1. The problem

Stressed syllables in European Portuguese contrast seven oral vowels, /i, u, e, o, ε, ɔ, a/, which are reduced to a simpler four-vowel set in unstressed syllables. Although /a/ never undergoes any phonological neutralization through vowel reduction, it has a raised allophone [ɐ] which occurs in several environments:

1. (1)

   1. a. In unstressed position

   2. b. In stressed position before nasals and palatals

Despite the lack of neutralization and the predictability of /a/-raising, a number of dialects show a surface contrast between [a] and [ɐ] in stressed syllables. However, this ‘contrast’ appears only in a very limited context, namely in the first person plural between the present indicative and the past perfect indicative of /a/-theme verbs. This is because the two tenses have segmentally identical person-number suffixes [muʃ] and are distinguished on the surface only by the vowel preceding the suffix's /m/. Few authors discuss how to analyse this distinction, and some, such as Mateus and d'Andrade (Reference Mateus and d'Andrade2000: 74, 77), give identical underlying forms for the relevant pairs, despite transcribing the distinction phonetically:

1. (2)

   1. a. /fal + a]_Stmos/ [fɐˈlɐmuʃ] ‘we speak’

   2. b. /fal + a]_St mos/ [fɐˈlamuʃ] ‘we spoke’

Carvalho (Reference Carvalho, Meisenburg and Selig2004: 14, fn. 2) even explicitly sets this phenomenon aside while providing an otherwise very insightful analysis of Portuguese verbal morphophonology:

According to Redenbarger (Reference Redenbarger1981), some scholars have proposed that examples such as (2) be dealt with by positing two separate phonemes /a/ and /ɐ/, but given the extremely limited contexts in which these surface realizations appear to be in opposition to each other, such an analysis overgenerates in terms of the presence of lexical contrasts. Nonetheless, no author to date has offered a satisfactory synchronic explanation of these facts. I will show that, by extending Carvalho's (Reference Carvalho, Meisenburg and Selig2004) analysis of Portuguese verbs, there is an elegant and independently motivated solution to the problem of the [a] ~ [ɐ] contrast.

2. Timing units and height

Let us first consider the phenomenon known as crasis (Portuguese crase), exemplified in (3b). Crasis occurs when two unstressed instances of /a/ come together across a word boundary and coalesce into a single vowel. Unstressed /a/ is typically raised to [ɐ] by phonetic vowel reduction, as in (3a), but crasis blocks this process, and the unstressed vowel is realized as [a].

1. (3) Crasis (Carvalho Reference Carvalho2011: 6, adapted)

   1. a. VR:   /a/ $ \rightarrow $ [ɐ] in unstressed syllables

   2. b. Crasis: [ɐ] + [ɐ'migɐ] $ \rightarrow $ [a'migɐ]  a amiga‘the friend (fem)’

      ['kazɐ] + [ɐ'zṵɫ] $ \rightarrow $ ['kaza'zṵɫ] casa azul ‘blue house’

      ['pagɐ] + [ɐ'kõⁿtɐ] $ \rightarrow $ ['paga'kõⁿtɐ] paga a conta‘pay the bill!’

      ['εɾɐ] + [ɐ'li] $ \rightarrow $ ['εɾa'li] era ali ‘it was there’

Carvalho (Reference Carvalho2011) suggests that if, in an element-based theory, [a] is represented with two resonance elements {A,A}, and [ɐ] is represented with a single resonance element {A}, then crasis can be analysed as the combination of the elements of two instances of [ɐ], creating [a].Footnote ² I have illustrated this in (4).

1. (4) Illustration of Carvalho's (Reference Carvalho2011) analysis of crasis

   

While this approach captures the compositional aspect of coalescence, it is essentially a notational variant of the two-phoneme analysis, with phonological elements standing in for phonological features; it still needs to assume that [a] and [ɐ] have distinct phonemic representations. I propose instead that the effect of crasis results not from the composition of phonological features, but from the double-linking of skeletal positions. In other words, crasis is phonological length.

1. (5)

The analysis in (5) is preferable to that in (4) for several reasons. First, it allows us to remain agnostic about the nature of segmental representations (i.e., whether contrasts are encoded with elements, features or something else). Second, the V-slots involved in coalescence are present in the underlying forms of the unstressed vowels in non-crasis position, since segments require skeletal slots in order to be realized at all; crasis is simply the preservation of both V-slots in a position where two of them have come together.

A two V-slot analysis of the blocking of unstressed /a/ raising also makes sense from a cross-linguistic perspective. It is a common phenomenon for long segments (i.e., those linked to two skeletal slots or their formal equivalents) to be exempt from processes that affect short segments (Hayes Reference Hayes1986). For Portuguese, we can thus say that unstressed /a/ undergoes raising to [ɐ] unless it is linked to two V-slots. The use of virtual long vowels as blockers of vowel reduction has precedent in other Romance varieties as well: Bucci (Reference Bucci2013) proposes that double linking of vowels to the CV-tier results in blocking of reduction processes in Corato Italian.

