4.1 Simple gemination

The alternations in (15) are the simplest manifestation of strengthening, namely gemination of a consonant in the strong grade. The phonological class of nouns represented in (15) contains many nouns with a diphthong or /a/ followed by a fricative or sonorant consonant, and the consonant alternates between single in the weak grade and geminate in the strong grade.

1. (15)

   

The derivation of the weak-grade form involves just basic prosodification, which we assume involves various simple and explicitly formalisable steps, such as morification of vowels, syllable projection and onset creation (see Kahn Reference Kahn1976 and Dell & Tangi Reference Dell and Tangi1992 for similar approaches, Clements & Hume Reference Clements, Hume and Goldsmith1995 for the related notion of single-operation rules and McCarthy Reference McCarthy2007 for cognate concepts in OT-CC). The syllable nucleus in this class of words is bimoraic for segmental reasons: it either contains a diphthong or the vowel /a/, which is usually long. Given a sequence of vowel segments, i.e. a diphthong, the first vowel is morified by (16a), and that mora then projects a syllable by (16b).

1. (16)

   

A further well-known process defining core syllabification assigns an unaffiliated consonant to the onset of a syllable (X′ denotes an X which is unassociated).

1. (17) Onset Creation

   

Moraic support for syllabically parsing the non-head elements of a diphthong may come by attachment to a consonantal mora in strong grade forms (e.g. [guolli]), but when there is no coda consonant (e.g. [guoli]) a mora is directly added at the end of the syllable and linked to the vowel. The following rule guarantees that both elements of a diphthong can be syllabified, by providing a mora that the second half of a diphthong can attach to – association to the mora and syllabic parsing of that mora come about by separate rules.Footnote ⁷

1. (18) Diphthong Morification

   

An unsyllabified mora is then parsed from the right into a syllable.

1. (19) Mora Syllabification

   

There are multiple sources for such a second mora, one being (18), others including lexically specified consonantal moras, as well as the nominative singular suffix . The context where there may be two unsyllabified moras will be discussed in sections concerned with the Q3 alternation.

Finally, the second vowel of a diphthong associates to the second mora in the syllable by Attachment to Mora.

1. (20) Attachment to Mora

   

These rules apply to /kuoli/, which underlyingly contains only segments, giving the syllabified surface output in (21).Footnote ⁸

1. (21)

   

The phonetic interpretation of this output is straightforward: the diphthong is long, since it is bimoraic, and the consonant is short, because it is neither moraic nor ambisyllabic.

In the strong grade, links to the consonant after the stress, which causes lengthening of that consonant. We posit the rule of -docking in (22) to associate to a consonant.

1. (22) –docking

   

Core syllables are erected by the application of (16) and (17), and binary left-headed feet are built on those syllables. In the strong grade, after the medial consonant is morified by (22), it syllabifies by Mora Syllabification (19), linking leftwards to the stressed syllable.Footnote ⁹

The derivation of [guolli] is parallel to that of [guoli] with respect to creation of the core CV syllable via Head Morification (16a), Syllable Projection (16b) and Onset Creation (17), but diverges from that of [guoli], because application of -docking to renders l moraic, which causes it to syllabify as a coda.Footnote ¹⁰ When that happens, the structural description of Diphthong Morification (18) is not satisfied, and that rule is not necessary to provide a position for syllabification of the vowel o. Instead, Attachment to Mora links o to the morphologically provided coda mora. The relevant differentiating steps in the derivation are those in (23), which starts after core syllabification.

1. (23)

   

The nuclear vowel may also be the monophthong /a/, which as we have noted is generally long (the exception is explained in §3.8). An example of such a noun is [gaaffe~gaafe] ‘coffee’. In the morphological weak grade, where there is no consonantal mora which /a/ could link to, a second mora is assigned to /a/, which results in the lengthening of /a/. Similar to the case of Diphthong Morification (18), the low front vowel becomes long by (24), which links /a/ to a second mora, inserting it if necessary.

1. (24) Low Front Vowel Lengthening

   

The formulation is to be interpreted as always linking a second mora to /a/, and inserting a second mora if there is not already a following mora that /a/ can link to. This rule completes the derivation of weak-grade [gaafe] (from /kafe/), starting in (25) after core syllabification.

1. (25)

   

The derivation of strong-grade [gaaffe] differs minimally from this, in that it is the nominative singular case suffix which links to /f/ and causes gemination, providing the second mora for /a/, resulting in (26).

1. (26)

   

4.2 Overlong geminates

Some stems (with the same segmental type as in the previous section) have an underlying medial moraic (geminate) consonant. The geminate appears as such in the weak grade, and in the strong grade the nominative singular suffix associates to this already moraic consonant. This additional mora indirectly results in overlength of the consonant; there is, additionally, a related vocalic length alternation associated with the nucleus.

1. (27)

   

This phonological class of nouns also includes the diphthongs /ea oa/, discussed below.

Ordinary syllabification accounts for weak grade forms such as [guossi], which is analogous to the strong grade of [guolli] in §4.1, except that the consonant's mora is lexically present. The second vocoid of [guossi] syllabifies by linking to this lexically specified consonant mora, just as o of strong grade [guolli] syllabifies by linking to .

The weak grade of nouns such as [gollli~goolli], with a length-alternating monophthong, requires further comment. The type of noun discussed in §4.1 contains a diphthong plus the inherently long monophthong /a/, therefore the nucleus of such nouns is always long. The present class of nouns is entirely complementary with respect to the class of monophthongal nuclei, in that the nucleus is never /a/. The question arises as to the source of vowel lengthening in weak grade [goolli]. Core syllabification plus Mora Syllabification would result in the following representation.

1. (28)

   

The surface form [goolli] results from a rule which lengthens monophthongs before a syllable-final continuant, by spreading a syllable-final consonant's mora to the vowel. Pre-continuant Lengthening (29) does not spread , a point discussed in §4.3. The relevant continuant consonants in the language are [l r m n ɲ v ð f s ʃ h] (we consider [h] further below).

1. (29) Pre-continuant Lengthening

   

Monophthongs are not lengthened before moraic stops, which occur in nouns such as [loddi~lotti] ‘bird (nom sg~acc sg)’, discussed in Bals et al. (Reference Bals, Odden and Rice2007).

This lengthening rule applies as in (30) (continuing from (28)).

1. (30)

   

Thus vowel length in weak grade [goolli] is the predictable result of spreading the mora of a coda continuant to the preceding vowel.

In the strong grade, is inserted after the head mora of the stressed syllable. The last mora, which is lexically linked to the medial consonant, is parsed into the coda, resulting in (31).

1. (31)

   

Pre-continuant Lengthening cannot apply between o and syllable-final lexical μ, since that mora is not adjacent to the mora of the o, and also cannot apply between o and , because that mora is not syllable-final.

