2.1 Syllable structure

Syllables in Huariapano are maximally [CGVC] (where G=glide) and minimally [V], as illustrated in (1).

1. (1)

   

Licit codas are nasals, glides and sibilant fricatives [s ʃ ʂ] (Parker specifies that [ʂ] is ‘retroflex alveopalatal’). Coda [h] is also permitted, but its distribution is non-contrastive and largely predictable (much more on this below).

Content words in Huariapano are minimally [CVC] or [CVː]. Vowel length is non-contrastive: apart from [CVː] words, where vowel length is clearly a reflex of a prosodic minimality condition, long vowels are unattested.

Coda nasals in Huariapano sometimes undergo a variable process of nasal coalescence, in which a /VN/ rhyme is realised as a single nasalised vowel [Ṽ]. These nasalised vowels are prosodic chimeras: they behave as closed [ṼN] rhymes for stress placement, but as open [Ṽ] for coda [h]-epenthesis (see §2.2 and §2.3 for details). I abstract away from these facts in this paper – see Parker (Reference Parker1998a) and González (Reference González2003) for discussion. Vowel nasality is not contrastive in Huariapano, which is atypical for a Panoan language (Shell Reference Shell1965, Loos Reference Loos, Dixon and Aikhenvald1999, González Reference González2003).

2.2 Stress placement

The phonetics of stress in Huariapano are unknown. Acoustic studies of the language are scarce, being limited to Parker (Reference Parker1998b). Furthermore, very few audio recordings have survived, making it almost impossible to directly investigate the phonetics of spoken Huariapano.

In §5 I discuss some methodological issues posed by the lack of good instrumental evidence for stress in Huariapano. For now, I simply adopt Parker's description of the Huariapano stress system without critique (though cf. §4.1.3).

2.2.2 Secondary stress

Unlike primary stress, secondary stress in Huariapano is entirely quantity-insensitive. There are two distinct patterns of secondary stress assignment. In the first pattern, secondary stress is assigned to odd-numbered syllables, counting from the beginning of the word. As (8c) shows, secondary stress does not occur on syllables adjacent to primary stress. This is because Huariapano absolutely prohibits stress clash.

1. (8)

   

This is the most frequent pattern of secondary stress, occurring in around 66% of eligible words in Parker's corpus. Following Parker (Reference Parker1998a), I will therefore refer to the pattern in (8) as ‘regular’ secondary stress assignment.

The other pattern of secondary stress in Huariapano targets even-numbered syllables, also counting from the beginning of the word.Footnote ³

1. (9)

   

I refer to this pattern of stress assignment as ‘irregular’ secondary stress, again following Parker (Reference Parker1998a). It occurs in the remaining 34% of relevant words in Parker's corpus. Since odd-syllable and even-syllable secondary stress are both fairly common, it might be more accurate to distinguish between ‘major’ and ‘minor’ patterns of stress assignment in Huariapano. With this point noted, for consistency with Parker's work I will continue to use the terms ‘regular’ and ‘irregular’ to distinguish these two types of rhythmic stress.

Whether a word manifests the regular or irregular pattern of secondary stress is determined idiosyncratically by the root morpheme of the word. This variation is lexical in nature, and cannot be predicted from the phonological content of the root or following suffixes. See Parker (Reference Parker1998a) and Bennett (Reference Bennett2012) for discussion and exemplification.

Table I outlines the attested patterns of secondary stress in Huariapano, with references to examples in this paper. Given that secondary stress is strictly binary (up to clash avoidance), secondary stress is sensitive to the position of main stress and to the number of syllables in each word.

Table 1

Schematic secondary stress patterns in Huariapano.

I will eventually argue that all of these stress patterns involve the exhaustive parsing of syllables into minimally bisyllabic feet (§4). In particular, I contend that some apparently underparsed syllables are actually integrated into a recursive metrical foot, giving complex structures like [( σ)(σ(σ))]. Medial lapse patterns like [σσσσ] will also prove important, as they support my claim that irregular secondary stress parses out left-to-right iambs, [(σ)σσ(σ)] (§4.1.3 and §4.5).

2.3 Coda [h]-epenthesis

The segment [h] has a narrowly circumscribed distribution in Huariapano. Onset [h] is only permitted word-initially.Footnote ⁴ Word-initial [h] is phonemic: it contrasts with Ø and with other consonants.

1. (10)

   

Coda [h] is permitted in Huariapano, but it must satisfy a number of phonotactic constraints.Footnote ⁵ Furthermore, whenever coda [h] is permitted in the language, it is obligatory. This provides an initial indication that coda [h] is an epenthetic segment.

