Download Chapter 4: Syllable Structure and more Exercises Phonetics and Phonology in PDF only on Docsity!
Chapter 4: Syllable structure
P´eter Szigetv´ari
Continuous human speech can be sliced into speech sounds, represented on paper by IPA symbols. Our question is whether these segments can be grouped into larger units that are not necessarily meaningful (ie the groupings are based purely on the sound shape of the elements), what these larger units are, how they are organized, and whether they are of real use in phonological theory.
4.1 Our topic
Although the term syllable sounds familiar, it will be useful to first identify what phonologists mean, and what they do not mean, by this term.
4.1.1 Syllables vs. morphemes
The existence of groups larger than segments but smaller than words is obvious: in the string catlike the first three segments, k, a, and t form a unit larger than a single segment, and so do the last few segments lajk. That their structure is similar: both have a vowel at the middle, surrounded by consonants at both edges. Units like kat and lajk are called syllables. Incidentally, these two units are meaningful, they are morphemes. But syllables and morphemes are not necessarily the same: for example, butter consists of two syllables, syllable of three, and category of four, but all three are a single morpheme, while cats is one syllable, but two morphemes. The sound string rokIt may be analysed as a single morpheme (rocket), or two morphemes (rock it), but since the phonological shape is identical, this string is two syllables in either case. syllabifi ation (the segmentation of sound strings into syllables) is related to the phonological shape of the word, while analysing into morphemes is dependent on meaning.
4.1.2 Syllabification vs. hyphenation
At a first glance, the terms syllabification and hyphenation may be synonymous. We must draw a clear distinction between the two, however.
Hyphenation is an orthographical device. The point of separation within a written word is a printer’s convention. For example, the word English is usually hyphenated as Eng-lish by British, but as En-glish by American printers, similarly to the different spelling conventions in the case of colour and color. Such conventions are to a large extent arbitrary.
Syllabification, on the other hand, is a theoretical issue: in normal speech syllable boundaries are not expressed by a pause (nor are morpheme boundaries, for that matter).
Accordingly, there is room for debate on where syllable boundaries fall: some analyst would argue that letter should be syllabified as in (1a) because this is what explains certain facts, while another comes to the conclusion that it should be as in (1b), since this is what fits her principles.
(1) Possible syllabifications of letter
a. lett.er b. le.tter
(Note that the word is hyphenated as let-ter, which suggests that the consonant belongs to both syllables, it is ambisyllabi .) This shows that hyphenation and syllabification are different businesses. Note also that hyphenation is indicated by hyphens, while syllable boundaries are represented by dots, as in (1).
4.2 The structure of the syllable
Accepting that the notion of the syllable is a useful one, let us see its structure.^1 (The lowercase Greek sigma, σ, represents the syllable by convention.) We may encode the fact that the consonants at the beginning of a syllable, as well as the vowels after them, are more closely related to each other than the consonant(s) to the vowel(s), by packing them into syllabic constituents. The syllabic constituent typically holding the vowel of a syllable is called the nu leus (or the peak). In some languages the nucleus may be occupied by certain consonants. The consonants preceding the nucleus occupy the onset, those that follow it are in the oda. For reasons to be explained in §4.4.1 and §4.5.1, the nucleus and the coda together form a fourth syllabic constituent, called the rhyme. Lines of a poem rhyme when they contain the same sounds from the last (stressed) vowel to the end. It is not an accident that this is also called rhyme: it is the identity of the rhyme of the last syllables that make them rhyme: e.g., cat and bat rhyme, but cat and cap, or cat and cut do not.
(2a) shows the tree diagram of the complex syllable grajnd (grind ) and (2b) that of kat (cat).
(2) Syllable trees for grind and cat
a. σ
R
O N C
g r a j n d
b. σ
R
O N C
k a t
The syllable grajnd in (2a) is special in that all of its constituents contain two segments, they are branching. It is more usual for syllabic constituents not to branch, but to contain only a single sound, as in cat in (2b).
(^1) Note that with this we already take it for granted that syllables do have an internal structure, although there is an alternative option, namely, that they do not.
(4) The sonority profiles of grind and ray
9 ◦ 8 ◦ 7 • • 6 • • 5 4 • 3 2 1 • • 0 g r a j n d r E j
Being monosyllabi (having one syllable), both of these words exhibit one peak in their sonority profile, here shown in white. bisyllabi (two-syllable long) words will obviously have two sonority peaks, as salad, and trisyllabi (three-syllable long) words three, as mineral, both shown in (5).