In order to be phonetically interpreted, material must be fully parsed into the representation, and in (31) is not syllabified, since the syllable already dominates two moras. Whether or not this mora remains present in the derivation is an empirical question (addressed below). What is important at the moment is that we do not rely on a third mora to represent consonantal overlength in the output. If a satisfactory analysis of the facts is possible within the confines of a model where syllables dominate no more than two moras, then there would be no evidence to support the less constrained theory, which posits that syllables may also dominate three moras. We assume that Stray Erasure, proposed in McCarthy (Reference McCarthy1979) and subsequent literature (Steriade Reference Steriade1982, Itô Reference Itô1989, Lin Reference Lin1997, van Oostendorp Reference Oostendorp2008), operates at least by the end of the phonological derivation, so that floating moras are not available to the phonetic interpretation component. Rather than rely on surface floating moras or trimoraic syllables to represent overlength,Footnote ¹¹ we rely on the distinction between a consonant which exhausts a mora (32a) and one which shares a mora (32b).

1. (32)

   

A mora implies some period of time, and when a consonant exhausts a mora, as in (32a), it has all of the timing of the mora, and is therefore ‘overlong’, resulting in a triply long consonant. In (32b), the timing entailed by μ is shared between [o] and [l], but in (32a), all of the timing entailed by the mora is implemented on [l], making it longer.

The derivation of the strong grade of nouns with a high vowel diphthong [ie uo] such as [gŭosssi] is similar to that of nouns with a monophthong: the nucleus is short in the strong grade. The mechanism for deriving this fact must be explained, since ordinarily one would expect a diphthong to have two moras in a language with a vowel-length distinction, especially given rules like (18) and (20). While the mid diphthongs [ea oa] are usually long (short [ĕa ŏa] are explained in §4.4), the morification of the high diphthongs [ie uo] is more variable in North Saami. The two vocoids of a diphthong are generally each parsed into a separate mora, but in certain contexts the more general V₂ morification process (20) is pre-empted by a rule parsing the mid vowel to the head mora in the syllable, which yields the distinction long (bimoraic) [ie uo] vs. short (monomoraic) [ĭe ŭo]. Full analysis of all contexts where short diphthongs are formed involves numerous segmental and prosodic facts, and is outside the scope of this paper. The basic generalisation in North Saami is that high vowel diphthongs ([ie uo]) are parsed as short in two contexts, depending on the nature of the following material, whereas mid vowel diphthongs are so parsed in a single, distinct context, before overlong [hh]. For the class of stems relevant to this section, high diphthongs are parsed as short in case the head mora is immediately followed by two non-head moras (thus [gŭosssi] ‘guest (nom sg)’, [mĭehhki] ‘sword (nom sg)’, [bĭemʔmɑt] ‘to feed’), which correspond to traditional Q3. The other context for creating short high diphthongs before so-called Q2 clusters is discussed in §4.7.

Given these conditions, the vowels [e] and [o] in [mĭellli] and [gŭosssi] respectively are linked to the head mora, pre-empting Diphthong Morification (18) and Attachment to Mora (20) as the means of syllabifying the second vocoid. The rule creating short diphthongs in this context, where only the moraic properties of the affected syllable are relevant, is formalised as follows.

1. (33) Short Diphthong Formation 1

   

Basic syllabification yields (34a). Rule (33) then links [o] to the head mora in (34b), forming a short diphthong. Unsyllabified deletes via Stray Erasure, and the surface form is (34c).

1. (34)

   

Since [ŭo] is associated to one mora, it is short, and since the consonant is moraic, it is geminate, but because no other segment is linked to the coda mora, [s] is phonetically extra-long.

Since short high diphthongs are created by (33), the question arises as to how our rules would handle an underlying form like /keas_ʼi/ in the nominative singular, where the initial vocoid of the diphthong is not high, and so could not undergo (33). Core syllabification and -docking would result in (35a). The unparsed consonantal mora will be deleted by Stray Erasure applied to prosodic elements, giving (35b). Floating /a/ can then associate to the syllable-final mora by Attachment to Mora, resulting in (35c).

1. (35)

   

Since the vowel /a/ is not parsed to the preceding mora by (20), which only targets diphthongs with a [+high] first vocoid, it remains liable to later syllabification to the second mora. In lieu of a specific rule addressing mid-vowel diphthongs (as exists for high-vowel diphthongs), our analysis predicts that stems with lexically moraic surface-geminating consonants preceded by mid-vowel diphthongs will not exhibit a surface alternation in the nucleus or consonant, and the output would be [geassi] for both the strong and weak grades.

The problem is diagnosing such underlying representations. There is a phonologically defined class of nouns, such as [geassi] ‘one who pulls’, [neavvu] ‘work tool’, [hearraa] ‘lord’, [ʃeaffɑ] ‘chef’, which do not have a surface consonant alternation in the nominative singular vs. accusative singular, with mid diphthongs in their non-final syllables. If these stems are underlyingly /keas_ʼi, neav_ʼu, hear_ʼaa, ʃeaf_ʼɑ/, they should not have a surface gradation alternation, since normal morification of diphthongs would link /a/ to the following consonant, eliminating the mora-exhausting association that is overlength. However, another way to look at such stems is that they are simply lexical exceptions to gradation. There are two facts which show that the invariant Q2 nature of some such nouns is not due to lexical exceptionality.

Non-alternating [geassi] is the agent nominalisation of /keasi/ ‘pull’, and agent nominalisations are actual exceptions to gradation – no agent nominalisation alternates between strong and weak grades in the nominal paradigm. It is an important grade-related generalisation about these nominalisations that the medial consonant is generally in the strongest grade, Q3, thus [tˢĭellli] from /tˢiellɑ+i/ ‘yell’, [vĭehhki] from /viehkɑ+i/ ‘run’, [stŏahhki] from /stoahkɑ+i/ ‘play’, each form being both nominative singular and accusative singular. Support for our prediction that a bimoraic parse of mid-vowel diphthongs overrides overlong gemination comes from the fact that roots with the diphthongs /oa ea/ before geminate consonants always have a surface (invariant) Q2 consonant rather than a Q3 consonant. Consider the distribution of surface vocoid and consonant lengths in the following set of agent nominalisations (which are both nominative singular and accusative singular).

1. (36)

   

The only context where the medial consonant in an agent nominalisation is surface Q2 rather than Q3 is when the consonant is a geminate (not aspirated or glottalised) and preceded by [ea oa]. This is the context where the burden of realising the quantity distinction rests solely on duration. There is no need to complicate the account of agent-nominalisation morphology with a disjunctive description of quantity properties, when our analysis automatically predicts that Q3 is realised the same way as Q2 in this context.

