribers corrupt a tribe through "this agreement, which is called sodalitas, a name more respectable than true" (per hanc consensionem quae magis honeste quam vere sodalitas nominaretur). Section 37 says that the name presents improper conduct as proper; section 46 says that the name presents a proper relationship as criminal. The same man uses the same word in the same debate, nine sections apart. In neither place does he say what specific acts the combination consists of; both places concern whether a name fits the thing it names, and whether it fits is decided by the speaker.⁶¹

Section 53 gives an example of how the prosecution proved that the defendant was "in league". Cicero reports the other side's words:

dubitatis, inquit, quin coitio facta sit, cum tribus plerasque cum Plotio tulerit Plancius?

Do you doubt, he says, that a coitio took place, when Plancius carried most of the tribes together with Plotius?

The prosecution adduced none of deliberation, division of tasks, or payment; it adduced only the fact that the two men won together in the result of the election. It is exactly this inference that Cicero rebuts: if the two had not won most of the tribes together, how could they have been elected together?⁶²

Cicero's reply when accused of being in league: we are the whole

In 56 BC the prosecution asked Cicero in its speech what this natio (breed) of yours, the optimates, was (quae esset nostra natio optimatium, Pro Sestio 96). Each time he mentions the word, he marks it as the other side's term: "so you said at the time" (96); "let them be, then, that natio of yours" (97); "it is not a natio, as you put it" (132).⁶³

What the word usually refers to can be shown in numbers. This corpus contains 60 works of Cicero, in which natio and its cognate forms occur at 158 places in all, spread over 31 works. I read each of them: 142 refer to some people or to peoples in general, 9 to a group of Romans, 3 to a Roman's origin, 2 are the name of the birth goddess Natio, and 2 are one of the "attributes of a person" listed in rhetoric.⁶⁴ So the word mainly names a people.

The nine places referring to a group of Romans must be looked at separately, because each is used differently. Two are applied to candidates: "that most attentive breed of candidates" (In Pisonem 55) and "a whole breed of candidates" (Pro Murena 69). Both use the superlative officiosissima, and neither is condemnation. The sentence in In Pisonem says that not even the candidates most willing to turn out came to meet Piso; for the construction "not even ..." to work, this class of people must be known as the most willing to turn out, so the sentence mocks Piso. The sentence in Pro Murena is one the defence itself adduces to prove that escorting was customary, and it adds a word of praise: "they let no one enter the city without honour".⁶⁵ Applied to these candidates, the word serves to classify: it presents a group of people who act for the same purpose as a breed. Another place is Cicero himself using it to rebuke others: "beware of those worse men; they have indeed become a large breed, but the leader and chief of them all is this man" (De haruspicum responso 57), referring to the band led by Clodius, with the leader named. Another refers to a philosophical school: the Stoic Balbus says to the Epicurean "that salt which your breed lacks" (De natura deorum 2.74). The remaining five are the breed of the optimates in Pro Sestio.⁶⁶

So what the prosecution used to question Cicero was a word of classification, not a word of disparagement; it presents a group of people as a kind, a breed, without saying they are bad. The word could be used to accuse precisely because, when Romans applied it to their own fellow citizens, it referred to the class of candidates, to someone's origin, or to an opponent's band, and never to the citizen body as a whole.

In replying to this question, Cicero in his very first sentence substitutes another word of classification. He does not say natio but "there have always been two kinds in this city of those who sought to engage in public affairs and to distinguish themselves in them" (duo genera semper in hac civitate fuerunt eorum qui versari in re publica atque in ea se excellentius gerere studuerunt): one kind wanted their words and deeds to please the many and were called populares; the other wanted their counsels approved by every best man and were regarded as optimates. In sections 96 and 97, genus and its cognate forms occur four times. The word the other side used was natio, meaning a breed; the word he put in its place was genus, meaning one of two kinds. The difference between the two words is this: natio picks a group of people out on their own and calls them a breed; genus divides everyone who takes part in public affairs into two kinds, with every person in one of them.⁶⁷

Only after changing the word does he go on to answer. He asks: who, then, are these "every best man"? numero, si quaeris, innumerabiles, neque enim aliter stare possemus: in number, if you ask, they are countless, or else we could not stand at all. Next he lists the members: the leaders of public counsel, those who follow them, men of the highest orders who are eligible to enter the senate, Romans of the municipalities and the countryside, men of business, and even among freedmen there are optimates. Finally he gives a negative definition: omnes optimates sunt qui neque nocentes sunt nec natura improbi nec furiosi nec malis domesticis impediti: all are optimates who are neither harmful, nor wicked by nature, nor mad, nor hampered by troubles at home.⁶⁸

In my view this is not an excuse but a refusal to accept the name. He does not argue over whether his side is good; he enlarges the people the name covers to nearly the whole city, so that it no longer refers to a separate "band".

