The curious case of Cj clusters in English*
The word-initial1 consonant clusters in English fall into two groups, those beginning with s — sometimes S — and those that do not. Clusters beginning with s or S have been troubling phonologists ever since they noticed the relevance of sonority in organizing sound strings. We will not be very much concerned with them here. The other clusters, those that do not begin with s or S, almost exclusively contain two consonants and are characterized by the regularities listed in (1). (1)
a. the second member is one of l r j w b. the first member is never a palatal/postalveolar (recall we have excluded S-initial clusters) c. the two members are usually not homorganic (if we take r to be palatal/ postalveolar) d. the first member is a plosive or a voiceless fricative before l r w e. only a limited set of vowels occur after a cluster ending in j f. these clusters occur significantly more often before a stressed vowel than before an unstressed vowel, except for those involving j
In this paper we are looking at the last three of these regularities. Implicit in (1d) is the claim that yod may occur not only after plosives and voiceless fricatives, but also after voiced fricatives and sonorants, as in view vj0w, music mj0wzIk, huge hj0wdZ.2 (1e) indicates that postconsonantal yod is (or at least was) in a closer relationship with the following vowel than the other postconsonantal sonorants. Finally, (1f) also proves j to behave exceptionally among the other nonnasal sonorants, r, l, and w.3 We will look for possible explanations for these facts about the phonotactics of English.
1
To avoid theoretical issues, we will look at word-initial clusters where available. Word medially the same clusters often (always?) mimic their behaviour, as far as phonotactic constraints go.
2
We here take h to be a sonorant. Nothing crucial hinges on this decision.
3
If h is a nonnasal sonorant, it ought to occur as the second member of consonant clusters. Perhaps it does: pit phIt, tit thIt, kit khIt. Nevertheless, we ignore this possibility here.
The Even Yearbook 12 (2016), Dept of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/even, c 2016, Péter Szigetvári
94 Péter Szigetvári
1 The oddity of CjV If we examine obstruent + liquid clusters in English, we find that restrictions hold between the two consonants, but not between the liquid and the following vowel. Consider the data in (2). (2) a. Cj— b. Cw— c. Cl— d. Cr— mute mj0wt (swoon sw0wn) flute fl0wt fruit fr0wt pure pjo: quart kwo:t flaw flo: frog frog (pure pj@:) quirk kw@:k blurb bl@:b brush br@S (piano pja:n@w) qualm kwa:m class kla:s grass gra:s dwell dwEl bless blEs dress drEs twin twIn bliss blIs brief brIjf A word-initial consonant + liquid cluster may be followed by any4 vowel, as illustrated in (2c) and (2d). We do find constraints with the two glides though. In (2b) swoon is in parentheses, because w0 is only possible after s, and due to their odd properties we have excluded s-initial clusters. Other consonants do not occur before w0, when they did, the glide was lost: cf two t0w, as opposed to the etymologically related twice, twin, twelve, between, Dutch twee, etc, all of which still contain w. Compared to the liquids and w, the examples in (2a) show that j is much more related to the following vowel. It occurs with 0, o, and @ for speakers of British English. With the other vowels it only occurs in recent loanwords: eg Myanmar mjánmá:, pied-à-terre pjÉjdatÉ:. Compared to the liquids and w, (1d) states that j is much less related to the preceding consonant: this position is not restricted to plosives and voiceless fricatives, cf (4) further below. These phonotactic facts may lead one to see j0 as a unit, possibly a rising diphthong. However, there are a number of reasons for being suspicious of such a conclusion. The most important argument against a diphthongal j0 is the fact that it may not occur after word-initial consonant clusters — except of course the notorious s-initial ones. So spew spj0w and skew skj0w are available, but blew is bl0w, not
lewd lj0wd. This indicates that the yod following a consonant is itself also a 4
Since the quality difference between the vowels of LOT and THOUGHT, O vs o:, on the one hand, and TRAP and PALM, a vs A:, on the other, is predictable from the length of the vowel, we ignore it here. This limits the number of distinctive vowels in British English to six: KIT, DRESS , TRAP , STRUT, LOT , and FOOT , ie I E a @ o 0, respectively. It is these six vowels that appear in (2).