Additional evidence shows that surface non-alternating nouns such as [neavvu] and [ʃeaffɑ] do in fact undergo gradation, whose effect is often (but not always) obscured by the regular bimoraic syllabification of mid-vowel diphthongs. Since the presumed Q3 quantity of the geminate is thwarted by parsing /a/ to the consonant's mora, and overlong geminates are representationally possible only if the consonant exhausts its mora, then the prediction of our analysis is that a process monophthongising /ea oa/ could eliminate the segment /a/ which deprives a Q3 geminate of its normal Q3 realisation. Monophthongisation is an active part of North Saami phonology, and occurs in the illative singular of stems ending in the back vowels /ɑ u/. The illative singular suffix is /i/, which causes short /ɑ/ to delete: illative singular is also a morphological category supplying the floating mora which triggers strong grade consonants (see (11)). When added to a stem ending in /ɑ u/, the illative singular causes monophthongisation of a preceding diphthong.

1. (37)

   

We can now contrast stems such as /healɑ/, with a mid diphthong and a lexically non-moraic consonant which manifests a length alternation between nominative singular and illative singular vs. accusative singular, with stems such as /neav_ʼu/ and /ʃeaf_ʼɑ/, which manifest a length alternation, but only in the illative singular, when the diphthong is eliminated.

1. (38)

   

As we predict, superficial ‘exceptions’ like [neavvu] and [ʃeaffɑ] are not actually exceptions, but are complications, namely that the realisation of Q3 as overlength of a geminate is overridden in a well-defined phonological context, which follows from the fact that /a/ in a diphthong links to the syllabically parsed mora of a would-be overlong consonant by Attachment to Mora.

4.3 Q3~Q1 gemination

While strong-grade Q3 geminates alternate with weak-grade Q2 geminates after certain nuclei, Q3 and Q1 alternate after some monophthongs, but not diphthongs or /a/. Such stems have an underlying non-moraic consonant (hence weak-grade Q1).Footnote ¹²

1. (39)

   

The class of such nouns is limited to stems with a monophthong nucleus which excludes /a/: that is, such nouns are the short monophthongal counterpart of the nouns in §4.1.

The appearance of Q3 in the strong grade rather than Q2 (the usual strong-grade quantity paired with Q1) is an automatic consequence of our analysis. In the weak grade, the surface form simply shows the direct prosodification of the underlyingly short consonants and vowels.

1. (40)

   

The consequence of adding to such a stem is that the medial consonant becomes moraic, thus geminate. This results in the syllabification in (41).

1. (41)

   

Pre-continuant Lengthening (29) does not apply here, since it is not triggered by . Its non-application can be guaranteed if lexically specified moraic consonants are syllabified in coda position before (29) applies, and syllabification of the output of -docking (22) takes place after (29), so that is not syllable-final at the stage when lengthening might apply, and forms like [gusssaa] do not have a syllable-final moraic continuant when (29) applies.

Under an analysis where overlong geminates are bimoraic in a trimoraic syllable, it would be difficult to explain the source of the spontaneous third mora here – there is not even a requirement in the language that stressed syllables be bimoraic (e.g. the first syllable of [dolaa]). Thus not only does our analysis account for overlong consonants without requiring three moras, but positing a third mora for the sake of representing overlength would be an ad hoc complication in the grammar.

4.4 Ordinary preaspiration

Similar patterns arise when the medial consonant is an underlying aspirated stop, but there are segmental complications. As seen in (42), such stems (which, like the case of the simple gemination stems in §3.1, all have a bimoraic nucleus with the vowel /a/ or a diphthong) have a preaspirated consonant in the strong grade and a voiced consonant in the weak grade. We analyse this as an alternation between (moraic) aspirated stops (realised as preaspirated) and non-moraic deaspirated stops, as indicated in the intermediate forms, which abstract away from certain segmental changes.

1. (42)

   

While the underlying segmental properties of the consonant are best revealed in the strong grade (aspiration is preserved), the prosody is best revealed in the weak grade – the consonant is underlyingly non-moraic. In the weak grade, the consonant syllabifies as the onset of the second syllable. Being non-moraic, it loses aspiration, and between vowels it is phonetically voiced, or lenites to [ð] in the case of /tʰ/.

1. (43)

   

Informally stated, there is a requirement in North Saami that laryngeal features be licensed by a mora – we discuss this below. Because of this, when a consonant is non-moraic, its laryngeal features are deleted, resulting in the intermediate forms koati, neapi. Assuming that intervocalic voicing of unaspirated stops is a phonological rule, the mapping from neapi to [neabi] involves the application of a voicing rule. The realisation of koati as [goaði] involves a further general process in the language, which spirantises a single unaspirated alveolar stop t to [ð] between vowels.

In the strong grade, links to the intervocalic consonant. The prosodic pattern of these stems is parallel to that of [guolli~guoli], discussed in §4.1.

1. (44)

   

The phonological account could end here, saying that because the consonant is moraic, aspiration is licensed, and is phonetically timed as preaspiration. We pursue a phonological treatment of preaspiration and deaspiration, and show the advantages of treating preaspiration as a phonological matter.

In the strong grade, stems with underlyingly non-moraic aspirated stops behave parallel to stems with underlyingly single consonants which alternate between singletons and geminates. Strong grade [goahti] and [guolli] morify the first vowel of the diphthong by core syllabification, and attaches to the post-tonic consonant, which syllabifies as the coda of the first syllable. The second element of the diphthong syllabifies by linking to the coda mora via Attachment to Mora (20). Strong grade [aahtˢi] and [gaaffe] lengthen their first-syllable vowel by Low Front Vowel Lengthening (24), linking to the low vowel. Weak grade [aad^zi] and [gaafe] similarly undergo (24), which inserts a mora for /a/ to attach to.

We posit that derivation (44) continues by splitting an aspirated stop into a sequence of h in the coda plus stop in the onset, resulting in (45).

1. (45)

   

This requires a segmental sequencing process, where the feature [spread glottis] (which is an unordered part of an ambisyllabic consonant) becomes a separate segment, ordered with respect to the remains of the consonant. We propose that this be accomplished by (46).

1. (46) Aspiration Linearisation

   

By this rule, a single segment (a feature structure dominated by a root node) splits into two segments, the first having only the feature [spread glottis], i.e. [h]. Prosodic reorganisation of /tʰ/ is phonologically necessary: were each of the segments resulting from segment-splitting to retain all of their original prosodic linkages, crossed lines would result, as in (47).

1. (47)

   

This is prevented by eliminating at least one of the crossing lines. Since *[ht] is not a word-initial onset in North Saami, we may assume that the association of [h] to σ₂ is eliminated. Since there are no word-final clusters at all, we may also assume that the association of [t] to μ₂ is eliminated. It is difficult to pin down the exact output of (46) by appeal to other phonological rules which refer to the precise derived prosodification of [h] or [t].