This dispute over the name ends only at section 132 of the same speech, which makes three things clear. First, he returns the word to the other side: "You asked me who the optimates are; you have your answer. It is not a natio, as you put it." From this we can see that the sentence in section 97, "let them be, then, that natio of yours", was a concession for the sake of argument, not an admission. Second, he points out whose word it is: quod ego verbum agnovi, I recognized this word, it is that man's expression; the questioner is not the source of the word, which comes from another man, the one who attacked Sestius most fiercely. Third, in that man's mouth the word occurs in a sentence meant to make this group disappear: "it was that man who wished this natio to be wiped out and cut down (hanc nationem deleri et concidi cupivit), that man who again and again reproached and accused Caesar of being soft and afraid of bloodshed, saying that as long as this natio lived (dum haec natio viveret), Caesar would never have a day's peace."⁶⁹

These three things show that to call a group of people a natio is the first step toward presenting them as something that should not exist; for the accused to escape this name, the way is not to prove that his side is good but to prove that no such group exists at all, as he puts it: "in number, if you ask, they are countless".

The man who in De officiis 1.85 called this charge "looking after a part" and the man who here declines to place himself under that charge are the same man. He condemns those who look after only a part of the people, yet of himself he says "we are the whole". This is not my inference: the two passages use the same set of words. 1.85 sets out two precepts for those who govern the state, the second being to care for the whole body of the state and not, in protecting one part, to abandon the rest; it goes on: "those who plan for one part of the citizens and abandon another bring into the city the most ruinous thing, namely sedition and discord; from this come some who seem to be populares and some who seem to be on the side of the best men, and few on the side of the whole." De officiis treats the two parties as a consequence of "looking after a part", and says too that those truly on the side of the whole are few; Pro Sestio 96, however, narrates the same two parties straightforwardly as two kinds that had always existed in the city, and in section 97 enlarges the optimates to nearly the whole city, assigning "being on the side of the whole" to his own side.⁷⁰

What Augustus called his enemies and himself in the Res Gestae

The first sentence of the Ankara copy of the Res Gestae is:

Annos undeviginti natus exercitum privato consilio et privata impensa comparavi per quem rem publicam a dominatione factionis oppressam in libertatem vindicavi

At the age of nineteen, on my own decision and at my own expense, I raised an army, by means of which I set free into liberty the state that was oppressed by the domination of a factio.⁷¹

The edited text of the work has 2,678 words, and factio occurs only once, at the opening. Augustus says that the other side is a factio, that the state was oppressed by its domination, and that he set the state free. Taken together these three statements form a complete argument, whose premise is precisely the definition in De re publica 3.44: if a whole community is controlled by a faction, it does not deserve to be called a state. Cicero, who wrote the definition, and Augustus, who used it, were some seventy years apart; between them came a civil war, and Cicero himself died in that war.

In the same work, names denoting the whole recur: "senate" twenty-six times, "Roman people" eleven times, "state" eight times, universus (all, entire) and its cognates five times, "the consent of all" once, fifty-one times in all; coniuratio does not occur once. None of the five instances of universus points to the enemy; they refer respectively to the whole city, the whole citizen body being of one mind, all the Roman knights, the consent of all, and the senate, the equestrian order, and the whole Roman people.⁷²

Section 25 has another sentence: iuravit in mea verba tota Italia sponte sua, the whole of Italy of its own accord swore to my words, and asked me to be its leader in the war at Actium; then the provinces of Gaul, Spain, Africa, Sicily, and Sardinia swore to the same words.⁷³ What the decree of 186 BC prohibited outright, leaving no route of application, was precisely a group of people swearing to one another, which is also what the word coniuratio refers to. In recording his achievements, Augustus uses iuro without a prefix and adds the two phrases tota Italia and sponte sua, meaning all, and of their own will. He makes no defence of his own side's combination; he only substitutes another name.