Cj clusters in English 95 consonant, it is part of the onset, the length of which is limited to two — again, if we exclude s-initial clusters. These peculiarities of the distribution of Cj in English follow from the historical development of words containing this cluster. This is what we briefly look at in the next section.
2 The source of Cj in English Old English did not have Cj clusters, although it did have Cl, Cr, and Cw, even Cn, but the latter have simplified to n (OE hnut is nut in Modern English). The majority of Cj clusters in present-day English come from two Middle English diphthongs and from French loanwords (Minkova 2014). Some examples are shown in (3). (3)
a. ME iw, eg new, Tuesday b. ME ew, eg dew, brew c. F y:, eg due, sure
Middle English iw and ew merged and the syllabicity of the resulting diphthong swapped from the first to the second element in most accents of English (not in Wales varieties of English;5 Penhallurick 2004 : 107f). That is, the falling diphthong iw became a “rising diphthong” ju:. Note that the vowel part of ju: is long, ie bimoraic, just like its source, iw. Since Old English y: has unrounded to i:, as in mice (also short y to i, as in fill), by the time French loans with y: and y were adopted in English this vowel was gone in earlier words, so it was broken into its components, i and u. Only long y: had the two moras that could accommodate both parts, the resulting vowel has merged with ju:. The short y of French loans could not preserve both labiality and palatality, it has lost the latter property and was adopted as short u in English, and in many accents later unrounded and lowered with it: eg buzzard, culprit, just. Since there was no restriction on the consonant — or, in fact, consonant cluster — occurring before iw, ew, and y:, it came to be a unique property of yod that its occurrence after consonants is not limited by sonority: it occurs after voiced fricatives and sonorant consonants too.6 What excludes some Cj clusters is the ban
5
Since this vowel is still a diphthong here, its palatal component remains in blew bliw.
6
Unlike the other glide, w, which does not occur after sonorants. For speakers who still retain word-initial hw (eg in what) it does. One reason for not taking h to be an obstruent is the simplification of OE hl, hr, and, for many, hw.
96 Péter Szigetvári on homorganicity (cf (1c)): yod does not occur after palatals and postalveolars, eg in chew, Jude, rule. Before consonant+unstressed vowel and before word-final r, unstressed ju: has reduced to j@ (eg regular rÉgj@l@, tenure tÉnj@). In British English stressed u: has lowered before r (eg cure kjo:), even unrounded for some speakers (eg cure kj@:), and split into a fronter and a backer vowel, the latter occurring before nonprevocalic l (eg mule mjuwl vs mute mj0wt). These developments have led to a larger set of vowels after Cj clusters. Let us now look at the constraints on the consonants before yod in English.
3 Consonants before yod Some of the variation in the range of consonants that do and do not occur before yod is well known (eg many Americans would pronounce new tune as noo toon, that is, n0w t0wn instead of nj0w tS0wn, and have due and do as homophones, whereas for British speakers due is usually homophonous with Jew). The following chart compares five varieties of English. On one side of the scale is Welsh English, in which the palatal element is preserved in all environments, obviously because it is manifested by a vowel. For the sake of uniformity, we will nevertheless represent this palatality by j in this chart. The other end of the scale is East Anglian English, where the palatal element is lost in all environments (Trudgill 2004 : 175f). Between the two is British and American English. We are experiencing changes in this area in Standard British English currently, so two varieties are included, a more traditional one labelled as RP and a more advanced one called Current British English (CUBE). These three varieties show that the distribution of postconsonantal yod primarily depends on the place of the preceding consonant. The words in (4e) show that the consonantal yod cannot be retained after consonant clusters — as has already been mentioned. It is noteworthy that the sinitial clusters are exceptional (eg slew slj0w,7 spew spj0w, smew smj0w). Yod is also missing after palatal and postalveolar consonants, as shown in (4c). English does not provide phonological evidence for distinguishing palatal and postalveolar as different places of articulation, in fact, the term palatalization is used to refer to an alveolar (t d s z) becoming postalveolar (tS dZ S Z, respectively). So we can safely merge the two groups under the label palatal. Classic RP retains the
7
Those who pronounce slew as sl0w also pronounce lewd as l0wd, ie they omit the yod after any l, not only after a cluster ending in l.