We now return to the nature of the laryngeal licensing requirement, in light of the proposal that aspiration is converted to a full coda segment [h]. In (45), [h] exists on the surface because it is prosodically licensed in the syllable via a coda mora. The linearisation rule does not require the aspirated segment to be moraic in order for the laryngeal feature to be segmentalised, so the question arises as to what would become of h when the stop is not moraic, that is, when (46) produces (48).

1. (48)

   

In lieu of any other syllabification process rescuing the stranded consonant, floating [h] simply deletes by Stray Erasure.Footnote ¹³ Thus, under our linearisation approach to aspiration, the laryngeal moraic licensing requirement reduces to the fact that without a mora for coda syllabification, the segmentalised laryngeal feature has no place in the phonological representation, and is therefore subject to Stray Erasure.

4.5 Long preaspiration

Parallel to the overlong~long gemination pattern analysed in §4.2, there is an alternation between long and regular preaspiration, with alternations in nuclear length. The stems of (49) have an underlyingly moraic stop, so their aspiration is always licensed and realised.

1. (49)

   
   Footnote ¹⁴

All diphthongs and monophthongs can occur in the first syllable nucleus of such stems.Footnote ¹⁵

The weak grade (e.g. [geahʧu]) syllabifies in a manner analogous to the strong grade of simple aspirates (e.g. [goahti]), the difference from (44) being that here the mora on the aspirated consonant is lexically present. In either case, the second member of the diphthong links to that mora by Attachment to Mora, and is thus syllabifiable independent of the first vowel. In the case of monophthongs, as in [heehke], the underlying short monophthong is lengthened by Pre-continuant Lengthening, which spreads the mora of the syllable-final continuant h to the preceding vowel. This provides further phonological evidence for the phonological treatment of preaspiration as a sequence of [h]+stop – aspiration behaves as a coda continuant with respect to the lengthening rule. We shall see in §4.7 that, in contrast, the segmentalised realisation of glottalised nasals as glottal stop+nasal yields a laryngeal segment that is, as expected, not treated as a continuant for purposes of vowel length.

The derivation of strong grade [mĭehhki] proceeds in a fashion analogous to that of [gŭosssi] in (34). Because of the presence of the lexical and inflectional moras, core syllabification and Short Diphthong Formation (33) create a short diphthong. The moraic stop syllabifies as the coda of the first syllable (as well as the onset of the following syllable), giving the first step in (50). Segmentalisation of aspiration realises the aspiration as the exclusive consonant of the coda, followed by Stray Erasure of the unparsed mora.

1. (50)

   

Note that when coda [h] exhausts the mora, preaspiration is long – the [h] receives all of the timing implied by the mora. When, as in weak-grade [miehki], the coda [h] shares a mora with the preceding vowel, [h] is phonetically shorter.

Since short diphthongs are formed only by a specific rule (like (33)) parsing the second element of a diphthong into the first mora of a syllable, we must now consider forms with an aspirated moraic stop and a mid diphthong, as found in examples such as [sĕahhkɑ] and [ŏahhpɑ]. It should be noticed that although [ea] and [oa] are generally long, there is one context in the language where they can be short, namely in the strong grade of lexically moraic aspirated stops. This indicates that at some point after segmentalisation of aspiration, something specific happens to the /a/ element of the diphthong, i.e. it associates with the mora to its left.

1. (51)

   

The rule of Leftward Capture (52) links an unsyllabified vowel to the mora to its left, just in case the free mora to its right associates with h.

1. (52) Leftward Capture

   

If the following mora is not free but is syllabified (as in the case of strong grade [goahti] or weak grade [seahkɑ]), then /a/ does not join with a mid vowel under a single mora. As we will see in §4.8, if the following free mora is linked to a consonant other than h (as in [boaɲʔɲi]), /a/ does not form a short diphthong.

4.6 Q3~Q1 aspiration

Parallel to the gemination alternation discussed in §4.3, there is an alternation between a strong-grade Q3 preaspirate and a weak-grade Q1 voiced stop after an underlying short monophthong.

1. (53)

   

All such stems have a short monophthong, excluding /a/.

The weak-grade form involves just basic CV-syllable prosodification, and deaspiration of the moraically unlicensed aspiration (since there is no coda mora).

1. (54)

   

With the case suffix added in the nominative singular, this renders the medial consonant moraic, which gives a syllabic position for [h] resulting from aspiration. The monomoraic nucleus has only a single vowel, so Diphthong Morification (18) would not motivate adding a second mora. Segmentalisation of aspiration as a coda [h] in the strong grade does not trigger Pre-continuant Lengthening (29) of the nuclear vowel, since, as we noted above, is not syllabified until after (29) applies, so the conditions required for (29) are not satisfied at this stage. Consequently, the vowel remains short, and later syllabification of coda [h] exhausts the second mora.

1. (55)

   

Because no other segment is associated with , all of its timing is realised on [h], making it phonetically overlong.

4.7 Preglottalised nasals: Q1~Q2

The plain nasal~preglottalised nasal alternation is seen in (56) after a long nucleus.

1. (56)

   

As mentioned above, the distribution of diphthong length is complex, and we begin with invariantly long nuclei – those which, for segmental reasons, always have long vowels. All mid-vowel diphthongs are long, as is /a/.

Parallel to the analysis of preaspirated stops, preglottalised nasals are treated as underlying nasals with a laryngeal feature, one that is mapped to a glottal stop segment, thus /mˀ/→[ʔm]. We start with the strong grade, where the laryngeal feature can be realised. Since an additional mora, , is available and associates to the medial consonant, the nasal is also parsed as the coda of the first syllable. Core syllabification and coda syllabification result in the first step of (57). Attachment to Mora then results in the syllabification of /a/ in the second step. Segmentalisation of the laryngeal feature finally realises glottalisation as a coda glottal stop.

1. (57)

   

At this point, it looks as though a generalised version of Aspiration Linearisation (46) could handle this final step: but below, we give evidence that there is a separate rule, one which specifically assigns glottal stop to syllable-final position (rather than postvocalic position), as specified by (46).

In the weak grade, when there is no mora for the consonant to associate with, it is exclusively in the onset.

1. (58)

   

When the laryngeal feature is sequenced as the segment [ʔ], since it cannot be syllabified due to the lack of a coda mora, it is deleted, and surface [deanu] results.

1. (59)

   

This analysis extends easily to nouns such as [laaʔnɑ~laanɑ], with the vowel /a/. In the weak grade, an additional mora must be assigned to the nuclear vowel /a/, and in the strong grade, provides the mora required by Low Front Vowel Lengthening (24). Thus there is significant parallelism between stems with medial glottalised nasal and those with preaspirated stops.

Differences also arise when the preceding nucleus is short. As seen in (60), monophthongs in this phonological class are invariantly short (except for /a/, covered above, which is long). We also indicate in our transcriptions that there is (surface-predictable) lengthening of the coda glottal stop, a fact whose explanation has significant implications for the phonological analysis.

