1 Introduction

Latin vowel weakening was a historical process, generally understood as an example of vowel reduction, that involved the apparent raising and fronting of short vowels in medial and final syllables, giving rise to alternations such as faciō ‘I make’~ perficiō (< *per-faciō) ‘I complete.’ The effects of weakening vary depending on several factors, including vowel quality and syllable structure. Weakening tends to produce greater alterations in open than in closed syllables (e.g., *per-fact- > perfect- ‘completed’ with a > e in a closed syllable, but *per-faciō > perficiō ‘I complete’ with a > i in an open syllable), running counter to an observed crosslinguistic trend that vowels exhibit shorter articulation, and hence are more likely to undergo reduction, in closed than open syllables (Maddieson 1985; Sen 2012; Weiss 2020: 127).

Taking into account phonetic (crosslinguistic) likelihoods as well as phonological (intralinguistic) evidence, I propose that the latter stage of weakening (e > i) should not in fact be described as reduction, but rather as open-syllable tensing (Storme 2019), an unrelated process. I assume a broad definition of reduction, encompassing both the loss of distinctions between vowel phonemes and changes in their phonetic realizations, particularly towards the center of the vowel space. In the case of weakening, as often happens, a trigger of synchronic (phonetic) reduction became obscured by later sound change, resulting in reduction as a historical (phonemic) change.

2 Current models

The history of Latin can be periodized into Archaic (before ca. 100 BC), further subdivided into Very Old (VOL, before ca. 300 BC) and Old (OL, ca. 300–100 BC); Classical (CL, ca. 100 BC–200 AD); and Late (LL, after ca. 200 AD). Weakening took place within VOL, probably between 500–300 BC, resulting in alternations which persisted through later, better-attested periods of the language. Synchronic reduction in VOL was triggered by the placement of stress on initial syllables. Crosslinguistically, reduction occurs most frequently in unstressed syllables (van Bergem 1995: 61, 91, 129) in languages with a prominent stress accent (Weiss 2020: 119, 126). In Latin, only initial syllables are universally protected from weakening, so weakening as reduction implies a strong initial stress pattern in VOL (Weiss 2020: 119, 126), differing both from the Proto-Indo-European pitch accent and from the later ‘penultimate rule’ or ‘Latin stress rule’ of CL.

Weakening was exclusively a development of short vowels. Diphthongs were affected, though their outcomes tend to be obscured by subsequent monophthongization in late OL to early CL (Sihler 1995: § 57): e.g., *en-cai̯dō > *incei̯dō > incīdō ‘I cut into’ shows the expected weakening outcome a > e, followed by regular monophthongization ei̯ > ī (Weiss 2020: 129), cf. *cai̯dō > caedō ‘I cut down, kill.’ Likewise, changes specifically affecting final syllables generally make weakening outcomes less obvious there than in medial syllables (Weiss 2020: 148–159). For this reason, unless otherwise noted, the majority of my examples in this article illustrate outcomes of monophthongs in medial syllables; after establishing the outcomes of weakening under these basic conditions, I briefly discuss subsequent changes that affect diphthongs and final syllables (see below, section 4) to provide a complete picture.

Latin at all periods distinguished five vowels, ‹a e i o u›, appearing in long and short forms; a sixth letter ‹y› was used primarily for transcribing upsilon in Greek loanwords. Grammarians of the Classical period, including Quintilian (fl. 1st c. AD) and Velius Longus (fl. 2nd c. AD), speak of an additional vowel quality, occurring as an allophone in weakening contexts before a single labial consonant (p b f m) and spelled either ‹u› or ‹i›, as in monumentum ~ monimentum ‘reminder.’ Quintilian describes this sound as ‘medius … U et I litterae sonus’ (‘a sound between u and i’) (Institutio Oratoria 1.4.8–9 in Russell 2002: 106–107), hence its name: sonus medius, the ‘middle sound.’ The Emperor Claudius (r. 41–54 AD) introduced a new letter ‹ⱶ› apparently to represent this sound: ‘Tiberius Claudius novam quandam litteram excogitavit … per quam scriberentur eae voces quae neque secundum exilitatem “i” neque secundum pinguitudinem “u” litterae sonarent’ (‘Tiberius Claudius thought up a certain new letter with which those vowels might be written whose sound conforms neither to the frontness of the letter i nor to the backness of the letter u’; Velius Longus De Orthographia 13.1.2 in di Napoli 2011, tr. mine), though archaeological evidence complicates this picture (see below, section 3).

The primary allophones of the five phonemic vowels can be defined in a feature phonological framework, specifying for height and backness, as follows:¹

Each sound is assumed to be maximally distinct from all others, given the features specified above: i.e., the graphemes ‹a e i o u› represent phonemes /a e i o u/ with primary allophones [a e i o u].² The sonus medius had by LL ceased to exist as an independent allophone, being no longer distinct from short /i/ in other contexts (Weiss 2020: 128); its realization before then is a matter of considerable debate, reviewed below (section 3) in the context of its historical development.