Cj clusters in English 97
(4) a. pure p beauty b few f music m b. Tuesday t duke d new n thuja T lewd l slew sl assume s presume z c. parachute S chew tS Jew dZ rule r yew j d. cube k gue g hue h e. blew bl glue gl flew fl true tr crude kr
WelshE
RP
CUBE
j j j j j j j j j j j j j j j j j j j j j j j j j
j j j j j j j j j j j j
j j j j tS dZ
GenAm EAnglia j j j j
; ; ; ; ;
; ; ; ; S; Z; ; ; ; ; ;
; ; ; ; ; ; ; ; ; ; ; ; ;
; ; ; ; ;
; ; ; ; ;
; ; ; ; ;
j j j
j j
j j j
j j j
; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ;
historical yod in all other contexts, ie after singleton nonpalatal consonants and after s-initial clusters.8 American varieties often go much further in eliminating yods: it is omitted not only after palatals, but after any coronal consonant, as shown in (4b). It is rather difficult to motivate this pattern with homorganicity, because that would also entail the simplification of clusters with r and l (cf Törkenczy 2008), which apparently does not occur in adult speech. What is more, we will shortly see that not all alveolar + yod clusters are simplified in General American, so whatever 8
There is no swj, but there is no wj either. There is also no snj. There simply was no input for these clusters in Middle English or French.
98 Péter Szigetvári
(5) a. copula p nebula b crofula f amulet m b. botulin, nature t module, procedure d monument n Matthew T cellular l insular, pressure s lazulite, closure z c. brochure S Quechua tS injure dZ virulent r d. accurate k argument g e. plurality pl influence fl altruist tr congruous gr
RP
CUBE
GenAm
j j j j j/tS j/dZ j j j j/ S j/ Z
j j j j tS dZ j j j j /S j/Z
j j j j tS dZ j j j /S /Z
; ; ; j;
; ; ; j;
; ; ; ; ; j;
; ; ; ;
; ; ; ;
; ; ; ;
j j
j j
j j
cause we construe for this type of yod-dropping, a well defined set of environments will have to be exempted from it. Some variation within a single variety is shown for Current British English in (4). Here we see that simple omission is not the only way of getting rid of a postconsonantal yod. Obstruents often fuse with yod, the result of which is that the yod disappears (at least from our transcriptions which use discrete symbols for “individual segments”), but palatality remains. Note that this process has been going on for centuries in English, yielding some lexicalized palatalizations, where only etymology (or spelling, the poor man’s etymological database) tells us what happened, like in sugar S0́g@, sure Só:, or mature m@tSó:. The distribution of postconsonantal yod described so far is only partial, it holds for yod before a stressed vowel. Before an unstressed vowel, we find a somewhat different picture. This is shown in (5). We cannot use word-initial clusters this time, since unstressed syllables containing CjV appear not to exist at the beginning of words. The two “extreme” accents, Welsh and East Anglian
Cj clusters in English 99 English, have been omitted this time, since they do not provide any novelty, their stressed and unstressed syllables are alike. As for the other three accents, there is no difference after noncoronal consonants, (5a, d) and non-s-initial clusters, (5e). In the other two environments we see more palatality before unstressed than before stressed vowels. We have seen that in a stressed syllable yod does not occur after r in any of these varieties of English. This makes r pattern with palatals/postalveolars, (5c). Before an unstressed vowel, however, some speakers have the yod in the few examples that contain this environment, like virulent, erudite, pulverulent, querulous, or ferrule.9 The differences between these three varieties in the distribution of yod after an alveolar consonant10 are almost all gone in (5b). We see that palatality is almost always present in this case, either as yod proper, or as a palatalized obstruent. In the latter case we know that the S, Z, tS, dZ of pressure, pleasure, virtue, module were Cj clusters earlier only from morphologically related alternants (like press, please) or, of course, from the spelling, ie etymology of these words. The generalization is valid in any case: in unstressed syllables that contained an earlier j0 the palatal element is only absent after clusters, as in (5e). So far we may conclude that the relative lack of constraints between yod and a preceding consonant, (1d), and the presence of constraints on yod and the following vowel, (1e), as compared to other consonant plus sonorant clusters, is the result of a historical development: the yod in Cj clusters was “part” of the following vowel in an earlier stage of English. We now turn to (1f): the difference we observe between stressed and unstressed syllables.