The restrictions on coda [h] are as follows. First, coda [h] is only allowed before a voiceless obstruent.

1. (11)

   

In this respect, coda [h] closely resembles preaspiration, as first pointed out by Parker (Reference Parker1998a). As I argue below, the interaction of [h]-insertion with syllable structure shows that coda [h] is nonetheless a full-fledged segment in Huariapano.

Second, [h] never appears in a coda cluster. This restriction is entirely expected, as complex codas are prohibited in Huariapano.

1. (12)

   

Patterns like (12) provide a strong argument that preconsonantal [h] is a true coda consonant. The impossibility of *[ʂohʂ.ˈka] indicates that [h] is a full segment, and not just subsegmental preaspiration on an obstruent, [ʰC]. Otherwise, the contrast between *[ʂohʂˈka] and [pohˈʂoj] would be inexplicable, given that the local phonetic environment for [h] is identical in the two cases. (The fact that [ʂ] belongs to a stressed syllable in [pohˈʂoj] is immaterial, because stress does not directly condition the distribution of [h]; see (16).)

Furthermore, preconsonantal [h] is clearly a coda rather than an onset. If [h] were an onset, then tautosyllabic [h]+obstruent clusters must be well-formed, as in [nɰhˈtɰno] ‘day (loc)’. This suggestion runs completely contrary to the observation that Huariapano does not allow obstruent clusters in onsets (1). It also fails to account for the blocking effect of a preceding coda consonant. If [ht] were an acceptable onset, forms like [nɰˈhtɰno] and *[majˈhtiʂo] should be equally grammatical, being parsed [nɰ.ˈhtɰ.no] and *[maj.ˈhti.ʂo] respectively. They are not, precisely because ungrammatical *[majh.ˈti.ʂo] would contain a coda cluster (cf. [maj.ˈti.ʂo] ‘hats’).

The third condition on coda [h] is that it never appears in word-final syllables. This too is expected if coda [h] is a kind of preaspiration: if coda [h] is sponsored by a voiceless obstruent within the same word, then word-final coda [h], which has no licensing obstruent, should be illicit. The ban on complex codas also prevents [h] from being licensed by a tautosyllabic obstruent in final position (13c).Footnote ⁶

1. (13)

   

Fourth, and most important, is that the distribution of coda [h] is rhythmic. Coda [h] only appears in odd-numbered syllables, counting from the left. Furthermore, when coda [h] is licensed in an odd-numbered syllable, it must appear there. These generalisations hold true regardless of where stress falls (with one small caveat; see (17) below).

1. (14)

   

In contrast, coda [h] never appears in even-numbered syllables, even when all other phonotactic restrictions on [h] are satisfied. Compare (14b) and (15): the penult of (14b) is an odd-numbered syllable, and so undergoes [h]-epenthesis; the penult of (15), though otherwise identical, resists epenthesis, because it is even-numbered.

1. (15)

   

The distribution of coda [h] in Huariapano is thus ‘rhythmic’, in the sense that it picks out every other syllable in the word. This is of course strikingly similar to the distribution of secondary stress, which also targets alternating syllables, and also counts from the left edge. This remarkable parallelism, which points to a metrical basis for coda [h], will form the crux of the theoretical issues addressed in this article.

Despite the commonalities between secondary stress and coda [h], stress itself does not directly condition [h]-insertion. Coda [h] may appear in stressed and unstressed syllables alike, and is insensitive to different degrees of stress.

1. (16)

   

Finally, coda [h] is prohibited in word-initial syllables that also bear primary stress, i.e. [#].

1. (17)

   

This is a joint effect of initial position and main stress; neither condition is sufficient on its own to block coda [h], as the examples in (16) show. The generalisation seems to be that word-initial syllables carrying main stress are somehow too prominent to license coda [h] (see Parker Reference Parker1998a, de Lacy Reference de Lacy2001, Bennett Reference Bennett2012).Footnote ⁷

To summarise, coda [h] is only permissible in Huariapano if it satisfies the phonotactic restrictions in (18).

1. (18)

   

Coda [h] occurs wherever these conditions are met. The distribution of coda [h] is thus non-contrastive, predictable and rule-governed: if coda [h] can appear in a particular syllable, it necessarily surfaces there.Footnote ⁸ For this reason, I follow Parker in assuming that coda [h] is always epenthetic. Alternations like (17) and (19) are therefore due to a productive phonological process of coda [h]-insertion.

1. (19)

   

Another argument for an epenthetic treatment of these facts is that coda [h] appears in assimilated loanwords from Spanish, even in the absence of a plausible source consonant.

