At the earliest stages of Latin poetry known to us, we can see the obsolescence of poetic meters based on word accent and the adoption/adaptation of Greek rhythms founded on syllable length. However, accentual rhythm never became entirely irrelevant to Latin verse-making, either because accentuation is affected by syllable length or because we can notice recurring accentual patterns in the extant Latin verses composed in the Greek mould. Metrists since Jackson Knight (1931, 1939) have argued that these patterns are “intentional.” I pursue a hypothesis in this camp while refining the description of the patterns. I proceed from the intuition that the verses that survive of Ennius’ Annales amount to a random sample, and the token counts of different accentual configurations reflect Ennius’ subconscious preferences for certain types over others. In other words, the more often a pattern occurs, the more favorable Ennius finds it (cf. Friedberg’s  “hierarchies of preferences”).
Complete lines from the Annales (411 verses) instantiate a range of accentual rhythms with differing strengths of attestation. This suggests that within Ennius’ mental poetic “grammar,” as he was conscious of the formal requirements of his adopted/adapted genre, he was subconsciously balancing preferences and bans on the alignment of word stresses to the positions of the hexameter. Out of the 3,072 accentuations that are logically possible from the combination of stressed or unstressed principes, stressed or unstressed contractible bicipitia (to use West’s terminology), and stressed or unstressed final position, ten patterns are instantiated by 204—almost half—of Ennius’s epic lines (the remaining verses of the corpus attest 110 other patterns):
The relative frequencies of such accented-hexameter types can be accounted for in a constraint-based analysis by a parsimonious set of preferences and bans. (a) The caesura should be “masculine.” (b) Adjacent metrical positions should not be stressed. (c) Adjacent metrical positions should not be unstressed. (d) Stresses should be aligned to the principes, and unstressed syllables to the contractible bicipitia and final position. (e) Under the Iambo-Trochaic Law, known from different domains of human cognition, accentual rhythm strongly tends to be taken as trochaic and quantities as iambic. (f) The accentual rhythms of the hemistichs should be parallel. A log-linear model using the Poisson regression, an implementation of Maximum Entropy, shows that an accentual type is less frequent the more it violates (a)–(d) by its non-masculine caesura, adjacent stresses, and adjacent unstressed syllables. Rather interestingly, an accentual type is more frequent if it is accentually iambic, violating (e), if its hemistichs do not have parallel rhythms, and if it respects (a), (b), and (c) simultaneously.
This account goes beyond Jackson Knight’s “heterodyne to the caesura, homodyne in the cadence.” Ennius carried out his rejection of uersibus quos olim Faunei uatesque canebant in a complex way. He avoided “regularity” and achieved interest by misaligning stresses to contractible bicipitia and unstressed to principles, going against (d), by arranging his stresses iambically in counterpoint to the trochaic durational rhythm of the hexameter, fully violating (e), and by minimizing the rhythmic parallelism of his hemistichs, flouting (f).
Greek and Latin Linguistics