(5) The sonority profiles of salad and mineral
9 ◦ 8 ◦ ◦ ◦ 7 ◦ 6 • 5 • • 4 • • 3 2 • 1 • 0 s a l @ d m I n @ r @ l
We have already mentioned that a nucleus may be occupied by a consonant. This is illustrated by the pronunciations of level (lEv@l and lEvl) in (6).
(6) The sonority profiles of level with schwa and with syllabic l
9 8 ◦ ◦ ◦ 7 6 5 • • • ◦ 4 3 • • 2 1 0 l E v @ l l E v l
The sonority profiles of lEv@l and lEvl both have two peaks. While in the first case, these two peaks are occupied by two vowels, E and @, in the case of lEvl, the second syllable has no vowel. Nevertheless, since the word-final l of this word constitutes a sonority peak itself, lEvl is also bisyllabic. A consonant which is a sonority peak (which is in the nucleus) is called a syllabic consonant.
If we believe that a given segment has to unambiguously belong to one of the neigh- bouring syllables, we need a way of deciding whether this should be the first or the second. That is, in the cases of (6), whether the syllabification is lE.v@l or lEv.@l. Neither of these options violates the sonority sequencing principle. To make a principled decision, we need further guidance.
4.3.2 Onset maximization
The onset maximization prin iple says: if a segment may belong to both the coda of the first and the onset of the second syllable, it belongs to the onset of the second syllable.
The principle gives preference to the syllabification lE.v@l over lEv.@l. A single in- tervo ali consonant (i.e., a consonant between two vowels) is always the onset of the second syllable, rather than the coda of the first: VCV is syllabified V.CV and never VC.V.^3
The situation with consonant clusters is more complex. The sonority profile of a consonant cluster may be falling, rising, or level, as (7) illustrates. (Examples: lp as in alpine, pl as in apply, mn as in chimney.)
(^3) Note that by maximizing the onset we inevitably minimize the coda. Thus the principle could equally be called “coda minimalization principle”. As we are going to see at the end of this chapter, there is a universal preference in human languages for onsets over codas.
4.3.3 Word edges
Many phonologists say that the onset-hood of a consonant cluster can be tested at the beginning of words. That is, if a given cluster occurs word initially (which is an empirically easily decidable question), then it also occurs syllable initially (which, as we have seen, is a theoretical issue). To take the example at hand: the syllabification of atlas cannot be a.tlas, because words in English do not begin with the cluster tl.
If we extend the argument in the other direction, and claim that word-initial clusters are all potential syllable-initial clusters, we will end up claiming that a string like dIst@ns (distance) is syllabified as dI.st@ns, since st is possible word initially, therefore syllable initially, so onset maximization lobbies for this division. Note, however, that sonority sequencing is violated by a syllable beginning with a fricative (whose sonority index is 2) followed by a less sonorous plosive (with a sonority index of 0). Further extending the word-edge test to the end of words, one is forced to discard onset maximization. It is a well-known fact of English that checked vowels (I E a o 0 and stressed @) do not occur at the end of words. Consequently, some analyst might be tempted to claim, they do not occur at the end of syllables either. Accordingly, a string like lEm@n (lemon) cannot be syllabified as lE.m@n, since this would create syllable-final E, which should be impossible, since word-final E is impossible, it must be syllabified as lEm.@n. To achieve this syllabification overriding the output of onset maximization, a oda apture rule is needed, which resyllabifies an onset consonant into coda position. Only stressed syllables are strong enough to capture the onset of the following syllable, and to force it into their own coda, so the noun r´ecord r´Eko:d is resyllabified to rec.ord, but the verb rec´ord rIk´o:d is not, it remains re.cord.
The same consideration underlies English hyphenation conventions, cf. lem-on, with short E in the first syllable vs. de-mon, having a free vowel dIjm@n, or the noun rec-ord, again with short E in the first syllable, as opposed to the verb re-cord, with reduced @ in the first syllable.
Although it offers a trivial explanation of the syllabification of atlas, the idea that word boundaries and syllable boundaries should coincide is far from obvious. On the practical side, it leads us into several difficult situations, two of which are presented below.