1. (20)

   

There are nevertheless a few cases where coda epenthesis fails to apply, despite satisfaction of the conditioning criteria in (18).

1. (21)

   

Words with exceptional non-epenthesis are a clear minority in Huariapano: Parker (Reference Parker1998a) reports that 93% of testable morphemes (115/124) show the expected pattern of [h]-insertion. For reasons of space I will not address these exceptions in any detail here; see Bennett (Reference Bennett2012) for arguments that the failure of epenthesis in such forms can be attributed to morphological and/or prosodic factors.

3.1 Stress placement

All prior analyses of Huariapano have assumed that primary stress is assigned in a right-aligned quantity-sensitive trochee. This assumption explains (i) why primary stress is normally limited to a word-final two-syllable window, (ii) why default primary stress falls on the penult in words ending in a light syllable and (iii) why primary stress shifts to word-final heavy syllables.

1. (22)

   

Secondary stress, in contrast, is quantity-insensitive: stress is placed by counting syllables, without any reference to moraic weight. In the regular pattern of parsing, stress falls on the first syllable and every other syllable that follows (up to clash, which is avoided). Since initial stress suggests a left-aligned trochaic foot, a natural assumption is that regular secondary stress parses out iterative, quantity-insensitive trochees from left to right.

1. (23)

   

Irregular secondary stress differs minimally from the regular, odd-syllable pattern: here, stress falls on every even-numbered syllable, again counting from the left. In Huariapano, this is roughly equivalent to counting even-numbered syllables, right to left, from the position of primary stress. (The two algorithms yield different results when primary stress is preceded by an even number of syllables; see §4.1.3.)

1. (24)

   

Past analyses have exploited this near equivalence: irregular secondary stress is taken to be exactly like regular secondary stress (that is, quantity-insensitive and trochaic), but with a non-default right-to-left direction of parsing.

1. (25)

   

Multiplanar approaches thus assume that lexical items may differ in the direction of parsing for secondary stress. Barring some sporadic exceptions, all other parameters of stress assignment remain fixed across lexical items.

3.2 Coda [h]-epenthesis

Recall that coda [h] has a rhythmic distribution: it appears only in odd-numbered syllables, counting from the left. This prosodic condition on [h]-insertion closely tracks regular secondary stress, which also targets odd-numbered syllables. As a consequence, stress and coda [h] coincide exactly in some forms, other phonotactics permitting; see (18) and §4.3. (σ _h is a syllable closed by coda [h].)

1. (26)

   

Inspired by this parallelism with regular secondary stress, multiplanar analyses assume that [h]-insertion is likewise conditioned by syllabic trochees, parsed left to right.

A key premise of this analysis is that [h]-epenthesis, like stress, targets foot heads. Taken with the preceding assumptions about parsing, this correctly limits coda [h] to odd-numbered syllables. It also has the virtue of providing a motivation for epenthesis: [h]-insertion converts open [CV] foot heads into closed [CVh], rendering them bimoraic, in accord with the cross-linguistic preference for heavy stressed syllables (the Stress-to-Weight Principle: Hammond Reference Hammond1986, Prince Reference Prince1990).

For these reasons, multiplanar frameworks adopt the view that coda [h]-epenthesis occurs in the heads of bisyllabic trochees, built left to right. Despite the tantalising similarities between [h]-epenthesis and secondary stress, the assumptions sketched above turn out to be jointly inconsistent with a single-tier treatment of Huariapano phonology. Instead, they lead to the conclusion that [h]-insertion depends on a distinct system of foot parsing, one which bears no relation to the feet behind stress assignment.

The problem is that there are many words in which stress and coda [h] do not coincide. For example, [h]-epenthesis and stress may diverge in even-parity words when primary stress falls on the ultima rather than an odd-numbered penult.

1. (27)

   

Mismatches also occur in odd-parity words. Clash avoidance can block stress assignment on an odd-numbered syllable, but such syllables are still targeted by epenthesis.

1. (28)

   

Examples (27) and (28) demonstrate that epenthesis can occur in syllables that should be left unparsed, given the strictly binary footing needed for secondary stress. Underparsing of this sort will occur whenever an odd number of syllables precede the main stress. Since these apparently unfooted syllables are nonetheless in an odd-numbered position, epenthesis applies unfettered.

The most dramatic mismatches between stress and epenthesis are found in words with irregular secondary stress. Words in this class bear stress on even-numbered syllables, counting left to right. This has no effect on [h]-epenthesis, which is restricted to odd-numbered syllables in all lexical items, regardless of where stress happens to fall. At least some of these mismatches occur in syllables that are unstressed, but nonetheless footed (a point that remains true under the alternative view of parsing I develop in §4).