As with patterns of historical reduction in English, Russian, and other languages (Weiss 2020: 126), weakening produces different outcomes across vowel qualities and environments. An important environmental factor is syllable structure, whether closed or open. As already noted, weakening is generally more ‘thoroughgoing’ in open than closed syllables (Weiss 2020: 127), resulting in a greater increase in vowel height. The unconditioned outcomes of a, e, and i are most transparent and can be formulated as follows:

• a > e in closed syllables, e.g., *per-fact- > perfect- ‘having been completed’

• a > i in open syllables, e.g., *per-faciō > perficiō ‘I complete’

• e > e in closed syllables (no change)

• e > i in open syllables, e.g., *com-premō > comprimō ‘I press together’

• i > i in closed and open syllables (no change)

These outcomes are traditionally analyzed as the result of two rules applied in sequence

1. a > e in all weakening environments

2. e > i in open syllables

so that the development a > i takes place over two stages, e.g., *per-faciō > *perfeciō > perficiō.³

Gaps in the data and later, unrelated changes make it more difficult to determine what happened to original o and u. Relatively few Latin words can be traced back to preforms containing short o or u in medial positions, where they would be subject to weakening in its most observable form. Original o surfaces regularly as i in open syllables, e.g., *homon- > homin- ‘man’ in oblique cases (cf. nom. homō), but scholars are divided regarding the outcome of o in closed syllables. Some propose o > e in closed syllables, citing *hosti-pot-is > *hospets > hospes ‘guest, host (nom.)’ with syncopation of a medial syllable (Buck 1933: § 101.2; Liesner 2014: §§ 14–15), but hosp-es, -itis might instead have developed by analogy ‘to other “agential” nouns of the artifex, artificis type’ (Weiss 2020: 131, n. 22), and no other secure examples appear to support this hypothesis. It seems more likely that in closed syllables o remained until the historical period, when it was raised to u (Leumann 1977: § 87; Sihler 1995: § 66.3b; Meiser 1998: § 53.1; Weiss 2020: 131), e.g., *homon-cul- > homuncul- ‘poor man.’ This raising o > u is not properly part of weakening, but took place later, in OL of the 3rd and 2nd c. BC (Sihler 1995: §§ 66.3b–c; Meiser 1998: § 53); forms in -us, -um are modern features (Wachter 1987: §§ 120b, 130a, 131) in inscriptions like the Scipio epitaphs of the late 3rd c. BC (CIL 1².6–9) and the Senatus Consultum de Bacchanalibus of 186 BC (CIL 1².581), where they coexist alongside more conservative forms. Weiss (2020: 131) further proposes that the development o > i in open syllables took place over two stages (similar to the development a > e > i)ː first o became e in open syllables, then e from all sources became i, e.g., *novo-tās > *novetās > novitās ‘novelty.’

As for u, in open syllables the most solid evidence indicates u > i, e.g., *caput- > capit- ‘head’ in oblique cases, *cornu-can- > cornicen- ‘trumpeter’ (Meiser 1998: § 52; Liesner 2014: § 14; Weiss 2020: 126). Leumann (1977: § 86.III) gives several potential counterexamples but notes that none is conclusive: arbutus ‘strawberry tree’ might be a loanword; coniug- ‘spouse’ and pecud- ‘sheep’ in oblique cases and tutud- ‘have crushed’ may have retained their root vowels by analogy to nom. coniunx, pecus and pres. tundō, respectively. In closed syllables u is regularly preserved, e.g., *ē-ruptus > ēruptus ‘burst forth’ (Weiss 2020: 127).

At this point two interpretations of the data are possible. During the first stage of weakening, the five short vowels may have collapsed into two: a e o > e and i u > i (o and u were affected only in open syllables); this view is implied, e.g., by Liesner’s presentation of the data (2014: § 14). Alternatively, Weiss (2020: 131) proposes that all five short vowels fell together as e in open syllables before the change e > i. In favor of the latter view, certain inscriptional forms appear to attest an intermediate e-stage for o and i (Leumann 1977: § 90a; Wachter 1987: § 184a; Parker 1988: 232, 234; Nishimura 2010: 232; Weiss 2020: 131):

• o > e in Apolenei (CIL 1².368), Apolene (CIL 1².384) for CL Apollinī ‘to Apollo’

• o > e in Esqelino (CIL 1².416) for CL Esquilinō ‘on the Esquiline (hill)’

• i > e in hec (CIL 1².9) for CL hic ‘he’

• i > e in Tempestatebus (CIL 1².9) for CL Tempestātibus ‘to the storm-goddesses’

The interpretation of Esqelino as evidence for o > e relies on a derivation from *ex-col- ‘dwell outside,’ because settlements on the hill were outside early city limits; the name may alternatively derive from aesculus ‘Italian oak,’ a species of tree common on its slopes (Richardson 1992: 146). In favor of *ex-col-, Harmon (1993) points out that initial ai̯ > e is unjustified in Latin of the relevant period; he also notes that the medial vowel in aesculus is most likely anaptyctic (*ai̯sclo- > *ai̯sculo-), in which case it would not have been introduced until the 3rd c. BC (Weiss 2020: 178–179), too late to be affected by weakening. The last two examples are questionable considering the late (3rd c. BC) date of the Scipio epitaph (CIL 1².9), as well as its inconsistent orthography (Wachter 1987: § 127e): cf., e.g., aediles cosol cesor ‘aedile, consul, and censor’ (l. 2) alongside consol censor aidilis ‘consul, censor, and aedile’ (l. 5), hec ‘he’ (l. 6) alongside hic ‘he’ (l. 5). It turns out, however, that regardless of whether the first stage of weakening resulted in a one- or two-vowel system, open-syllable tensing explains the second stage of weakening (e > i) better than reduction. The conclusions presented in the remainder of this article are equally valid assuming one- or two-vowel reduction for the first stage of weakening.