1. (60)

   

The consonantal alternation here is exactly the same as that in (57) and (58). The nuclear vowel does not lengthen even when it is followed by a coda consonant, since stops do not trigger Pre-continuant Lengthening. Nothing causes a sharing of the coda mora with the preceding vowel, so the vowel remains short. Thus the surface representation of [joʔʔɲaa] is as in (61).

1. (61)

   

The structural difference between [deaʔnu] (57) and [joʔʔɲaa] (61) makes the prediction that there should be a difference in the duration of glottal stops after long vowels and diphthongs vs. those after short vowels. This is indeed the case, though it is a non-contrastive phonetic detail that a glottal stop is longer after a short nucleus, and this length is perceptually somewhat less salient than the degree of overlength associated with exhausting the mora in the case of overlong [h] or other fricatives. In a set of 71 words of the structure (C)V(V)ʔNV with long and short nuclei, glottal stop had a mean duration of 318 ms after a short nucleus and 235 ms after a long nucleus. Total nucleus+[ʔ] duration is indistinguishable for words like [laaʔnɑ] and [lɑʔʔɲaa], being on average 417 ms.

Nouns of this phonological type (medial glottal stop+nasal), with a high-vowel diphthong, have an invariant long nucleus in some contexts and an alternating-length nucleus in others. The nucleus is always long when the vowel of the second syllable is /ɑ/.

1. (62)

   

This is as expected: it is parallel to the data in (56).

High diphthongs are short in the strong grade when the vowel of the second syllable is either /i/ or /u/.Footnote ¹⁶

1. (63)

   

The data of (63) illustrate the second context where high diphthongs can be short, namely before a consonant cluster when the vowel of the following syllable is not /ɑ/.

It is unclear what the best way is to capture the fact that /ɑ/ does not trigger this short diphthong formation rule. In other areas of the phonology, there is evidence for associating /ɑ/ with prosodic weakness, for example, the epenthetic vowel in the essive case is [i] in odd-numbered (stressed) syllables ([ʧaalim-i-n] ‘writing’, [vuogɑs-i-n] ‘suitable’, [ɑhhkið-i-n] ‘boring’) but [ɑ] in even-numbered (unstressed) syllables ([beaivɑðɑhhk-ɑ-n] ‘sunshine’, [nuorɑmuss-ɑ-n] ‘youngest’, [monemuʤʤ-ɑ-n] ‘kidney’). Rather than refer directly to the presence of two consonant segments, we refer to this fact indirectly via the requirement that the consonant following the unparsed diphthong element must be exclusively associated to one syllable – that is, it is not geminate (the notation in (64) indicates that there is no association between C and any item to the right of μ).

1. (64) Short Diphthong Formation 2

   

This rule pre-empts Attachment to Mora, which would otherwise parse the diphthong of guoʔmi as long, by associating [o] to the mora of [ʔ].

There are other contexts where Short Diphthong Formation 2 applies, especially in all stems with obvious medial consonant clusters and the requisite following vowel context, illustrated below and documented more extensively in Bals et al. (Reference Bals, Odden and Rice2007).

1. (65)

   

The rule does not parse a diphthong as short before /i u a/, even though a consonant cluster appears to follow before geminate consonants ([guolli], [guossi]) and preaspirated stops ([muohtu], [miehki]). What distinguishes the stems which are subject to (64) from those such as [guolli, muohtu], which are not, is that at this point in the derivation, stems which shorten the diphthong have consonant clusters, but stems like [guolli, muohtu] have single consonants shared between syllables.

1. (66)

   

This correlation is obvious in the case of [guolli] vs. [gŭowllu], following standard assumptions regarding the monosegmentality of geminates. The difference between [gŭoʔʔmi] and [muohtu] follows from the fact that segmentalisation of glottalisation applies before and feeds (64), whereas segmentalisation of aspiration follows (64); thus a preaspirated consonant is a single consonant when (64) applies. This provides one piece of evidence that segmentalisation of aspiration and segmentalisation of glottalisation are separate processes, the latter being formalised in (70) below. We consider below additional data showing a difference in surface realisation in Q3 preaspiration vs. preglottalisation.

4.8 Interrupted nasals: Q2~Q3

Parallel to the Q2~Q3 alternations with geminates and preaspirated stops, there is an alternation between interrupted nasals and glottalised nasals. In the strong grade a glottal stop is flanked by identical nasals, and in the weak grade we find a glottal stop followed by a nasal.

1. (67)

   

We also include the verbal forms [bĭemʔmɑt] ‘to feed’ (strong grade) and [bieʔmɑt] ‘feed (2sg pres)’, which are morphologically related to the noun [bĭemʔmu], to further illustrate the connection between diphthong length and the vowel of the following syllable.

We derive the [NʔN~ʔN] alternation from one underlying segment, a glottalised nasal – parallel to our derivation of [hC] sequences from monosegmental /Cʰ/ – via segmentalisation of glottalisation. The prosody of this alternation is parallel to the [hhC~hC] preaspiration alternation, in that [NʔN] behaves like the overlong version of a glottalised nasal, but the rule segmentalising the glottal feature as a glottal stop is not the same as the one segmentalising aspiration. Whereas aspiration is sequenced to appear after the vowel, and thus entirely before the stop, glottalisation is sequenced to appear at the end of the syllable, which causes splitting of the nasal.

The weak grade of nouns having this alternation is identical to the strong grade of the Q1~Q2 nouns such as [deaʔnu~deanu] discussed in the preceding section, being distinguished only by the fact that the mora resulting in surface [ʔN] is lexically present. The more interesting and phonologically distinct case is the strong grade, where both lexical and inflectional moras are present. In the case of [gunʔni], initial morification and syllabification yield (68).

1. (68)

   

The nasal is associated to three positions: , the following lexical mora and σ₂. The rule segmentalising [constricted glottis] does not directly eliminate these associations, and just positions the glottal stop at the end of σ₁, not after the vowel, as is the case with preaspiration. The emerging glottal stop associates to the syllable-final mora, so that it is followed by no other segmental material in the syllable. Because existing associations to prosodic tiers are preserved, this causes a splitting of the nasal. This is a process very similar in effect to the mitosis of a single multiply linked /b/ into two b segments in Tiberian Hebrew [sibbeːß] ‘he surrounded’ from /sb/, discussed under the rubric of Tier Conflation (see McCarthy Reference McCarthy1986). Splitting of n into two segmental instances is the result in North Saami, with the first nasal being postvocalic in the first syllable and the second nasal linking to the onset of the second syllable.

1. (69)

   

Stray Erasure eliminates the unassociated , which plays no further role in the grammar. Closer scrutiny of the details of diphthong length below provides evidence that the unsyllabified mora is actively deleted in the phonology, supporting the application of Stray Erasure.