Finally, it makes sense from the point of view of phonetic implementation. Barnes (Reference Barnes2006), in examining Russian vowel reduction, finds that the height at which unstressed low vowels are realized is gradient. Vowels can be realized along a continuum from [a] to [ə], as a function of phonetic duration. The pattern is that vowels of longer duration are realized at lower height, while vowels with shorter duration are realized higher; the apparent presence of discrete categories in different positions is due to those positions imposing different phonetic durations for independent prosodic reasons. If the same duration-to-height correlation holds for Portuguese, then independently motivated increased duration of unstressed /a/ would cause the blocking of raising to [ɐ].Footnote ³ This is exactly what we find in crasis: two unstressed instances of /a/, each with one V-slot on the skeletal tier, coalesce into a single melody associated to two V-slots. Despite being in unstressed position, the phonological length is spelled out as a single unstressed /a/ with the duration of a stressed one, and thus it is realized as [a], with the same height as would be expected in stressed position.

3. A unified solution

Standard descriptions of Portuguese morphology (e.g., Mateus and d'Andrade Reference Mateus and d'Andrade2000) assume that the verb stem consists of a root and a theme vowel, which can be one of /a, e, i/. This is followed by a tense–aspect–mood (TAM) suffix and a person–number (PN) suffix:

1. (6) Structure of the Portuguese verb

   Verb Stem

   [Root + Theme V] + TAM Suffix + PN Suffix

The indicative forms of the present tense and past perfect tense are said to have phonologically null TAM suffixes, but are distinguished by having different sets of PN suffixes, as seen in (7) and (8). Note, however, that the first person plural suffixes of these two tenses are homophonous.Footnote ⁴

1. (7) Present indicative PN suffixes (Mateus and d'Andrade Reference Mateus and d'Andrade2000: 74)

   1st sing. /o/ 1st plur. /mos/

   2nd sing. /s/ 2nd plur. (none)

   3rd sing. $\emptyset $ 3rd plur. /N/ (nasal autosegment)

1. (8) Past perfect indicative PN suffixes (Mateus and d'Andrade Reference Mateus and d'Andrade2000: 77)

   1st sing. /i/ 1st plur. /mos/

   2nd sing. /ste/ 2nd plur. (none)

   3rd sing. /u/ 3rd plur. /ɾaN/ (nasal autosegment)

Carvalho (Reference Carvalho, Meisenburg and Selig2004) proposes a templatic analysis of the Portuguese verb, where the right edge of the verb stem contains an empty V-slot and the theme vowel is a floating vowel melody. Floating melodies associate by convention to V-slots from right to left. Thus when there is no other melody occupying the stem-final V-slot, the theme vowel associates to it and is realized there:

1. (9) Association of theme vowel melody (Carvalho Reference Carvalho, Meisenburg and Selig2004: 24, adapted)

   1. a.

Other suffixes, such as the present tense first person singular suffix, can consist of only a floating melody. For example, in (10), the stem-final V-slot receives the floating /o/ of the first person singular, and the theme vowel's /e/ melody is left to associate to the V-slot of the stem. Carvalho uses this mechanism in order to account for a kind of ‘harmony’ wherein stem vowels alternate between the upper and lower mid vowels, depending on the suffix appended, causing underlying /ε/ and /ɔ/ to be realized as [e] and [o] respectively:

1. (10) Stem vowel harmony (Carvalho Reference Carvalho, Meisenburg and Selig2004: 24, adapted)

   1. a.

However, Carvalho (Reference Carvalho, Meisenburg and Selig2004) still assumes that both the present tense and the past perfect tense have null TAM morphemes, and as a result, he is forced to set aside the apparent contrast of [a] and [ɐ]. While Carvalho uses morphemes consisting of both melodies and skeletal positions, as well as morphemes consisting of only floating melodies, he does not specifically posit any consisting only of skeletal positions with no melodic content. If we allow for this possibility, then we can arrive at an elegant unified analysis of both crasis and the [a]~[ɐ] distinction. I propose that the past perfect is in fact non-null, but has no segmental melodic content – it consists only of a single V-slot on the skeletal tier:

1. (11) ‘past’

   /–V/

In the present tense forms of first person singular /a/-theme verbs, the theme vowel associates to the extra timing slot of the stem much as seen in (9), and there is no TAM morpheme present in the underlying form. Because the /a/ occurs in stressed position before a nasal, it is predictably realized as the raised allophone [ɐ]:

1. (12)

In the analogous past perfect forms, however, the timing tier has two adjacent V-slots, because of the addition of the past perfect TAM morpheme between the stem and the PN suffix, and the floating theme vowel thus associates to both of these. Just as being associated to two V-slots blocks /a/-raising with crasis, the two V-slots here block raising before the nasal:

1. (13)

In addition to its theoretical elegance, this analysis makes empirically testable predictions about the realization of prenasal stressed /a/ in words like (13): if these vowels are phonologically long, then they should have longer durations than their phonologically short counterparts. This will serve as a focus for future phonetic investigations of European Portuguese.

4. Summary

I have demonstrated that the apparent central vowel contrast in European Portuguese can be elegantly accounted for with a simple addition to Carvalho's (Reference Carvalho, Meisenburg and Selig2004) analysis of the Portuguese verb. This is accomplished through a unified approach to the blocking of /a/-raising, where the different allophones correlate with the presence/absence of two separate skeletal positions. These V-slots are clearly observable in the non-coalesced forms of words undergoing crasis (where two separate segments are realized), and can originate from the past perfect suffix, which derives virtual long vowels and results in an apparently morphologically-conditioned surface contrast between [a] and [ɐ]. Most importantly, this analysis is not only parsimonious, but it makes a testable prediction about the relative phonetic durations of phonologically short and long vowels, which will be tested in future research.