3 The sonus medius

The outcomes described so far represent unconditioned changes, dependent only on the quality of the vowel and syllable weight. Weakening produces different outcomes or is checked entirely in various phonological conditioning environments, of which Weiss gives a comprehensive list (2020: 127–129). Particularly illuminating to a phonetic perspective on the Latin pattern is weakening to the sonus medius before a single labial consonant, e.g., *mone-ment- > monument- ~ moniment- ‘reminder,’ *ponti-fac-s > pontufex ~ pontifex ‘high priest’ (orig. ‘bridge-maker’). The appearance of the sonus medius is uncertain and seems to be influenced partly by the quality of the surrounding vowels (Weiss 2020: 128): spellings in ‹i› are favored when an i-vowel directly follows the rounding labial (cf. occipiō ‘I take up,’ occupō ‘I grasp’) or when the vowel of the initial syllable is unrounded (cf. prīncip- ‘leader,’ aucup- ‘bird-catcher’). The conditions under which it does appear are nevertheless illustrative of its pronunciation within the Latin phonological system.

A few examples have been raised to suggest that the sonus medius appears in closed syllables as well (Weiss 2020: 128, n. 8): condumnari ‘to be condemned’ (CIL 1².582) for CL *com-damn- > condemn-, surruptum ‘stolen’ (Plautus Aulularia 39) for CL *sub-rapt- > surrept-,⁴ and Gk. θρίαμβος ‘hymn to Dionysus’ > triumpus, later triumphus ‘victory parade.’ This last, the only example with more than one secure attestation, most likely had an intermediate stage in Etruscan: the change b > p is characteristic, and a > u is attested in some other borrowings, as in Gk. Πρίαμος ‘Priam’ > Etruscan Priumne (Warren 1970: 111). An exclamatory form triumpe appears repeatedly at the end of the Carmen Arvale (Warmington 1940: 250–253); the language of the Carmen has been dated before the turn of the 6th c. BC (Warren 1970: 110),⁵ making triumpus an implausible product of weakening, which is not supposed to have begun until the 5th c. (Weiss 2020: 130). It seems better to conclude that the sonus medius appeared exclusively in open syllables,⁶ while a wealth of examples confirm regular development to e in closed syllables, as in CL condemnō ‘I condemn’ and surrept- ‘stolen.’

The debates on the pronunciation of the sonus medius furnish elegant and compelling evidence regarding the pronunciation of the regular weakening products e and i. Most scholars agree that the Latin spellings ‹i u› for the sonus medius represent something other than [i u], but the agreement ends there. From Quintilian’s description of a sound ‘medius … U et I,’ some have thought that the sonus medius was pronounced [y] with features [+close,–open] like both [i] and [u], [+front,–back] like [i], and [+round] like [u] (Buck 1933: § 110.4a; Leumann 1977: § 92B; Meiser 1998: § 52.3; Liesner 2014: § 14). Sihler writes that ‘during the period of orthographic hesitation,’ i.e., when spelling varied between ‹i› and ‹u›, ‘the vowel in question was in all likelihood … a central or back unrounded vowel’ (1995: § 69b), that is [ɨ] or [ɯ]. Weiss suggests [ɨ] and [ʉ] as possibilities (2020: 72), Nishimura favors [ʉ] or more open [ɵ] (2010: 222), and Allen identifies the sound as [ʉ] when written ‹u› but [ɨ] when written ‹i› (1978: 57–58).

The exception to this guessing game is Buck, who does not assign a unique (non-cardinal) phonetic value to the sonus medius. He writes: ‘It is more probable, though not the usual view, that we have to do with an ordinary u in the early maxumus (as obviously in the persistent occupō) and with an ordinary i in the later maximus, the alleged intermediate sound being imaginary, suggested by the fluctuation in spelling’ (1933: § 110.4a). Though others have acknowledged Buck’s hypothesis as a possibility (Leumann 1977: § 92B; Sihler 1995: § 69b) or supported it (Parker 1988: 235, n. 44), a phonetic reading of the sonus medius is both more plausible and intellectually safer than an orthographic reading. To agree with Buck, we should have to disregard Quintilian’s testimony, which is both clear and, as rare evidence from a speaker, difficult to ignore: ‘et medius est quidam U et I litterae sonus (non enim sic “optimum” dicimus ut “opimum”)’ (‘there is also a sound between u and i (we do not pronounce optimus and opimus in the same way)’; Institutio Oratoria 1.4.8–9 in Russell 2002: 106–107).⁷ To consider matters from another perspective, if Buck is indeed correct and the sonus medius represents not a separate allophone but fluctuation between cardinal [i] and [u], it is only ourselves that we are fooling: overanalysis is a lighter sin than underanalysis.