Glottal stop creation can be derived by (70), which separates [constricted glottis] from the segment bearing it, linearising it into syllable-final position.

1. (70) Glottalisation Linearisation

   

This rule differs from Aspiration Linearisation (46) both in terms of ordering ((70) precedes Short Diphthong Formation 2 (64), but (46) follows (64)), and the position to which the laryngeal feature is moved – postvocalically in the case of aspiration, but syllable-finally in the case of glottalisation.

The derivation of [bĭemʔmu] is similar, except that the nucleus contains a diphthong with a high vowel. Because of the prosodic structure following the diphthong – the two non-head moras required for Short Diphthong Formation 1 and the coda stop+full vowel condition for Short Diphthong Formation 2 – either rule is applicable, and we expect the diphthong /ie/ to be short.

1. (71)

   

When the diphthong has a mid-vowel first element, as in [boaɲʔɲi], /a/ would not join with the head mora in a short diphthong, because only a high vocoid conditions general parsing of V₂ to the head mora, and /a/ in a mid diphthong links to the head mora only before long preaspiration. After basic prosodification and glottal stop linearisation, structure (72) results.

1. (72)

   

Stray Erasure eliminates floating μ₂, and subsequently Attachment to Mora applies, giving the surface output in (73).

1. (73)

   

We observe that three segments are associated to the final mora: the theory has no direct limit on the number of segments that can be under a mora, only indirect limits derived from the nature of the rules parsing segments to moras.Footnote ¹⁷

4.9 The underlying representation of laryngealised consonants

In light of the fact that deaspiration and deglottalisation are explained by assuming a sequential representation of such consonants, interacting with Stray Erasure, a natural move to consider is that even underlyingly, preaspirated and preglottalised consonants are sequences of segments. There are a number of arguments against such a move.

The most straightforward way to implement this would be to assume that, segmentally, stems have their Q2 form, e.g. /koahti/, /seah_ʼkɑ/, /lɑʔɲaa/, /pieʔ_ʼmu/. If a coda mora is available, it will map to the laryngeal, preventing its deletion. For stems such as /seah_ʼkɑ/, which have an underlyingly moraic consonant, we may presume that the mora is lexically attached to the laryngeal. One crucial aspect of morification under this account would be that a consonant cannot be parsed to the coda by association to a vowel's mora: the coda laryngeal is morified either lexically (for stems like /seah_ʼkɑ/ and /pieʔ_ʼmu/) or by association of (in the strong grades of /koahti/ and /lɑʔɲaa/). Representative examples of preaspiration are given in (74).

1. (74)

   

The first challenge which an underlying sequence account faces is the problem of relating Q2 and Q3 preglottalisation, i.e. the alternation between [boaɲʔɲi] and [boaʔɲi]. Given the underlying stem /poaʔɲi/, the problem is explaining the copying of a nasal to get [boaɲʔɲi]. The input–output relation involved through core syllabification, -docking (22) and Mora Syllabification (19) is given in (75).

1. (75)

   

It is crucial that the required nasal insertion rule precede Stray Erasure – it is specifically triggered by a bimoraic glottal stop, but Stray Erasure will eliminate one of the moras. The formalisability of such a rule depends on certain theoretical assumptions. The process could be stated as a direct consonant copying rule such as (76).

1. (76)

   

Subsequently, the inserted nasal must link to a syllabified mora, after Stray Erasure. This of course runs counter to research within autosegmental phonology which has sought to limit the power of rules, so that apparent insertion of an identical segment is the result of a spreading rule (see Clements Reference Clements1985, Hayes Reference Hayes1986, McCarthy Reference McCarthy1988, Clements & Hume Reference Clements, Hume and Goldsmith1995).

Construing this process as a spreading rule faces a major difficulty, which is that such a rule would involve crossed association lines, involving the relationship between the target mora and the root node of the glottal stop.

1. (77)

   

Thus it would be necessary to construe the hypothesised nasal spreading as being the result of multiple rules spreading lower-level segment constituents, such as nasality spread (since intervening glottal stop would have no specification for nasality) and place spreading (assuming that glottal stop is placeless).

A quite significant problem with treating preaspirated and preglottalised consonants as underlying consonant clusters is that actual underlying clusters behave rather differently from preaspirated and preglottalised consonants. While a full analysis of gradation in stems with consonant clusters is beyond the scope of this paper, the gradation pattern of such stems can be considered briefly. As seen in (78), whenever the stem has a medial cluster, in the strong grade the first consonant is somewhat longer (it is the only consonant linked to ) and the second consonant is a singleton, whereas in the weak grade the first consonant is shorter and C₂ is geminated, i.e. both consonants of the cluster are linked to a coda mora, and the second consonant is also in the onset of the following syllable.Footnote ¹⁸

1. (78)

   

A typical representation of this distinction is that in (79).

1. (79)

   

In other words, when the stem contains a consonant cluster, an extra mora is always mapped to the syllable, and in the weak grade it is assigned to the rightmost consonant, so that the cluster is always syllabifiable, and is thus phonetically realised. This is manifestly not the pattern followed by preaspirated and preglottalised consonants: if no lexical or suffixally determined mora is available for parsing derived ʔ or h, then that segment is deleted, e.g. weak grade /muotʰu/→[muoðu] and /tuomˀu/→[duomu]. If stems of the type /muotʰu/ and /duomˀu/ had underlying clusters, we would expect both consonants to always be parsed, as is the case with any other stem with a consonant cluster, and the gradation alternation should involve gemination of the final consonant in the weak grade, again as is the case with any other stem with a consonant cluster. Thus, if preglottalised and preaspirated consonants were underlying consonant clusters, we would incorrectly predict the following.

1. (80)

   

5.1 Laryngeal properties

Our analysis of preaspirated consonants in North Saami is they are underlyingly aspirated stops which split into [hC] clusters, with [h] being syllabified in coda position just in case there is a prosodic position (mora) for it. The treatment of preaspirated consonants has received a fair degree of attention in the literature (Garnes Reference Garnes1976, Thráinsson Reference Thráinsson1978, Clements & Keyser Reference Clements and Keyser1983, Hermans Reference Hermans, Hulst and Smith1985, Hayes Reference Hayes1990, Buckley Reference Buckley1994, Keer Reference Keer1999, Morén Reference Morén1999, Ringen Reference Ringen1999, Helgason Reference Helgason2002, Hansson Reference Hansson2003, Silverman Reference Silverman2003, Kehrein & Golston Reference Kehrein and Golston2004), but there is no consensus as to their proper representation or origin. Discussions have been significantly influenced by the facts of Icelandic, the best-studied example of preaspiration. It is generally agreed that in Icelandic, the difference between preaspiration and postaspiration is predictable and can be subsumed under one phoneme type, namely aspirated stop, which alternates prosodically in how aspiration is realised (e.g. [jøːtʰYYn] ‘giant (nom sg)’, [jøhtnɪ] ‘giant (dat sg)’). Transcriptions of forms disagree in a manner that indicates a lack of clarity as to what the phonological output is. One can find transcriptions of the type [jøhtnɪ] (Thráinsson Reference Thráinsson1978), which implies full-segment status, [jøʰtnɪ], which implies ‘property of the stop’ status (Einarsson Reference Einarsson1945), or even [jøøtnɪ] (Hansson Reference Hansson2003). Hansson (Reference Hansson2003: 64) proposes that languages can differ in their treatment of preaspiration: ‘Faroese preaspiration appears to be a genuinely sub-segmental property (like postaspiration), whereas in Icelandic it is clearly a full-fledged moraic segment [h]’.