This is not merely a rhetorical coincidence. With the same verb coniurare, an oath taken through state procedure was a levy or an oath of allegiance, and an oath taken without state procedure was a capital crime. Caesar records that the senate decreed that the men of military age of all Italy should take the oath, and the levy was then held throughout the province; Suetonius records that Octavian publicly exempted the people of Bononia from the obligation of swearing for his side together with all Italy. That one town could be granted exemption shows that this was an institution in which everyone had to take part. The other side is Livy's record: twenty-five slaves were crucified for having sworn an oath in the Campus Martius, and the informer received his freedom and a reward of twenty thousand asses of heavy bronze. The two kinds of use do not differ in the word; what decided the outcome was whether the oath had gone through state procedure.⁷⁴

Roman law: association required public permission case by case

By the time of the Severan jurists, association was governed by general rules. Book forty-seven, title twenty-two of the Digest is entitled "On associations and corporate bodies", and seven of its eight fragments deal with the conditions for admission to association and the consequences of violation. A fragment from the second book of Marcian's Public Prosecutions extends the standard applied in 186 BC to one cult to all associations:

nisi ex senatus consulti auctoritate vel caesaris collegium vel quodcumque tale corpus coierit, contra senatus consultum et mandata et constitutiones collegium celebrat

Unless formed by authority of a decree of the senate or of the emperor, any society or any such body that has come together holds an association contrary to the decree of the senate, the mandates, and the constitutions. (47.22.3.1)⁷⁵

Roman law dealt with association among citizens not by defining which kinds of alliance were criminal, but by laying down that a group, to exist lawfully, had to obtain public permission case by case. Provincial governors were not to tolerate associations, soldiers were not to form associations in camp, the humble might pay a monthly fee but meet only once a month, and a person might belong to only one lawful association. Those who organized an unlawful association without authority received the same punishment as those who occupied public places or temples under arms.⁷⁶

The last rule in this title, however, is exactly the opposite. Gaius, Commentary on the Twelve Tables, book four:

Sodales sunt, qui eiusdem collegii sunt: quam graeci hetaireian vocant. his autem potestatem facit lex pactionem quam velint sibi ferre, dum ne quid ex publica lege corrumpant. sed haec lex videtur ex lege solonis tralata esse.

Sodales are those who belong to the same society, which the Greeks call hetaireia; and the law gives them power to make whatever agreement they wish among themselves, provided they do not violate anything in the public law; this law appears to have been taken over from the law of Solon. (47.22.4)⁷⁷

Immediately after comes the list from Solon's law: demes, phratries, cult associations, sailors, those who share a common table, those buried in the same tomb, Dionysiac societies, those who go out for plunder, those who go out for trade; whatever these people agree among themselves is valid, unless public documents forbid it. Nine kinds of combination are listed under the same rule, among them administrative divisions, kinship groups, religious groups, occupational groups, companions in common meals and common burial, raiding parties and trading parties, all recognized alike, whatever their nature.⁷⁸

Both rules turn on whether there is a public act; what differs is which side bears the burden of proof. The earlier rule recognizes association first, and the public authority prohibits only when it wishes to prohibit; the later rule requires public permission first, and without permission the group cannot be formed. The rule in the Twelve Tables that tacitly permitted association was not abolished; it is kept at the end of the same title, and it is the only provision in the whole title stating how an association's internal affairs are to be handled. An association must first pass through a public act; once formed, its internal affairs are handled by the association itself, and this rule of internal self-management comes from the law that tacitly permitted association. The compilers put both rules in the same title without reconciling them.⁷⁹

The Romans did not prohibit association from the beginning. In 186 BC, the method of prohibition by default with licences case by case applied only to one cult, named in a specific decree; some four hundred years later this method had become a general rule applying to all associations. Between the two periods the standard of judgment did not change; what changed was the range to which the standard applied.⁸⁰

By the early second century, what a group was called no longer affected the decision. Pliny the Younger reported to Trajan a great fire at Nicomedia, saying that the whole city had not a single pump and not a single bucket, and asked whether he might establish a guild of craftsmen of at most a hundred and fifty men, promising to admit only craftsmen. Trajan refused:

Quodcumque nomen ex quacumque causa dederimus iis qui in idem contracti fuerint, hetaeriae eaeque brevi fient

Whatever name we give, for whatever reason, to those who have been brought together for the same purpose, they will become hetaeriae, and quickly.⁸¹

hetaeria is a Greek loanword that occurs at only two places in the entire Latin corpus, both in this book of Pliny's letters. When a Roman provincial governor wanted to prohibit a kind of association, Latin had no ready word for it, and so he borrowed a Greek one; when those prohibited described their association themselves, they still used the Latin coire.⁸² The rule by which the Romans recognized association came from the law of the Greek Solon, and the name they used when prohibiting association also came from Greek.