Assuming that the sonus medius represents an allophonic pronunciation, and given its variational distribution as described above, a description of its features can be reasoned with some accuracy; from there one can extrapolate toward the realizations of the other weakening products. To begin with, the feature of the sonus medius that distinguishes it from the regular weakening product i must be [+round]: the sonus medius appears exclusively before one of the labial consonants (p, b, f, or m), and the quasi-vowel harmonic effects that favor spellings in ‹u› are of a rounding type (Weiss 2020: 128). It is reasonable to assume, given its distribution, that roundedness is the only distinguishing feature of the sonus medius, i.e., all its other features are shared with the regular weakening product i. Variation in spelling between ‹i› and ‹u› indicates that the sonus medius was [+close,–open], as these features are shared uniquely by cardinal [i] and [u]. Given the features defined above (section 2) for the cardinal vowels, as well as the distinguishing feature [±round], the sonus medius and weakening product i must have had one of the following feature profiles (represented for the purposes of this discussion by a reasonable choice of IPA symbols): [±round], [+close,–open], and [+front,–back] as close front [i] and [y]; [±round], [+close,–open], and [–front,–back] as close central [ɨ] and [ʉ]; or [±round], [+close,–open], and [–front,+back] as close back [ɯ] and [u].

Close back [ɯ] and [u] are easily eliminated, as there is no reason to suggest that the regular weakening process involved backing. Among the remaining two possibilities, front [i] and [y] have been historically the most popular, but formidable counterevidence can be found in the distribution of the Latin letter ‹y›. This letter was used to transcribe the sound represented by Ancient Greek upsilon, which has been reconstructed as front rounded /y/ (Allen 1968: 62–66). Latin writers, however, almost never used ‹y› to represent the sonus medius. The reverse holds true as well: when Roman names were transcribed into Greek, the sonus medius was never represented by upsilon (Allen 1978: 58; Weiss 2020: 72, n. 10), but by a variety of letters representing close or rounded vowel sounds: Maximus ~ Μάξιμος, Decumus ~ Δέκομος, Postumius ~ Ποστούμιος (Weiss 2020: 72, n. 10). Furthermore, Quintilian (Institutio Oratoria 12.10.27–28) mentions explicitly that Latin lacks the sound made by Greek upsilon:

… iucundissimas ex Graecis litteras non habemus (vocalem alteram, alteram consonantem), quibus nullae apud eos dulcius spirant: quas mutari solemus quotiens illorum nominibus utimur; quod cum contingit, nescio quo modo velut hilarior protinus renidet oratio, ut in ‘zephyris’ et ‘zopyris.’

We lack the two most pleasing of the Greek letters, one vowel and one consonant, the sweetest sounds in their language. We borrow these when we use Greek words, and when this happens, the language at once seems to brighten up and smile, as in zephyrus and zopyrus.

Russell 2002: 296–297

The consonant is zeta, transcribed as Latin ‹z›; the vowel is upsilon, ‹y›. This passage occurs at a distance from the description of the sonus medius. Quintilian does not appear to have made a connection, as some modern scholars have, between the sonus medius and upsilon.

‹Y› for the sonus medius appears in scattered inscriptions, e.g., in lachrymans for standard lacrimāns ‘weeping’ (AE 1995), lachrymis for lacrimīs ‘with tears’ (CIL 2.13; CIL 6.6051), and contybernali for contubernālī ‘to (his) companion’ (CIL 6.11273; CIL 9.2608). Spellings in lachrym- are probably Hellenizing, with ‹y› matching Gk. δάκρυον ‘tear(s),’ while ‹ch› for c is an invention. CIL 9.2608 has contybernali optimae ‘to (his) excellent companion,’ with the sonus medius appearing puzzlingly as ‹y› in one word and ‹i› in the next; Hellenization is possible here too, especially considering the cognomen Eutyches at the bottom of the inscription. Hübner suggests that ‹y› at CIL 6.11273 may be a variant letterform of ‹v› (1885: lxvi). In any of these cases ‹y› may have been used imprecisely, as a stand-in for an intermediate close vowel; these examples can be compared with cases where ‹y› is used inappropriately for one of the other close vowels, especially where a Greek origin is imagined (Hübner 1885: lxvi; Allen 1978: 53; Graßl 2005: 241).

Claudius was a deliberate philologist and would presumably not have invented a new letter for the sonus medius if it sounded the same as Greek upsilon (or for that matter Latin ‹y›). Archaeological evidence, however, complicates an understanding of the relevant passage from Velius Longus (see above, section 2): in all secure cases but one, the Claudian letter ‹ⱶ› is actually used to transcribe a short upsilon (Oliver 1949). Graßl has found ‹ⱶ› for the sonus medius once in an inscription containing the Greek name Pilⱶmuso ‘Philomusus’ (2005: 241), but as he points out, its appearance here can perhaps be attributed to Hellenization in the face of uncertainty (2005: 241, n. 3). Additionally, the possibility that the letter was not actually used as Claudius intended (Oliver 1949) renders any interpretation of the passage from the De Orthographia effectively moot.