A theory of prelaryngealisation (subsuming preaspiration and preglottalisation under one theory) must address two aspects of the problem: what are the representations, and how is one type of representation mapped to another? This presupposes, of course, that two types of representation are possible. One line of theorising on laryngeal properties of consonants, seen in Golston & Kehrein (Reference Golston and Kehrein1998) and Kehrein & Golston (Reference Kehrein and Golston2004, ms), is that laryngeal features are not properties of specific segments, but are unordered properties within a prosodic constituent, and distinctions such as [pha] vs. [pʰa] are purely orthographic.Footnote ¹⁹ In contrast, standard representational theories since before SPE allowed for the possibility of a contrast between bisegmental [ph] and monosegmental [pʰ]. See Kehrein & Golston (Reference Kehrein and Golston2004) for restrictiveness arguments in support of an unordered properties model, based on the lack of contrast between [ʰp] and [pʰ], or [ph] and [pʰ], within a particular prosodic position. Golston & Kehrein (Reference Golston and Kehrein1998: 335) state that ‘there is no contrast between pre- and postglottalization or between pre- and postaspiration, just as there is no contrast between pre- and postnasalization’. Ladefoged & Maddieson (Reference Ladefoged and Maddieson1996: 73) observe: ‘we do not know of any language in which it is necessary to regard preaspiration as a feature required for distinguishing underlying forms’.

Kehrein & Golston (ms: 2) ‘assume that laryngealized sounds (tʰ, t’) are phonologically and phonetically indistinct from laryngeal clusters (th, tʔ) within an onset or coda'. The equivalence of [ph hp pʰ ʰp] is reduced in Kehrein & Golston (Reference Kehrein and Golston2004) to the representational assumption that laryngeal features are unordered with respect to supralaryngeal properties of segments.Footnote ²⁰

1. (81)

   

The claim in Kehrein & Golston (ms: 1) is that the cross-linguistic differences in pre- and post- realisation of aspiration and glottalisation ‘result from three prosodic considerations: sonority sequencing, syllable-final fortition, and the avoidance of syllable-final glottalization’. Thus the postaspiration realisation in Icelandic [phaːr] ‘pair’ follows the standard increasing-sonority requirements of onsets, and preaspiration in [jøht.nɪ] reflects decreasing-sonority requirements of codas.

One of the most significant challenges for restrictive theories of laryngeal representation such as those of Kehrein & Golston is posed by Kashaya (Buckley Reference Buckley1994), which possesses, on the evidence of behavioural patterns, preaspirated/preglottalised and postaspirated/postglottalised consonants, as well as combined pre- and postlaryngealisation, and [hC/ʔC] clusters.

1. (82)

   

Golston & Kehrein (Reference Golston and Kehrein1998) addresses this language, and their response to the facts is instructive of how the theory relies on a particular notion of ‘contrast’. It is correctly noted that Kashaya does not present equivalently rich phonetic outputs within a syllable position. Preaspirated stops are realised as [h.C] clusters, thus preaspirated stops neutralise with /hC/ clusters. Preaspiration is also phonetically absent in utterance-initial position, exactly where /h/ could not be syllabified in the coda of a preceding syllable. Furthermore, /hC/ clusters are underlyingly in complementary distribution with preaspirated consonants – both are present in Buckley's analysis, but do not underlyingly contrast. The distinction between hC and preaspiration is thus a fact about deeper phonology, not surface phonetics. Kashaya phonetic forms can indeed be entirely described in terms of a contrast [h.t] vs. [.th].

Golston & Kehrein's analysis of Kashaya underscores an important difference in the goals of the present paper, compared to the former theory. Their theory concentrates on ‘contrast’ as a property of phonetic outputs. As Kehrein & Golston (Reference Kehrein and Golston2004: 325) put it, quoting Ladefoged & Maddieson (Reference Ladefoged and Maddieson1996: 2), ‘our study focuses on the elements ‘that are known to distinguish lexical items within a language’, i.e. on laryngeal contrasts involving voicing, aspiration and glottalisation that account for minimal pairs'. Our concern, on the other hand, is the proper representation of the phonological units at all levels of the phonology. We would also be concerned with accounting for the morphophonemic patterning of Kashaya roots beginning with preaspirated consonants, even though they are always mapped to plain stops or hC sequences.

As we have shown above, the representational nature of preaspiration in North Saami changes in the course of the derivation, and thus the presumption of behavioural uniformity throughout a language is falsified. In this respect, our analysis of preaspiration in North Saami is parallel to typical generative phonological analyses of Icelandic preaspiration, involving the mapping from a more abstract single segment /Cʰ/ with a laryngeal feature to a surface cluster [hC]. It is difficult to see how the theory of Golston & Kehrein could reconcile surface patterns of aspiration realisation which favour a sequence analysis of preaspiration with the failure of preaspirated and preglottalised consonants to behave like true clusters such as /lt jt wt nt/ in terms of C₂ gemination, and with the failure of preaspirated consonants (but not preglottalised nasals) to behave like clusters under Short Diphthong Formation 2 (64). That is, preaspirated stops in North Saami act both like single segments and like clusters, whereas the Golston & Kehrein theory of prosodic laryngeal representations does not countenance such dual behaviour.

North Saami also poses a problem for the claim of Kehrein & Golston (ms: 19) that ‘aspiration and glottalisation never uncouple from sonorants, but always surface in whatever syllable the sonorant surfaces in’, and (ms: 15):

Their theory predicts uniform onset and intervocalic postlaryngealisation, thus [nˀa.nˀa] and [nʰa.nʰa], never *[ˀna.ˀna] or *[ʰna.ʰna]. It is unclear how such irrevocable coupling of laryngeal feature and sonorant is to be accomplished. However such a tautosyllabicity requirement might be enforced, this forbidden pattern is exactly what happens in North Saami.Footnote ²¹

We have shown that a crucial factor governing the appearance of preglottalisation in nasals is whether the consonant is moraic, which clearly indicates that the glottal stop would have to be in the coda. Given the claim that the laryngeal feature is always in the same syllable as the sonorant, it follows that the nasal would also have to be in the coda, hence under Golston & Kehrein's restrictions, [lɑʔɲaa] must be parsed as [lɑʔɲ.aa]. This is a highly improbable conclusion, since the language does not allow any onsetless syllables except at the beginning of the word. It is also unclear how to reconcile the statement that laryngeal features ‘always surface in whatever syllable the sonorant surfaces in’ with forms like [bĭemʔmu], since the correspondents of the nasal appear in two different syllables.