If it is impossible for a syllable to end in a sound that words cannot end in, then we cannot syllabify words like p´arIS (parish), f´ErIj (ferry), m´Ir@ (mirror ), h´@rIj (hurry),or s´or@w (sorrow ), since the syllable boundary before the r in these words is inhibited by the fact that words cannot end in a checked vowel, while r is also impossible word finally, so the boundary cannot be after it either. Another difficulty with the variable syllabification of intervocalic consonants is with l-darkening and r-dropping. Both of these phenomena are formulable by syllabic constituency: both processes occur in the syllable coda, but not in the onset (see §4.4.1 and §4.4.2 below). Yet neither process is sensitive to the “length” of the preceding vowel— whether it’s free or checked—, which is unexpected if the syllabic affiliation of a consonant depends on the status of this vowel: VCV is VC.V if the first vowel is checked.
In addition to these empirical counterarguments, it is not even a theoretical necessity that syllables begin and end exactly where words do. Another suprasegmental unit, the
rhythmic foot ignores word boundaries, and easily incorporates segmental material from two neighbouring morphemes, or words. Consider, for example, the sentence Amanda should ignore the children, its morphological and metrical structure shown in (9).
(9) A m´anda # should # ig n´ore # the # ch´ıldren
A foot contains a stressed syllable and the following unstressed syllables. If there are two stressed syllables next to each other, the first foot will only contain a single syllable. If two stressed syllables are separated by many unstressed syllables, the first foot will be very long.
In this sentence, foot boundaries ( ) and word boundaries (#) do not coincide, except for the last foot/word. This is because when determining the foot boundaries word boundaries are not taken into consideration. Therefore the two coincide only when two consecutive words or the last word of a sentence (ch´ıldren) begin with a stressed syllable. We also see that the sentence begins with a degenerate foot, one which does not begin with a stressed syllable (A-), thus is an incomplete foot, consisting only of an unstressed syllable, which is otherwise the ending of a foot. In fact, there is no reason to assume that foot and word boundaries should coincide: feet are defined on purely phonological (rhythmic) grounds, while words are units of meaning or syntax.
Given that the syllable is also a purely phonological category, one may easily hypoth- esize an analogous situation where it is not an absolute necessity that syllable boundaries coincide with word boundaries. In this case, word-initial (and -final) consonant clusters which seem to violate sonority sequencing are analysable as the end or the beginning of a “degenerate” syllable. (10) shows two such “offending” words, skat (scat) and taks (tax ).
(10) Syllables violating sonority sequencing
9 ◦ ◦ 8 7 6 5 4 3
1 0 • • • • s k a t t a k s
Although these words behave as monosyllabic, they both contain two sonority peaks. This contradiction is solved by supposing that the s in skat is the end (the coda), and in taks it is the beginning (the onset) of a “degenerate” syllable (see §4.5.1.3 for more details).^4
(^4) There still remain a number of unsolved issues in this analysis. For exmaple, we have no explanation for why only s occurs in such “degenerate” syllables in English.
An onset l is followed by a vowel in all cases except one, when it is followed by j. In a branching onset, l usually occupies the second position (pl, bl, fl, kl, gl), but there is one cluster in which it is the first: lj (e.g., lj0wd lewd or valj0w (value). Thus, an l that precedes a j is the only preconsonantal l that is in an onset. Syllabic l occupies the nucleus, which is in the rhyme. Coda l, which is also in the rhyme, is either followed by a consonant (the onset of the next syllable), or is word final.
The fact that l behaves similarly in both the nucleus and the coda is one reason to claim that these two syllabic constituents form a group, called the rhyme.
4.4.2 The distribution of [r]
In rhotic accents of English (like General American), r is found both before vowels and before consonants, while in nonrhotic accents (like Standard Southern British English) r occurs only before vowels. The distribution reminds us of that of clear and dark l, therefore the syllabic formulation may also be the same, given in (13).
(13) Distribution of r (attempt 1)
In nonrhotic accents, r occurs only in the onset.
The statement in (13) suggests that there is no syllabic r, that is, no r in the nucleus, in nonrhotic accents. This is not the case, however, we do find syllabic r in SSBE, too, as the words in (14a) show.
