DHQ: Digital Humanities Quarterly
2015
Volume 9 Number 3

Comparative rates of text reuse in classical Latin hexameter poetry

Abstract

This paper presents a quantitative picture of the interactions between poets in the Latin hexameter tradition. The freely available Tesserae website (tesserae.caset.buffalo.edu) automatically searches pairs of texts in a corpus of over 300 works of Latin literature in order to identify instances where short passages share two or more repeated lexemes. We use Tesserae to survey relative rates of text reuse in 24 Latin hexameter works written from the 1st century BCE to the 6th century CE. We compare the quantitative information about text reuse provided by Tesserae to the scholarly tradition of qualitative discussion of allusion by Latinists.

The detection and interpretation of allusion currently represent the dominant mode of study of Latin poetry.[1] The typical goal of intertextual study is to describe how links between texts affect the meaning of both the specific passages that contain them and the poems as a whole. Although intertextual associations may be signalled in many different ways (including similarity of action, character, or theme), verbal repetition, or text reuse, is the best studied and often the strongest type of signal. Philogical commentaries, copiously detailed collections of information on individual books of Latin epic poems, have been the traditional means for Latin poetry scholars to collect and present interpretations based on studies of text reuse. An example from Parkes’ recent commentary on the fourth book of Statius’ Thebaid demonstrates the practice of translating the evidence of verbal repetition into interpretation:

[Statius, Thebaid 4.260] audaci Martis percussus amore [“struck by a bold desire for warfare”[2]]: … The collocation percussus amore [“struck by a desire”] is not uncommon (compare e.g. Verg. G. 2.476, Hor. Epod. 11.2 amore percussum, and Nem. Cyn. 99) but Statius may be specifically recalling the ephebe Euryalus’ reaction to Nisus’ planned expedition at Verg. A. 9.197: magno laudum percussus amore [“struck by a great desire for glory”]…. Like Parthenopaeus, Euryalus is eager to brave danger for the chance of glory (A. 9.205–6), with similarly fatal results.  [Parkes 2012, 164]

This exemplary note builds its interpretation on the evidence of the repetition of two key lexemes, the verb percutio (“I strike”) and the noun amor (“desire”).[3] The cooccurence of these lexemes in the Statian passage signifies for most readers a link to the passage from Vergil. The discovery of such verbal links has been facilitated in recent years by digital tools such as the freely available Tesserae web interface (tesserae.caset.buffalo.edu), a search program developed by Neil Coffee and a team at the University at Buffalo. Tesserae allows users to search pairs of texts (an earlier “source” text paired with a later “target” text) in a corpus of over 300 poetic and prose works, in order to discover every instance where short passages (either lines of verse or grammatical periods) share two or more repeated lexemes. Thus, a Tesserae search that pairs the Thebaid with the Aeneid permits the user to discover the allusion discussed by Parkes by identifying the repetition of the lexemes percutio and amor. The Tesserae scoring system signals the potential interpretive significance of the match by assigning it a high score, 8 out of approximately 11.[4]
In addition, Tesserae identifies a second potential match (score = 7) between Thebaid 4.260 and another passage from the Aeneid:

Statius, Thebaid 4.260 prosilit audaci Martis percussus amore (“Parthenopaeus leapt up, struck by a bold desire for warfare”).

Vergil, Aeneid 7.550 accendamque animos insani Martis amore (“I’ll inflame their minds with a desire for mad warfare”).

The words in the Aeneid are spoken by Allecto, a demon of the underworld, and we may thus once more translate this evidence of verbal repetition provided by Tesserae into literary interpretation.[5] Parthenopaeus’ desire to fight in the Theban war in Statius is not only fatal, like the desire of Vergil’s Euryalus to participate in Nisus’ expedition; it is also infernal, like the war provoked by Vergil’s Allecto. This is consistent with Statius’ characterization of the Theban war as destructive and impious throughout the Thebaid. Such new avenues for specific intertextual interpretation are the typical results of Tesserae searches. Previous examples of comparable results can be found in a study of verbal reuse of Vergil’s Aeneid by the epic poet Lucan [Coffee et al. 2012]. Coffee et al. hand-ranked all Tesserae results from a comparison of Lucan Bellum Civile 1 (target) and Vergil’s Aeneid (source) on a 5–point scale of interpretive significance. They concluded that the Tesserae search had identified 25% more interpretively significant instances of text reuse than the standard philological commentaries on Bellum Civile 1 [Roche 2009] [Viansino 1995].
The interpretation of specific allusions relies partly on the characterization of the overall intertextual relationship between texts, which is often hampered by a haphazard approach to gathering data. This paper presents a more consistent, quantitative picture of the interactions between poets in the Latin hexameter tradition. We use Tesserae to generate a statistical analysis of relative rates of text reuse in 24 Latin hexameter works written from the 1st century BCE to the 6th century CE. We then compare the quantitative information about text reuse provided by Tesserae to the scholarly tradition of qualitative discussion of allusion by Latinists. Statistical analyses of certain aspects of Latin poetry are not new. Drobisch’s studies beginning the 1860s represented the birth of the modern statistical studies of metrical aspects of the epic hexameter, a tradition which has reached a high-water mark in the recent work of Ceccarelli [Ceccarelli 2008] [Drobisch 1866]. Counts of individual lexical items in Latin poetry, usually in an effort to determine whether particular words should be considered “poetic” or “unpoetic”, are best represented by the tradition of Axelson’s work [Watson 1985] [Axelson 1945]. Yet scholars have not typically evaluated instances of verbal reuse in quantitative terms, as it has simply not been possible for human readers to count such instances accurately. The speed, consistency, and comprehensiveness of Tesserae searches now enable the interpreter to quantify the reuse of phrases on a scale beyond the capacities of ordinary human reading.
Powerful and productive as the Tesserae interface is, the following limitations must be clearly understood. They bear on analysis of specific passages, and to a lesser extent on our large-scale study:
1. Text reuse does not give the full, complex picture of intertextuality in Latin hexameter, where allusions may be signalled by similarity of action, character, theme, and so on.
2. Not all text reuse features the repetition of two or more lexemes. At its current stage of development, Tesserae focuses on pairs of lexemes and so cannot reliably identify repetition of single significant words. It would accordingly be unable to flag, for example, the very common word arma (“warfare”). This word takes on a new intertextual significance in poems written after the Aeneid, a foundational epic poem that begins with the words Arma uirumque cano… (“I sing of arms and the man…”) [Fowler 1997, 20]. There is accordingly need of a sensitive human interpreter to uncover the metapoetic significance, for example, of the opening word of Ovid’s Amores, Arma graui numero uiolentaque bella parabam / edere… (“I was beginning to sing of arms and violent wars in a serious meter...”)
3. The Latin poets wrote for an audience of Roman elites that were literate in Greek [Hutchinson 2013], and so created numerous translingual calques on Greek phrases. To remain with the example of Vergil, the Aeneid adapts numerous lines and phrases from Homer’s Iliad and Odyssey. Some foundational studies have uncovered these calques using traditional philological methods [Knauer 1964] [Nelis 2001], but such studies have not been pursued systematically across the Latin corpus. A feature of Tesserae currently in development searches for such translingual allusions between Latin and Greek poetry, but is not yet a reliable tool.
4. Repetitions with verbal variations that seem slight to a human reader are determinative for Tesserae. For example, Tesserae will locate the following correspondence based on the repetition of the lexemes Acheron and moueo:

Silius Italicus, Punica 2.536 quis Acheronta moues, flammam immanesque chelydros… (“[The weapons] with which you rouse the underword — flame and monstrous serpents…”).

Vergil, Aeneid 7.312 flectere si nequeo superos, Acheronta mouebo (“If I cannot sway the gods above, I will rouse the underworld”).

But Tesserae cannot yet locate the equally significant allusion:

Silius Italicus, Punica 2.367 …aeternum famulam liberque Acheronta uidebo (“…An eternal slave; I will see the underworld as a free man”).

The change from the verb moueo (“I move”) to uideo (“I see”) means the phrase no longer contains two repeated lexemes. This means that Tesserae will inevitably miss some of the variations on a verbal motif that form a component of the Latin poets’ creative art. That said, the majority of allusions identified via traditional reading are repeated phrases. So though Tesserae cannot uncover allusions of this type, the majority of such allusions are typically missed by human readers as well.
5. The Tesserae scoring system provides a measure of interpretive significance that correlates with human-generated measures [Forstall et al. 2014]. Numerous passages of Latin poetry that human readers have traditionally thought of as linked through allusion are also high-scoring lexeme matches, and these correspondences form the basis for scholarly confidence in the scoring system. Yet the score assigned to any given lexeme match does not generate by itself the kind of sensitive assessment of significance that a scholarly reader of Latin poetry brings to the identification of parallel passages. In order to be significant, the allusion must be placed in a larger scholarly narrative of the passage’s compositional goals. A human reader must be able to make a plausible interpretation of the allusion before it can be recognized as an allusion rather a chance repetition [Farrell 2005]. Tesserae’s usefulness comes in discovering potential allusive connections through lexeme matching and ordering them by the rarity and proximity of the paired lexemes. Subjective interpretation of these connections is still required for any meaning-making exercise [Drucker 2009].
Within these acknowledged limitations, Tesserae can be an extraordinarily powerful tool for representing the large-scale reuse of text in a literary tradition. Focusing as it does on repetition of phrases, the most commonly studied marker of allusion, Tesserae can provide a large-scale view of intertextual relationships that models traditional scholarly practice. The program can generate provisional answers to questions of particular relevance to the study of the Latin hexameter genre. Tesserae enables us to undertake the first large-scale statistical study of intertextuality in classical literary studies. Classicists have used new digital tools since their inception, and several techniques of digital text analysis were pioneered on Latin literary corpora, from Fr. Busa’s Index Thomisticus to the Packard concordance of Livy [Bodard and Mahony 2010] [McCarty 2005]. Studies of intertextuality, however, have generally been confined to pairs or very small sets of texts, and have traditionally relied on broad but subjective classification of intertextual data (synonyms, similar motifs, images, etc.), rather than objective parameters such as lexeme matches, lexeme frequency, and lexeme proximity. The Tesserae scoring system, however, represents the first opportunity to quantify the study of intertextuality using a large set of poems and objective parameters. Our object of study is the entire super-genre of Latin hexameter poetry, in which we privilege the system of relationships between texts rather than any integral text itself.
Latin poetry scholars have traditionally divided the “super-genre” of hexameter into several subgenres, including satire, epic, and didactic [Hutchinson 2013]. Is it possible to quantify the verbal cohesiveness and distinctiveness of these genres? What other general factors affect text reuse across the entire hexameter tradition? Can the well-known influence of Vergil and Ovid on their epic successors be quantified? In particular, can it be determined how frequently one predecessor’s text is reused compared to another’s? For example, is Statius’ Thebaid more “Vergilian” in terms of text reuse than another contemporary epic poem, Silius Italicus’ Punica? Most specialist readers of these Flavian epic poets would correctly guess that the answer is no, but would perhaps not be so confident in making assertions about the two poems’ relative rates of reuse of other, earlier poets such as Ovid, Lucan, or Manilius. Which works in the classical hexameter tradition provide the most significant verbal resources for the hexameter epics of late antiquity? This study offers preliminary answers to such questions from a quantitative perspective by surveying the relative rates of text reuse in 24 Latin hexameter works written from the 1st century BCE to the 6th century CE.

2. METHODS

a. Text Selection

Our analysis included every possible source–target pair from a set of 24 Latin hexameter texts written from the 1st century BCE to the 6th century CE (Table 1[6]). This set included every hexameter text available on the Tesserae website,[7] excluding hexameter poems from polymetric collections (such as Catullus’ poems or Statius’ Silvae), hexameter works with non-hexameter prefaces (such as Claudian’s In Rufinum[8]), and four very short minor texts.[9]

b. Data collection and scoring

Using the Tesserae Batch Processing option (http://tess-dev.caset.buffalo.edu/html/batch.php), we recorded the number of “hits” (phrases sharing at least two matching lexemes) in each source–target pair (searches conducted on 2 May 2014). Hits may include exact matches of inflected forms, such as Vergil, Georgics 1.493 exesa inueniet scabra robigine pila ~ Statius, Thebaid 3.582 tunc fessa putri robigine pila (lemmata: robigo, pilum). Matches may also occur among differently inflected forms of the same lexeme, such as Vergil, Georgics 2.64 solido Paphiae de robore myrtus ~ Statius, Thebaid 4.300 hi Paphias myrtos a stirpe recuruant (lexemes: Paphius, myrtus).[10]
We used a set of search parameters that capture the most instances of interpretively significant text reuse while excluding many instances of less significant reuse. These were:
• phrases as the search unit
• lemma as the matching feature
• 20 stop words, determined by frequency in the Tesserae corpus
• scores calculated by stem
• a maximum distance of 10, calculated by frequency
• no score cutoff[11]
We then partitioned the results by score. Tesserae assigns each matched phrase a score (rounded to the nearest integer) according to the following formula, which reflects the observation that instances of text reuse featuring rare words in close proximity are often more interpretively significant than instances featuring common words spaced farther apart [Forstall et al. 2014] [Coffee et al. 2013].
• $$Score = \ln\left( \frac{\sum_{}^{}{\frac{1}{f(t)} + \sum_{}^{}\frac{1}{f(s)}}}{d_{t} + d_{s}} \right)$$
• f(t) is the frequency of each matching term in the target text
• f(s) is the frequency of each matching term in the source text
• dt is the distances in the target text
• ds is the distances in the source text
Examples of hits of different scores are listed in Table 2.