Only two kinds of combination could be called legitimate

The Romans had no legitimate name for a horizontal political association. Putting the approval rates of the words together, the first four words, which refer to persons or approved groups, all have items judged approval: societas 0.2416, sodalis 0.1796, collegium 0.1364, coeo 0.1009; the three pejorative words have almost none: factio 0.0194, seditio 0.0098, coniuratio 0.0000.⁸³ The generic composition of this set of figures is as described above: the approval cells of sodalis come mainly from first-person poetry, Ovid 0.4762, Horace 0.3750, Martial 0.2381, against Plautus 0.0455 and Cicero 0.0645; for societas, coeo, and collegium the side quoting others is higher. factio and coniuratio together have only 3 cells judged approval.

The condemnation side can be computed in the same way. The condemnation rate of coniuratio is 114/210 = 0.5429 and that of collegium 8/176 = 0.0455, the former 11.94 times the latter; societas is 64/592 = 0.1081, and coniuratio is 5.02 times it. These three words are 0.7465, 0.0781, and 0.1181 in Cicero, and 0.5778, 0.0000, and 0.0743 in Livy.⁸⁴

There is only one kind of human combination that can be called legitimate without further argument: the fellowship of all mankind. Among the approval sentences read item by item, there are seven places where a word denoting the whole modifies a combination, and that combination is mankind itself, for example "the whole fellowship of man with man" (De finibus 3.69), "that fellowship among men, of all with all, is the widest" (De officiis 1.51), "once good faith is abolished, the whole fellowship of mankind is gone" (Livy 6.41.11), "so that the whole human race could come together with one another" (Quintilian). There is not one such use among the condemnation sentences. All seven places come from philosophical and rhetorical discursive writing, and the combination they speak of is all mankind, not any particular group of Romans.⁸⁵

The fellowship of all mankind and groups authorized by the public authority both had ready expressions; for a particular group of Romans forming a horizontal political association for common affairs, there was no corresponding word. This is not because I failed to find one: Cicero himself listed the degrees of fellowship between people one by one, eleven degrees in all, and this degree is not among them.

De officiis 1.53 to 1.57 lists gradus societatis hominum, that is, the degrees into which human fellowship is divided. Moving inward from the widest layer, nearer are those of the same race, the same people, the same language; further in is belonging to the same city, for citizens have many things in common: forum, temples, colonnades, roads, laws, rights, courts, votes, and also customs, friendships, and the business and accounts that many have contracted with many; a closer layer is kinship; further down are marriage, children, a household, brothers, cousins, marriages and in-laws. 1.55 has a sentence in the superlative: sed omnium societatum nulla praestantior est, nulla firmior, quam cum viri boni moribus similes sunt familiaritate coniuncti, of all fellowships none is more excellent, none firmer, than when good men of similar character are joined in intimacy. 1.57 has another: omnium societatum nulla est gravior, nulla carior quam ea, quae cum re publica est uni cuique nostrum, of all fellowships none is weightier, none dearer, than the one that each of us has with the state.⁸⁶

The relations Cicero lists degree by degree are all blood, the city itself, the household, or private ties between individuals. A group of citizens who are not kin forming a horizontal political association for the same public business is not in this scale.

In this scale, the layer closest to such an association is friendship. Cicero set limits for friendship, and also wrote down what it should be called once the limits are crossed. De officiis 3.43 first sets the limit: At neque contra rem publicam neque contra ius iurandum ac fidem amici causa vir bonus faciet, a good man will do nothing for a friend's sake against the state, or against his oath and good faith. The conclusion of 3.44 is a single sentence: nam si omnia facienda sint, quae amici velint, non amicitiae tales, sed coniurationes putandae sint, for if everything a friend wishes must be done, such ties should be considered not friendships but conspiracies.⁸⁷ Once friendship makes unlimited demands on a person, its name changes from amicitia to coniuratio.