Nevertheless, one must at this point abandon [i] and [y] in the face of growing evidence to the contrary and turn to the remaining choice: the sonus medius was most likely a rounded close central sound [ʉ], while the regular weakening product ‹i› was its unrounded counterpart [ɨ]. This reading most closely aligns with those of Weiss (2020: 72), Allen (1978: 57–58), and Nishimura (2010: 222), who says of the sonus medius that ‘we should be satisfied with its central feature as the most plausible’ (2010: 232, n. 45). This conclusion comes with an addendum regarding the variation in spelling of the sonus medius. Allen believes that in Archaic Latin the sonus medius was pronounced [ʉ] when written ‹u› and [ɨ] (i.e., same as the regular weakening product) when written ‹i› (1978: 57–58). Absent evidence to the contrary, I am inclined to agree that differences in spelling represent differences in pronunciation. Allen supports his view with a quotation from Velius Longus, in whose time ‘usque i littera castigavimus illam pinguitudinem, non tamen ut plene i litteram enuntiaremus’ (‘We have corrected the former broadness by (a movement in the direction of) i, but not so far as to pronounce it fully as i’; De Orthographia 4.2.5 in Allen 1978: 58). Quintilian corroborates and dates the fixing of spellings to the 1st c. BC: ‘Iam “optimus” “maximus” ut mediam I litteram, quae veteribus U fuerat, acciperent C. primum Caesaris inscriptione traditur factum’ (‘Again, an inscription of Gaius Caesar is said to be the first authority for writing optimus maximus with i in the middle syllables, instead of u, as in older texts’; Institutio Oratoria 1.7.21 in Russell 2002: 194–195). Medial /i/ must have lost its allophonic pronunciation sometime in the centuries after these authors, because both the sonus medius and the regular weakening product ‹i› surface as [i] in the Romance languages (Weiss 2020: 128, n. 8); for example, Latin lacrima ~ lacruma ‘tear’ becomes It. lacrima and Sp. lágrima, both with a close front unrounded vowel. The most favorable reading of the evidence is thus the following: during the Archaic period, the sonus medius appeared as an allophone of the regular weakening product i in certain rounding environments; later (if Quintilian is to be believed, during the 1st c. BC) CL speakers fixed medial weakening products as either phonemic /u/ or (mostly) /i/, but they continued to pronounce these sounds as [ʉ] and [ɨ], at least in some environments (perhaps only where they remained unaffected by CL stress); finally, in LL these sounds merged with cardinal [u] and [i] (or [ʊ] and [ɪ]), respectively.

4 The phonetics of reduction

If the phonetic realization of the weakening product ‹i› was a close central unrounded vowel [ɨ], then what was the phonetic realization of the weakening product ‹e›? When primary evidence presents an incomplete picture, one may consider how to fill in the details by other means. The short vowel resulting at least from a, e, and o the first stages of weakening can be represented as an allophone of /e/, as it remained in closed syllables until the Classical period. Rix provides the insight that this sound may have been realized in context as a central vowel, [ə] or [ɪ] (1966: 160–161).⁸ This suggestion is attractive from a typological perspective, as reduced vowels tend to take on a central articulation (Lindblom 1963: 1780; van Bergem 1995: 7), and resolves some problems in the phonetic interpretation of the data: e.g., fronting movements /a/ > /e/ and /o/ > /e/ are crosslinguistically more marked and lack parallels in the history of Latin, but centralization [a e o] > [ə] is common and easy to motivate across theoretical backgrounds, as, e.g., by Lindblom (1963) with articulatory undershoot theory and Crosswhite (2001) in Optimality Theory. Thus the intermediate e-stage was probably rather a schwa-stage, with movement towards not the front but the center of the vowel space (Meiser 1998: § 52; Nishimura 2010: 225).

Under this assumption, either all five vowels collapsed into a single central articulation [ə], or a two-way distinction of height was maintained [ə ɨ]. Reduction to a vertical two-vowel system is attested (Miller 1972: 483) but less common than full reduction to [ə], making the two-vowel interpretation comparatively unlikely; this kind of argumentation is however inconclusive.

A summary of weakening and related changes may now be presented, taking into account both possibilities. In the first stage, either all five short vowels or only a, e, and o fell together (i.e., were reduced) to [ə]; if i and u did not merge with the other vowels, they fell together as [ɨ]. In the second stage of change, /e/ [ə] in weakening environments was raised to [ɨ]; before a labial consonant this sound was sometimes additionally rounded to the sonus medius, pronounced [ʉ]. The weakening product [ə] was interpreted as an allophone of /e/ and spelled ‹e›,⁹ [ɨ] was interpreted as an allophone of /i/ and spelled ‹i›, and [ʉ] (where it appeared) was interpreted as an allophone of /u/ and spelled ‹u›. To put the weakening rules in another format,

1. 1. [a e] > [ə] /e/ in non-initial (unstressed) syllables;

      [o] > [ə] /e/ (‹e› in Apolene(i), Esqelino?) in non-initial (unstressed) open syllables

   2. [i] > [ə] /e/ (‹e› in hec, Tempestatebus?) or [ɨ] /i/ in non-initial (unstressed) syllables;

      [u] > [ə] /e/ or [ɨ] /i/ in non-initial (unstressed) open syllables

2. [ə] > [ɨ] /i/ ‹i› in non-initial open syllables (sometimes [ʉ] /u/ ‹u› before a labial)

while [ə], which remained in closed syllables, was spelled ‹e›.