c. Weighing of counts

We thus obtained for each pair a count of the number of hits at each score (from 2 to 11). Hits scoring 6 and lower were excluded from the analysis, since it has been shown that these are unlikely to be instances of interpretively significant text reuse [Forstall et al. 2014]. We were left with five data points for each pair, C7, C8, C9, C10, and C11 (counts of score 7, 8, 9, 10, and 11; Table 8 and 9). In order to convert these five counts into a single useful “composite count”, C, we took advantage of the strongly linear relationships between counts of every score except for the rare C11 hits. Because the mean correlation was strongest between C9 and the other counts (mean R2 = 0.879; mean ρ = 0.931), the smallest amount of error was introduced by converting all counts into C9, using a combination of linear regressions and principal component analysis.
First, we used a series of linear regressions to characterize the relationship between C9 and the other four counts and obtain an initial composite count, Cregr.[12]Second, we applied principal component analysis (PCA) to the five counts, first correcting for their very different scales by dividing each count by its standard deviation, in order to obtain a second composite count, Cpca.[13] Noting the similar weights in the formulae for Cregr and Cpca, we chose the average weights for the final formula for composite counts, which we considered to be the “observed count”, Cobs:
• $$C_{regr} = 0.057C_{7} + 0.225C_{8} + C_{9} + 6.168C_{10} + 212.062C_{11}$$
• $$C_{pca} = 0.057C_{7} + 0.225C_{8} + C_{9} + 6.404C_{10} + 243.426C_{11}$$
• $$C_{obs} = 0.057C_{7} + 0.225C_{8} + C_{9} + 6.286C_{10} + 227.744C_{11}$$

d. Relative intensity of reuse

The resulting observed counts could not be directly compared to one another, since the total lengths of the texts were different for each source–target pair. For instance, we expected to obtain a much higher Cobs value for the pair Ovid, Metamorphoses (78098 words) – Silius Italicus, Punica (76292 words) than for the pair Horace, Ars Poetica (3090 words) – Claudian, De Bello Gildonico (3165 words), simply because there is much more space for text reuse in the longer texts. Indeed, we found that Cobs was correlated with the lengths of both source and target texts, Ws and Wt; the correlations were strongly linear when the variables were converted to a logarithmic scale (cobs, ws, and wt).
Thus, we could use a multiple regression to determine (in logarithmic scale) an expected count, cexp, for any given length of source and target text, ws and wt. We obtained the model (R2 = 0.979):[14] $$c_{exp} = - 19.591 + 1.311w_{s} + 1.208w_{t}$$ We then subtracted the expected count for each source–target pair from the observed count to obtain a residual, which we considered to be a measure of the relative intensity of text reuse for each pair: $$r + c_{obs} - c_{exp}$$
A positive value of r for a given pair indicates that the observed intensity of text reuse was higher than would be expected for an “average” pair of texts with those particular word counts — that is, for a pair of texts with no particularly strong or weak intertexual relationship. A negative value of rindicates that the observed intensity of text reuse was lower than average. The further the value deviates from zero, the stronger the evidence for an intensity of reuse above or below average. Thus, we sorted all pairs by their r values, presented in both standardized and non-standardized forms (Table 3).[15] We also presented the (non-standardized) r values graphically, partitioning the pairs by source text (Figure 07) and target text (Figure 12), and presented various subsets of the data to aid discussion (Figures 13–15, Tables 5–7).
It should be reiterated that r is not a measure of the number of phrases reused for each pair (for which Cobs is the most direct measure), but a measure of the intensity of text reuse that takes into account the lengths of the source and target texts in each pair. For instance, the very high Cobs value of 7407.3 for the pair of the longest texts in our data set, Ovid, Metamorphoses (78098 words) – Silius Italicus, Punica (76292 words), actually reflects only moderately intense text reuse (r = 0.146), whereas the very intense reuse (r = 1.280) of Vergil’s Georgics (14154 words) by Vergil’s later poem, the Aeneid (63719 words) corresponds to a lower Cobs value (1974.8) because the texts are shorter.

e. Centrality

For each of our 24 chosen texts, we determined the mean value of r for all pairs involving that text (23 pairs each time), and sorted the texts by the results (Table 4). We considered this to be a measure of the “centrality” of each of our chosen texts within the 24–text set: that is, how often each text reuses earlier texts and is reused by later texts. A text strongly influenced by its predecessors and influential to its successors would have a higher mean r than a text more peripheral to the literary tradition of Latin hexameter poetry.

3. RESULTS AND DISCUSSION

We have kept two objectives in mind in interpreting our data set. First, we attempt to test whether the results of the automated search and statistical analysis match the conclusions reached by traditional scholarship. Second, we endeavor to identify unexpected results that suggest avenues for future research. We achieve those two objectives when interpreting both general (sections 3.a-b) and specific trends (section 3.c).

a. Statistical outliers and centrality

Three pairs with standardized residuals near or above |3| may be considered statistical outliers (GeorgAen, standardized r = 4.571; MetMos, 3.830; ArsGild, –2.977). These results reflect several phenomena that we will discuss: the influence of author on text reuse (GeorgAen, section 3.b), the influence of genre (ArsGild, 3.b), and the importance of Ovid (among others) to late antique hexameter (MetMos, 3.c.iv). For a further 11 pairs, standardized residuals near or above |2| indicate intensity of text reuse markedly above or below average; these results also reflect phenomena that we will discuss.[16] These standard statistical thresholds should not be relied upon naively, however: for instance, several pairs for which we would expect a strong intertextual engagement (such as texts written by the same authors) had standardized r values well below 2.
The centrality scores conformed to expectations (Table 4). The high centrality of the Aeneid (0.133) reflects the importance of Vergil’s works to the subsequent hexameter tradition, while the high centrality of the Ilias Latina (0.186) reflects multiple reuse facilitated by its intense reuse of the Aeneid (see section 3.c.i). The high centrality of the Georgics (0.279) stems from a combination of these factors. All four of Claudian’s works had positive centrality. This reflects not only Claudian’s extensive reuse of his predecessors, but also the influence of authorship on text reuse: each of Claudian’s works had high r values when paired with the other three works, thus increasing their centrality. The low centrality of the works of Horace, Persius, Juvenal, and Lucretius reflects the influence of genre in our data set comprising mainly epic/panegyric texts. Perhaps the most unexpected result is the high centrality of the Achilleid (0.117), which reflects both intense reuse of earlier epic sources and intense reuse by later epic targets. Because the Achilleid is a very short text, certain considerations must be kept in mind (see section 3.c.i).

b. General trends

Unsurprisingly, the most important influence on text reuse intensity was authorship. In all 13 cases where a pair of texts was written by the same author, the reuse intensity was higher than average (r > 0.000), markedly so in 5 of the cases (standardized r > 2.000); see Figure 13 and Table 5. Vergil showed the highest intensity of text reuse within his own poems, followed by Claudian, while Horace and Statius reused their own poems with less intensity. Though drawing on a very different data set (a relatively small corpus of Latin hexameter poems), the results are nevertheless broadly comparable to Jockers’ study of the author signal in a corpus of 3500 nineteenth-century novels written in English. Jockers observes that of five “signals” (author, decade, genre, gender, and text), the author signal is the strongest.[17]
A secondary influence on text reuse intensity was genre. Although categorizing Latin poetry by genre is difficult, we may obtain a rough idea of the influence of genre by partitioning the texts of our data set into three genres: didactic, epic/panegyric, and satiric (Figure 14).[18] Within the small didactic and satiric genres, reuse intensity was higher than average for 5 of 6 pairs (r > 0.000; the exception, HSat – JSat, was slight: r = −0.007). Within the much larger (and more diverse) epic/panegyric genre, reuse intensity was higher than average in 66 of 78 pairs; the 12 remaining pairs had only slightly lower than average reuse intensity (standardized r ≥ −0.446). In contrast, pairs comprising texts from different genres tended to display lower than average reuse intensity. The trend was clearest for pairs composed of one epic/panegyric and one satiric text: 37 of 39 pairs had lower than average reuse intensity.[19] The results conform to the expectations of traditional reading, as epic and satire are the most distant hexameter genres from one another in style and subject matter. Genre is also perhaps the best explanation for the trends seen in the “centrality” measure (Table 4). Since 13 of 24 texts in our data set belong to the epic/panegyric genre, we would expect each of them to be more central than texts belonging to smaller genres. This is true in most cases; the most notable exception is the Georgics, which had the highest centrality score by far, despite belonging to the didactic genre. We discuss this exceptional text in section 3.c.i.
Time period appeared to have no influence on text reuse intensity. This is not surprising, since the technical and aesthetic constraints of hexameter poetry discouraged changes in diction or syntax over time. However, it is possible that a future study which controls for much more salient influences such as authorship and genre may discover a subtle influence of time period.

c. Specific observations

The 276 pairs in our data set represent a generically and chronologically diverse collection of texts. Different scholars will accordingly highlight various aspects of the data. We only offer a handful of specific observations here. As with the general trends we observed, these specific results both confirmed that our analysis falls in line with the results of traditional scholarship and identified several possible avenues for future inquiry. For instance, Virgil’s Aeneid predictably emerged as a major influence on subsequent poetry of all periods. Lucretius’ De Natura Rerum was not a prominent verbal resource for later authors. The four Flavian epics were closely related, and late antique poets reused material from previous works in expected ways. The congruence of these results with traditional scholarship supports our contention that several unexpected results are indicators of potential for fruitful further research. For instance, Virgil’s Georgics and the anonymous Ilias Latina scored high in reuse intensity in almost every case. This is probably an indication of frequent multiple allusions to both these texts and the more prominent Aeneid and Metamorphoses (section 3.c.i). The relationship between the Flavian epics and the Aeneid appears to be more “creative” or “original” than often allowed, although these terms must be used carefully (see 3.c.ii). Horace’s Ars Poetica seems to have an unexpected influence on Manilius’ Astronomicon, suggesting that didactic sensibility may cut across genre (see 3.c.iii). Finally, Ausonius’ Mosella, usually considered primarily “Vergilian” in nature, also shows close links with Ovid’s Metamorphoses (3.c.iv).

i. Vergil’s Georgics and the Ilias Latina

The influence of Vergil’s Aeneid on the subsequent tradition of Latin hexameter is well established and reflected in our results. The work had a high centrality score (0.133) and higher than average reuse intensity (r > 0.000) when paired with 13 of 18 subsequent target texts (the exceptions are BC and the non-epic texts Ars, PSat, JSat, and HE); see Tables 4 and 6 and Figure 11. However, the results for Vergil’s early work, the Georgics, are even more exceptional. Its centrality score was more than twice as high (0.279) and it had higher than average reuse intensity (r > 0.000) when paired with 16 of 20 subsequent target texts (the exceptions are the non-epic texts Ep, Ars, PSat, and JSat). These results may seem surprising at first. Although the Georgics is an important text, few would argue that its influence on subsequent Latin literature eclipses that of the Aeneid. But two factors must be kept in mind. First, recall that r is not a measure of the number of phrases reused for each pair (for which Cobs is the most direct measure), but a measure of the intensity of text reuse that takes into account the lengths of the texts in each pair. Because the Aeneid is much longer than the Georgics (63719 vs 14154 words), it requires values of Cobs over 7 times higher, and thus the reuse of many more phrases, in order to achieve the same residual when paired with any subsequent target text. Subsequent target texts use many more phrases from the Aeneid than from the Georgics in total,[20] and the influence of the Aeneid on subsequent literature is therefore more obvious to the reader. Yet the intensity of the reuse is greater for the shorter Georgics.
The second factor arises from Vergil’s extensive reuse in the Aeneid of his own phrases from the Georgics, which resulted in the highest r value in our data set (1.280), one of three statistical outliers (standardized r = 4.571). Because Vergil’s two texts share many phrases, subsequent target texts that reuse phrases from one Vergilian text will often automatically reuse the same phrase from the other Vergilian text. In practice, subsequent epic poems that reuse phrases from the epic Aeneid will often automatically reuse the same phrase from the Georgics. A similar phenomenon explains the unexpected results for the Ilias Latina. Although no scholar would argue that this minor poem, a rough compression and translation of the Iliad, exerted any discernable influence on Latin literature in antiquity,[21] it had a higher centrality score than the Aeneid (0.186) and higher than average reuse intensity (r > 0.000) when paired with every subsequent target text (Tables 4 and 6 and Figure 11). However, the Ilias Latina also had markedly higher than average reuse intensity (standardized r > 2.000) when paired with both the Aeneid and Ovid’s Metamorphoses, two foundational texts for later Latin literature.[22] This suggests that when a subsequent target text reuses phrases from either the Aeneid or the Metamorphoses, it will often automatically reuse the same phrase from the Ilias Latina and thereby increase the r value when paired with that poem.
The high scores for both the Georgics and Ilias Latina demonstrate that allusion in Latin literature is not always a case of a target text reusing a phrase from a single, specific source text. On the contrary, an allusion to, say, the Aeneid often necessarily entails an allusion to the Georgics, the Ilias Latina, or some other text(s). While scholars routinely privilege one source text at the expense of the others for the sake of interpretation, the automatic searches of Tesserae do not. This egalitarian interpretive practice is not very suitable in the case of the Ilias Latina, a minor text rightly subordinated to the sources it reuses, but it is more suitable in the case of the Georgics, where readers will more often hit upon compelling interpretations by treating the Georgics as a source text on par with the Aeneid.[23] Tesserae encourages this kind of interpretation not only by presenting all texts as equal in value, but also by offering the option to perform multi-text searches (http://tesserae.caset.buffalo.edu/multi-text.php), where matches between a source–target pair are presented alongside every other instance of the matching phrase in a user-selected set of texts.