(14) Syllabic r in SSBE
a. k´amr@ (camera),s´Igr´Et (cigarette) b. *d´Ezrt (d´Ez@t desert) or *´ofr (´of@ offer ) c. b´arn (barren, baron), v´Etrn (veteran)
Crucially, in SSBE syllabic r does not occur before a consonant or word finally, forms like those in (14b) are impossible. In fact, not just syllabic but any r occurs not only before vowels, but before any syllabic segment, as (14c) shows. (This also likens the distribution of clear l and r, since l is clear before syllabic r too: e.g., s´ElrIj (celery).^6 ) Thus it is not (only) the syllabic affiliation of the r that decides whether it is pronounced or not, but the syllabic affiliation of the following segment. An r, whether in the onset or in the nucleus, is pronounced if immediately followed by a segment that is in a nucleus. This distribution is different from that of clear and dark l (recall, syllabic l is dark, irrespective of what follows), and hints at a mixed type of conditioning: it is not the host of the r that matters, but the host of the following segment. We will not speculate any further here on possible solutions of this anomaly. The distribution of r can thus be given as (15).
(^6) Notice the paradox that both l (and m and n) may be followed by syllabic r and r may be followed by syllabic l (and m and n). This is appear to go against sonority sequencing. However, we cannot deal with this problem here.
(15) Distribution of r (attempt 2)
In nonrhotic accents, r occurs only before a segment in a nucleus.
4.4.3 Aspiration
(16a) contains the environments where voiceless plosives (here exemplified by p) are aspi- rated, (16b) contains cases where aspiration is missing. (For uniformity’s sake, aspiration is marked by the superscript h even when a sonorant follows, where the transcription pl is also used in other publications.) ˚
(16) Aspirated and unaspirated p
a. pain ph´Ejn plain phl´Ejn apace @ph´Ejs complain kh@mphl´Ejn pagoda ph@g´@wd@ placenta phl@s´Ent@
b. Spain spEjn splay splEj leper l´Ep@ explain Ikspl´Ejn specific sp@s´IfIk lap l´ap
We see that voiceless plosives are pronounced as ph, th, kh^ when followed by an optional approximant (l, r, j, or w) and a stressed vowel, unless preceded by s. If the plosive is word initial, the following vowel may be unstressed. This rather complicated description seems to call for a reformulation along syllabic lines. Phonologists do not hesitate to provide the alternative shape of the rule, running as in (17).
(17) The aspiration rule: syllabic formulation
Voiceless plosives are aspirated at the beginning of a word or a stressed syllable.
This formulation of the rule explains the optional approximant on the one hand, and the aspiration-inhibiting effect of s on the other. A syllable-initial voiceless stop is followed either by a vowel, or by an approximant which in turn is followed by a vowel. Furthermore, if sC clusters are thought to be tautosyllabic, then a voiceless plosive preceded by s is not at the beginning of a syllable. Note that in the Jones/Gimson transcribing tradition this is the reason why the stress mark is placed before the s in explain Ik"spleIn.
But in §4.3.3, we argued that sC clusters were in fact not tautosyllabic, that is, the s of a word-initial sC cluster is a “degenerate” syllable. The elegance of the syllabic formulation of the aspiration rule (in (17)) appears to undermine this view. However, explaining the absence of aspiration by the tautosyllabicity of the pre-plosive s is flawed. It is not only s that inhibits the aspiration of the following voiceless plosive, but any other fricative. The appearances mislead us only because fricative+voiceless plosive clusters within a morpheme are overwhelmingly sC clusters. The only other instance of this cluster is ft, but then ft does not occur word initially and is very rarely followed by a stressed vowel. But when it is, the plosive may be unaspirated: e.g., in kaft´an kaftan.
4.4.4 Yod-dropping
The distribution of yod (j) is not only more easily formulated by making reference to syllabic constituents, but also provides arguments about syllabic constituency. For our current purposes an examination of stressed syllables with yod is sufficient, therefore we will not be concerned with the somewhat different distribution of yod in unstressed syllables.
After consonants morpheme internally, the distribution of yod is rather constrained in English: it can only occur if followed by 0 (or its pre-R version, o). Accordingly, sequences like CjE, or Cja only occur in loanwords (e.g., pied-`a-terre, piano). This suggests that the yod and the following vowel share a syllabic constituent (most probably the nucleus), since, recall, phonotactic constraints hold within syllabic constituents (see §4.2). Surprisingly, a postconsonantal yod is also in a phonotactic relationship with the preceding consonant. English accents vary to some extent with respect to the distribution of postconsonantal j, we examine a conservative version of Standard Southern British English (also known as Received Pronunciation). (18) collects some of the relevant data.