4 The role of stress In English the consonantal position before a stressed vowel is “stronger” than that before an unstressed vowel. Here we do not pursue why this is so, but refer the interested reader to Harris (1997) or Scheer & Ségéral (2001). For our purposes it is enough to assert that the type of clusters discussed in this paper are in a more favourable position before a stressed than before an unstressed vowel. This
9
Since this is the only consonant of this group that may allow yod to follow it, even if only before an unstressed vowel, one may be tempted to reclassify it with alveolars, so that it can be maintained that these two consonants must not be homorganic. However, r may palatalize the preceding alveolars (eg disregard dISr@ga:d, string StrIN), like yod (eg stupid StS0wp@d, misuse mISj0wz).
10
Let it be admitted that we are glossing over the fact that T is not alveolar, but dental. This does not make it very different from the rest of this group.
100 Péter Szigetvári difference is not categorical though: consonant + approximant clusters occur both before stressed and unstressed vowels, but with different frequency. The figures in (6) illustrate this fact. The data are based on the CUBE searchable online pronunciation dictionary (http://cube.elte.hu), containing 103 000 entries. The numbers indicate the number of entries containing the given string, potentially several times (like in program). We cannot always unambiguously separate stressed from unstressed vowels in this resource, so we compare all prevocalic with preschwa sequences — excluding both STRUT and GOAT from our definition of schwa. A further distorting factor is syncope, which occurs before unstressed vowels, hence creating consonant + approximant clusters of the type examined here before schwa: eg interest Ínt(@)r@st, lateral lát(@)r@l, etc. But even with these cases, the figures of preschwa Cr, Cl, and Cw is around 10% of all the occurrences of these clusters, as shown in (6). Note the relatively much higher ratio of preschwa Cj. (6) Cr Cl Cw Cj
@
V
ratio
2263 1061 339 1210
16510 10274 3433 3596
14% 10% 10% 34%
The figures in (7) show the occurrence of all consonants in word initial or postvocalic — that is, not postconsonantal — position. We see that these ratios are generally at least twice as large. This means that a Cr, Cl, Cw cluster is only half as preferable before schwa as either of the members of the cluster. It is noteworthy that two sets of consonants stick out of the general picture: compared to other consonants voiced fricatives, S, and N are far more common before schwa, while the glides are vanishingly rare in this position. We know that in Old English voiced fricatives were the foot-internal variants of fricatives, foot initially their voiceless counterparts occurred. In other words, a voiced and a homorganic voiceless fricative were allophones (Minkova 2014 : 88ff). Apparently, this historical fact can still be detected in the phonotactic trends of English. A similar explanation holds for N, which is impossible prevocalically in some accents of English to this day (cf Wells 1982 : 365f, Clark 2004 : 139). The fact that glides are almost absent before schwa is an artefact of a prevalent tradition which interprets vowel+glide sequences as diphthongs in English. If we include diphthongal offglides in our calculations, that is, if we count the glides in lion láj@n and vowel váw@l as glides proper, we get the results shown in (8).
Cj clusters in English 101
(7)
@
V
ratio
l r j w h
3980 3144 17 67 117
15232 15595 840 3477 4474
26% 20% 2% 2% 3%
p t k b d g tS dZ
2127 4319 4196 2245 1991 1270 337 735
9127 13394 13030 8927 10063 4152 1719 3668
23% 32% 32% 25% 20% 31% 20% 20%
f T s S
1019 250 2529 3388
6493 1210 11753 5241
16% 21% 22% 65%
v D z Z
2262 404 1422 240
5613 541 3383 356
40% 75% 42% 67%
m n N
3325 3452 146
12571 11400 204
26% 30% 72%
@
V
ratio
6211 1082
9943 5592
62% 19%
(8) j w
This lets w join the group of regular consonants, with ratios ranging from f’s 16% to t’s 32%. The high ratio of preschwa j is a result of words in Ij@ (eg alien ÉjlIj@n), without this pattern, the ratio is down to 14%.11 11
not that there is any reason to exclude Ij@ from our calculations
102 Péter Szigetvári Even if we do not compare clusters to singleton consonants, the high ratio of preschwa Cj in (6) compared to other clusters — 34% against 10–14% — requires an explanation. This is what we will now attempt to give.