ii. Post-Vergilian classical epic

Scholarly interest in post-Vergilian classical epic (the Metamorphoses, Bellum Civile, Argonautica, Thebaid, Achilleid, and Punica) has roughly tracked the chronology of the epics themselves, with attention paid first to the Metamorphoses and last to the Punica. Similarly, the assumption has often been made that the earlier epics (Metamorphoses and Bellum Civile) responded to Vergil’s influence in more creative and original ways, while the four later epics of the Flavian period tended to imitate Vergilian epic less creatively.[24] To compare this assumption to the results of our study, we must bear in mind the nature of the text reuse that Tesserae can discover. At its current stage of development, Tesserae identifies only matching phrases with exact repetition of two or more lexemes. It cannot detect allusions signaled by similarity of action, character, or theme, or text reuse involving single significant words or verbal variations. That is, Tesserae preferentially detects exactly the sort of allusions that may be classified as less “creative”. Thus a high residual indicates not only higher than expected text reuse, but also potentially a less “creative” allusive relationship.
Bearing this in mind, the results do not fully support the assumption of declining creativity over time (Figure 15 and Table 7). In contrast, although the intensity of text reuse of both the Georgics and Aeneid by the Argonautica, Thebaid, and Achilleid was higher than average (0.160 ≤ r ≥ 0.299), it was not as high as the intensity of reuse of any of Vergil’s three works by the Metamorphoses (0.323 ≤ r ≥ 0.560). The intensity of reuse of Vergil by the Bellum Civile was even lower: in fact, the intensity of reuse of the Aeneid was slightly lower than average (r = −0.026).[25] Thus, it would seem that the intertextual engagement with Vergil’s texts by Lucan, Valerius Flaccus, and Statius are either less intense or more “creative” (or both) than often assumed.
The notable exception is the Punica of Silius Italicus, which had much higher than average intensity of text reuse when paired with the Georgics (r = 0.433) and Aeneid (r = 0.540). This is consistent with the assumption of an uncreative intertextual relationship, and inconsistent with recent claims about the Punica’s originality.[26] It must be acknowledged, however, that “originality” and “creativity” are subjective concepts, which are not directly measured by r values. A high r value for a given pair indicates only that the number of matching phrases of two or more lexemes was greater than expected for an “average” pair of texts with the same word counts. It does not indicate, for instance, a paucity of other kinds of subtler intertextuality (text reuse with verbal variation, or similarities of action, theme, or character). Nor does it take into account the context into which the lexemes are redeployed: a poet may, for instance quote a predecessor’s words exactly, but in a completely different and original context.
Other observations may be made about the results for the four Flavian epics. The high r values for the epics when paired with the Georgics (0.160 ≤ r ≥ 0.433) may be influenced by factors discussed in section 3.c.i, but scholars have begun to interpret the relationship between these texts more aggressively (Pagán 2015), and our results support this line of inquiry. The Metamorphoses and Bellum Civile have often been interpreted as important texts for the Flavian epics; however, although the intensity of text reuse for the eight relevant pairs was usually higher than average (r ≥ −0.075), it was usually only moderately so, approximately on par with the intensity of reuse for the epics when paired with the Eclogues, a text rarely argued to be important to Flavian epic. Again, this does not argue against a strong intertextual engagement between the Metamorphoses, Bellum Civile, and Flavian epics; it may instead suggest that future investigations should focus on allusions not signalled by the obvious text reuse that Tesserae discovers.
The intertextual relationship between the four Flavian epics has been the subject of recent study, and this line of inquiry is supported by our results. The intensity of the Thebaid’s reuse of the Argonautica was slightly higher than average, on par with the Thebaid’s reuse of the Metamorphoses (r = 0.064, 0.037). The intensity of the Achilleid’s reuse of the Argonautica was much higher than average, on par with the Achilleid’s reuse of the Aeneid (r = 0.279, 0.289).While the intertextual relationship between the Thebaid and Argonautica has been well studied, the relationship between the Achilleid and Argonautica has not;[27] future work in this vein could be productive. Unsurprisingly, the intensity of reuse of Statius’ Thebaid by Statius’ later Achilleid was higher than average (r = 0.141), but it was lower than 11 of the 12 remaining intra-author pairs (Figure 13 and Table 5). This low reuse cannot be explained purely by the divergent subject matter and style of Statius’ two epics: Vergil’s Eclogues and Aeneid are at least as divergent, but had a higher r value (0.224). Finally, the r value for the pair AchilleidPunica was very high (0.410).[28] This was unexpected. Research on the intertextual relationship between Statius’ and Silius’ works has focused on the pair ThebaidPunica,[29] but these results suggest more attention should be paid to the Achilleid. In all discussion of the Achilleid, however, we should keep in mind that it is much shorter than the other three Flavian epics; therefore, the considerations that applied to the Georgics in section 3.c.i apply here.

iii. Didactic and satiric hexameter

Hardie’s study of the reception of Lucretius makes a strong and well-received case for the fundamental contribution of the De Rerum Natura to succeeding poetry from the Augustan poets through Milton’s Paradise Lost [Hardie 2009]. No reader would dispute the conceptual and formal importance of the DRN to the Latin hexameter tradition. Features of later hexameter poetry such as sententiae, multiple explanations, and similes from the natural world all bear the marks of the Epicurean poet’s mode of argumentation. Yet the vocabulary of the DRN was not mined as extensively as the other foundational works of Republican and Augustan poetry, as can be seen from our results (centrality = −0.151, r < 0.000 when paired with 21 of 23 succeeding target texts; Figure 11). The only positive r values resulted from pairings with other didactic works: Vergil’s Georgics (r = 0.230) and Manilius’ Astronomica (r = 0.023). While these results are consistent with the observed influence of genre on text reuse (section 3.b), the low r values overall demonstrate the difference between the importance of Lucretius’ poem as a conceptual resource and its importance as a verbal resource.
Volk’s study of the Astronomica makes a series of valuable observations about Manilius’ thematic adaptations of Lucretius, Vergil, and Ovid [Volk 2009]. Those thematic adaptations were accompanied by verbal reuse only for Vergil in our results. Vergil’s Georgics yielded the highest reuse intensity (r = 0.342), followed by the Eclogues (r = 0.307). Unexpectedly, Horace’s Ars Poetica had the next highest r value (0.213). As the Ars is one of the shortest poems in our data set, the considerations that applied to the Georgics in section 3.c.i apply here. Yet there may be hitherto unexplored verbal connections between the poem on composing poetry and the poem of the stars, likely in the addresses of the didactic narrator. The intensity of reuse of the DRN was higher than average, but only negligibly so (r = 0.023). The intensity of text reuse of the Astronomica by later texts was low, suggesting a limited influence on the language of subsequent classical hexameter tradition.
The inclusion of the Satires of Horace, Persius, and Juvenal (HSat, PSat, JSat) in this study permits us to begin investigation of the influence of genre on text reuse in Latin hexameter. As mentioned above (section 3.b), the author signal is a stronger determinant than the genre signal for intensity of text reuse, as evidenced by higher r values for pairs of texts written by Horace than inter-author pairs within the satiric genre.[30] But the importance of genre was especially marked when pairing epic/panegyric with satiric texts, where 37 of 39 pairs had lower than average reuse intensity (r < 0.000), including the lowest r values in our data set (Figure 14).[31] These results indicate a strong separation between the genres, related to satire’s pedestrian vocabulary and everyday concerns, which contrast with the more elevated style and subject matter of epic.

iv. Late antiquity

The tremendous influence of Vergil and Ovid on the hexameter poems of late antiquity has been well recognized in prior scholarship, but has been typically studied from the perspective of theme, character, and subject. The present study permits some initial quantification of the intensity of text reuse between these poems and those occurring earlier in the hexameter tradition.
Prior scholarship has identified Ausonius’ Mosella as primarily Vergilian in character, with several secondary influences, but has not heretofore been able to quantify the nature of Ausonius’ reuse of his predecessors’ texts.[32] In our study, the intensity of text reuse of Ovid’s Metamorphoses by the Mosella was markedly higher than average (standardized r > 2.000). This pairing had the highest r value of any two independently authored texts (r = 1.073), and second only to Vergil’s reuse of the Georgics in the Aeneid (r = 1.280). The intensity of reuse of Vergil’s works was decidedly lower (Georgics, r = 0.260; Aeneid, r = 0.115). The intensity of reuse of Statius’ Achilleid and Silius Italicus’ Punica was slightly above average (r = 0.104 and 0.028), but lower than that of Manilius’ Astronomica and the Ilias Latina (r = 0.130 and 0.120; for the latter, see section 3.c.i). Intensity of reuse was lower than average (r < 0.000) for Lucretius’ De Rerum Natura, Lucan’s Bellum Civile, Valerius Flaccus’ Argonautica, and Statius’ Thebaid. The centos entirely composed of phrases adapted from Vergil’s works that appear in this period represent a new level of engagement with the foundational texts of the genre [McGill 2005]. Ausonius’ Cento Nuptialis, the best known of the centos, is available on Tesserae, but was excluded in this study, since its artificially high reuse rates of Vergil’s works would have produced extreme outliers that would have distorted our results.
As observed above (section 3.b), the works of Claudian are evidence for the strength of the author signal. Four of the top fifteen r values in our data set were derived from pairing works of Claudian (HonStil, HonGild, GildStil, and RaptHon; 0.461 ≤ r ≥ 0.716). The lower position of the De Raptu Proserpinae among the pairings of Claudian’s works (RaptHon, RaptGild, RaptStil; 0.243 ≤ r ≥ 0.461) may suggest that Claudian’s self-reuse is strongest among works in a similar genre (panegyric rather than mythological epic). We are hesitant to draw firm conclusions, however, about the relative importance of the author and genre signals with so few data. Claudian’s rates of reuse of his Augustan predecessors present a similar story to that told in the scholarly literature [Ware 2012, 9–10]. For instance, Vergil’s Georgics (r = 0.538) and Aeneid (r = 0.326) had high reuse intensity when paired with Claudian’s mythological De Raptu Proserpinae. The intensity of reuse of Statius’ Achilleid was also high (r = 0.426), which accords with the importance of Statius as an intermediary between the Augustans and the poets of late antiquity. As Kaufmann observes, “Claudian, possibly inspired by Ausonius, [was] the trendsetter for the increased interest in Statius’ poetry by the later poets”  [Kaufmann 2015]. An unexpected but plausible result is the importance of Lucan’s Bellum Civile to Claudian’s historical panegyrics, Gild (r = 0.351) and Hon (r = 0.278).
We also included Juvencus’ Historia Evangelica, a fourth-century Christian epic, and Corippus’ Johannis, a sixth-century historical epic, in the data set. Both the Johannis’ high rates of reuse of Vergil and Claudian and the HE’s low rates of reuse of classical pagan poetry (with the exception of the Georgics and Ilias Latina) conform to the expectations set by the scholarly literature.[33]

4. CONCLUSIONS

We chose to begin by studying a selected corpus of Latin hexameter poems because relationships between works in this “super-genre” have been the most closely studied of all intertextual relationships in ancient literature. We are able to compare the information about the relative rates of reuse of texts in Table 3 to a long tradition of qualitative discussion of allusion by Latinists. We provisionally conclude that a majority of the results conform to the statements typically made by poetry scholars about the significance of various intertextual relationships in the Latin hexameter tradition. For instance, the author signal is one of the strongest determinants of intensity of text reuse, the works of Ovid and Vergil are the most important verbal resources for the later works of the tradition, and satiric hexameter is strongly separated from the other hexameter genres in terms of reuse. If it is accepted that the high level of correlation between our quantified results and the scholarly tradition’s qualitative assessments provides a strong vote of confidence for our methodology, then we can begin to explore the significance of unexpected findings. These include (a) the importance of Vergil’s Georgics to the later tradition, (b) the indications of multiple reuse visible in the Ilias Latina, (c) the relatively low reuse of Vergil by Lucan, Valerius, and Statius, and (d) the intense reuse of Ovid’s Metamorphoses by Ausonius’ Mosella.
This is a first step in algorithmic criticism of the hexameter super-genre [Ramsay 2011]. As observed in the Introduction, Tesserae has some limitations which reflect its current state of development, and others which reflect the nature of Latin poetry. In this initial study, we confirmed the value of the lexeme-matching approach by comparing it to the traditional critical narrative of relationships among Latin hexameter poems. Our goal is to model a system of relationships between texts that can frame critics’ discussions of the role of individual poems within the tradition. As Drucker observes, “on the surface, a model seems static. In reality it is, like any ‘form,’ a provocation for a reading, an intervention, an interpretive act”  [Drucker 2009, 16]. In Drucker’s terms, Tesserae modeling is a dynamic rather than static approach to textual analysis. New data sets can easily be constructed, whether by using different Tesserae parameters or changing the texts in the group under analysis. These future analyses will produce new and different perceptions of the system of relationships among Latin literary texts in other genres, or between other genres and the hexameter super-genre.