Subsequent changes unrelated to weakening brought the evolution of the CL vowel system to completion. Among these were changes affecting diphthongs and short vowels in final syllables, which render the outcomes of weakening less obvious in many cases, though the process originally affected all non-initial short vowels in the same way. Given the scope of this article, the following discussion is necessarily brief. I do not attempt to provide explanations, as I do elsewhere in this article, for conditioned changes unique to diphthongs and final syllables. I hope that it may prove a fruitful area of further research to investigate these processes more closely in their proper phonological and phonetic perspective.

VOL had exclusively falling diphthongs ai̯ au̯ ei̯ eu̯ oi̯ ou̯, which underwent closed-syllable weakening as sequences of short vowel + semivowel + consonant or word boundary. As expected, ai̯ au̯ underwent weakening to ei̯ eu̯, while original ei̯ eu̯ oi̯ ou̯ remained unchanged. Subsequently, the following monophthongizations occurred in all positions¹⁰ (Sihler 1995: §§ 57, 61, 70, 75; Weiss 2020: 127–128): ei̯ > ī, e.g., *en-cai̯dō > *incei̯dō > incīdō ‘I cut into’; eu̯ > ou̯ > ū, e.g., *com-clau̯dō > *concleu̯dō > *conclou̯dō > conclūdō ‘I close up’; and oi̯ ou̯ > ū, e.g., *com-moi̯n- > commūn- ‘common.’ Meanwhile ai̯ and au̯ were preserved in initial syllables as ae and au, producing, e.g., *cai̯dō > caedō ‘I cut down’ and *clau̯dō > claudō ‘I close’; in a very few words oi̯ was also preserved in initial syllables as oe (Sihler 1995: § 59), giving, e.g., *moi̯n-i-a > moenia ‘walls.’ These changes occurred late in OL, more than a century after weakening was completed, so that the Senatus Consultum de Bacchanalibus of 186 BC (CIL 1².581) still has quei for quī, plous for plūs, etc. Diphthongs in final syllables mostly show the same developments as in medial syllables (Sihler 1995: § 75; Weiss 2020: 128), e.g., 1sg. perf. act. ind. -ai̯ (cf. Faliscan peparai ‘I gave birth’) > -ī. The exception is oi̯ (Sihler 1995: § 75; Weiss 2020: 128), which yields (ei̯ > ẹ̄ >) ī in final syllables rather than expected ū, e.g., masc. nom. pl. -ī alongside Gk. -οι.

Short monophthongs in final closed syllables appear largely to have undergone closed-syllable weakening as expected, though many examples are obscured by a subsequent change e > i before a single non-nasal consonant (Sihler 1995: § 71). The following outcomes are observed: a > e in final syllables before m, n, or a consonant cluster, e.g., *cornu-can > cornicen ‘trumpeter’ (cf. can- ‘to sing’), *prīsmo-cap-s > prīnceps ‘foremost’ (cf. cap- ‘to take’); a > i in final syllables before a single non-nasal consonant, e.g., *re-de-dat > reddit ‘he/she returns’ (cf. dat ‘he/she gives’); e and i developed similarly to a, while o and u remained unchanged until the Classical period. (Recall that raising o > u, which among other forms engendered the characteristic CL -us, -um alongside Gk. -ος, -ον, was an OL innovation; see above, section 2) Apparent exceptions to these rules can generally be explained as the result of unrelated changes, for example shortening of long vowels¹¹ (Sihler 1995: § 71; Weiss 2020: 133) or apocope (Sihler 1995: §§ 73–74; Weiss 2020: 157–158).

In final open syllables, i.e., in absolute final position, the outcome of a is difficult to reconstruct but may be preserved in ita ‘thus’ (< PIE *ith₂, cf. the Skt. quotative particle iti); e, i, and o were deleted in some contexts and in others developed into e; and u may have been deleted, as in PIE *maḱsu > mox ‘soon,’ or developed into e, as in *rītu > rīte ‘correctly,’ *tūtu > tūte ‘you (sg. emph.),’ and *-nu > -ne in interrog. nōnne (Weiss 2020: 157–159). If, alternatively, ita derives from *itā (cf. Avestan iθa) by iambic shortening (Weiss 2020: 159), then all five short vowels may have merged as e in absolute final position (Sihler 1995: § 71), suggesting regular open-syllable weakening to i followed by a lowering rule i > e. This lowering rule must have been independent of and superordinate to open-syllable tensing as proposed above (see below, section 5), as those developments would rather have favored e > i. Tensing is moreover independently attested in absolute final position for Dutch and Standard French (Storme & Lancien 2019), further suggesting that the inferred change i > e would be problematic if considered as a laxing process. On the other hand, lowering of absolute final vowels also occurs across languages irrespective of tense-lax distinctions; for example, according to at least one recent analysis, the English reduced vowel [ɨ] is lowered to [ə] in absolute final position (Flemming & Johnson 2007). Something similar may have occurred with Latin i, which (as I have argued) was realized as [ɨ] in non-initial syllables until the Classical period.