1. (29)

   

These facts rule out any direct correspondence between stress and coda [h]-epenthesis: mismatches include both cases where coda [h]-epenthesis fails to apply in an otherwise eligible stressed syllable and cases where it applies in unstressed syllables (see e.g. (28a)).

We are thus faced with a conundrum: coda [h]-epenthesis in Huariapano appears to be foot-based, but the feet required to determine the locus of epenthesis are not isomorphic to the feet that determine surface stress assignment. This dilemma leads to the central proposal of multiplanar frameworks, the claim that the phonology of Huariapano makes use of two distinct metrical tiers. One of these tiers determines stress assignment, while the other determines the location of coda [h]-epenthesis.

1. (30)

   

These tiers are ‘metrical’ in having characteristics typical of foot structure. Both tiers host rhythmic alternations, and both show evidence of head prominence (stress assignment on the stress tier; augmentation of foot heads with coda [h] on the rhythm tier). On this view of Huariapano prosody, words are parsed into metrical constituents in two different phonological planes: stress feet ‘( )’ on the stress tier, and epenthesis feet ‘{ }’ on the rhythm tier.

1. (31)

   

The rhythm tier is thus ‘process-specific’ in that it conditions only a single phonological pattern, while also being autonomous from, and inconsistent with, the structural parse needed for stress placement.

4.1 Stress placement: uniform parsing, variable headedness

4.2 Coda [h]-epenthesis targets foot-initial syllables

As discussed in §3.2, multiplanar frameworks assume that coda [h]-epenthesis derives bimoraic foot heads, where the ‘headedness’ relevant for epenthesis is determined on the rhythm tier, independent of stress. This idea has some typological backing, given that there are other languages with trochaic footing that require stressed syllables to be heavy (e.g. Lahiri & Dresher Reference Lahiri and Dresher1999, Mellander Reference Mellander2003; cf. Hayes Reference Hayes1995: 83–84).

On the other hand, footing on the rhythm tier in Huariapano does not otherwise care about syllable weight: foot parsing for epenthesis is quantity-insensitive, just like secondary stress. This argues against the view that coda [h]-epenthesis is motivated by pressure for bimoraic foot heads (as Parker Reference Parker1998a points out).

Here, I offer a different take on the motivation behind coda [h]-epenthesis. If feet are always bisyllabic, and parsing proceeds from left to right, odd-numbered syllables will normally be foot-initial (with one important complication, to be discussed in §4.5).

1. (43)

   

Odd-numbered syllables are, of course, precisely those syllables that are eligible for [h]-epenthesis. Drawing on this result, I contend that [h]-epenthesis occurs in foot-initial syllables, whether or not those syllables are stressed. This derives the parallelisms between coda [h]-epenthesis and secondary stress, because both phenomena are conditioned by exactly the same underlying metrical structure (cf. §3.2).

By assuming that foot boundaries are fixed in words of a given length, regardless of where stress falls, this single-tier analysis also explains why stress and epenthesis only sometimes coincide. Stress and epenthesis align in trochaic feet, where stress is foot-initial, but diverge in iambic feet, where stress is foot-final.

In §6, I justify this analysis on typological grounds. For now, I show that it derives the full distribution of coda [h]-epenthesis in Huariapano without the need for disjoint footing.

4.3 When stress and epenthesis align

As a first illustration, consider cases in which coda [h] is limited to stressed syllables. This occurs in even-parity words (so that exhaustive parsing into bisyllabic feet is possible) in which all feet are trochaic (so that stress and epenthesis coincide on odd-numbered syllables, in foot-initial position).

This configuration requires penultimate primary stress, regular secondary stress (L→R trochees) and an even-parity span preceding the main stress. This pattern is exemplified in (44). Words with final primary stress may involve a mismatch between stress and epenthesis, (σ _h ); I consider them in the next section.

1. (44)

   

Each stressed syllable is foot-initial in (44), and therefore correctly predicted to undergo epenthesis if eligible. This is the simplest pattern to account for, as it involves a perfect correspondence between stress and epenthesis. To fully motivate the claim that [h]-epenthesis targets foot-initial syllables, we now turn to discrepancies between stress and epenthesis.

4.4 Mismatches under iambic footing

Mismatches between stress and [h]-insertion arise in iambic feet. This occurs when main stress is word-final (under pressure from quantity-sensitivity), or when secondary stress follows the irregular (even-numbered) pattern.