Tables and Figures

 Text Abbreviation (name of work) Date (approximate) Length (words) Lucretius, De Rerum Natura DRN before 55 BCE 49099 Vergil, Eclogues Ecl 42–39 BCE 5617 Horace, Satires HSat 40–30 BCE 14215 Vergil, Georgics Georg 36–29 BCE 14154 Horace, Epistles Ep 23–20 BCE 9906 Vergil, Aeneid Aen 29–19 BCE 63719 Horace, Ars Poetica Ars 14 BCE 3090 Ovid, Metamorphoses Met 2–8 CE 78098 Manilius, Astronomica Astr after 9 CE 27353 Persius, Satires PSat before 62 CE 4457 Lucan, Bellum CivileM BC 64–65 CE 51065 [Italicus], Ilias Latina Ilias 60–70 CE 6597 Valerius Flaccus, Argonautica Arg before early 90s CE 37250 Statius, Thebaid Theb 92 CE 62504 Statius, Achilleid Ach 95 CE 7204 Silius Italicus, Punica Pun before 96 CE 76292 Juvenal, Satires JSat after 96 CE 24884 Juvencus, Historia Evangelica He 330 CE 19854 Ausonius, Mosella Mos 370 CE 2957 Claudian, De Raptu Proserpinae Rapt 395–397 CE 6991 Claudian, De Quarto Consulatu Honorii Augusti Hon 397 CE 3965 Claudian, De Bello Gildonico Gild 398 CE 3165 Claudian, De Consolatu Stilichonis Stil 399–400 CE 7583 Corippus, Johannis Joh 6th c. CE 29046
Table 1.
 Score 11 Pun 13.752 miscuerint Italis Piraeo litore leges Met 6.444 Cecropios intrat Piraeaque litora tangit Ilias 401 instat et exstructos morientum calcat aceruos Met 5.85 sternit et exstructos morientum calcat acervos Score 10 Theb 10.228 cum fetura gregem pecoroso vere novavit Ecl 7.35 si fetura gregem suppleverit, aureus esto Astr 2.807 per latera atque imum templi summumque cacumen Aen 6.678 dehinc summa cacumina linquunt Score 9 Astr 1.753 nec mihi celanda est famae vulgata vetustas Aen 12.608 Hinc totam infelix volgatur fama per urbem JSat 10.99 an Fidenarum Gabiorumque esse potestas HEp 1.11.7 Gabiis desertior atque / Fidenis vicus Score 8 Theb 7.262 arma patris pinuque iubas imitatur equinas, / terribilis silvis Ecl 2.31 Mecum una in silvis imitabere Pana canendo Astr 4.897 pars sua perspicimus genitique accedimus astris Aen 9.641 sic itur ad astra, / dis genite et geniture deos Score 7 Theb 7.447 ipsa loco mirum natura favebat Ecl 3.68 ipse locum, aëriae quo congessere palumbes Astr 4.96 quin etiam infelix virtus et noxia felix Aen 9.799 Quin etiam bis tum medios invaserat hostis
Table 2.
Randomly selected examples of hits from Tesserae searches scoring 11, 10, 9, 8, and 7.
 Source Target r Standardized r Georg Aen 1.280 4.571 Met Mos 1.073 3.830 Met Ilias 0.719 2.565 Hon Stil 0.716 2.555 Ilias Joh 0.663 2.368 Hon Gild 0.634 2.264 Ecl Georg 0.603 2.153 Aen Ilias 0.594 2.119 Gild Stil 0.575 2.054 Georg Met 0.560 1.999 Aen Pun 0.540 1.928 Georg Rapt 0.538 1.921 Rapt Hon 0.461 1.644 Ilias Gild 0.457 1.631 Georg Pun 0.433 1.546 Ilias Pun 0.433 1.545 Gild Joh 0.427 1.525 Ach Rapt 0.426 1.520 Ach Pun 0.410 1.462 Rapt Gild 0.404 1.442 Ilias Ach 0.396 1.414 Mos Hon 0.395 1.411 Ilias Arg 0.389 1.387 Ach Joh 0.375 1.337 Hon Joh 0.372 1.327 Georg Joh 0.355 1.266 Ep Ars 0.354 1.262 Georg BC 0.351 1.253 BC Gild 0.351 1.252 Aen Met 0.350 1.249 Georg Astr 0.342 1.221 Ach Stil 0.331 1.183 Aen Rapt 0.326 1.162 Rapt Stil 0.324 1.157 Ecl Met 0.323 1.151 Georg Ilias 0.310 1.107 Ecl Astr 0.307 1.096 Rapt Joh 0.302 1.079 Aen Theb 0.299 1.067 Georg Arg 0.297 1.061 Aen Ach 0.289 1.033 Arg Ach 0.279 0.996 BC Hon 0.278 0.992 Aen Joh 0.269 0.961 Georg Gild 0.268 0.957 HE Joh 0.267 0.951 Ach Gild 0.263 0.939 Georg Mos 0.260 0.928 HSat Ep 0.259 0.925 Aen Arg 0.255 0.910 BC Stil 0.253 0.903 Ilias Theb 0.252 0.899 BC Rapt 0.250 0.893 Mos Stil 0.247 0.883 Mos Joh 0.243 0.866 Ach Hon 0.238 0.851 Met BC 0.238 0.850 Georg Hon 0.232 0.829 DRN Georg 0.230 0.823 Ars Stil 0.228 0.813 Ecl Aen 0.224 0.800 Ecl Ilias 0.224 0.799 Ilias HE 0.223 0.795 Ars Astr 0.213 0.762 HSat PSat 0.210 0.750 Stil Joh 0.199 0.711 Georg Theb 0.186 0.664 Georg HE 0.172 0.614 Georg Ach 0.160 0.570 Ilias Stil 0.157 0.560 Mos Rapt 0.153 0.546 Ilias Rapt 0.152 0.543 Met Pun 0.146 0.520 Theb Ach 0.141 0.503 PSat JSat 0.137 0.490 Georg Stil 0.132 0.472 BC Joh 0.132 0.471 Astr Mos 0.130 0.465 Theb Rapt 0.128 0.457 Pun Rapt 0.128 0.457 BC Pun 0.126 0.450 Astr Ilias 0.125 0.446 Ilias Mos 0.120 0.428 Aen Mos 0.115 0.409 Ach Mos 0.104 0.373 Astr Joh 0.101 0.361 BC Ach 0.097 0.348 Arg Gild 0.097 0.346 Ecl Stil 0.094 0.335 Ecl Pun 0.093 0.332 Aen Gild 0.092 0.328 Met Rapt 0.089 0.319 Ars JSat 0.078 0.278 Met Ach 0.077 0.276 Arg Pun 0.076 0.272 HE Gild 0.075 0.267 Arg Rapt 0.067 0.239 Aen Hon 0.064 0.229 Arg Theb 0.064 0.229 Pun Hon 0.060 0.214 Theb Pun 0.057 0.204 JSat Hon 0.053 0.188 Theb Hon 0.052 0.185 Pun Joh 0.051 0.182 Astr BC 0.048 0.171 Astr Ach 0.047 0.168 Ecl Ach 0.047 0.167 Ilias JSat 0.045 0.160 Mos Gild 0.042 0.150 HSat Ars 0.041 0.146 HSat Georg 0.037 0.133 Met Theb 0.037 0.131 Ilias Hon 0.036 0.130 Astr Hon 0.036 0.130 BC Arg 0.035 0.125 Aen Stil 0.030 0.108 BC Ilias 0.028 0.100 Pun Mos 0.028 0.099 Arg Hon 0.024 0.087 DRN Astr 0.023 0.082 Pun Gild 0.018 0.065 Met Stil 0.017 0.060 Ars HE 0.013 0.047 Aen Astr 0.011 0.039 Ars PSat 0.008 0.028 Ep JSat 0.003 0.010 Ecl Arg -0.003 -0.012 Met Arg -0.006 -0.020 Theb Stil -0.006 -0.022 Astr Stil -0.007 -0.026 HSat JSat -0.007 -0.026 Ecl Rapt -0.009 -0.032 Georg JSat -0.009 -0.033 HE Rapt -0.010 -0.036 Astr Pun -0.013 -0.046 DRN Aen -0.015 -0.054 HE Stil -0.016 -0.058 Astr Rapt -0.016 -0.058 DRN Ilias -0.017 -0.062 Ecl JSat -0.021 -0.076 Met Hon -0.022 -0.077 Aen BC -0.026 -0.091 Arg Joh -0.030 -0.106 Arg Stil -0.039 -0.138 Ep PSat -0.040 -0.141 DRN Hon -0.042 -0.149 Georg Ep -0.045 -0.162 Aen HE -0.047 -0.169 DRN Ars -0.052 -0.184 Met Gild -0.055 -0.197 Ecl Mos -0.058 -0.206 Pun Stil -0.059 -0.212 Met Astr -0.064 -0.228 Ecl HSat -0.066 -0.237 Ach JSat -0.068 -0.242 BC Theb -0.075 -0.268 Ach HE -0.076 -0.273 Astr Gild -0.077 -0.276 HE Mos -0.079 -0.283 JSat Gild -0.081 -0.288 Theb Gild -0.081 -0.289 Met Joh -0.085 -0.302 Ecl HE -0.089 -0.318 Ecl BC -0.089 -0.319 Astr HE -0.090 -0.321 Ep Stil -0.090 -0.323 DRN Ach -0.091 -0.324 DRN Pun -0.092 -0.329 JSat Mos -0.094 -0.334 Ecl Joh -0.098 -0.351 Ars BC -0.100 -0.355 DRN Joh -0.101 -0.361 HSat Ach -0.102 -0.364 JSat Joh -0.104 -0.372 DRN Rapt -0.111 -0.394 DRN Mos -0.111 -0.397 HE Hon -0.112 -0.400 Ars Met -0.112 -0.401 JSat Rapt -0.113 -0.405 Ep Astr -0.114 -0.406 Georg Ars -0.114 -0.408 Astr Arg -0.114 -0.409 JSat Stil -0.117 -0.418 Ep Hon -0.117 -0.418 Ep Mos -0.121 -0.433 Theb Joh -0.125 -0.446 Theb Mos -0.128 -0.459 Ars Mos -0.131 -0.467 Ep Rapt -0.133 -0.476 DRN Ecl -0.134 -0.480 Arg HE -0.137 -0.490 Astr JSat -0.139 -0.496 DRN Ep -0.139 -0.497 PSat Stil -0.139 -0.498 BC Mos -0.142 -0.506 Ecl Theb -0.145 -0.518 Pun HE -0.154 -0.550 DRN Met -0.164 -0.585 Georg PSat -0.164 -0.587 DRN Stil -0.170 -0.606 Ars Ilias -0.171 -0.609 Ecl Ep -0.171 -0.610 DRN HSat -0.183 -0.654 PSat Mos -0.185 -0.661 Ars Ach -0.187 -0.668 Ep Ach -0.188 -0.670 Astr Theb -0.189 -0.673 HSat Ilias -0.195 -0.697 Ars Pun -0.195 -0.697 BC JSat -0.203 -0.724 HSat Pun -0.205 -0.731 PSat Arg -0.211 -0.754 BC HE -0.215 -0.768 Ep Aen -0.216 -0.772 DRN HE -0.221 -0.789 PSat Pun -0.230 -0.823 HSat Astr -0.236 -0.842 Met JSat -0.236 -0.843 HSat HE -0.242 -0.866 HSat Gild -0.243 -0.866 Aen JSat -0.243 -0.867 HSat Aen -0.243 -0.869 HSat Stil -0.244 -0.870 Ep Ilias -0.245 -0.873 Met HE -0.246 -0.877 HSat Mos -0.253 -0.903 Ep Gild -0.253 -0.903 Ep Met -0.261 -0.931 Ars Rapt -0.272 -0.971 JSat HE -0.273 -0.973 PSat Ach -0.273 -0.976 Ep Arg -0.279 -0.995 DRN Arg -0.283 -1.009 Ep BC -0.288 -1.028 Arg Mos -0.290 -1.035 Theb HE -0.293 -1.045 DRN BC -0.297 -1.059 Pun JSat -0.298 -1.062 PSat BC -0.300 -1.071 Ep Joh -0.301 -1.075 Arg JSat -0.303 -1.083 Ars Arg -0.304 -1.084 PSat Hon -0.315 -1.126 Ecl PSat -0.316 -1.127 DRN PSat -0.316 -1.128 HSat Joh -0.322 -1.148 HSat Met -0.326 -1.163 HSat BC -0.326 -1.165 DRN JSat -0.330 -1.179 Ep HE -0.336 -1.198 Ep Pun -0.338 -1.208 Ecl Hon -0.341 -1.219 Ars Joh -0.348 -1.242 Ars Hon -0.351 -1.254 PSat Theb -0.354 -1.264 Aen Ars -0.356 -1.269 DRN Theb -0.363 -1.296 Ecl Ars -0.370 -1.320 HSat Hon -0.376 -1.342 Met PSat -0.379 -1.353 HSat Theb -0.387 -1.381 Ars Theb -0.390 -1.393 HSat Arg -0.404 -1.442 PSat Ilias -0.406 -1.448 Ecl Gild -0.422 -1.507 Theb JSat -0.434 -1.548 PSat Gild -0.444 -1.584 Ep Theb -0.451 -1.609 PSat Joh -0.453 -1.617 PSat HE -0.454 -1.621 Astr PSat -0.468 -1.669 DRN Gild -0.484 -1.726 HSat Rapt -0.485 -1.731 Aen PSat -0.537 -1.917 PSat Rapt -0.579 -2.065 Ars Gild -0.834 -2.977