This criterion differs from the one in Sallust. Memmius says that the same combination is friendship among good men and faction among bad men; which name applies depends on a judgment of whether those combining are themselves good or bad, and the one who judges is the speaker. Cicero in De officiis 3.44 says that which name applies depends on what the combination demands: whether it requires you to act against the state or against your oath, whether it requires you to do everything you are asked. The former criterion judges the people; the latter judges whether the combination's demands on people have limits. The Romans had at least these two ways of making a horizontal political association into a charge, and the two differ from each other.⁸⁸

Cicero in court had to enlarge the optimates on the spot to countless people, and Augustus had to avoid applying factio to his own side throughout his work and write names denoting the whole fifty-one times, and the reason in both cases is the same: when a particular group of Romans formed a horizontal political association for common affairs, there was no ready word by which to call it.

Being called a part was not necessarily a crime. The following three passages can be compared. Speaking on behalf of the tax-farming company of Bithynia, Cicero calls it "by its very rank and the birth of its members, the largest part of the city" (ordine ipso et hominum genere pars est maxima civitatis, Ad familiares 13.9.2): he concedes that it is a part, and says it is the largest part. Pro Rabirio 21 speaks of "joining with all good men in a fellowship not only of safety but also of danger" (cum bonis denique omnibus coire non modo salutis verum etiam periculi societatem), and here those combining are called "all good men". Tacitus' Calgacus tells the Britons before battle "you have all come together" (nam et universi coistis, Agricola 30.1), and the speaker is an enemy of Rome.⁸⁹ These three passages either place a part within a larger whole or come from the mouth of someone who is not a Roman.

Cicero also has an appeal that speaks of a group of Romans uniting for common public business. He says to the tribunes: conspirate nobiscum consentite cum bonis communem rem publicam communi studio atque amore defendite, breathe with us, be of one mind with the good, defend the common state with common zeal and love (De lege agraria 1.26). He uses the imperative, and he places himself within this union. At the end of the Republic a Roman once used this verb for a horizontal political association; in the imperial period it became the term of opprobrium for a "plot".⁹⁰

6. Results measured on the approval–disapproval axis, and their limits

The conclusions above come from sentence-by-sentence judgment. The praise and blame of the same words can also be measured from the relations of co-occurrence between words. The program first computes a set of coordinates for a word from the words it often appears with; it then takes a set of generally recognized approving and disapproving adjectives as reference. A word near the approving side gets a positive value; a word near the disapproving side gets a negative value. This scale from praise to blame is called below the approval–disapproval axis. The 86 reference adjectives were not chosen by me; they were registered from the passages on praise and blame in the Rhetorica ad Herennium, De inventione, and De officiis. Five models are trained on each period of the corpus, and a result is counted only if all five models give the same sign.⁹¹

This method must first be calibrated, that is, used to judge Latin words whose praise or blame others have already annotated, to see how many it gets right. In 2020 Sprugnoli and colleagues asked two experts in Latin language and culture to assign polarity scores to 1,144 adjectives and nouns; I took the five hundred or so of these that carry polarity as the calibration list. In the Republican period the agreement rate is 85.5% for adjectives (69 words) and 80.6% for nouns (67 words); in the early-to-middle Imperial period, after removing the words the experts judged only slightly positive or slightly negative, the rates are 82.1% and 82.9%. Both periods met the standard set in advance. The Augustan period gives 74.0% and 71.2%, and neither of the two calculations met the standard, so the results for that period can be listed only for reference.⁹² The late antique period was added this time, corresponding roughly to the early third to the early sixth century AD. The corpus's original author table assigns only 12 authors to this period, among them Tertullian, Augustine, Jerome, and Ammianus, with Augustine alone accounting for 42%; the corpus also had 87 author numbers that had never been assigned to any period, all dated within this one, and I added them too, making 99 authors and 7.5 million words in all, with Augustine's share falling to 23%. Trained and calibrated in the same way, the agreement rate is 86.4% for adjectives (118 words) and 86.7% for nouns (113 words), the highest of the four periods.¹⁰⁷