4.5 Mismatches due to underparsing

I have so far proceeded on the assumption that footing is strictly and maximally binary in Huariapano. However, there is reason to believe that Huariapano exploits a somewhat richer inventory of metrical structure – in particular, a limited amount of recursive footing. The evidence comes from the distribution of coda [h] in words with an odd number of syllables preceding the main stress.

In words with an odd-parity span, we find that strictly binary footing undergenerates epenthesis in some antepenultimate syllables. As (47) illustrates, the problem is wholly independent of the secondary stress pattern of the word.

1. (47)

   

The examples in (47) should have unfooted antepenults, given a strictly binary system of parsing. Unfooted syllables are clearly not foot-initial, so the ‘trapped’ antepenults in (47) should not be targets for [h]-epenthesis. Nevertheless, these odd-numbered antepenults do contain a coda [h]. This suggests that the assumption of strictly binary parsing needs to be amended in some way.

As hinted above, this issue disappears if we assume that the antepenults in (47) are in fact recursively adjoined to the foot to their right (the foot bearing primary stress).

1. (48)

   

This refinement solves the undergeneration problem posed by strictly binary footing. Prosodically ‘trapped’ antepenults end up being parsed after all, but only as an adjunct in a recursive foot. Importantly, such antepenults are initial in the higher recursive foot derived by adjunction. These syllables are thus correctly predicted to be epenthesis sites, as in (48). Since recursively parsed antepenults are dependent adjuncts rather than prosodic heads, we also correctly expect that they should be unstressed.

I claim that Huariapano exploits recursive adjunction as a last-resort strategy to ensure exhaustive parsing. Without recursive footing, antepenultimate syllables in odd-parity spans would be prosodically trapped, given the inviolable prohibition on degenerate feet. Recursive adjunction thus serves to foot otherwise unfootable syllables. Importantly, there is good evidence that Huariapano prefers exhaustive parsing of words: the existence of iterative secondary stress, which results from the maximal parsing of syllables into feet.

An additional wrinkle is that penults in words of this prosodic shape – which are also foot-initial, under the current set of assumptions – are not eligible for coda [h]-epenthesis. This is consistent with the basic descriptive generalisation that [h]-insertion is blocked in even-numbered syllables.

1. (49)

   

The lack of penultimate coda [h]-epenthesis in (49) can be explained if epenthesis only targets syllables at the edges of maximal feet (Selkirk Reference Selkirk1980, Jensen Reference Jensen2000, Ito & Mester Reference Ito, Mester, Honma, Okazaki, Tabata and Tanaka2003, Yu Reference Yu2003).

1. (50)

   

The intuition here is that epenthesis is limited to syllables that are strictly foot-initial. Syllables at the left edge of a non-maximal foot (σ(σσ)) are also medial within the superordinate maximal foot; as such, they do not qualify as ‘foot-initial’ in the most stringent sense. In this respect, coda [h]-epenthesis in Huariapano is demarcative: it is a segmental cue to the boundaries between successive feet, much like the fortition processes found in Yupik languages (§6.2).

1. (51)

   

In words that necessitate recursive adjunction of trapped antepenults, penults will not be eligible for epenthesis, but antepenults, which are initial in Ft_max, will. Recursive footing thus reconciles the distribution of coda [h] with the claim that [h]-epenthesis is an augmentation process that targets foot-initial syllables.

4.6 Trisyllabic words

I argued in §4.1.1 that trisyllabic words with final stress contain a bisyllabic iamb, e.g. [ʂɰ(roj.ˈʃin)] ‘soul, spirit’. Now that the case for recursive footing has been made, a fuller structural parse for these examples would include recursive adjunction of the antepenult, e.g. [(ʂɰ(roj.ˈʃin))].

While I believe this analysis is correct, there is some question as to whether words like [ʂɰ.roj.ˈʃin] might have carried an initial secondary stress, consistent with the alternative parse [(ˌʂɰ.roj)(ˈʃin)] (note 13). Interestingly, these two parses – recursive (σ(σ)) and non-recursive ()(σ) – predict exactly the same epenthesis sites under my analysis. Coda [h] never appears in final syllables, so in either case epenthesis should target only the initial syllable in an [σσH] word. (Sadly, none of the attested [σσH] words allow us to test this prediction, given the additional phonotactic restrictions listed in (18).)

1. (54)

   

The message is simple: whether [σσH] words have initial secondary stress or not, my single-tier account of Huariapano makes the same predictions regarding the position of epenthesis. These predictions are consistent with the distribution of coda [h] elsewhere (non-final, odd-numbered syllables).