Table 3.
Intensity of text reuse for 276 pairs of hexameter texts from the 1st century BCE to the 6th century CE, determined by comparing composite counts of high scoring results in Tesserae searches with expected counts based on text lengths. Reuse intensity is presented as both non-standardized and standardized residuals.
 Text Mean r Georg 0.279 Ilias 0.186 Aen 0.133 Ach 0.117 Stil 0.105 Rapt 0.088 Hon 0.086 Joh 0.078 Met 0.073 Mos 0.057 Pun 0.044 Gild 0.032 BC 0.006 Astr -0.006 Ecl -0.018 Arg -0.036 Theb -0.096 HE -0.102 JSat -0.120 Ars -0.146 DRN -0.151 Ep -0.153 HSat -0.187 PSat -0.270
Table 4.
Centrality scores for 24 hexameter texts from the 1st century BCE to the 6th century CE, determined by calculating for each text the mean text reuse intensity for all 23 pairs involving that text.
 Horace Vergil Statius Claudian Source Target r Source Target r Source Target r Source Target r Ep Ars 0.354 Georg Aen 1.280 Theb Ach 0.141 Hon Stil 0.716 HSat Ep 0.259 Ecl Georg 0.603 Hon Gild 0.634 HSat Ars 0.041 Ecl Aen 0.224 Gild Stil 0.575 Rapt Hon 0.461 Rapt Gild 0.404 Rapt Stil 0.324
Table 5.
Intensity of text reuse for pairs of hexameter texts written by the same author.
 Georg Aen Ilias Target r Target r Target r Aen 1.280 Ilias 0.594 Joh 0.663 Met 0.560 Pun 0.540 Gild 0.457 Rapt 0.538 Met 0.350 Pun 0.433 Pun 0.433 Rapt 0.326 Ach 0.396 Joh 0.355 Theb 0.299 Arg 0.389 BC 0.351 Ach 0.289 Theb 0.252 Astr 0.342 Joh 0.269 HE 0.223 Ilias 0.310 Arg 0.255 Stil 0.157 Arg 0.297 Mos 0.115 Rapt 0.152 Gild 0.268 Gild 0.092 Mos 0.120 Mos 0.260 Hon 0.064 JSat 0.045 Hon 0.232 Stil 0.030 Hon 0.036 Theb 0.186 Astr 0.011 HE 0.172 BC -0.026 Ach 0.160 HE -0.047 Stil 0.132 JSat -0.243 JSat -0.009 Ars -0.356 Ep -0.045 PSat -0.537 Ars -0.114 PSat -0.164
Table 6.
Intensity of text reuse for pairs of hexameter texts with Vergil’s Georgics, Vergil’s Aeneid, or the Ilias Latina as source text.
 Met BC Arg Theb Ach Pun Source r Source r Source r Source r Source r Source r Georg 0.560 Georg 0.351 Georg 0.297 Aen 0.299 Aen 0.289 Aen 0.540 Aen 0.350 Met 0.238 Aen 0.255 Georg 0.186 Arg 0.279 Georg 0.433 Ecl 0.323 Aen -0.026 BC 0.035 Arg 0.064 Georg 0.160 Ach 0.410 Ars -0.112 Ecl -0.089 Ecl -0.003 Met 0.037 Theb 0.141 Met 0.146 DRN -0.164 Ars -0.100 Met -0.006 BC -0.075 BC 0.097 BC 0.126 Ep -0.261 Ep -0.288 Ep -0.279 Ecl -0.145 Met 0.077 Ecl 0.093 HSat -0.326 DRN -0.297 DRN -0.283 DRN -0.363 Ecl 0.047 Arg 0.076 HSat -0.326 Ars -0.304 HSat -0.387 DRN -0.091 Theb 0.057 HSat -0.404 Ars -0.390 HSat -0.102 DRN -0.092 Ep -0.451 Ars -0.187 Ars -0.195 Ep -0.188 HSat -0.205 Ep -0.338
Table 7.
Intensity of text reuse for select pairs of hexameter texts with post-Vergilian epics as target text (Ovid’s Metamorphoses, Lucan’s Bellum Civile, Valerius Flaccus’ Argonautica, Statius’ Thebaid and Achilleid, and Silius Italicus’ Punica).
 Source Target C7 C8 C9 C10 C11 Cobs r DRN Ecl 911 193 28 1 0 129.9 -0.134 DRN HSat 2171 552 100 5 0 380.0 -0.183 DRN Georg 2643 894 169 8 0 571.8 0.230 DRN Ep 1414 358 70 4 0 256.7 -0.139 DRN Aen 11060 3350 790 92 0 2755.7 -0.015 DRN Ars 407 74 16 2 0 68.6 -0.052 DRN Met 12958 3726 827 100 0 3036.4 -0.164 DRN Astr 5380 1458 331 10 0 1030.2 0.023 DRN PSat 601 84 16 2 0 81.9 -0.316 DRN BC 8160 2318 464 22 0 1591.4 -0.297 DRN Ilias 1089 254 39 3 0 177.4 -0.017 DRN Arg 5904 1416 351 15 0 1102.3 -0.283 DRN Theb 9682 2443 520 44 0 1901.1 -0.363 DRN Ach 1117 260 42 3 0 183.4 -0.091 DRN Pun 12722 3892 907 105 0 3171.5 -0.092 DRN JSat 3685 955 188 5 0 645.5 -0.330 DRN HE 3353 883 132 4 0 548.0 -0.221 DRN Mos 377 86 14 1 0 61.2 -0.111 DRN Rapt 1028 235 49 2 0 173.4 -0.111 DRN Hon 565 124 27 1 0 93.6 -0.042 DRN Gild 387 65 9 0 0 45.8 -0.484 DRN Stil 1074 282 49 1 0 180.3 -0.170 DRN Joh 4842 1393 306 13 0 978.6 -0.101 Ecl HSat 201 33 6 0 0 25.0 -0.066 Ecl Georg 374 98 5 0 0 48.5 0.603 Ecl Ep 130 27 1 0 0 14.5 -0.171 Ecl Aen 1222 250 53 4 0 204.4 0.224 Ecl Ars 39 3 0 0 0 2.9 -0.370 Ecl Met 1482 325 55 12 0 288.5 0.323 Ecl Astr 501 93 24 1 0 79.9 0.307 Ecl PSat 56 7 0 0 0 4.8 -0.316 Ecl BC 801 146 23 2 0 114.3 -0.089 Ecl Ilias 138 19 1 0 0 13.2 0.224 Ecl Arg 693 116 13 1 0 85.1 -0.003 Ecl Theb 1106 175 29 1 0 138.1 -0.145 Ecl Ach 128 22 0 0 0 12.3 0.047 Ecl Pun 1312 261 45 7 0 222.9 0.093 Ecl JSat 413 83 9 0 0 51.4 -0.021 Ecl HE 330 56 5 0 0 36.5 -0.089 Ecl Mos 54 3 0 0 0 3.8 -0.058 Ecl Rapt 117 20 0 0 0 11.2 -0.009 Ecl Hon 51 5 0 0 0 4.1 -0.341 Ecl Gild 30 5 0 0 0 2.8 -0.422 Ecl Stil 137 26 0 0 0 13.7 0.094 Ecl Joh 524 99 5 0 0 57.3 -0.098 HSat Georg 583 176 20 0 0 93.0 0.037 HSat Ep 490 126 19 0 0 75.4 0.259 HSat Aen 2638 674 111 3 0 432.7 -0.243 HSat Ars 151 23 1 0 0 14.8 0.041 HSat Met 3107 780 112 7 0 509.6 -0.326 HSat Astr 1090 282 31 0 0 156.9 -0.236 HSat PSat 196 45 6 0 0 27.4 0.210 HSat BC 1983 424 58 6 0 304.8 -0.326 HSat Ilias 239 56 3 0 0 29.3 -0.195 HSat Arg 1419 329 31 1 0 192.6 -0.404 HSat Theb 2400 551 67 6 0 366.3 -0.387 HSat Ach 254 81 3 0 0 35.8 -0.102 HSat Pun 3151 839 108 13 0 559.1 -0.205 HSat JSat 1167 292 37 1 0 175.9 -0.007 HSat HE 769 190 19 0 0 105.8 -0.242 HSat Mos 87 20 1 0 0 10.5 -0.253 HSat Rapt 218 49 0 0 0 23.5 -0.485 HSat Hon 125 18 2 0 0 13.2 -0.376 HSat Gild 126 19 0 0 0 11.5 -0.243 HSat Stil 257 68 3 0 0 33.0 -0.244 HSat Joh 1052 287 30 0 0 154.9 -0.322 Georg Ep 413 114 6 0 0 55.3 -0.045 Georg Aen 4150 1276 275 42 4 1974.8 1.280 Georg Ars 140 16 1 0 0 12.6 -0.114 Georg Met 4460 1415 251 28 1 1228.6 0.560 Georg Astr 1578 473 75 1 0 278.1 0.342 Georg PSat 158 34 2 0 0 18.7 -0.164 Georg BC 2876 794 140 18 0 596.6 0.351 Georg Ilias 372 111 2 0 0 48.3 0.310 Georg Arg 2341 584 108 2 0 386.1 0.297 Georg Theb 3433 965 144 14 0 645.8 0.186 Georg Ach 428 83 3 0 0 46.2 0.160 Georg Pun 4728 1489 245 32 0 1052.0 0.433 Georg JSat 1104 329 31 1 0 174.6 -0.009 Georg HE 975 313 33 0 0 159.3 0.172 Georg Mos 159 28 2 0 0 17.4 0.260 Georg Rapt 379 126 15 0 0 65.1 0.538 Georg Hon 204 42 3 0 0 24.1 0.232 Georg Gild 162 39 1 0 0 19.1 0.268 Georg Stil 353 96 6 0 0 47.8 0.132 Georg Joh 1677 473 94 1 0 302.8 0.355 Ep Aen 1658 420 56 5 0 276.9 -0.216 Ep Ars 134 22 0 0 0 12.6 0.354 Ep Met 2052 545 67 5 0 338.6 -0.261 Ep Astr 860 205 15 0 0 110.4 -0.114 Ep PSat 116 25 1 0 0 13.3 -0.040 Ep BC 1437 278 46 1 0 197.2 -0.288 Ep Ilias 150 39 0 0 0 17.4 -0.245 Ep Arg 951 211 34 0 0 136.0 -0.279 Ep Theb 1496 372 32 2 0 214.0 -0.451 Ep Ach 188 43 0 0 0 20.5 -0.188 Ep Pun 1803 460 79 3 0 304.7 -0.338 Ep JSat 705 219 21 0 0 110.7 0.003 Ep HE 513 114 5 0 0 60.1 -0.336 Ep Mos 75 14 0 0 0 7.5 -0.121 Ep Rapt 150 41 3 0 0 20.8 -0.133 Ep Hon 94 19 1 0 0 10.7 -0.117 Ep Gild 57 17 0 0 0 7.1 -0.253 Ep Stil 199 47 2 0 0 24.0 -0.090 Ep Joh 742 164 19 0 0 98.4 -0.301 Aen Ars 539 85 15 1 0 71.3 -0.356 Aen Met 21610 6172 1364 250 7 7156.5 0.350 Aen Astr 6658 1763 437 35 0 1435.2 0.011 Aen PSat 735 131 21 0 0 92.6 -0.537 Aen BC 13863 3157 815 99 0 2942.2 -0.026 Aen Ilias 2361 670 112 10 0 460.9 0.594 Aen Arg 12214 3071 647 99 0 2660.2 0.255 Aen Theb 18667 4816 1166 190 3 5196.9 0.299 Aen Ach 2196 511 87 8 0 378.1 0.289 Aen Pun 26063 7011 1720 323 7 8415.5 0.540 Aen JSat 5113 1252 299 19 0 993.2 -0.243 Aen HE 4656 1236 255 19 0 919.3 -0.047 Aen Mos 623 114 28 3 0 108.2 0.115 Aen Rapt 1674 494 83 14 0 378.1 0.326 Aen Hon 910 182 41 2 0 146.7 0.064 Aen Gild 773 151 24 2 0 114.9 0.092 Aen Stil 1761 411 73 7 0 310.4 0.030 Aen Joh 9050 2482 550 59 0 1997.9 0.269 Ars Met 681 83 15 2 0 85.3 -0.112 Ars Astr 213 58 8 0 0 33.3 0.213 Ars PSat 37 4 0 0 0 3.0 0.008 Ars BC 398 56 10 1 0 51.7 -0.100 Ars Ilias 59 3 0 0 0 4.1 -0.171 Ars Arg 268 33 6 0 0 28.8 -0.304 Ars Theb 515 57 7 0 0 49.4 -0.390 Ars Ach 50 7 0 0 0 4.4 -0.187 Ars Pun 642 109 15 0 0 76.3 -0.195 Ars