The measurements show that the mean of the thirteen words fell from +0.087 at the end of the Republic to +0.048 in the early-to-middle Imperial period. Under four alternative algorithms the order did not change; when the praise and blame labels were randomly shuffled two hundred times and the calculation repeated, the actual value for the Republican period was still higher than every one of the two hundred.⁹³ Word by word, ten of the thirteen words are positive at the end of the Republic, the three most favourable being consensus (agreement) +0.209, concordia (concord) +0.193, and coniunctio (union) +0.178; only seditio and coniuratio are negative, and both only at about −0.03. By the early-to-middle Imperial period, the decline is mainly in the words denoting unanimity: consensus fell from +0.209 to +0.076, coetus and socius turned from positive to negative, while collegium is above +0.11 in all three periods and did not decline.⁹⁴

In late antiquity the mean of the thirteen words fell to +0.004. This mean by itself says nothing: with the praise and blame labels randomly shuffled two hundred times, the actual value ranks 117th and cannot be distinguished from random results. What deserves attention is that the gaps between the words' positions on the approval–disapproval axis grew larger. For the first time the three terms of opprobrium all fall clearly on the disapproving side: factio −0.143, seditio −0.142, coniuratio −0.101, with the five models agreeing in sign; and when any two of the disapproving adjectives are removed and the calculation repeated, the sign does not change even once. Meanwhile concordia (+0.162) and collegium (+0.130) remain clearly on the approving side. Computed with only the original 12 authors, the three terms of opprobrium are likewise all below −0.11, so adding authors did not change the conclusion.¹¹⁰ In the first three periods, factio was always positive (+0.049 to +0.115) and coniuratio lay between −0.032 and +0.068; the terms of opprobrium and the neutral words differed little on this axis, and only in this period did they separate clearly. The generic composition of this period differs from that of the first three and must be considered together with the authors: of the 176 occurrences of factio, the Historia Augusta accounts for 20% and Augustine for 12%, with no single source above a quarter; of the 855 occurrences of societas, Augustine still accounts for 36% and the astrological writer Firmicus Maternus for 21%. So how much of this change comes from ideas themselves, and how much from imperial history, Christian polemic, and other genres, cannot yet be distinguished.¹⁰⁸

The approval–disapproval axis measures the overall tendency of the evaluative words around a word; it cannot determine what role the word plays in a particular sentence. coniuratio is the clearest counterexample. History books often treat it as the standard Roman term for condemning the forming of factions, yet its value in the early-to-middle Imperial period is +0.012, close to zero; when any two of the 32 disapproving adjectives are removed and the calculation repeated, the sign changes in seven cases out of ten.⁹⁵ Sentence-by-sentence judgment shows something else: of the 211 items read, 59 occur in indictments, statutes, trials, or decrees of the senate, 43 equate coniuratio with a criminal charge, and 0 cells use it for one's own side. A word can be a criminal charge and at the same time appear often with neutral or even sympathetic evaluative words in early-to-middle Imperial texts: in the run of sentences in which Tacitus narrates the conspiracy of Piso, the narrator is not necessarily unsympathetic to the conspirators. The two methods do not measure the same thing.

These data have four further limits. First, the reference words themselves carry Cicero's influence: of the 86 adjectives, 75% of the occurrences on the approving side come from Cicero, and 62% on the disapproving side. Part of the approval measured in the Republican period is simply Cicero's own approving usage. Second, the strictest standard of reliability requires a word to occur at least eight thousand times for its coordinates to be stable, and none of the thirteen words reaches it; under a looser standard (at least eight times), ten words reach it. The values reported above follow the looser standard, and for coniuratio, socius, and sodalis the proportion of sign reversals in one or another period exceeds one in ten, so the signs in those cells are not reliable. Third, the Augustan period did not meet the calibration standard and can serve only for reference, probably because that period's corpus has only 1.11 million words, more than half of them from Livy alone, but I have not tested this separately.⁹⁶ Fourth, the late antique period is dominated by Christian writers, and word meanings had also changed: societas in Augustine, for example, mostly refers to the human community in a theological sense. The earliest feasibility report assessing this corpus had already stated that this period should be used only for reference and should not be drawn on the same curve as the first three periods and compared directly, unless word senses are distinguished item by item; adding authors this time reduced the weight of a single author but did not distinguish word senses item by item, so when this period is compared with the first three, the differences of genre and of word sense have not been removed.

So on this question the approval–disapproval axis can serve only as corroborating evidence. It shows that the approval attached to the words denoting unanimity weakened, and that the three terms of opprobrium leaned clearly toward disapproval on this axis only in late antiquity; the table of nearest neighbours below shows what factio was associated with in each period. All of this is only corroborating evidence; which kind of alliance the Romans treated as a crime must still be answered by sentence-by-sentence judgment.