5 Reduction and OST

The remainder of this article aims to situate weakening within a typology of historical change, particularly to answer how the general pattern of weakening (V > e in closed syllables, V > e > i in open syllables) can be reconciled with the crosslinguistic trend that vowels in closed syllables are more likely to reduce (Maddieson 1985). This kind of question has been largely neglected in previous treatments of the material, with Sen (2012) being a notable exception.

The first stage of weakening shows unilateral movement towards the center of the vowel space: low central /a/, mid front /e/, and (in open syllables) mid back /o/ collapse in mid central [ə], and close front /i/ and (in open syllables) close back /u/ either do the same or merge in close central [ɨ]. Reduction towards the center of the vowel space can be explained simply in terms of articulatory undershoot: when a speaker’s brain sends signals to the articulators (tongue, lips, and other structures of the mouth), the articulators move to produce the desired sound, but their movements are physically and physiologically constrained. Especially in rapid speech, the articulators may undershoot (i.e., fail to reach) their target positions (Lindblom 1963: 1778). Depending on the articulatory context of the undershoot event, i.e., the nature of the surrounding consonants and vowels, the sounds produced may tend to be more central than their ideal articulations (Lindblom 1963: 1780). This model predicts well the situation in Latin during the first stage of weakening: in unstressed syllables the articulators (primarily the tongue) occasionally failed to meet their targets for the short peripheral vowels [a e i o u] and instead produced central sounds, [ə] and possibly [ɨ]. Over time these centralized articulations became phonetically licensed, then obligatory, resulting in phonemic (historical) change.

By contrast, the second stage of weakening, whereby [ə] > [ɨ] in open syllables (sometimes [ʉ] with rounding), cannot be described so easily as reduction. For one thing, movement away from the schwa, towards the periphery of the vowel space, is unlikely according to a reduction model: in the undershoot theory, it is difficult to imagine why articulators aiming for (say) any of the mid or low vowels would consistently pass through close [ɨ].¹² Additionally, this development takes place exclusively in open syllables, while one might rather expect that vowels in closed syllables are more likely to reduce.¹³

This expectation stems originally from Maddieson’s (1985) theory of closed-syllable vowel shortening (CSVS). Based on data from numerous languages across multiple language families, Maddieson finds that vowels tend overwhelmingly towards a longer articulation before single consonants than before geminates; these conditions are taken as representative of a distinction between open and closed syllables (1985: 210). This tendency coexists in many cases with systems of phonemic length (1985: 212), and in at least two cases (Tamil and Swedish) CSVS is identified as the historical cause of an incipient phonemic contrast (1985: 210–211). A few apparent counterexamples to CSVS, including Japanese, ‘do not seem to … seriously challenge the validity of the claim that CSVS is found across the broad generality of languages’ (Maddieson 1985: 214–216). Reduction as undershoot happens because the articulation time is too short, so one would expect that if vowels are relatively shorter in closed syllables, they are also more likely to be reduced there (Sen 2012: 4). Here the Latin pattern, with weakening more prevalent in open syllables, presents a puzzle.

With regards to this issue, Miller suggests that Archaic Latin stresses were complemented by secondary stress on non-initial closed (heavy) syllables (1972: 487). Across languages, stress is commonly attracted to heavy syllables (as in the CL stress pattern), and vowels under stress are less likely to be reduced (van Bergem 1995: 61, 91, 129). Sen, however, has shown convincingly through analyses of the distribution of syncope (2012: 30–31) and the behavior of closed syllables in second position (2012: 31–37) that secondary stresses did not exist in Archaic Latin.

Another proposed explanation is based on Steriade’s (1999) Licensing by Cue theory, which Sen summarizes as follows: ‘We could argue that in Latin, more vowel features were licensed in closed syllables because the language’s coda allophones provided conditions where they were more robustly cued than when the vowel was before an onset’ (2012: 37). Contra this approach, Sen points out (2012: 38) that coda consonants in Latin appeared to affect the qualities of preceding vowels, rather than providing cues for their features, as with R-Lowering (Parker 1988) and developments before dark l (Buck 1933: § 80.6). He goes on to consider, following a description by Burzio (2007) of reduction patterns in English, that influence may have run in the opposite direction: vowels in closed syllables might have retained contrasts that were lost in open syllables (e vs. i) because of the need to provide perceptual cues for coda consonants, ‘which did not benefit from release cues’ (2012: 40). This theory, however, he considers unlikely as an explanation for the Latin weakening pattern, because unlike reduction in English, closed-syllable weakening in Latin is insensitive to the features of coda consonants (Sen 2012: 41).