JSat 229 39 4 0 0 25.9 0.078 Ars HE 164 27 3 0 0 18.5 0.013 Ars Mos 24 1 0 0 0 1.6 -0.131 Ars Rapt 49 5 0 0 0 3.9 -0.272 Ars Hon 28 1 0 0 0 1.8 -0.351 Ars Gild 15 0 0 0 0 0.9 -0.834 Ars Stil 66 15 0 0 0 7.2 0.228 Ars Joh 207 29 2 0 0 20.4 -0.348 Met Astr 8646 2366 466 39 0 1738.9 -0.064 Met PSat 797 161 22 6 0 141.6 -0.379 Met BC 16737 3936 966 131 6 5000.8 0.238 Met Ilias 2497 662 74 14 1 681.8 0.719 Met Arg 12279 2899 622 75 1 2677.2 -0.006 Met Theb 19745 5002 1015 165 4 5220.1 0.037 Met Ach 2330 639 78 7 0 399.3 0.077 Met Pun 24950 6621 1564 284 5 7407.3 0.146 Met JSat 6383 1685 328 37 0 1305.5 -0.236 Met HE 5307 1420 235 20 0 984.3 -0.246 Met Mos 814 135 26 6 1 368.5 1.073 Met Rapt 1971 548 97 9 0 389.8 0.089 Met Hon 992 260 29 5 0 175.8 -0.022 Met Gild 775 192 23 3 0 129.5 -0.055 Met Stil 2073 597 84 10 0 400.0 0.017 Met Joh 9569 2604 515 29 0 1831.5 -0.085 Astr PSat 305 59 2 0 0 32.8 -0.468 Astr BC 6009 1465 239 21 0 1045.0 0.048 Astr Ilias 732 161 17 0 0 95.2 0.125 Astr Arg 3871 867 171 3 0 606.8 -0.114 Astr Theb 6224 1506 257 16 0 1053.1 -0.189 Astr Ach 760 166 17 0 0 97.9 0.047 Astr Pun 8545 2134 439 30 0 1597.4 -0.013 Astr JSat 2253 554 91 3 0 363.6 -0.139 Astr HE 1916 533 61 0 0 290.7 -0.090 Astr Mos 314 59 5 0 0 36.3 0.130 Astr Rapt 657 151 17 0 0 88.6 -0.016 Astr Hon 407 70 8 0 0 47.1 0.036 Astr Gild 304 47 4 0 0 32.0 -0.077 Astr Stil 731 159 21 0 0 98.7 -0.007 Astr Joh 3184 870 154 4 0 557.3 0.101 PSat BC 564 79 12 1 0 68.4 -0.300 PSat Ilias 63 7 0 0 0 5.2 -0.406 PSat Arg 434 63 12 0 0 51.1 -0.211 PSat Theb 698 113 11 1 0 82.7 -0.354 PSat Ach 60 14 0 0 0 6.6 -0.273 PSat Pun 891 150 28 1 0 119.1 -0.230 PSat JSat 343 79 7 0 0 44.4 0.137 PSat HE 197 33 0 0 0 18.7 -0.454 PSat Mos 31 3 0 0 0 2.5 -0.185 PSat Rapt 62 5 0 0 0 4.7 -0.579 PSat Hon 30 6 0 0 0 3.1 -0.315 PSat Gild 28 2 0 0 0 2.1 -0.444 PSat Stil 77 16 0 0 0 8.0 -0.139 PSat Joh 250 46 5 0 0 29.7 -0.453 BC Ilias 1387 294 25 4 0 195.8 0.028 BC Arg 8571 1837 391 48 0 1597.4 0.035 BC Theb 14282 3067 631 85 0 2674.1 -0.075 BC Ach 1607 303 48 4 0 233.5 0.097 BC Pun 18677 4366 957 147 1 4161.6 0.126 BC JSat 4549 1061 211 10 0 773.3 -0.203 BC HE 3518 851 138 8 0 581.4 -0.215 BC Mos 463 76 19 0 0 62.7 -0.142 BC Rapt 1458 334 47 9 0 262.3 0.250 BC Hon 920 160 22 4 0 135.9 0.278 BC Gild 580 133 23 4 0 111.3 0.351 BC Stil 1645 357 59 9 0 290.2 0.253 BC Joh 7466 1754 336 23 0 1303.1 0.132 Ilias Arg 1150 237 30 1 0 155.5 0.389 Ilias Theb 1818 380 45 3 0 253.6 0.252 Ilias Ach 234 36 0 0 0 21.5 0.064 Ilias Pun 2581 582 70 6 0 386.6 0.396 Ilias JSat 465 107 17 0 0 67.7 0.279 Ilias HE 484 106 10 0 0 61.6 0.433 Ilias Mos 50 12 0 0 0 5.6 0.076 Ilias Rapt 158 32 0 0 0 16.3 0.045 Ilias Hon 88 10 0 0 0 7.3 -0.303 Ilias Gild 77 18 0 0 0 8.5 0.223 Ilias Stil 177 35 0 0 0 18.0 -0.137 Ilias Joh 977 212 29 3 0 151.6 0.120 Arg Theb 11371 2503 535 45 0 2032.8 -0.290 Arg Ach 1304 297 31 2 0 185.1 0.152 Arg Pun 14178 3236 650 68 0 2618.2 0.067 Arg JSat 2895 682 118 4 0 462.5 0.036 Arg HE 2731 614 121 0 0 415.7 0.024 Arg Mos 357 50 4 0 0 35.7 0.457 Arg Rapt 1099 224 31 0 0 144.4 0.097 Arg Hon 538 96 11 1 0 69.7 0.157 Arg Gild 430 82 14 0 0 57.1 -0.039 Arg Stil 1120 226 22 1 0 143.3 0.663 Arg Joh 4622 1069 171 9 0 733.0 -0.030 Theb Ach 2283 517 58 2 0 317.8 0.141 Theb Pun 22806 5275 1076 165 2 5062.7 0.057 Theb JSat 4733 1232 170 13 0 800.1 -0.434 Theb HE 4047 989 184 10 0 701.3 -0.293 Theb Mos 612 95 20 1 0 82.8 -0.128 Theb Rapt 1844 500 53 5 0 302.6 0.128 Theb Hon 972 200 28 2 0 141.3 0.052 Theb Gild 727 130 17 1 0 94.2 -0.081 Theb Stil 1836 395 60 6 0 291.8 -0.006 Theb Joh 7759 1900 303 22 0 1313.5 -0.125 Ach Pun 2631 614 91 7 0 423.9 0.410 Ach JSat 559 115 10 0 0 67.9 -0.068 Ach HE 448 87 6 0 0 51.3 -0.076 Ach Mos 68 10 0 0 0 6.1 0.104 Ach Rapt 207 45 2 0 0 24.0 0.426 Ach Hon 112 16 0 0 0 10.0 0.238 Ach Gild 68 13 1 0 0 7.8 0.263 Ach Stil 226 45 1 0 0 24.1 0.331 Ach Joh 881 194 27 1 0 127.4 0.375 Pun JSat 6378 1625 296 26 0 1190.6 -0.298 Pun HE 5321 1403 262 26 0 1046.0 -0.154 Pun Mos 755 133 21 5 0 125.6 0.028 Pun Rapt 2074 573 76 11 0 392.9 0.128 Pun Hon 1168 247 31 5 0 185.0 0.060 Pun Gild 957 163 31 2 0 135.1 0.018 Pun Stil 2204 544 73 6 0 359.4 -0.059 Pun Joh 11414 2598 525 43 0 2034.1 0.051 JSat HE 1386 361 47 1 0 213.9 -0.273 JSat Mos 191 43 5 0 0 25.6 -0.094 JSat Rapt 464 131 15 0 0 71.1 -0.113 JSat Hon 300 67 10 0 0 42.3 0.053 JSat Gild 222 42 6 0 0 28.2 -0.081 JSat Stil 592 143 12 0 0 78.1 -0.117 JSat Joh 2486 594 106 3 0 401.0 -0.104 HE Mos 165 35 2 0 0 19.3 -0.079 HE Rapt 430 102 11 0 0 58.6 -0.010 HE Hon 244 43 3 0 0 26.7 -0.112 HE Gild 202 44 3 0 0 24.5 0.075 HE Stil 470 117 11 0 0 64.3 -0.016 HE Joh 2321 639 105 8 0 432.1 0.267 Mos Rapt 62 5 1 0 0 5.7 0.153 Mos Hon 44 5 0 0 0 3.6 0.395 Mos Gild 34 0 0 0 0 2.0 0.042 Mos Stil 77 11 0 0 0 6.9 0.247 Mos Joh 263 43 10 0 0 34.8 0.324 Rapt Hon 104 27 0 0 0 12.0 0.243 Rapt Gild 100 13 0 0 0 8.7 0.461 Rapt Stil 193 53 0 0 0 23.0 0.404 Rapt Joh 845 198 21 0 0 114.0 0.302 Hon Gild 59 8 0 0 0 5.2 0.634 Hon Stil 133 38 0 0 0 16.2 0.716 Hon Joh 472 89 11 0 0 58.1 0.372 Gild Stil 100 21 0 0 0 10.5 0.575 Gild Joh 402 74 6 0 0 45.7 0.427 Stil Joh 862 182 24 0 0 114.4 0.199
Table 8.
Results of Tesserae searches of 276 pairs of hexameter texts from the 1st century BCE to the 6th century CE, sorted chronologically by source text. Results include: raw counts of score 7, 8, 9, 10, and 11; composite counts calculated from the raw counts using a combination of linear regressions and principal component analysis; and text reuse intensity, determined by comparing the composite counts with expected counts based on a text lengths.
 Source Target C7 C8 C9 C10 C11 Cobs r DRN Ecl 911 193 28 1 0 129.9 -0.134 DRN HSat 2171 552 100 5 0 380.0 -0.183 Ecl HSat 201 33 6 0 0 25.0 -0.066 DRN Georg 2643 894 169 8 0 571.8 0.230 Ecl Georg 374 98 5 0 0 48.5 0.603 HSat Georg 583 176 20 0 0 93.0 0.037 DRN Ep 1414 358 70 4 0 256.7 -0.139 Ecl Ep 130 27 1 0 0 14.5 -0.171 HSat Ep 490 126 19 0 0 75.4 0.259 Georg Ep 413 114 6 0 0 55.3 -0.045 DRN Aen 11060 3350 790 92 0 2755.7 -0.015 Ecl Aen 1222 250 53 4 0 204.4 0.224 HSat Aen 2638 674 111 3 0 432.7 -0.243 Georg Aen 4150 1276 275 42 4 1974.8 1.280 Ep Aen 1658 420 56 5 0 276.9 -0.216 DRN Ars 407 74 16 2 0 68.6 -0.052 Ecl Ars 39 3 0 0 0 2.9 -0.370 HSat Ars 151 23 1 0 0 14.8 0.041 Georg Ars 140 16 1 0 0 12.6 -0.114 Ep Ars 134 22 0 0 0 12.6 0.354 Aen Ars 539 85 15 1 0 71.3 -0.356 DRN Met 12958 3726 827 100 0 3036.4 -0.164 Ecl Met 1482 325 55 12 0 288.5 0.323 HSat Met 3107 780 112 7 0 509.6 -0.326 Georg Met 4460 1415 251 28 1 1228.6 0.560 Ep Met 2052 545 67 5 0 338.6 -0.261 Aen Met 21610 6172 1364 250 7 7156.5 0.350 Ars Met 681 83 15 2 0 85.3 -0.112 DRN Astr 5380 1458 331 10 0 1030.2 0.023 Ecl Astr 501 93 24 1 0 79.9 0.307 HSat Astr 1090 282 31 0 0 156.9 -0.236 Georg Astr 1578 473 75 1 0 278.1 0.342 Ep Astr 860 205 15 0 0 110.4 -0.114 Aen Astr 6658 1763 437 35 0 1435.2 0.011 Ars Astr 213 58 8 0 0 33.3 0.213 Met Astr 8646 2366 466 39 0 1738.9 -0.064 DRN PSat 601 84 16 2 0 81.9 -0.316 Ecl PSat 56 7 0 0 0 4.8 -0.316 HSat PSat 196 45 6 0 0 27.4 0.210 Georg PSat 158 34 2 0 0 18.7 -0.164 Ep PSat 116 25 1 0 0 13.3 -0.040 Aen PSat 735 131 21 0 0 92.6 -0.537 Ars PSat 37 4 0 0 0 3.0 0.008 Met PSat 797 161 22 6 0 141.6 -0.379 Astr PSat 305 59 2 0 0 32.8 -0.468 DRN BC 8160 2318 464 22 0 1591.4 -0.297 Ecl BC 801 146 23 2 0 114.3 -0.089 HSat BC 1983 424 58 6 0 304.8 -0.326 Georg BC 2876 794 140 18 0 596.6 0.351 Ep BC 1437 278 46 1 0 197.2 -0.288 Aen BC 13863 3157 815 99 0 2942.2 -0.026 Ars BC 398 56 10 1 0 51.7 -0.100 Met BC 16737 3936 966 131 6 5000.8 0.238 Astr BC 6009 1465 239 21 0 1045.0 0.048 PSat BC 564 79 12 1 0 68.4 -0.300 DRN Ilias 1089 254 39 3 0 177.4 -0.017 Ecl Ilias 138 19 1 0 0 13.2 0.224 HSat Ilias 239 56 3 0 0 29.3 -0.195 Georg Ilias 372 111 2 0 0 48.3 0.310 Ep Ilias 150 39 0 0 0 17.4 -0.245 Aen Ilias 2361 670 112 10 0 460.9 0.594 Ars Ilias 59 3 0 0 0 4.1 -0.171 Met Ilias 2497 662 74 14 1 681.8 0.719 Astr Ilias 732 161 17 0 0 95.2 0.125 PSat Ilias 63 7 0 0 0 5.2 -0.406 