In fact, this congruity holds for all odd-parity words ending in a heavy syllable: the non-recursive parse (55a) has exactly the same left-edge Ft_max boundaries as the recursive parse (55b) (setting aside the final syllable, which is ineligible for epenthesis anyway).

1. (55)

   

It may be that recursive feet are limited to odd-parity words with penultimate stress: given the ban on degenerate feet, there is no way to exhaustively parse a string like [σσσ(ĹL)] without recursive adjunction. Parsing schemas consistent with my analysis are given in (56); form (56b.ii) shows that recursive footing is needed in any case to predict epenthesis in pretonic antepenults.

1. (56)

   

This concludes the heart of my reanalysis of Huariapano. To recap, I have made the major claims about coda [h]-epenthesis in (57).

1. (57)

   

4.7 Is epenthetic [h] moraic?

At various points I have alluded to the idea that coda [h]-epenthesis is domain-initial strengthening at the level of the foot. So far, nothing has been said about how [h]-epenthesis contributes to the salience of the syllables that it targets. One obvious possibility is that epenthetic [h] is moraic, with coda [h]-insertion ensuring that foot-initial syllables will be heavy (cf. Parker Reference Parker1994, Reference Parker1998a, Reference Parkerb, Smith Reference Smith2005).

Though seemingly reasonable, this assumption proves untenable. If coda [h] is moraic, then epenthesis creates (HĹ) and (H) iambs – feet that are very badly formed from the perspective of structural markedness (Prince Reference Prince1990, Hayes Reference Hayes1995, Prince & Smolensky Reference Prince and Smolensky2004).Footnote ²⁰

1. (58)

   

Indeed, many languages actively avoid parsing heavy syllables into the weak, unstressed branch of a foot (e.g. Hayes Reference Hayes1981, Reference Hayes1995, Kager Reference Kager and Roca1997, Bennett Reference Bennett2012, Norris to appear, etc.). Since primary stress is quantity-sensitive in Huariapano, any account of the language that assumes […(H )#] footing should be viewed with scepticism.

Given these difficulties, I suggest that coda [h] is never moraic in Huariapano. My view is that the non-moraic nature of coda [h] stems from a more general property of the language: only word-final consonants sponsor an independent mora. Several important consequences follow from this assumption. Limiting moraic codas to final position means that [CVC#] ultimas are the only heavy syllables in Huariapano (setting aside long vowels, which are exclusive to monosyllabic [CVː] words). This is a valuable result, because it derives the fact that only primary stress is sensitive to syllable weight.

Heavy syllables, being restricted to the ultima, always belong to the right-aligned foot that carries main stress. Non-head feet are then trivially quantity-insensitive, because they never contain heavy syllables. Assuming positionally restricted coda weight thus obviates the need for specialised constraints that enforce quantity-sensitivity for primary stress, but not secondary stress (cf. Parker Reference Parker1998a, McGarrity Reference McGarrity2003, Pruitt Reference Pruitt2012).

It also follows that there are no moraic [h]'s in Huariapano, since (i) only word-final codas sponsor a mora, and (ii) [h] is never word-final.Footnote ²¹ By restricting moraic codas to final position, we derive the fact that only primary stress interacts with syllable weight, while also guaranteeing that coda [h]-epenthesis does not create prosodically ill-formed feet.

There is ample precedent for assuming that moraic codas may be limited to specific positions (see especially Rosenthall & van der Hulst Reference Rosenthall and van der Hulst1999). Of particular relevance are languages with the same finality condition on coda weight that I propose for Huariapano. In Goroa, for example, closed syllables only attract stress when in final position (Hayes Reference Hayes1981). In tandem with these observations, work in Government Phonology and related frameworks has often observed that word-final consonants may behave differently than word-medial codas for a range of phonological phenomena (see Kaye Reference Kaye1990, Harris & Gussmann Reference Harris, Gussmann and Cyran1998, Piggott Reference Piggott1999, Gussmann Reference Gussmann2002: ch. 5 and citations there). Finally, contextual syllable weight is attested in other Panoan languages, making its occurrence in Huariapano somewhat less surprising (Elías-Ulloa Reference Elías-Ulloa2006, Reference Elías-Ulloa2009). I conclude that there is both language-internal and typological support for the claim that only word-final codas are moraic in Huariapano.