Sen proposes instead that Latin belongs to a small but robust class of languages for which vowels are longer in closed than in open syllables (2012: 42), in a pattern than can be called closed-syllable vowel lengthening (CSVL). In addition to Japanese (Maddieson 1985: 214) he includes in this class Finnish and Turkish (Sen 2012: 5). I am not however convinced, even if this class exists, that Latin was a member of it. Few other Indo-European languages show CSVL consistently across environments, and several Romance languages (Spanish, Catalan, French, and Italian) have been shown not to exhibit characteristics of CSVL (Sen 2012: 49). Furthermore, although lengthening in some pre-coda contexts, as before nasals (Sen 2012: 45), is well-attested and phonetically motivated, lengthening before clusters, as Sen proposes for Latin, is less securely documented crosslinguistically (Sen 2012: 23–24, 26).

In light of these doubts I prefer a new, alternative explanation: the second stage of weakening was caused not by reduction, but by the unrelated phenomenon of open-syllable tensing (OST). OST and its frequent counterpart, closed-syllable laxing (CSL), describe a distribution where lax vowels in closed syllables alternate with tense vowels in open syllables (Storme 2019: 303–304).¹⁴ Although the terms tense and lax have been used effectively to describe documented phonological realities (as notably in English), there is disagreement about how they should be defined or whether they can be subsumed in traditional categories, for example height and backness (Ladefoged & Maddieson 1996: 302–303). Roughly speaking, lax vowels are closed to the center of the vowel space, while tense vowels are closer to its periphery (Ladefoged & Maddieson 1996: 306). OST and CSL have been documented more widely than closed-syllable vowel lengthening: Storme (2019: 307) surveys fifteen languages and dialects across four families that exhibit some combination of OST and CSL, including Dutch and three dialects of French. In a study of Southern French, where lax [ɛ œ ɔ] in closed syllables alternate with tense [e ø o] in open syllables, Storme (2017) found that OST-CSL interactions were a more likely explanation for the distribution of mid vowel allophones than reduction.

The results from French suggest that ‘vowel quality can be affected by syllable structure independently of vowel duration’ (Storme 2017: 22). In contrast to reduction, which is motivated by articulatory undershoot, Storme proposes that OST and CSL are motivated by the need for perceptual cues for features of coda consonants: perceptual studies of Southern French confirm the hypothesis that ‘consonant-place contrasts are more distinct after non-low lax than after non-low tense vowels’ (2019: 306). In this way OST-CSL effects echo Burzio’s (2007) theory regarding English reduction. Contra Sen (2012: 41), Storme’s analysis of OST-CSL effects in French and other languages show that the effects of coda consonants on preceding vowels are more widespread than Burzio proposes and need not surface with sensitivity to coda-consonant features, as they do in English.

The second stage of Latin vowel weakening, whereby [ə] > [ɨ] in open syllables, looks like the historical outcome of an OST process. Though not prototypical, it is not unheard of for a language to display OST without corresponding CSL. Storme finds OST without CSL in Dutch and Standard French (2019: 310), making it seem plausible that Latin displays this pattern also: e > i (tensing) in open syllables, but closed syllables preserve a contrast between e and i.

The question remains whether a (synchronic) change [ə] → [ɨ], as I propose occurred in late VOL, can accurately be described as a tensing alternation. As I have mentioned, the definitions of tense and lax vary across languages, but in this case one should allow for comparisons of Latin with languages whose phonetics are more easily accessible. Few languages are described with a contrast between [ə] and [ɨ], but several cases offer support for the idea that [ə] > [ɨ] could represent tensing. In the history of Romanian, phonemic /ə/ was raised to [ɨ] in stressed syllables preceding a nasal (Renwick 2012: 30–32), later merging with other sources into a new phoneme /ɨ/ (Renwick 2012: 27).¹⁵ The Romanian data are not described in terms of tense and lax, but pre-nasal position has been shown to be a potential tensing environment, as in some dialects of North American English (Labov 2008: 174–175). In the northern dialect of Welsh, underlying /ə/ alternates with [ɨ] in final syllables (Hannahs 2013: § 4.2), which historically were stressed (Hannahs 2013: § 3.2) and hence also likely targets for tensing. Russian /i/, which appears as [ɨ] after hard (non-palatalized) consonants in stressed syllables, is lowered to ‘lax’ [ɨ̞] in unstressed syllables (Jones & Ward 1969: 38–39). In Madurese, an Austronesian language of Indonesia, phonemic /ə/ appears as [ɨ] after voiced and voiceless aspirated stops; this alternation has been analyzed as tongue root advancement (Trigo 1991: 129–131), a phonetic feature commonly associated with tensing. Vietnamese â appears only in closed syllables, while its counterpart ơ appears in both open and closed syllables (Hoang 1965: 31); the phonetic representations of these vowels are disputed, but they have been described as central, with ơ closer than â (Hoang 1965: 24). Finally, a close analogue to the Latin pattern has been described in Tera, a Chadic language of Nigeria, where non-low vowels, including phonemic /ɨ/, are lowered in closed syllables (Tench 2007: 231).

Taken together these examples present a convincing case that OST-CSL effects could induce an alternation [ə] ~ [ɨ]. Given the broad distribution of OST-CSL effects across languages, it seems likely that the second stage of weakening (e > i) represents OST rather than closed-syllable vowel lengthening.