BC Ilias 1387 294 25 4 0 195.8 0.028 DRN Arg 5904 1416 351 15 0 1102.3 -0.283 Ecl Arg 693 116 13 1 0 85.1 -0.003 HSat Arg 1419 329 31 1 0 192.6 -0.404 Georg Arg 2341 584 108 2 0 386.1 0.297 Ep Arg 951 211 34 0 0 136.0 -0.279 Aen Arg 12214 3071 647 99 0 2660.2 0.255 Ars Arg 268 33 6 0 0 28.8 -0.304 Met Arg 12279 2899 622 75 1 2677.2 -0.006 Astr Arg 3871 867 171 3 0 606.8 -0.114 PSat Arg 434 63 12 0 0 51.1 -0.211 BC Arg 8571 1837 391 48 0 1597.4 0.035 Ilias Arg 1150 237 30 1 0 155.5 0.389 DRN Theb 9682 2443 520 44 0 1901.1 -0.363 Ecl Theb 1106 175 29 1 0 138.1 -0.145 HSat Theb 2400 551 67 6 0 366.3 -0.387 Georg Theb 3433 965 144 14 0 645.8 0.186 Ep Theb 1496 372 32 2 0 214.0 -0.451 Aen Theb 18667 4816 1166 190 3 5196.9 0.299 Ars Theb 515 57 7 0 0 49.4 -0.390 Met Theb 19745 5002 1015 165 4 5220.1 0.037 Astr Theb 6224 1506 257 16 0 1053.1 -0.189 PSat Theb 698 113 11 1 0 82.7 -0.354 BC Theb 14282 3067 631 85 0 2674.1 -0.075 Ilias Theb 1818 380 45 3 0 253.6 0.252 Arg Theb 11371 2503 535 45 0 2032.8 0.064 DRN Ach 1117 260 42 3 0 183.4 -0.091 Ecl Ach 128 22 0 0 0 12.3 0.047 HSat Ach 254 81 3 0 0 35.8 -0.102 Georg Ach 428 83 3 0 0 46.2 0.160 Ep Ach 188 43 0 0 0 20.5 -0.188 Aen Ach 2196 511 87 8 0 378.1 0.289 Ars Ach 50 7 0 0 0 4.4 -0.187 Met Ach 2330 639 78 7 0 399.3 0.077 Astr Ach 760 166 17 0 0 97.9 0.047 PSat Ach 60 14 0 0 0 6.6 -0.273 BC Ach 1607 303 48 4 0 233.5 0.097 Ilias Ach 234 36 0 0 0 21.5 0.396 Arg Ach 1304 297 31 2 0 185.1 0.279 Theb Ach 2283 517 58 2 0 317.8 0.141 DRN Pun 12722 3892 907 105 0 3171.5 -0.092 Ecl Pun 1312 261 45 7 0 222.9 0.093 HSat Pun 3151 839 108 13 0 559.1 -0.205 Georg Pun 4728 1489 245 32 0 1052.0 0.433 Ep Pun 1803 460 79 3 0 304.7 -0.338 Aen Pun 26063 7011 1720 323 7 8415.5 0.540 Ars Pun 642 109 15 0 0 76.3 -0.195 Met Pun 24950 6621 1564 284 5 7407.3 0.146 Astr Pun 8545 2134 439 30 0 1597.4 -0.013 PSat Pun 891 150 28 1 0 119.1 -0.230 BC Pun 18677 4366 957 147 1 4161.6 0.126 Ilias Pun 2581 582 70 6 0 386.6 0.433 Arg Pun 14178 3236 650 68 0 2618.2 0.076 Theb Pun 22806 5275 1076 165 2 5062.7 0.057 Ach Pun 2631 614 91 7 0 423.9 0.410 DRN JSat 3685 955 188 5 0 645.5 -0.330 Ecl JSat 413 83 9 0 0 51.4 -0.021 HSat JSat 1167 292 37 1 0 175.9 -0.007 Georg JSat 1104 329 31 1 0 174.6 -0.009 Ep JSat 705 219 21 0 0 110.7 0.003 Aen JSat 5113 1252 299 19 0 993.2 -0.243 Ars JSat 229 39 4 0 0 25.9 0.078 Met JSat 6383 1685 328 37 0 1305.5 -0.236 Astr JSat 2253 554 91 3 0 363.6 -0.139 PSat JSat 343 79 7 0 0 44.4 0.137 BC JSat 4549 1061 211 10 0 773.3 -0.203 Ilias JSat 465 107 17 0 0 67.7 0.045 Arg JSat 2895 682 118 4 0 462.5 -0.303 Theb JSat 4733 1232 170 13 0 800.1 -0.434 Ach JSat 559 115 10 0 0 67.9 -0.068 Pun JSat 6378 1625 296 26 0 1190.6 -0.298 DRN HE 3353 883 132 4 0 548.0 -0.221 Ecl HE 330 56 5 0 0 36.5 -0.089 HSat HE 769 190 19 0 0 105.8 -0.242 Georg HE 975 313 33 0 0 159.3 0.172 Ep HE 513 114 5 0 0 60.1 -0.336 Aen HE 4656 1236 255 19 0 919.3 -0.047 Ars HE 164 27 3 0 0 18.5 0.013 Met HE 5307 1420 235 20 0 984.3 -0.246 Astr HE 1916 533 61 0 0 290.7 -0.090 PSat HE 197 33 0 0 0 18.7 -0.454 BC HE 3518 851 138 8 0 581.4 -0.215 Ilias HE 484 106 10 0 0 61.6 0.223 Arg HE 2731 614 121 0 0 415.7 -0.137 Theb HE 4047 989 184 10 0 701.3 -0.293 Ach HE 448 87 6 0 0 51.3 -0.076 Pun HE 5321 1403 262 26 0 1046.0 -0.154 JSat HE 1386 361 47 1 0 213.9 -0.273 DRN Mos 377 86 14 1 0 61.2 -0.111 Ecl Mos 54 3 0 0 0 3.8 -0.058 HSat Mos 87 20 1 0 0 10.5 -0.253 Georg Mos 159 28 2 0 0 17.4 0.260 Ep Mos 75 14 0 0 0 7.5 -0.121 Aen Mos 623 114 28 3 0 108.2 0.115 Ars Mos 24 1 0 0 0 1.6 -0.131 Met Mos 814 135 26 6 1 368.5 1.073 Astr Mos 314 59 5 0 0 36.3 0.130 PSat Mos 31 3 0 0 0 2.5 -0.185 BC Mos 463 76 19 0 0 62.7 -0.142 Ilias Mos 50 12 0 0 0 5.6 0.120 Arg Mos 357 50 4 0 0 35.7 -0.290 Theb Mos 612 95 20 1 0 82.8 -0.128 Ach Mos 68 10 0 0 0 6.1 0.104 Pun Mos 755 133 21 5 0 125.6 0.028 JSat Mos 191 43 5 0 0 25.6 -0.094 HE Mos 165 35 2 0 0 19.3 -0.079 DRN Rapt 1028 235 49 2 0 173.4 -0.111 Ecl Rapt 117 20 0 0 0 11.2 -0.009 HSat Rapt 218 49 0 0 0 23.5 -0.485 Georg Rapt 379 126 15 0 0 65.1 0.538 Ep Rapt 150 41 3 0 0 20.8 -0.133 Aen Rapt 1674 494 83 14 0 378.1 0.326 Ars Rapt 49 5 0 0 0 3.9 -0.272 Met Rapt 1971 548 97 9 0 389.8 0.089 Astr Rapt 657 151 17 0 0 88.6 -0.016 PSat Rapt 62 5 0 0 0 4.7 -0.579 BC Rapt 1458 334 47 9 0 262.3 0.250 Ilias Rapt 158 32 0 0 0 16.3 0.152 Arg Rapt 1099 224 31 0 0 144.4 0.067 Theb Rapt 1844 500 53 5 0 302.6 0.128 Ach Rapt 207 45 2 0 0 24.0 0.426 Pun Rapt 2074 573 76 11 0 392.9 0.128 JSat Rapt 464 131 15 0 0 71.1 -0.113 HE Rapt 430 102 11 0 0 58.6 -0.010 Mos Rapt 62 5 1 0 0 5.7 0.153 DRN Hon 565 124 27 1 0 93.6 -0.042 Ecl Hon 51 5 0 0 0 4.1 -0.341 HSat Hon 125 18 2 0 0 13.2 -0.376 Georg Hon 204 42 3 0 0 24.1 0.232 Ep Hon 94 19 1 0 0 10.7 -0.117 Aen Hon 910 182 41 2 0 146.7 0.064 Ars Hon 28 1 0 0 0 1.8 -0.351 Met Hon 992 260 29 5 0 175.8 -0.022 Astr Hon 407 70 8 0 0 47.1 0.036 PSat Hon 30 6 0 0 0 3.1 -0.315 BC Hon 920 160 22 4 0 135.9 0.278 Ilias Hon 88 10 0 0 0 7.3 0.036 Arg Hon 538 96 11 1 0 69.7 0.024 Theb Hon 972 200 28 2 0 141.3 0.052 Ach Hon 112 16 0 0 0 10.0 0.238 Pun Hon 1168 247 31 5 0 185.0 0.060 JSat Hon 300 67 10 0 0 42.3 0.053 HE Hon 244 43 3 0 0 26.7 -0.112 Mos Hon 44 5 0 0 0 3.6 0.395 Rapt Hon 104 27 0 0 0 12.0 0.461 DRN Gild 387 65 9 0 0 45.8 -0.484 Ecl Gild 30 5 0 0 0 2.8 -0.422 HSat Gild 126 19 0 0 0 11.5 -0.243 Georg Gild 162 39 1 0 0 19.1 0.268 Ep Gild 57 17 0 0 0 7.1 -0.253 Aen Gild 773 151 24 2 0 114.9 0.092 Ars Gild 15 0 0 0 0 0.9 -0.834 Met Gild 775 192 23 3 0 129.5 -0.055 Astr Gild 304 47 4 0 0 32.0 -0.077 PSat Gild 28 2 0 0 0 2.1 -0.444 BC Gild 580 133 23 4 0 111.3 0.351 Ilias Gild 77 18 0 0 0 8.5 0.457 Arg Gild 430 82 14 0 0 57.1 0.097 Theb Gild 727 130 17 1 0 94.2 -0.081 Ach Gild 68 13 1 0 0 7.8 0.263 Pun Gild 957 163 31 2 0 135.1 0.018 JSat Gild 222 42 6 0 0 28.2 -0.081 HE Gild 202 44 3 0 0 24.5 0.075 Mos Gild 34 0 0 0 0 2.0 0.042 Rapt Gild 100 13 0 0 0 8.7 0.404 Hon Gild 59 8 0 0 0 5.2 0.634 DRN Stil 1074 282 49 1 0 180.3 -0.170 Ecl Stil 137 26 0 0 0 13.7 0.094 HSat Stil 257 68 3 0 0 33.0 -0.244 Georg Stil 353 96 6 0 0 47.8 0.132 Ep Stil 199 47 2 0 0 24.0 -0.090 Aen Stil 1761 411 73 7 0 310.4 0.030 Ars Stil 66 15 0 0 0 7.2 0.228 Met Stil 2073 597 84 10 0 400.0 0.017 Astr Stil 731 159 21 0 0 98.7 -0.007 PSat Stil 77 16 0 0 0 8.0 -0.139 BC Stil 1645 357 59 9 0 290.2 0.253 Ilias Stil 177 35 0 0 0 18.0 0.157 Arg Stil 1120 226 22 1 0 143.3 -0.039 Theb Stil 1836 395 60 6 0 291.8 -0.006 Ach Stil 226 45 1 0 0 24.1 0.331 Pun Stil 2204 544 73 6 0 359.4 -0.059 JSat Stil 592 143 12 0 0 78.1 -0.117 HE Stil 470 117 11 0 0 64.3 -0.016 Mos Stil 77 11 0 0 0 6.9 0.247 Rapt Stil 193 53 0 0 0 23.0 0.324 Hon Stil 133 38 0 0 0 16.2 0.716 Gild Stil 100 21 0 0 0 10.5 0.575 DRN Joh 4842 1393 306 13 0 978.6 -0.101 Ecl Joh 524 99 5 0 0 57.3 -0.098 HSat Joh 1052 287 30 0 0 154.9 -0.322 Georg Joh 1677 473 94 1 0 302.8 0.355 Ep Joh 742 164 19 0 0 98.4 -0.301 Aen Joh 9050 2482 550 59 0 1997.9 0.269 Ars Joh 207 29 2 0 0 20.4 -0.348 Met Joh 9569 2604 515 29 0 1831.5 -0.085 Astr Joh 3184 870 154 4 0 557.3 0.101 PSat Joh 250 46 5 0 0 29.7 -0.453 BC Joh 7466 1754 336 23 0 1303.1 0.132 Ilias Joh 977 212 29 3 0 151.6 0.663 Arg Joh 4622 1069 171 9 0 733.0 -0.030 Theb Joh 7759 1900 303 22 0 1313.5 -0.125 Ach Joh 881 194 27 1 0 127.4 0.375 Pun Joh 11414 2598 525 43 0 2034.1 0.051 JSat Joh 2486 594 106 3 0 401.0 -0.104 HE Joh 2321 639 105 8 0 432.1 0.267 Mos Joh 263 43 10 0 0 34.8 0.243 Rapt Joh 845 198 21 0 0 114.0 0.302 Hon Joh 472 89 11 0 0 58.1 0.372 Gild Joh 402 74 6 0 0 45.7 0.427 Stil Joh 862 182 24 0 0 114.4 0.199
Table 9.
Results of Tesserae searches of 276 pairs of hexameter texts from the 1st century BCE to the 6th century CE, sorted chronologically by target text. Results include: raw counts of score 7, 8, 9, 10, and 11; composite counts calculated from the raw counts using a combination of linear regressions and principal component analysis; and text reuse intensity, determined by comparing the composite counts with expected counts based on a text lengths.