(18) The distribution of postconsonantal yod in RP
a. compute k@mpj´0wt rebuke rIbj´0wk confuse k@nfj´0wz revue rIvj´0w amuse @mj´0wz
c. agglutinate @gl´0wtInEjt peruse p@r´0wz assure @S´o: eschew IstS´0w adjudicate @dZ0wdIkEjt
b. enthuse InTj´0wz consume k@nsj´0wm exude Igzj´0wd minute majnj´0wt volute v@lj´0wt obtuse @btj´0ws deduce dIdj´0ws
d. acute @kj´0wt ambiguity ´ambIgj´0wItIj exhume Ekshj´0wm
The words in (18a) contain a labial consonant before the yod, those in (18b) a dental or alveolar consonant, and those in (18d) a velar or glottal one. In all these classes, the yod remains intact. It is only in the type in (18c) that the yod is lost (or, put alternatively, cannot occur).
Apart from agglutinate, all the words in the yodless group contain a postalveolar consonant before the 0 (or o). The accepted syllabic explanation for the absence of the yod in this position is that postalveolar consonants cannot form a branching onset with yod in second position. This hypothesis is supported by at least two pieces of evidence: on the one hand, postalveolar consonants can precede yod, provided the two are not in the same onset. Unfortunately, this only occurs if the two are in separate morphemes, e.g., in fresh urine (-Sj-), church use (-tS j-). It is a stronger argument that branching onsets generally inhibit homorgani clusters: after labials the labial glide does not occur (*pw, *bw, *fw, mw), after dentals and alveolars there is no l (tl, *dl, *Tl),
r is impossible after postalveolars (*tSr, *Zr).^7 If so, we expect the glide yod not to be possible exactly after postalveolars.
The case of yod following l is more complicated. Yod may follow l only if the l is “lone”, that is, it is not preceded by a consonant. This is a strange rule: apparently the yod can “see through” the preceding l to check the properties of the sound before it. If, however, we suppose that the onset of the syllable can host a maximum of two consonants, the phenomenon falls into place: a consonant together with the following l occupy both available slots, not leaving any place for the yod, as is shown in (19b). On the other hand, a single l occupies only the first slot, the second being vacant for receiving the yod, as is shown in (19a).
(19) The impossibility of yod after consonant+l
a. N O N
@ l j u @lj0wd allude
b. N O N
@ g l j u @gl0wtInEjt agglutinate
Intriguingly, a yod appears after an sl cluster in the accent under examination: sleuth is slj0wT. That is, the sl cluster behaves like a single l. If we are to maintain the explanation of the presence and absence of yod in the Vl and #l environments, as opposed to the Cl environment, then we must conclude that the sl cluster’s syllabic affiliation is similar to that of single onset l’s and not to that of branching onset Cl’s. This is a strong piece of evidence for the claim made in §4.3.3, namely, that syllable and word boundaries do not necessarily coincide. In a word like sleuth the l belongs to an onset, but the s before it does not, at least not to the same onset as the l.
The same situation prevails with any consonant cluster of the shape sC. Thus if yod may appear after a given consonant Cx and sCx is also a possible cluster, then yod may appear also after the sCx cluster. (20) gives an example for each cluster in question. (The Cxj clusters not appearing here do not have an sCx counterpart, hence obviously also lack their sCxj pair. We take the nonexistence of snj- to be an accidental gap.)
(20) Cj and sCj clusters in English
Cxj sCxj beauty bj- sputum sbj- mew mj- smew smj- duty dj- student sdj- new nj- — lewd lj- slew slj- argue gj- rescue sgj-
(^7) Note that the fact that there are English words beginning with Sr does not warrant that the cluster in them is a branching onset.
of the vowels in a no-contrast system is immaterial: the vowels of such a language are systematically analysed as short. Thus the implication becomes trivial.
In §4.5.3, we are going to see further implicational relationships of syllable types. But before that we have to get acquainted with traditional names for some syllable types: open, losed, light, and heavy syllables.