This proposal raises an important question regarding the function of epenthesis in Huariapano: if coda [h] isn't moraic, how does it ‘augment’ foot-initial syllables? My claim is that coda [h]-epenthesis enhances the prominence of foot-initial syllables by increasing their raw segmental content. Put differently, epenthesis adds to overall syllable duration (and thus perceptual salience), but does so in a non-moraic fashion (Gordon Reference Gordon2002b).Footnote ²² This is not a novel idea: Beckman (Reference Beckman1998), Hall (Reference Hall2001), Bye (Reference Bye2005), Bye & de Lacy (Reference Bye, de Lacy, de Carvalho, Scheer and Ségéral2008) and Ryan (to appear) have proposed that there is independent pressure to maximise the amount of segmental material contained in prominent syllables, irrespective of moraic weight (cf. the constraint *Head/CV from González Reference González2003, and similar ideas in Munshi & Crowhurst Reference Munshi and Crowhurst2012). Coda [h]-epenthesis in Huariapano, then, is a prosodically determined but non-moraic strengthening process. In §6 I provide other cases of non-moraic augmentation in foot-initial syllables, further supporting this view of rhythmic [h]-insertion in Huariapano. See Bennett (Reference Bennett2012) for an OT implementation of these proposals using positional markedness constraints (Zoll Reference Zoll1998, Smith Reference Smith2005).

A so-far unanswered question is why [h], rather than some other segment, is chosen as the epenthetic consonant in Huariapano. I assume with Parker (Reference Parker1994, Reference Parker1998a) that [h] is chosen as the epenthetic segment for two reasons: first, [h] has no oral place features, and is thus relatively unmarked (e.g. de Lacy Reference de Lacy2006), and second, coda [h] is licensed by a following voiceless obstruent, as it is a species of (heterosyllabic) preaspiration. See Parker (Reference Parker1998a) for an OT formalisation of these views.

6.1 Canela

In Canela (Jê; Central/NE Brazil), intervocalic consonants lengthen before stressed vowels, provided the pretonic vowel is short (Popjes & Popjes Reference Popjes and Popjes1971, Reference Popjes, Popjes, Derbyshire and Pullum1986). Vowel length is contrastive, though it doesn't carry a high functional load.

1. (62)

   

Gemination fails after a long vowel because [CVːC] syllables are banned outright. This is a good indication that consonant lengthening (62b) derives a true ambisyllabic geminate: otherwise, the blocking effect of a preceding long vowel would be rather puzzling.

Gemination thus provides a closing coda for pretonic open syllables, when permitted by general constraints on syllable shape. Stress is typically word-final in Canela (and uniformly so in nouns and verbs), which points toward iambic footing, e.g. [(kuh.ˈhe)]. Taken together, these observations indicate that gemination always closes a syllable in foot-initial position. This is exactly parallel to [h]-epenthesis in Huariapano: foot-initial open syllables are augmented with a closing coda.

The bottom line is that foot structure provides a rationale for why stressed onsets lengthen at all in Canela: foot-initial strengthening. Other explanations fall short on this point. In particular, gemination cannot be driven by a pressure to augment stressed syllables (cf. Bye & de Lacy Reference Bye, de Lacy, de Carvalho, Scheer and Ségéral2008). As argued above, it is clear that geminates are ambisyllabic in Canela, not ‘pure’ onset geminates belonging only to the stressed syllable (Topintzi Reference Topintzi2008). But it follows from this that gemination doesn't alter the structure of the stressed syllable itself, which is [C(X)] in any case. If gemination is triggered by prosodic structure in Canela, it must depend on footing, not just syllable structure or stress alone (cf. Giavazzi Reference Giavazzi2010).

Canela thus provides a striking case of coda augmentation in foot-initial syllables, of the same general sort proposed for Huariapano in §4. I suspect that other instances of this phenomenon are waiting to be identified: for example, Karo and Kaapor (two unrelated Amazonian languages) also have patterns of stressed onset gemination that are amenable to an analysis in terms of foot-initial strengthening (Bennett Reference Bennett2012).

6.2 Yupik

Yupik languages (Eskimo-Aleut; Alaska and Siberia) are well known for having fortition processes that mark foot-initial syllables (Jacobson Reference Jacobson1985, Leer Reference Leer1985a, Reference Leerb, c, Hayes Reference Hayes1995, van de Vijver Reference Vijver1998, etc.). This fortition may involve subphonemic consonant lengthening (with concomitant devoicing), or in the case of Norton Sound Yupik, neutralising changes in consonant manner. Fortition is clearly conditioned by metrical structure rather than stress: since footing is iambic in the Yupik languages, the foot-initial syllables that undergo fortition may be either stressed () or unstressed (LĹ).

