Identification of Latin Poets by Sound

Abstract

L.C. Curran: Identification of Latin Poets59 Identification of Latin Poets by Sound Leo C. Curran "Most authors writing in the same language at the same time use most words at roughly the same rate."1 Anyone who gives the matter any thought would probably agree with this statement and with another that it entails: that the authors also use most sounds at roughly the same rate. However, the degree of approximation is "rough" enough that it is not impossible to distinguish certain Latin poets from each other on the basis of differences in sound preferences, i.e., relative frequency in their works of all the sounds of Latin. My thesis is that, in the case of certain authors, a book of Latin poetry of the late republic or early empire can be assigned not only to its author but also to the individual work of which it is a part solely by means of one simple measure: the relative frequency of its sounds. For example, applying this test to the text of Aeneid 1 will tell us not only that Virgil wrote it, but that it is from the Aeneid rather than the Georgics. By sound alone, it is possible to distinguish one poet's unique voice from another's and to measure the modulation of that voice from one of the author's works to another of them. The poets can be recognized by their sounds, by their voices, just as painters can be recognized by their brush strokes or violinists by their bowing. There are unique voices in Latin poetry in a very concrete sense: they can be measured. In order to make the results of the study as conclusive as possible, I have worked with very large samples, the complete text of a work or the complete text of an individual book from a work.2 The body of Latin poetry I have analyzed totals 39,577 lines containing over two million alphabetical characters. The only work not represented in its entirety is Ovid's Metamorphoses, of which I had only two of the 15 books. The 48 texts are listed in Table 1, column 2. The A.P.A. Registry of Greek and Latin Texts in Machine-Readable Form was the source of the texts.3 1 A. Kenny, The Computation ofStyle (Oxford 1982) 69. 2 Cf. D. Crystal, The Cambridge Encyclopedia ofLanguage (Cambridge 1987) 67. 3 I am grateful to Professor S.V.F. Waite for his assistance in obtaining the machine-readable texts. CM. Birch entered Catullus from Mynors' Oxford Classical Text. L. Roberts entered Lucretius from Bailey's editio major. T.F. Brunner entered the Amores, the Ars Amatoria, the Medicamina Faciei Femineae, and the Remedia Amoris from Kenney's Oxford Classical Text, the 60Syllecta Classica 2 (1990) The study rests upon counts of sounds m the texts. I use the less precise word sound instead oíphoneme, because the 31 sounds studied are not quite a oneto -one match with the phonemes of Latin. Therefore, I will use the term phoneme only in this section of the paper, elsewhere, I will use the term sound and will not enclose the letters representing tie sounds within the slashes of phonemic notation. There are additions to the list of Latin phonemes: I include non-Latin sounds in words borrowed from Greek: Y, Z, CH, PH, and TH. There are omissions: the very rare diphthong ei (which occurs primarily in contracted deinde and dein) and, of greater moment, tie distinction between long and short vowels. What I call the sound A, for example, encompasses the phonemes /a:/ and /a/, i.e., both the long and the short vowel. I could have added a scanning routine to my program, but that would have been only a partial solution to the problem of distinguishing long vowels from short. Since a scanning program would determine the length of syllables, but not vowels, it would treat as long tie vowel in every syllable that is long by position. Latin has many short vowels in syllables long by position that would trip up a scansion program. For example: It would count e infero as short but inferí as long...