4.5.1 Syllable weight
Syllables may end in a consonant, or in a vowel. A syllable that ends in a consonant is a losed syllable (viz., closed by that consonant, e.g., the first syllable of panda pand@ is closed by the coda n); one which does not is open (e.g., the second syllable of panda, or the first syllable of paddock pad@k is open).
One may be tempted to identify syllable weight with graphic positions. There is indeed some similarity between the notions of free graphic position and open syllables on the one hand, and overed graphic position and closed syllables on the other. For example, the word cinema contains three open syllables, and its three vowels are each in free graphic position, while map is a closed syllable, and its vowel is in covered graphic position. But equating the two pairs is unwarranted, as the above example clearly shows: the first syllable of pa.d@k is open, but the first vowel of paddock is in covered position. And vice versa: the vowel of make is in free position, but mEjk is a closed syllable— see §4.5.1.2 though. It is also difficult not to notice that stop+liquid clusters, which do not “cover” the vowel letter preceding them (e.g., maple), are reminiscent of branching onsets, the type of consonant cluster which does not “close” the preceding syllable. The parallel, however, is again incomplete: tl and dl are stop+liquid clusters, but not branching onsets.
In many respects, however, it is not open and closed syllables that pattern together, rather, open syllables with a short vowel are opposed to open syllables with a long vowel and to closed syllables. An open syllable with a short vowel is called a light syllable, a syllable which is either closed or has a long vowel (diphthongs in English all count as long) is called a heavy syllable. Thus, the above cinema sI.n@.m@ is three light syllables, tambour tam.bo: is two heavy syllables. Sometimes a third category is also identified: a syllable that contains both a long vowel and a coda consonant closing it, as the first syllable of thirsty T@:s.dIj, or a short vowel but two coda consonants, as in lamp, is called superheavy. In most cases, a heavy and a superheavy syllable are equivalent. In English, as in many other languages, stress assignment is sensitive to the heavy–light distinction (cf. chapter 6). The relevant kinds of syllable are illustrated in (22). The syllables in (22a–b) are open, those in (22c–d) are closed. The syllable in (22a) is light, those in (22b–d) are heavy, that in (22d) is also called superheavy. The sequence oo in (22b) and (22d) is a way of representing a long vowel, i.e., it is equivalent to o:.
(22) Branching in light, heavy, and superheavy syllables
a. σ
R
O N
p o
b. σ
R
O N
p o o
c. σ
R
O N C
p o t
d. σ
R
O N C
p o o t
We can see that the relevant notion for syllable heaviness is branching in or below the rhyme. While in a light syllable neither the nucleus nor the rhyme branches, (22a), in a heavy syllable either the nucleus, (22b), or the rhyme branches, (22c). If both branch, as in (22d), we have a superheavy syllable.
Interestingly, the onset has no bearing on the weight of the syllable. Only the rhyme counts: an onsetless closed syllable (like Eg egg) is heavy, just like one with a single onset consonant (like pEg peg), or a cluster, that is, a branching onset (like grEg Greg).
4.5.1.1 Moras
Syllable weight is measured in moras. Each segment in a nucleus is worth one mora, and in some languages (English among them) coda consonants are also moraic. Some phonol- ogists would therefore represent the syllable types in (22) without syllabic constituents, with only moras (usually symbolized by the lowercase Greek mu, μ).
(23) Moras in light, heavy, and superheavy syllables
a. σ
μ
p o
b. σ
μ μ
p o o
c. σ
μ μ
p o t
d. σ
μ μ μ
p o o t
The weight of a light syllable is one mora, (23a), that of a heavy is two, (23b–c). Super- heavy syllables contain three moras, (23d). Note that this definition of syllable heaviness avoids the awkward formulation of the previous approach: “branching somewhere in the rhyme.”
A way of justifying the phonological existence of moras is the phenomenon of om- pensatory lengthening. In this change, which is frequently found in the history of various languages, the loss of a segment is made up for by the lengthening of another, neighbouring segment. A case in point is the lengthening of the broad vowels in nonrhotic accents of English. Before coda r’s were lost, the first syllable of party was closed, hence heavy (par.ti). The weight of this syllable was retained even after the r was lost and it became open. The change is illustrated in (24).
4.5.1.2 Extrametrical consonants
Examining the place of stress in English verbs (also see chapter 6), we find a curious pattern. (25) contains the relevant data. In these words diphthongs are treated as long vowels, according to the traditional analysis.