Here I focus on Chugach Alutiiq, as spoken on Prince William Sound. According to Leer (Reference Leer1985b), fortis consonants are lengthened and realised with ‘preclosure’, which may involve preglottalisation (Leer mentions a ‘slight hiatus [and] gap in breath flow’ before fortis consonants). Fortition only affects short consonants: there are no fortis–lenis alternations for geminates. Though lengthened, fortis consonants remain audibly shorter than true geminates, so fortition is non-neutralising. Lastly, fortis consonants resist postvocalic voicing, which normally affects singleton voiceless obstruents (though voicing is only partial in such cases).

The distribution of fortis consonants is illustrated in (63) (examples from Leer Reference Leer1985b; ù indicates a fortis consonant).

1. (63)

   

To reiterate a point made above, fortition is clearly conditioned by metrical structure rather than stress. Fortis consonants are found in stressed and unstressed syllables alike. Fortition is likewise indifferent to the stress profile of adjacent syllables. Fortis onsets occur after both stressed and unstressed syllables in (63), as well as word-initially. The presence of stress on the following syllable is similarly irrelevant. Fortition depends only on the position of foot boundaries: it is foot-initial strengthening par excellence.

This pattern of fortition is particularly interesting, because it is non-quantitative, having no effect on syllable weight. In that respect Yupik-type fortition is akin to coda [h]-insertion in Huariapano, though the subphonemic strengthening in (63) is clearly a lower-level process. Other instances of non-quantitative, foot-initial strengthening include allophonic stop aspiration in English (Jensen Reference Jensen2000, Davis & Cho Reference Davis and Cho2003) and phonetic lengthening in Japanese affricates (Shaw Reference Shaw, Trouvain and Barry2007).

As a closing observation, it is worth mentioning that initial consonant fortition is also attested for stressed syllables and word-initial syllables, two uncontroversially prominent positions (Lavoie Reference Lavoie2001, Smith Reference Smith2005, Giavazzi Reference Giavazzi2010). This is as expected, if foot-initial syllables belong to the class of strong positions. For general discussion of domain-initial articulatory strengthening, see Fougeron & Keating (Reference Fougeron and Keating1997) and Keating et al. (Reference Keating, Cho, Fougeron, Hsu, Local, Ogden and Temple2003).

6.3 Russian

In most Central and Southern dialects of Russian, unstressed [a] is permitted only in immediately pretonic syllables (Halle & Vergnaud Reference Halle and Vergnaud1987, Crosswhite Reference Crosswhite2000, Reference Crosswhite2001, Padgett & Tabain Reference Padgett and Tabain2005, Iosad Reference Iosad2012, among many others). This non-uniform pattern of vowel reduction is plausibly foot-based: assuming that footing is iambic in these varieties of Russian, we can conclude that the reduction of unstressed [a] (to [ə]) is inhibited in foot-initial syllables, (σ).Footnote ²⁵

1. (64)

   

It is relevant here that [a] is a highly sonorous vowel, and as such tends to be licensed in phonologically strong positions (e.g. de Lacy Reference de Lacy2004, Reference de Lacy2006, Reference de Lacy and de Lacy2007a). If foot-initial position counts as phonologically prominent, as I propose, then the retention of underlying /a/ in pretonic syllables amounts to the preservation of sonorous vowels in a position of phonological strength – a typologically familiar pattern.

Indeed, in some dialects of Russian the mid vowels /e o/ actually lower to [a] when pretonic, thereby becoming more sonorous even at the cost of neutralisation.

1. (65)

   

This phenomenon can be interpreted as another case of foot-initial augmentation, given that it involves an active increase in the sonority of pretonic vowels.

Interestingly, these patterns of vowel allophony go against the clear typological preference for low-sonority vowels in unstressed, footed syllables (see e.g. Kenstowicz Reference Kenstowicz1996, Gouskova Reference Gouskova2003, de Lacy Reference de Lacy2004). In Dutch, for example, we find a mirror-image skew in vowel reduction: unstressed, but footed, syllables, ( σ)σ, are more prone to reduction than unfooted syllables (Kager Reference Kager1989: 312–317). This apparent discrepancy vanishes once we recognise that, in iambic systems, the push towards foot-initial prominence may trump the preference for low-sonority vowels in weak footed syllables. I conclude that the exceptional behaviour of pretonic vowels in Russian stems from the fact that foot-initial syllables are phonologically strong.

Similar conclusions can be drawn from positionally restricted contrasts in other iambic languages. In San Martín Itunyoso Trique, for example, pretonic syllables host a greater range of consonant, vowel and tone contrasts than other unstressed syllables. See DiCanio (Reference DiCanio2008: chs 2, 5) for details.