Nearest neighbours

The approval–disapproval axis gives a word only one signed number and does not show which words surround it. So I also listed the nearest neighbours of the thirteen words in each of the four periods: in each period each of the five models takes the 20 words closest to the word; rare words occurring fewer than 20 times in that period are removed; only words listed by at least four of the five models are kept; and the first ten are taken in order of closeness. A word's nearest neighbours are sensitive to small changes in the corpus, so below I use only neighbours filtered in this way, and draw no conclusion from one word's neighbours alone.¹⁰⁹

The nearest neighbours of factio differ in the three periods. In the Republican period they are penes (in the power of), potentia (power), dominatus and dominatio (rule, despotism), that is, "the state falling under the power of a few", which is exactly the content of the definition in De re publica 3.44. In the early-to-middle Imperial period they are optimates, conspiratio (conspiracy), largitio (lavish giving of money), and seditiosus (seditious). In late antiquity they are seditiosus (seditious), conspiratio, coniuratio, insimulo (to accuse falsely), transfuga (deserter), satelles (henchman), latrocinium (brigandage), and damnator (one who condemns). factio originally referred to the domination of a few, and later appeared in the same kind of context as treason, desertion, false accusation, and conviction.

Among the nearest neighbours of consensus in the Republican period are optimates and uniuersus (whole, entire). Section 5 showed that Cicero, when accused of being in league, described the optimates as nearly the whole city; the Republican corpus is dominated by Cicero, and these two neighbours reflect precisely his usage, which links "agreement" and "the whole" with the optimates.

The nearest neighbours of societas in the Republican period are communitas (community), cognatio (blood kinship), affinitas (kinship by marriage), caritas (affection), patrocinium (patronage), clientela (clientship), and pactio (agreement). That is, in the Republican period this word was surrounded by kinship, patronage, and contractual relations, not by political combination. This agrees with what Section 5 says: among the degrees of fellowship Cicero lists there are blood, the city, and private ties, but no degree of horizontal political association.

In all four periods the nearest neighbours of pars are words of part, division, and proportion such as partitio, diuisio, dimidius, totum, and portio, without a single political word. So the polarity value of pars reported above measures the general sense of "part", not "party". In the Augustan period the neighbours of many words are personal and place names from Livy's history; for example, the neighbours of societas and amicitia are a string of names of Greek cities and kings, which also shows that more than half of that period's corpus comes from Livy.

7. What has not yet been checked or tested

Two variant readings in the decree of 186 BC have not yet been checked against a critical edition. I collated the decree using three digital transcriptions: Ernout's 1947 edition, Schumacher's 1988 edition, and EDCS-15100127 in the EDCS database. The three agree in their letters, with two substantive variants: line 6 cosoloretur/cosoleretur, and line 27 ubi/ubei. EDCS reads with Schumacher. However, these three belong to one tradition of editions and are not independent readings of the bronze; to settle these two places one must consult the critical text of CIL I² or ILLRP. Neither variant affects any judgment in the preceding sections. In line 13 only the stem CONIOVRA is inscribed on the bronze, and the ending is supplied in the same way by all three transcriptions; the phrase "no swearing to one another" cited above rests on the stem, not on the supplied ending. I have not examined the fine distinctions among the four verbs of swearing in the legal language of the second century BC, and distinguish them here only by the literal sense of their stems. Between the licensable and the unlicensable groups of clauses, the placement of line 15, "no performing rites in secret", cannot be determined: it forms a chain with the clauses on performing rites that follow it, and whether that single nisei clause (meaning "unless") governs the whole chain or only the last sentence cannot be read from the Latin. After the five clauses of lines 10 to 14 there is no nisei at all, so the dividing line itself is unaffected.⁹⁷

The 158 places of natio are divided into five classes, and I have read each of them only once. No second person judged them independently, and no agreement rate was computed. What can be cited directly are the texts and section numbers of the nine places applied to Romans, all of which have been checked one by one. The two places of the goddess Natio are not counted as the common noun. Within the class "some people or peoples in general", I did not distinguish whether the general references in Tusculan Disputations 1.36 and 1.108 and De natura deorum 1.62 include the Romans; Philippics 10.20, "all nationes can endure slavery, our city cannot", clearly excludes Rome. As for what each of the 178 words denoting the whole modifies, I likewise read each only once and classified them in the same way, without computing an agreement rate.⁹⁸