(25) Verbal stress patterns
penultimate stress final stress a. cons´ıder k@ns´Id@ b. lam´ent l@m´Ent bother b´oD@ sal´ute s@l´0wt c. c´ancel k´ans@l d. agr´ee @gr´Ij dev´elop dIv´El@p def´y dIf´aj
The data in (25a) show that verbs with a light ult are stressed on their penult. Verbs that end in a closed syllable, that is, contain a heavy ult, also have penultimate stress, as in (25c). So far it is only verbs with a superheavy ult, those in (25b), that are stressed on their ult, thus the division seems to be between light and heavy syllables on one side and syperheavy syllables on the other side. The data in (25d), however, upset this generalization, since here plain heavy (i.e., not superheavy) syllables are stressed. Although there is a difference between the heavy syllables of (25c) and of (25d), the former are heavy by virtue of a closed syllable, the latter of a long vowel, it is not particularly neat to include such a distinction in a stress rule. In fact, this cannot even be done in a moraic theory, where both -cel, -lop and -gree, -fy are equally bimoraic.
By applying a little trick, the whole picture becomes much simpler. If we disregard the final consonant in each of these words, the choice boils down to the difference between heavy and light syllables. (Note that for this to work we must not take diphthongal offglides to be consonants.) The analysis of the types of verb in (25a) and (25d) remains, because these verbs are vowel final. The verbs in (25b) and, especially, in (25c) change their category. Without the final consonant, the ult of cancel is light (-s@-), so the verb will follow the pattern of bother. The verbs that end in a superheavy syllable (lament -mEnt, salute -l0wt) will still contain a heavy syllable without the last consonant (-mEn-, -l0w-), accordingly they will have stress on their ult. In fact, by ignoring the last consonant, we get rid of most superheavy syllables of English altogether, since these mostly occur word finally. A word-final consonant which does not count in stress calculation is said to be extrametri al.
The notion of extrametricality is, nevertheless, a problematic one. We have seen above (in §4.3.3) that short lax vowels cannot occur word finally. Now if word-final consonants are extrametrical, lax vowels are expected not to occur in the C# context. Yet they do (e.g., cat, dog, nut, regret), so one has to make arbitrary claims about why at the point of stress assignment the consonant is invisible, while at the point when phonotactic constraints are obeyed the consonant becomes visible.
4.5.1.3 Degenerate syllables
An alternative is to assume that a word-final consonant is part of a “degenerate” syl- lable, similarly to a word-initial s. In this way, the consonant is visible throughout the
calculations, but does not contribute to the weight of the last syllable. (26) shows the representation of the word splendid along these lines.
(26) The syllables of splendid
σ σ σ σ
R R R
C O N C O N O
s b l E n d I d ︸ ︷︷ ︸ ︸ ︷︷ ︸ degenerate degenerate syllable syllable
In (26) the degenerate syllables occur at the edges of the word. This is a general, perhaps exclusive pattern. If so, then it is a totally misguided idea to base syllabification (notably the decision of what is a possible onset) on what occurs at the beginning of the word and what does not, as suggested in §4.3.3.
4.5.2 Closed syllable shortening
The phenomenon of losed syllable shortening may also be explained by mora counting. English exhibits some traces of this phenomenon (which indicates that it must have been active at some stage of the history of the language), but there are large portions of the vocabulary that do not show it.
A verb like keep kIjp is represented as in (27a). Its past form kept kEpt is shown in (27b), while a hypothetical past form— occurring in child language— *kIjpt is in (27c).
(27) Moras in keep and its past forms
a. σ σ
μ μ
k I j p
b. σ σ
μ μ
k E p t
c. σ σ
μ μ *μ
k I j p t
The word-final consonant is extrametrical, hence the rhyme of keep weighs two moras. With the addition of the past tense suffix the word has two consonants at the end, only the second of which can be extrametrical. If the stem-final p were simply pushed into the rhyme without any further changes, the rhyme would end up with three moras, (27c), a superheavy rhyme, which is (or rather was, at least when this form developed) impossible. As a result, the nucleus has to shorten, hand over one of its two moras to the now coda p. This is closed syllable shortening. The change of the quality of the vowel (Ij to E) is clear evidence that this change is a historical relic in the phonology of English.
