S YNCOPE IN E NGLISH : F ACT OR F ICTION ?
1. Syncope in English: the facts (?)
As it has been introduced above, traditional (generative) descriptions (dating back, at least, to Zwicky 1972a-b and Hooper 1978; see also Algeo 1974, Kaisse 1985, etc.) make a crucial distinction between post-stress vs. pre-stress syncope. This convention is well illustrated by one of the most recent handbook-style introductions to the topic, in Harris (2011):
Syncope in English, which is both lexically and phonetically variable, targets unstressed syllables in two environments […] (a) a word-initial unfooted syllable […] and (b) between a stressed and an unstressed syllable where the consonant following the targeted vowel is a sonorant and more sonorous than the consonant preceding […] The effect of the second pattern is to contract a trisyllabic sequence into a bisyllabic trochaic foot.
As it is apparent even in the above quote, the conditions on syncope seem to be more strict after a stressed vowel than before it. For post-stress syncope (cf. (b) in the quote) to take place, the consonant following the schwa must be a sonorant, and it must be more sonorous than the one preceding it. That is, there seems to be a strict sonority constraint.5 In addition, the following vowel must be unstressed. That is how the underlined vowels in camera, family, different, separate (adj) can be deleted, but not in vanity (the /t/ is not a sonorant), felony (nasals are less sonorous than liquids), or separate (v) (the third syllable contains a full vowel). The literature also acknowledges the existence of lexicalized cases, of which every is
5 Authors usually assume a sonority/strength hierarchy along the following lines: vowels – glides – r – l – nasals – fricatives – plosives. Hooper explicitly claims that post-tonic syncope does not apply before obstruents, not even in sC clusters. She concludes that the constraints on schwa deletion are not governed by language-specific syllable structure conditions but are governed by universal constraints by which sonorants in second position are favoured over obstruents.
29
the most oft-quoted one6, but our Britney example qualifies, too, as well as a handful of others (see Section 2).
Pre-stress syncope, on the other hand, is allegedly not that restricted: the consonants surrounding the target schwa do not necessarily obey the sonority constraint. Thus, it is taken to be either phonotactically unconstrained (e.g., Zwicky), or less constrained, applying on a relative scale (e.g., Hooper7). The underlined vowel can not only be elided in words like terrain or parade but also in suppose, suffice, potato, etc. It is usually considered to be a typical fast-speech phenomenon, which is only attested at very fast tempos and in casual styles. In addition, it is assumed in classical descriptions (as well as the Harris quote above, cf. (a)) that it only occurs in word-initial syllables.
These traditional claims about the facts, however, have been criticized by nearly everybody who has taken the data under closer scrutiny, at least as early as Dalby (1986). The main objection is that they are based on mere intuitions and introspection (see, e.g., Davidson 2006), and do not faithfully reflect the tendencies found in spontaneous speech as extractable from, e.g., corpus data (Dalby 1986, Głowacka 2001, Davidson 2002, 2006, Patterson et al.
2003, Carlotti et al. 2009, Turcsán and Carlotti 2010, etc.). Such corpus studies typically only partially support the traditional descriptions8; for instance, it has been shown (esp. by Dalby 1986) that in certain registers, and not necessarily in very fast speech, following obstruents do in fact favour syncope and the reverse of the expected sonority effect is found. Carlotti et al.
(2009) are surprised to find that syncope before /r/, which they consider to be the least marked environment, actually decreases as speech tempo gets faster. In this respect, Dalby (1986)’s study is particularly instructive, as it finds that, e.g., in fast reading the rate of pre-obstruent syncope increases, whereas in slow reading obstruents and sonorants show almost identical behaviour, and in conversations stops favour syncope. Although the overall result is that a maximal sonority difference between the members of the secondary cluster increases the rates of schwa deletion9, in certain cases the effect is just the opposite of what the traditional description suggests, that is, sonorant-schwa-obstruent sequences have the highest deletion rate, while the reverse order may have the lowest one. It has also been noticed (e.g., Dalby 1986, Głowacka 2001, etc.) that besides their manner of articulation, the place of articulation and voicing of the consonants straddling the elision site also have an influence.
In addition, contrary to the general claim that pre-stress syncope applies at faster speech tempos, Dalby is unable to clearly identify a speech rate at which only post-tonic syncope occurs to the exclusion of pre-tonic syncope. More recently, Turcsán and Carlotti (2010) also conclude that the difference between fast vs. slow speech syncope patterns is only quantitative. Similarly, Davidson (2002), an acoustic analysis of word-initial pre-stress syncope attempting to rule out as many disturbing factors as possible10, establishes a distinction between rate-dependent vs. rate-independent speakers, both of whom observe phonotactics in such a way that they only delete when the resulting cluster is either found in
6 As a reviewer points out, the word every is in fact not the most fortunate example, since it contained a vr sequence as early as the Old English period (from æfre ‘ever’). To illustrate the early application of syncope, however, s/he provides the example empty, from Old English æmettig.
7 Hooper explains this by assuming that stressed syllables tolerate freer clustering. For the opposite view, see Section 3 below.
8 It has been pointed out to me that the difference between the classical descriptions and the results of corpus studies may also be due to the fact that the two investigate two different speaker samples: the individual vs. a population, resp. While this is true, it still does not invalidate the comparison, since as far as I understand the classical descriptions were also meant to represent a population, describing syncope in “English”.