Notes

[1]  See, for example, [Hutchinson 2013], [Farrell 2005], and [Hinds 1998] for points of entry to the study of intertextuality in Latin literature.
[2]  All translations are by the authors.
[3]  Because Latin is a highly inflected language, the same lexeme may occur in many different inflected forms. For example, percutio may appear as percussus (“struck”), percutimus (“we strike”), percusserant (“they had struck”), etc. Traditional literary interpretation may privilege specific morphological forms, such as the opening words of Vergil’s Aeneid (arma uirumque), which are frequently adapted by later poets, but more often the various inflected forms of a lexeme may be considered to be the same. Tesserae converts all inflected forms to a single lemma (e.g., percussus and percussum are treated as percutio) and so does not permit analysis of individual inflected forms.
[4]  See section 2.b for discussion of the scoring system.
[5]  Recent commentaries (such as [Steiniger 2005], [Micozzi 2007], and [Parkes 2012]) note the verbal parallel with Aeneid 7.550, but do not offer a literary interpretation of the link. Their reticence is symptomatic of the scholarly tendency to privilege certain allusions (here, Aeneid 9.197) over others in interpretation. The impartial automatic searches of Tesserae encourage an interpretive style that is both less hierarchical and less committed to authorial intention.
[6]  The dates of texts mostly follow those found in Brill’s New Pauly, and depart in some cases from the dates used by the Tesserae to assign source and target text status for each pair (http://tesserae.caset.buffalo.edu/blog/authors−and−text−dates/). Where necessary, we manually corrected for the switched source and target. Some dates are uncertain; see, e.g., [Zissos 2008, xiv–xvii] on Valerius Flaccus’ Argonautica, or [Gruzelier 1993, xviii–xix] on Claudian’s De Raptu Proserpinae. Alternative datings would affect our results in some cases, since the calculation of the variable cexp depends on which text in a pair is considered the source and which the target. But the overall effect of any plausible change in dating would be small.
[7]  The Tesserae repository is extensive but not complete. Relevant hexameter texts unavailable for the study at this writing include, for example, Ennius’ Annales, the Appendix Vergiliana, the Eclogues of Calpurnius Siculus, and the various Latin versions of Aratus’ Phaenomena.
[8]  We included Claudian’s De Raptu Proserpinae because it is an important text and because its pentameter preface is short compared to the text as a whole (69 out of 6991 words), and therefore unlikely to noticeably affect our results.
[9]  Ausonius’ Precationes, Ordo Urbium Nobilium, and Cento Nuptialis (see section 3.c.iv), and Claudian’s In Consulatum Olybrii et Probini.
[10]  False lemma matches also sometimes occur, such as Vergil, Georgics 4.308 ossibus umor ~ Statius, Thebaid 4.698 oraumor. Here ossibus (“bones”) and ora (“faces”) are inflected forms of two different lexemes, both of which share the lemma os. Since such false matches occur infrequently, we did not expect them to affect the results significantly.
[11]  These parameters are explained at http://tess−dev.caset.buffalo.edu/html/help_advanced.php.
[12]  The regressions yielded the following formulae:
• $$C_{9} = - 21.191 + 0.057C_{7}$$
• $$C_{9} = - 17.943 + 0.225C_{8}$$
• $$C_{9} = 36.958 + 6.168C_{10}$$
• $$C_{9} = 84.259 + 212.062C_{11}$$
We omitted the intercepts, which provide no useful information, and thus obtained a formula for a composite count: $$C_{regr} = 0.057C_{7} + 0.225C_{8} + C_{9} + 6.168C_{10} + 212.062C_{11}$$
[13] The first principal component had weights $$0.458{\widetilde{C}}_{7} + 0.462{\widetilde{C}}_{8} + 0.465{\widetilde{C}}_{9} + 0.463{\widetilde{C}}_{10} + 0.383{\widetilde{C}}_{11}$$ This led to, in original scale, the composite count (which accounts for 90.1% of the total variability): $$C_{pca} = 0.458\frac{C_{7}}{4341} + 0.462\frac{C_{8}}{1118} + 0.465\frac{C_{9}}{253} + 0.463\frac{C_{10}}{39} + 0.383\frac{C_{11}}{1} = 10^{- 3}\left( 0.106C_{7} + 0.413C_{8} + 1.839C_{9} + 11.775C_{10} + 447.617C_{11} \right)$$ Further rescaling it such that the weight for C9 became 1, we obtained: $$C_{pca} = 0.057C_{7} + 0.225C_{8} + C_{9} + 6.404C_{10} + 243.426C_{11}$$
[14]  This model was the best of several considered, namely:
• $$c_{exp}\ w_{s} + w_{t}$$
• $$c_{exp}\ w_{s} + w_{t} + w_{s} \times w_{t}$$
• $$c_{exp}\ w_{s} + w_{t} + {w_{s}}^{2} + {w_{t}}^{2} + w_{s} \times w_{t}$$
• $$c_{exp}\ w_{st},\ \text{where}\ w_{st} = w_{s} + w_{t}\ \text{(treated as a single variable)}$$
[15]  The r values are also sorted chronologically by source and target in Table 8 and 9. Standardized residuals have been adjusted by the standard deviation of the entire set in order to detect statistical outliers. Standardized residuals greater than |2| are normally considered unusual; standardized residual greater than |3| are normally considered statistical outliers.
[16]  Author (HonStil, Hon Gild, EclGeorg,GildStil, section 3.b), genre (PSat - Rapt, 3.b), multiple reuse (MetIlias, IliasJoh, Aen Ilias, Georg Met, Georg Rapt, 3.c.i), and the influence of Vergil (Aen Pun, 3.c.ii).
[17]  Jockers observes, “the strength of the author signals in this experiment in fact trumps the signals of individual texts — something intuition does not prepare us for. The classifier [program] is more likely to identify the author of a given text segment than it is to correctly assign that same text segment to its novel of origin.”  [Jockers 2013, 93]
[18]  The didactic genre comprised: DRN, Georg, and Astr. The epic/panegyric genre comprised: Aen, Met, BC, Ilias, Arg, Theb, Ach, Pun, Rapt, Hon, Gild, Stil, and Joh. The satiric genre comprised: HSat, PSat, and JSat. This partitioning excludes five texts (Ecl, Ep, Ars, HE, Mos) that do not fit into any of the three genres. Including Horace’s Epistles and Ars Poetica in the satiric genre would not alter our conclusions: in fact, the lowest r value in our data set would then comprise an epic/panegyric–satric pair, ArsGild (r = −0.834).
[19]  The exceptions were slight: Ilias – JSat (r = 0.053) and JSatHon (r = 0.045).
[20]  The average Cobs value for the Aeneid paired with all subsequent target texts is 1876.6, compared to 284.6 for the Georgics.
[21]  Its influence grew later on: the text is quoted in the late antique commentary on the Thebaid ascribed to Lactantius Placidus, and became popular in the Middle Ages [Curtius 1953, 49–51].
[22]  This is consistent with scholarly observation; see New Pauly s.v. Ilias Latina [Courtney 2016].
[23]  For example, see the discussion of the “many mouths” topos [Gowers 2005].
[25]  Given the low residual, it is remarkable that Tesserae searches reported in [Coffee et al. 2012] identified 25% more interpretively significant instances of verbal reuse in the pair AeneidBellum Civile 1 than the standard philological commentaries. Similar studies for pairs with more intense text reuse (e.g., Aeneid Metamorphoses) would presumably be even more successful.
[26]  E.g.,

Compared with other writers of Latin epic, [Silius] tends to eschew signposting his intertexts by the technique of “quotation”, that is, by repeating complete phrases or other word collocations from earlier poems. He prefers to signal the intertextual connection by alternative means, in particular, by coincidence of situation and detail rather than wording and, occasionally, by more explicit hints.  [Wilson 2004, 225]

[27]  Parkes on the Achilleid and Argonautica is an exception [Parkes 2009]. For the Thebaid and Argonautica, see [Lovatt 2015], with bibliography.
[28]  The relative dating of these two epics is uncertain. This study has treated the Achilleid as the source, but the two epics may well have been composed concurrently and influenced one another [Ripoll 2015].
[29]  Marks argues for “bi-directional influence” between the two works [Marks 2014].

[30]  EpArs (r = 0.354), HSatEp (0.259), HSatArs (0.041) vs. HSat – PSat (0.210), HSat – JSat (−0.007), PSat – JSat (0.137).
[31]  HSatRapt (r = −0.485), Aen – PSat (r = −0.537), PSatRapt (r = −0.579). The lowest pair, ArsGild (−0.834), was one of three statistical outliers (section 3.a); although we did not class Horace’s Ars Poetica as a satire, it shares has stylistic features of the genre.
[32]  Gruber’s comments are representative of a long tradition of Ausonius commentary: “Sprachlich und thematisch ist Vergil stets gegenwärtig. In jahrzehntelanger Lehrtätigkeit, in deren Mitte der Vergilerklärung stand, hat Ausonius diesen Dichter so verinnerlicht, daβ ihm nicht nur seine Worte, sondern die gesamte Thematik seiner Werke zur Verfügung stehen. Aber auch Lukrez, Horaz und Ovid gehören zum sprachlichen Fundus. Von den Autoren der frühen Kaiserzeit ist vor allem Statius sprachliches und thematisches Vorbild. Dazu kommen Lukan, Silius Italicus, Valerius Flaccus, und Martial”  [Gruber 2013, 27–28]. One of the goals of the present study is to place on an objective footing such statements of the relative importance of a given text as an overall verbal resource for its successors.
[33]  Hofmann (New Pauly s.v. Corippus, Flavius Cresconius) calls Corippus “the last great practitioner of the Roman epic… in his use of language and his narrative skill,” and cites Vergil and Claudian as the poet’s primary classical influences. Juvencus’ Historia Evangelica differs from all other texts in the data set due to its Biblical subject matter, and it should accordingly come as no surprise that exhibits both low rates of reuse and low centrality. Schmidt (New Pauly s.v. Iuvencus, C. Vettius Aquilinus) lists only Vergil as a relevant source for Juvencus. See [Green 2006, 11–14], who observes “roughly speaking, allusions to Vergil outnumber allusions to all other writers combined by at least five to one” (11 n. 63).

Works Cited

Axelson 1945 Axelson, B. Unpoetische Wörter, ein Beitrag zur Kenntnis der lateinischen Dichtersprache. Gleerup, Lund (1945).
Bodard and Mahony 2010 Bodard, G., & S. Mahony. 2010. Digital research in the study of classical antiquity. Ashgate, Burlington VT (2010).
Ceccarelli 2008 Ceccarelli, L. Contributi per una storia dell’esametro latino. Herder, Roma (2008).
Coffee et al. 2012 Coffee, N., J.-P. Koenig, S. Poornima, R. Ossewaarde, C. Forstall, and S. Jacobson. “Intertextuality In the Digital Age.” Transactions of the American Philological Association, 142 (2012): 383–422.
Coffee et al. 2013 “The Tesserae Project: Intertextual Analysis of Latin Poetry,” Literary and Linguistic Computing, 28 (2013): 221–28.
Courtney 2016 Courtney, Edward. 2016. “Ilias Latina.” Brill’s New Pauly, ed. Hubert Cancik and Helmuth Schneider. Leiden: Brill Online.
Curtius 1953 Curtius, E. R. European Literature and the Latin Middle Ages. Princeton University Press, Princeton (1953).
Dominik 2010 Dominik, W. J. “The Reception of Silius Italicus in Modern Scholarship.” In A. Augoustakis, ed., Brill’s Companion to Silius Italicus. Brill, Leiden (2010), pp. 425-447.
Drobisch 1866 Drobisch, M. W. “Ein statistischer Versuch über die Formen des lateinischen Hexameters”, Berichte über die Verhandlungen der Königlich-Sächsische Gesellschaft, Philologisch-Historische Classe, 18 (1866): 75–139.
Drucker 2009 Drucker, J. SpecLab: digital aesthetics and projects in speculative computing. University of Chicago Press, Chicago (2009).
Farrell 2005 Farrell, J. “Intention and Intertext”, Phoenix 59 (2005): 98-111.
Forstall et al. 2014 Forstall, C., N. Coffee, T. Buck, K. Roache, and S. Jacobson. “Modeling the Scholars: Detecting Intertextuality through Enhanced Word-Level N-Gram Matching,” Literary and Linguistic Computing (2014): doi: 10.1093/llc/fqu014.
Fowler 1997 Fowler, D. “On the Shoulders of Giants: Intertextuality and Classical Studies”, Materiali e discussioni per l’analisi dei testi classici 39 (1997): 13–34.
Gowers 2005 Gowers, E. Vergil’s Sibyl and the ‘many mouths’ cliché, Classical Quarterly 55 (2005): 170-182.
Green 2006 Green, R. Latin Epics of the New Testament: Juvencus, Sedulius, Arator. Oxford University Press, Oxford (2006).
Gruber 2013 Gruber, J. D. Magnus Ausonius, “Mosella”: kritische Ausgabe, Übersetzung, Kommentar. Walter de Gruyter, Berlin (2013).
Gruzelier 1993 Gruzelier, C. Claudian: De raptu Proserpinae. Oxford University Press, Oxford (1993).
Hardie 2009 Hardie, P. Lucretian Receptions: History, The Sublime, Knowledge. Cambridge University Press, Cambridge (2009).
Hinds 1998 Hinds, Stephen. Allusion and Intertext: Dynamics of Appropriation in Roman Poetry. Cambridge University Press, Cambridge (1998).
Hutchinson 2013 Hutchinson, G.O. Greek to Latin: frameworks and contexts for intertextuality. Oxford University Press, Oxford (2013).
Jockers 2013 Jockers, M. Macroanalysis: Digital Methods and Literary History. University of Illinois Press, Urbana (2013).
Kaufmann 2015 Kaufmann, H. “Papinius Noster: Statius in Roman Late Antiquity.” In W. J. Dominik, C. E. Newlands, and K. Gervais (eds.), Brill’s Companion to Statius. Brill, Leiden.
Knauer 1964 Knauer, G. Die Aeneis und Homer: Studien zur poetischen Technik Vergils mit Listen der Homerzitate in der Aeneis. Vandenhoeck & Rupprecht, Göttingen (1964).
Lovatt 2015 Lovatt, Helen. 2015. “Following after Valerius: Argonautic Imagery in the Thebaid.” In: Brill’s Companion to Statius, ed. Dominik, Gervais, and Newlands. Leiden: Brill. Pp. 408-424.
Marks 2014 Marks, R.D. “Statio-Silian Relations in the Thebaid and Punica 1–2”, Classical Philology 109 (2014): 130-139.
McCarty 2005 McCarty, Willard. Humanities Computing. Palgrave, London (2005).
McGill 2005 McGill, S. Vergil Recomposed: The Mythological and Secular Centos in Antiquity. Oxford University Press, Oxford (2005).
Micozzi 2007 Micozzi, L. Il catalogo degli eroi: saggio di commento a Stazio, Tebaide 4, 1–344. Edizioni della Normale, Pisa (2007).
Nelis 2001 Nelis, D. Vergil's Aeneid and the Argonautica of Apollonius Rhodius. Leeds: Francis Cairns, (2001).
Pagán 2015 Pagán, V. E. “Georgics 2.497 and Thebaid 1.19–20: Allusion and Inspiration.” In W. J. Dominik, C. E. Newlands, and K. Gervais (eds.), Brill’s Companion to Statius. Brill, Leiden.
Parkes 2009 Parkes, R. “Sed tardum (Ach. 1.47): Valerius Flaccus’ Argonautica as Prequel to Statius’ Achilleid.” Materiali e discussioni per l’analisi dei testi classici 63 (2009): 291-297.
Parkes 2012 Parkes, R. Statius, Thebaid 4. Oxford University Press, Oxford (2012).
Ramsay 2011 Ramsay, S. Reading machines: toward an algorithmic criticism. University of Illinois Press, Urbana (2011).
Ripoll 2015 Ripoll, F. “Statius and Silius Italicus.” In W. J. Dominik, C. E. Newlands, and K. Gervais (eds.), Brill’s Companion to Statius. Brill, Leiden.
Roche 2009 Roche, P. Lucan De Bello Civili Book 1. Oxford University Press, Oxford (2009).
Steiniger 2005 Steiniger, J. P. Papinius Statius: Thebais 4, 1–344. Franz Steiner Verlag, Stuttgart (2005).
Viansino 1995 Viansino, G. Lucano. La guerra civile. Mondadori, Milano (1995).
Volk 2009 Volk, K. Manilius and his Intellectual Background. Oxford University Press, Oxford (2009).
Ware 2012 Ware, C. Claudian and the Roman Epic Tradition. Cambridge University Press, Cambridge (2012).
Watson 1985 Watson, P. “Axelson Revisited: The Selection of Vocabulary in Latin Poetry.” Classical Quarterly 35 (1985): 430-448.
Wilson 2004 Wilson, M. “Ovidian Silius”, Arethusa 37 (2004): 225-249.
Zissos 2008 Zissos, A. Valerius Flaccus. Argonautica. Book 1. Oxford University Press, Oxford (2008).