We conclude that the unordered cluster theory of laryngeal properties is not tenable at all levels of the grammar, but we leave open the possibility that the theory may be correct in its claims about phonetics.

5.2 Length and syllable structure

A central argument of this paper is that, despite surface ternary durational differences in consonants, it is unnecessary to posit trimoraic syllables, and that limiting syllables to two moras is enlightening especially when it comes to understanding surface length trade-off relations between vowels and coda consonants.

Non-linear theories of length generally treat length as association to two prosodic positions (see Davis Reference Davis and van Oostendorp2011, Odden Reference Odden and van Oostendorp2011). Moraic theory somewhat complicates that relation for intervocalic geminates, in that linkage to a mora in one syllable and onset position in the other represent intervocalic geminates. The representation of strictly coda geminates has been most problematic: see Odden (Reference Odden, Lehiste and Ross1996) for discussion.Footnote ²² The fundamental question of interest to us is how a moraic coda consonant can be integrated into a syllable that already contains a long vowel, since such syllables have impelled others to posit trimoraic syllables.

Morén (Reference Morén1999) proposes a representation for weight-bearing coda vs. weightless coda, which as he points out, are ‘sometimes referred to as final geminates’ (1999: 22), e.g. in Hungarian, where final geminates (moraic, weighted consonants) are dominated by a mora, whereas a non-geminate consonant directly links to the syllable.

1. (83)

   

This contrasts with the type of representation assumed in Hayes (Reference Hayes1989), where a weightless coda C is represented as in (84a) and a weight-bearing coda C is as in (84b).Footnote ²³

1. (84)

   

We take as our point of theoretical departure the principle that a maximally bimoraic syllable suffices universally, though we do not address every proposal for trimoraic syllables, e.g. the proposal of Hayes (Reference Hayes1989) for trimoraic syllables in Estonian. Certain kinds of syllable structure pose a challenge for a moraic analysis which limits syllables to two moras, namely syllables which have long vowels plus a geminate consonant coda, i.e. [XVːCː] syllables. An example of such a language is Hungarian, which has words like áll [aːll] ‘stand’ vs. ál [aːl] ‘false’. Morén's (Reference Morén1999) moraic account of Hungarian relies on trimoraic syllables.

1. (85)

   

A similar analysis is given to Finnish in Sherer (Reference Sherer1994) and Kiparsky (Reference Kiparsky, Cairns and Raimy2011). The question is whether facts such as the existence of [aːll] in Hungarian are also consistent with restricting syllables to two moras.

The assumption that a coda geminate consonant is dominated by a mora does not automatically lead to the conclusion that the consonant is dominated by its own mora. Following the analysis which we have motivated for North Saami, a coda geminate after a long vowel can share a mora with the preceding vowel, thus [aːll] can be represented as (86).

1. (86)

   

Mora-sharing where the onset and first vowel are dominated by one mora has been proposed in a number of moraic subtheories, such as Hyman (Reference Hyman1985), Zec (Reference Zec1988) and Buckley (Reference Buckley1994), but is surprisingly rare in the literature with respect to vowels and following tautosyllabic consonants. Such mora-sharing representations have been employed in work by Maddieson (Reference Maddieson1993), Hubbard (Reference Hubbard1995) and Broselow et al. (Reference Broselow, Chen and Huffman1997), but only to handle differences in how languages can differ in vowel and consonant duration patterns.Footnote ²⁴Hubbard (Reference Hubbard1995) argues that Luganda and Runyambo differ in their phonetic realisation of long vowels before NCV sequences, proposing that in Luganda the nasal is (in surface representations) exclusively in the onset of the following syllable, as in (87a), whereas in Runyambo the nasal is also in the onset but shares a mora with the preceding long vowel, as in (87b).

1. (87)

   

The motivation for this distinction is a phonetic difference, i.e. that in Runyambo long vowels before NC are proportionally not lengthened as much as they are in Luganda. This follows from the representation, so that in Runyambo, but not in Luganda, part of the timing of the second mora is assigned to the consonant. Broselow et al. (Reference Broselow, Chen and Huffman1997) give a similar analysis for the phonetic difference between Hindi and Malayalam.

In Malayalam (88a), a long vowel before a coda consonant is shorter than a long vowel with no coda consonant, whereas in Hindi (88b), long vowel duration is not affected by a coda consonant. This fact is attributed to different representations of coda consonants in Hindi vs. Malayalam: Malayalam coda consonants share a mora with a preceding vowel, whereas Hindi coda consonants have an independent mora.Footnote ²⁵ The possibility that coda consonants in Malayalam are non-moraic is not considered by Broselow et al. (Reference Broselow, Chen and Huffman1997).

1. (88)

   

Ham (Reference Ham1998) similarly investigates the phonetic connection between length and duration given the possibility of mora-sharing.

The above authors have identified and exploited a representational power of moraic theory: a mora may dominate one segment, as in the second mora of (88b), or it may dominate two segments, as in the second mora of (88a). The phenomena which have motivated these representational differences are details of phonetic implementation, and are differences between languages, not within a language. In principle, if the representational theory allows a mora to dominate either one segment or two, then it follows that such a difference could exist within a single grammar, and could be available to categorial phonology as well as phonetic implementation. North Saami is such a language, where representations similar to (88b) in having a consonant that exhausts the mora result in Q3 consonants, and ones similar to (88a) in having a consonant which shares a mora with a preceding vowel result in Q2 consonants.

By allowing both one-to-one and one-to-many mappings between mora and segment, the theory sanctions more possibilities than would be the case if only one-to-one mappings between mora and segment are allowed.Footnote ²⁶ Then the question arises why such contrasts are not common – why are more languages not like North Saami? We believe that the answer lies in how such representations would come about. The surface contrasts arise through interaction between rules of morification and syllabification plus a floating mora present in various affixes. Very simply, if a language lacks a source for prosodic opacity (a surface-unpredictable distinction between one-to-one and one-to-many mappings), then it will not have such contrasts on the surface.

It is an empirical question whether languages like Hungarian and Finnish, which are claimed to have trimoraic syllables based on surface forms such as Hungarian [aːll] and Finnish [aluun] ‘beginning (gen sg)’, actually require such syllables, a question which can only be answered by a detailed investigation of the phonology of such languages. That investigation must take into consideration the possibility that coda consonants may be moraic and yet share a mora with the preceding vowel. We have demonstrated in the preceding sections that this is not only conceivable, but enlightening for North Saami.