5 High vowel gliding Let us recall, in §2 we saw that Cj clusters of present-day English are reflexes of two Middle English diphthongs and French loans containing long y:, as shown in (3), repeated in (9a–c) for convenience. (9d–f) complete the sources of Cj in English. (9)
a. ME iw, eg new, Tuesday b. ME ew, eg dew, brew c. F y:, eg due, sure d. high vowel gliding, eg million, opinion, Slovakia e. loanwords, eg banyan, canyon, cognac, fjord, piano, pinyin f. morpheme concatenation, eg churchyard, egg yolk, light-year
(9e, f) are mentioned only to make the list exhaustive; note that in loanwords, the yod can be followed by vowels other than 0, o, or @ (cognac kónjak, pinyin pÍnjIn), and, of course, any consonant — even a palatal — can precede the yod across a word boundary, and any vowel can follow it, as is the case for word-initial yod. The most productive source of word-internal Cj clusters in English currently is high vowel gliding (HVG), illustrated in (9d). HVG is a case of syncope. Syncope is most likely in English if an unstressed vowel (namely schwa) is followed by a singleton sonorant consonant which in turn is followed by another unstressed vowel (eg literal lÍt/@r@l, catholic káT/@lIk).12 Syncope is blocked if the second vowel of this string is stressed (eg iterate Ít*/@rÈjt, catholicity káT*/@lÍs@tIj). Syncope is also sensitive to the cluster preceding the syncopated vowel: this may be a falling-sonority cluster (eg adultery @d´@lt/@rIj, company k´@mp/@nIj), a level-sonority cluster, involving two obstruents, which we here take to be of equal sonority (eg victory vÍkt/@rIj, history hÍst/@rIj, excellent Éks/@l@nt), but not a rising-sonority cluster (eg burglary b´@:gl*/@rIj). Examples like tolerant tól/@r@nt prove that the cluster is to blame for the impossibility of syncope in burglary, not the l and r around the schwa to undergo syncopy. In HVG the sonorant consonant after the syncopated vowel is restricted to glides and the syncopated vowel is an unstressed high vowel (I before j and 0 before w), but the rest of the phenomenon is identical, the constraints hold in 12
The judgements on where syncope (including HVG) is possible come from Wells 2008.
Cj clusters in English 103 exactly the same manner. So HVG occurs before an unstressed vowel (eg million mÍl/Ij@n, mania mÉjn/Ij@, graduate grádZ/0w@t), but not before a stressed vowel (eg humiliate hj0wmÍl*/IjÈjt, maniac mÉjn*/Ijàk, graduate grádZ*/0wÈjt).13 As before, the cluster preceding the syncopated vowel may show a falling sonority profile (eg champion tSámp/Ij@n, Albion álb/Ij@n), or a level sonority profile (eg axiom áks/Ij@m, Caspian kásp/Ij@n, Actium ákt/Ij@m, Sogdian sógd/Ij@n), but again, clusters with a rising sonority profile may not be followed by HVG (eg atrium Éjtr*/Ij@m, Cypriot sÍpr*/Ij@t, nuclear nj0́wkl*/Ij@). And just like for syncope “proper”, if r and l are not part of a rising-sonority cluster, HVG is possible: area É:r/Ij@, Lilian lÍl/Ij@n. HVG involving 0w is much rarer than that involving Ij, as seen in (10). This is because unstressed 0w is preceded by palatality, often in the guise of an explicit yod, which creates a rising-sonority cluster (eg annual ánj0w@l), which — as we have just seen — inhibits HVG. When this palatality is incorporated in the preceding obstruent, HVG is available (eg actual áktS/0w@l). The data in (10) give absolute numbers of sequences that may undergo syncope (or HVG).14 These figures are only approximate: the vowel following the approximant is always schwa here (although it could also be unstressed I), and we examine only those environments that may allow syncope, that is, which either have no cluster before the syncope site, or have a falling-sonority (RC)15 or an obstruent+obstruent (TT) cluster, after which — as we have seen — syncope may occur. (10)
VC @r @l Ij 0w
395 121 891 86
@
RC 290 93 2688 30
@
TT 104 25 182 67
sum
@
789 239 3761 183
We see that there are relatively many potential inputs for HVG that result in a Cj cluster. Since HVG, like any other case of syncope, occurs predominantly before an unstressed vowel, furthermore in many cases syncope/HVG creates Cj, there are many instances of Cj clusters before an unstressed vowel. As we have
13
The word milliard mÍl/Ijà:d appears to behave exceptionally.
