¹ The consonants other than semivowels are p t c, b d g, f s h, m n η, r 1. Word boundary is a phoneme, but word stress is not entirely a function of it, since such unpredictable stresses as adhú:c adhuc must be recognized. On the whole, however, stress is predictable in terms of word boundary and the particular sequence of vowels and consonants preceding it. We leave it unmarked in this paper. Phonemic symbols are printed in roman type, without special marks of identification.

² See also Robert A. Hall Jr., Classical Philology 41.84-90 (1946). Forms in italics in the present paper are to be understood primarily as transliterations from the records, more or less normalized, but sometimes (where explicitly stated) as reflections of modern printing conventions.

³ With specific restrictions due to conditioned sound change. The differentiation of IE i and u into Lat. i i̯ j and u u̯ v respectively is recent, and the conditioning factors are still discernible in the fragmentary and sometimes nearly complementary distribution of the three entities in each set. For the details see Roland G. Kent, The sounds of Latin² 140-4 etc., and the other standard manuals. See also §6 below.

In discussing the distribution of sounds, we do not distinguish sequences that occur across a compounding seam (as bj in abjectus) from others (as lv in vol vit). The evidence of word accent, consonant assimilation, and absence of elision or special types of it all indicate that the boundary between members of a compound was not phonologically marked. Vowel shortening, contraction, and then again restoration of unshortened or uncontracted forms create variants where hiatus is involved: dēst, dēest, etc.

⁴ This list implies that there is no contrast between post-archaic ae oe and instances of contraction spelled ai ei (see dein above) oi (and ui in cui, huie, as well as in fluitai). Indeed there is no reason to believe that proin, when pronounced as a monosyllable, had a different diphthong from poena. That there should be a difference in spelling is easy enough to understand, in view of the full form proin (de). What is still rather obscure is why the archaic spelling ai oi was changed to ae oe except before j. A minute subphonemic change in the height of the semivowel hardly explains it. One is reminded of the parallel fluctuations in Greek inscriptions. For the whole problem see Edgar H. Sturtevant, The pronunciation of Greek and Latin² 123-9, 132-8, 146-7.

Manu Leumann, Lateinische Grammatik 80, would write se: .

⁵ Kent, in TAPA 43.40 (1912), has called attention to this and to a small number of very similar instances. He has shown that reiectus with a long first syllable () does not warrant the assumption that there is an automatic change to in all compounds in which a final vowel is followed by (in our terms) an initial j.

In view of the small number of examples, there is even a possibility that biiugis was different from (say) diurnus only in spelling. If so, it might be mechanically possible to get along with only one semivowel in each series (rounded and unrounded). The unrounded semivowel would pattern like vowel length in always going with a preceding vowel, while the rounded semivowel would behave like an ordinary consonant. The instance in §6 would then have to be treated as an entirely foreign tour de force, or as an example of -evv- parallel to -ejj- in peius.

⁶ See Kent, TAPA 43.37; Eduard Schwyzer, Griechische Grammatik 1.158, 198; also 219 on eu(h)oe.