The works of Cicero in this corpus number 60; it is not the complete works. It lacks De legibus and some other works. I am studying the vocabulary of law and association, yet De legibus is not in the corpus.⁹⁹

The man who questioned Cicero in court is not named in the speech. No member of the prosecution is named anywhere in Pro Sestio; section 96 says only "you asked in your speech", and section 132 separates the source of the word and the questioner into two people. Identifying these two as Vatinius and Clodius is the usual view, but it is a judgment of secondary scholarship, not words in this speech, so above I wrote only "the prosecution".¹⁰⁰

For the comparison of De officiis 3.44 with the passage in Sallust, I have so far read only these two passages. How many passages in Roman authors support each of the two criteria has not yet been counted. In De officiis the word societas occurs at 47 places, the great majority meaning "human fellowship", but they have not yet been counted separately by sense.¹⁰¹

I read only the Latin part of the Res Gestae. The Ankara copy is a bilingual inscription in Latin and Greek, and I did not look up the Greek rendering of dominatione factionis. The Greek rendering of that sentence could show directly how a contemporary of Augustus understood factio. The law of Solon quoted at Digest 47.22.4 is transliterated into Latin letters in the corpus, and I did not check the forms of the Greek original.¹⁰²

In the sentence-by-sentence judgment table, 42 cells need to be completed from the original text, or have their limits stated, before they are cited. Of these, 2 cells require returning to the original files to settle their references. If the original sentence is only one word, the script that extracts neighbouring sentences matches on the first eight words; when a single word cannot be matched, it is matched to a different work. So it is not only the neighbouring sentences that are suspect; the reference itself must be checked. A further 3 cells are suspected misclassifications, 26 cells are misalignments between sentence division and verse lines, and 1 cell is the index at the back of a modern edition that got mixed into the corpus; these four kinds total 32 cells. The remaining 9 cells must be noted, when cited, as coming from the same piece of material, and 1 further cell has already been dealt with separately. This list does not alter the values in that table; any alteration would require another round of judgment and recomputation of the affected ratios.¹⁰³

The method of judging from a single sentence does not work for three kinds of sentence: ironic sentences, verse running over the line, and fragments cut out because of abbreviations. Two further cells show best why the method fails. Florus writes that Spartacus drew slaves to come over to him and soon gathered more than ten thousand men. This is the association Romans most abhorred in the whole corpus, and the information needed to judge its nature lies entirely in the two preceding sentences; judged by this sentence alone, it is neutral. The other cell describes the elders gathering in the Forum to devote themselves to death when the Gauls took Rome. This is the most solemn gathering in the Romans' own writing, and the information needed to judge its nature likewise lies in the preceding sentence; judged by this sentence alone, it too is neutral.¹⁰⁴

Two common words were not included in the sentence-by-sentence judgment. coetus occurs 272 times and conventus 233 times, both more often than most of the fifteen words judged, yet neither was in the original list of forty-one words. coetus occurs 8 times in the Digest, 4 of them in statutes carrying penalties, including the element of the crime of treason "forming a coetus or conventus, or summoning people to revolt" (48.4.1.1); it occurs 53 times in the Theodosian Code, half as an honorific for the senate and half for illegal assemblies of heretics. Of the 233 occurrences of conventus, about two in ten refer not to a human combination but to the senses "agreement" and "being met", so this number cannot be used directly as a count of an association word. To bring these two words into that table, two new passes of judgment would be needed.¹⁰⁵

I have done only part of the comparison after separating by genre. The 58 items of coniuratio from the imperial period come almost entirely from Tacitus' and Suetonius' narratives of the conspiracies faced by successive emperors; of the 107 items from the Republican period, nearly half come from Cicero's speeches and defences. The generic composition of the two periods differs, and how much of the gap between them comes from change in ideas and how much from differences of genre has not been separated. The frequency of coetus per ten thousand tokens rose from 0.2690 in the Republican period to 0.7501 in the early-to-middle Imperial period, while of those 174 occurrences in the early-to-middle Imperial period, 52 come from three Flavian epics, 0.2989; the share of epic in this rise has likewise not been computed separately.¹⁰⁶