14
We did not check each and every of these cases if they really do.
15
There are less instances of VC than of falling-sonority clusters before Ij, because diphthongal offglides and the second part of long vowels are counted as sonorant consonants. While this may distort the ratio of the first two columns, it does not influence the sum in the last column.
104 Péter Szigetvári seen syncope/HVG-created Cj clusters (like Tenniel tÉn/Ij@l) are remarkably similar to “lexical” ones (like tenure tÉnj@), except that the former only occur before an unstressed vowel, while the latter is three times as frequent before a stressed vowel, as shown in (6). But the comparison in (6) also reveals that other “lexical” C+approximant clusters are much less frequent before an unstressed than before a stressed vowel. We contend that this asymmetry in the phonotactic patterning of English is due to the fact that syncope/HVG produces a pattern that supports “lexical” Cj@. Speakers do not drop the j in tenure tÉnj@, because it is identical to the result of HVG in a hypothetical form tenia tÉn/Ij@. Because this pattern is relatively common, “lexical” Cj clusters are also acceptable before an unstressed vowel. Approximants other than j do not need the support provided by syncope/ HVG-created clusters (and, as the figures in (10) suggest, this support would be much weaker anyway), since these clusters do not show any sign of simplification in English, neither before unstressed, nor before stressed vowels. Cr, Cl, Cw clusters are stable, unlike Cj clusters — as we have shown in §3.
6 Conclusions We have shown in this paper that Cj clusters in English pattern differently than other consonant + approximant clusters in several respects. The consonant before yod may not only be a plosive or a voiceless fricative (like that before r, l, and w), but also a voiced fricative (eg view vj0w) or a sonorant (eg mute mj0wt, hue hj0w). The range of vowels occurring after a Cj cluster on the other hand is more limited, it may be 0 (as above), u (eg mule mjuwl), o (eg cure kjo:), or @ (eg accurate ákj@r@t). We have argued that both of these oddities are due to the fact that the yod of Cj clusters has developed from earlier diphthongs or long y:, which had not been constrained with respect to the preceding consonant in any way. The other peculiarity of Cj clusters was their significantly higher frequency before schwa, as compared to other consonant + approximant clusters. To explain this distribution, we have shown that the most abundant current source of Cj clusters, high vowel gliding, occurs almost exclusively before schwa. Therefore, the large number of words containing Cj@ lend support to the retention of preschwa Cj also in those cases where it is “lexical”, ie where it developed from a historical diphthong or long y:. This kind of support is only relevant in the case of Cj clusters, since the other approximants are not subject to loss after consonants.
Cj clusters in English 105 REFERENCES
Clark, Urszula. 2004. The English West Midlands: phonology. In Kortmann & Schneider 2004 : 134–162. Harris, John. 1997. Licensing Inheritance: an integrated theory of neutralization. Phonology 14 : 315–370. Kortmann, Bernd and Edgar W. Schneider. 2004. A Handbook of Varieties of English. Berlin & New York: Mouton de Gruyter. Minkova, Donka. 2014. A Historical Phonology of English. Edinburgh University Press. Penhallurick, Robert. 2004. Welsh English: phonology. In Kortmann & Schneider 2004 : 98–112. Scheer, Tobias and Philippe Ségéral. 2001. La Coda-Mirroir. Bulletin de la Société de linguistique de Paris 96 : 107–152. Törkenczy, Miklós. 2008. Level of generalisation: the role of contrast in an English phonotactic constraint. The Even Yearbook 8. Trudgill, Peter. 2004. The dialect of East Anglia. In Kortmann & Schneider 2004 : 163–177. Wells, John C. 1982. Accents of English. Cambridge University Press. Wells, John C. 2008. Longman Pronunciation Dictionary (3rd ed.). Harlow: Longman. Péter Szigetvári Eötvös Loránd University szigetvari@elte.hu