9 Głowacka (2001: 78), however, arrives at the opposite conclusion (for British English).
10 Most importantly, she uses a strict definition of schwa deletion to rule out any gestures that could correspond to the presence of a vowel: any part of the interconsonantal interval which includes a voice bar and/or formant structure is considered part of the vowel. In addition, there should be no aspiration on the consonant before the deletion site, as that may be taken as a devoiced vowel.
HUSSE 10 Proceedings Katalin Balogné Bérces
30
English or conforms to universal syllable unmarkedness. Notice that Davidson’s conclusion concerning speech rate – viz., that it does not necessarily have an effect on syncope, since for rate-independent speakers syncope is a general characteristic of their dialect – chimes with Dalby’s, and at the same time, the role of (universal) phonotactics she identifies is at best unsupportive of the traditional conception of unrestricted pre-stress syncope. Although in a later study, Davidson (2006) somewhat modifies her view, on the basis of this and other corpus studies we can detect a weak tendency in pre-tonic syncope to obey some kind of a universal unmarkedness (which is more in line with Hooper than with Zwicky).
The claim that pre-stress syncope is identical to word-initial syncope (cf. Harris’ (a) in the above quote) has also been challenged. Dalby, for instance, finds almost as high a percentage of deletions in word-marginal pre-stress locations as in medial ones in all three speech styles he investigates (relatively formal Television English, slow and fast reading).
What is even more surprising is that his word-medial pre-stress percentages are always (somewhat) higher than the corresponding word-initial data (although the difference is either non-significant or only marginally significant).
In sum, the traditional description of the two subtypes of English syncope has hardly received empirical echo in experimental investigations. However, unfortunately, even the results of such corpus studies are sometimes difficult to interpret, since, besides the segmental constraints mentioned above, syncope seems to be highly variable along other dimensions such as tempo, style, dialect11, intraspeaker variation12, and word frequency13, and it is close to impossible to control the experiments for all of them. In addition, the potential interference with syllabic consonant formation14, which also results in the loss of the schwa, further complicates the picture, and raises the issue of the method of data evaluation: a phonetic study aimed at schwa deletion as such may classify cam’ra (< camera) and butt’n (< button) identically, which may undesirably skew the results. (See also Carlotti et al. 2009’s criticism of Dalby’s methods.) These complications sometimes even lead to contradictory data, which is frequently, but not always, attributable to differences in data collection or other aspects of methodology (e.g., Patterson et al. 2003 attribute their results being somewhat divergent from Dalby’s to the difference between dialogue vs. monologue type texts; Dalby considers all cases of deleted schwas while Davidson excludes the ones bearing any phonetic traces (cf.
footnote 10), etc. – see also Kürti 1999).
11 For example, as demonstrated in Turcsán and Carlotti (2010), pre-tonic syncope shows marked regional differences, namely, in their corpus, speakers from Santa Barbara, California, syncopate the most, followed by speakers from Lancashire and Ayrshire. See also, e.g., Davidson (2006) for individual variability. Speakers of different dialects also disagree on whether or not in words like police the schwa is fully deleted and the result is a monosyllabic word, cf. the exchange between Geoffrey S. Nathan and A. F. Gupta on LinguistList (http://linguistlist.org/issues/11/11-681.html).
12 E.g., it does not hold true that a word that is syncopated by a speaker at a slower rate will be syncopated by the same speaker at a faster rate, cf., e.g., Carlotti et al. (2009)’s results.
13 More frequent words are more prone to undergo syncope (cf. Fidelholtz 1975, Hooper 1978, Patterson et al.
2003), making it a popular topic for usage-based models (e.g., Bybee 2001, Pierrehumbert 2002). Cf. summary vs. summery, nursery vs. cursory.
14 “In English, synchronic/dynamic syncope is always preceded – historically/derivationally – by SCF, for any C1əC2~C1C2 alternation there exists an intermediate C1Ç2 stage; the reverse [...] does not hold. This fact points to a strong connection between CÇ and syncope-created consonant clusters.” (Szigetvári 2002: 139) It has to be added, though, that Szigetvári bases this claim on post-stress syncope only, where he also excludes examples like vegetable. For him, it is “obvious that their development followed a different path: the obstruent following the
“trace” of the schwa could never have been syllabic” (ibid: 148). (This is also true for many examples of pre-stress syncope.) Most probably on the same grounds, Nádasdy (1994), too, distinguishes between the two subtypes, and only considers cases like vegetable as examples of syncope, dubbing schwa deletion in family and the like Sonorant Desyllabification.
31
On the basis of the comparison of the corpus studies referred to above on the one hand, and the comparison of their results to the standard descriptions on the other, we can draw two major conclusions concerning the facts of schwa deletion in English.
First, contrary to the classical view, there is no clear difference between pre-stress and post-stress syncope in terms of speech tempo. Faster speech seems to boost both in basically the same way, thus the difference caused by speech rate is only quantitative.
Second, the role that the phonotactics of the consonants straddling the elision site plays (henceforth, the cluster effect) is ambivalent. While traditional formal analyses insist on a strict strength/sonority restriction in post-tonic syncope but not word-initially, experimental evidence does not support this; rather, there appears to exist a complicated, variable system determined by manner (sonority) primarily and place/voicing secondarily, which boils down to a weak tendency to observe universal syllable structure principles in both subtypes of syncope.