2.1.1 Two classes of suffixes

English stems show different phonological behaviour depending on the types of suffix which follow them. Stems followed by Level II suffixes, i.e. inflectional suffixes such as -ing (present participle), -ed, -s and derivational suffixes of Germanic origin such as -er, -ness, -less, -ing (gerundive), behave phonologically as if they constitute a word-level constituent on their own: the stems (e.g. sing in singing and parent in parenting) and their corresponding independent lexical words have the same surface representations, as shown in (2). In (2a), the stem-final and lexical word-final consonant cluster /ŋg/ is simplified to [ŋ], and in (2b) the stem and the lexical word (henceforth LWd) counterpart share initial stress.

1. (2)

   

In contrast, stems followed by derivational suffixes of Latin origin such as -al, i.e. Level I suffixes, show differences in stress placement and segmental behaviour as compared to related independent lexical words.Footnote ¹ For example, unlike LWd-final consonants, stem-final /ŋg/ consonant sequences do not undergo cluster simplification before Level I suffixes, as shown in (3).

1. (3)

   

In addition, stress-placement patterns may differ for Level I affixed stems in comparison with their corresponding independent words. In (3), the main stress of the LWd occurs on the initial syllable, but on the final syllable when the stem is followed by the Level I suffix.

As we discuss in §2.1.2, differences in the phonological behaviour of stems can be accounted for by postulating word-level boundaries before Level II suffixes but not before Level I suffixes. However, there are also other ways to account for these differences, as we will see in §2.1.3. In §§2.4 and 2.5 we discuss experiments designed to test the word boundary proposal.

2.1.2 A morphological word boundary proposal

Chomsky & Halle (Reference Chomsky and Halle1968) assign different boundary markers for Level I vs. Level II affixation (‘+’ for Level I and ‘#’ for Level II), and claim that phonological rules are sensitive to the presence of a # boundary. For example, word-stress assignment rules do not apply to a string of segments that contains # (Chomsky & Halle Reference Chomsky and Halle1968: 85), while the presence of a + boundary does not block their application. Thus, ˈparent in ˈparent#ing and its corresponding independent word in (2b) above have the same stress pattern, because word-stress assignment rules only apply within parent in both cases.

Selkirk (Reference Selkirk1982a, 1984) reinterprets the differences between Level I and Level II suffixes in terms of different levels of morphological constituents, i.e. morphological word and Root. Stems to which Level II affixes attach have morphological word (henceforth MWd) status, while stems to which Level I suffixes attach form a Root-level constituent only, one level below MWd. Furthermore, a combination of Root+Level I suffix also forms a Root-level constituent, according to the morphological rules adopted in Selkirk (Reference Selkirk1982a: 95ff, 1984: 76), as shown in (4).

1. (4)

   

According to Selkirk (Reference Selkirk1984), stress-related differences between Level I and Level II cases can be accounted for by proposing that stress-assignment rules, i.e. rules of ‘metrical grid construction’, apply cyclically only to designated morphosyntactic domains, i.e. Root-level domains. Thus Level II suffixes are opaque to stress assignment. Furthermore, Selkirk (Reference Selkirk1982a: 90) states that the morphological category Root is the domain of syllabification, so that a stem-final consonant, for example [g] in diphthong-al in (3), is syllabified with the following Level I suffix. However, the same stem-final consonant [g] in sing-ing is in MWd-final position in (2a), and cannot be syllabified with the following Level II suffix across a MWd boundary. The MWd-final cluster /ŋg/ in (2a) therefore undergoes simplification because such a sequence is illegal in a syllable-final position.

2.1.3 The PWd boundary proposal

Following the Indirect Reference Hypothesis that phonological rules only refer to phonological structure but not directly to morphosyntactic structure (Selkirk Reference Selkirk1986, Inkelas Reference Inkelas1990), Selkirk's (Reference Selkirk1984) MWd vs. Root distinction for stems with Level I vs. Level II suffixes can be translated into representational differences in prosodic constituent structure.Footnote ² MWd boundaries that coincide with the edges of Level II stems should have prosodic counterparts, i.e. the edges of some prosodic domain α, from which Level II stems are excluded. In contrast, Root-level stems and their Level I suffixes coexist within the same prosodic domain α. Aronoff & Sridhar (Reference Aronoff, Sridhar, Richardson, Marks and Chukerman1983) and Szpyra (Reference Szpyra1989), among others, propose that this prosodic domain α is a PWd.

A word-level phonological (prosodic) constituent was originally proposed by Selkirk (Reference Selkirk1980a) to account for main word stress assignment in English. According to her, binary branching prosodic constituent structure above the syllable, i.e. the foot and the prosodic word, determines weak–strong relations among syllables. On this view, the location of main stress is ultimately derived from rules relating to these constituents: in English, the leftmost syllable in a foot is strong, i.e. stressed, and a strong syllable dominated by the strongest foot of a PWd bears the main word stress. The fact that Level II suffixes do not affect the placement of main word stress on stems is due to their stems forming an independent PWd.

The PWd has also been proposed to be the domain for segmental rule application (Selkirk Reference Selkirk1980a, Reference Selkirk, Aronoff and Keanb, Nespor & Vogel Reference Nespor and Vogel1986). There are even rules that apply only to PWd junctures, such as voicing assimilation and deaspiration in Sanskrit (Selkirk Reference Selkirk, Aronoff and Kean1980b). The consonant-cluster simplification observed in English Level II-suffixation cases is also captured as a PWd juncture phenomenon.

The adjunction of Level II stems to a PWd-level constituent can be described in Optimality Theory terms by a highly ranked constraint requiring a string of segments that correspond to Level II suffixes to be immediately preceded by a PWd base, introduced by McCarthy & Prince (Reference McCarthy and Prince1993b: 67ff) as Suffix-to-PWd, and modified as in (5).Footnote ³

1. (5) LevelII-Suffix-to-PWd

   The terminal string of segments in the output representation, corresponding to a Level II suffix in the input, is immediately preceded by a PWd. That is, the base of Level II suffixation is a PWd.

If LevelII-Suffix-to-PWd is undominated in English, Level II suffixes should be preceded by a PWd boundary regardless of their segmental content and organisation. However, Raffelsiefen (Reference Raffelsiefen, Downing, Hall and Raffelsiefen2005: 234) claims that stems with a single obstruent suffix, e.g. girl-s, should have no PWd break between the right edge of the stem and the suffix, as shown in (6a). According to her, the representation in (6b), which satisfies LevelII-Suffix-to-PWd, should be ruled out, on the assumption that all segments must be parsed into syllables, i.e. Parse(seg) (‘every segment is parsed into a syllable’) is undominated.

1. (6)

   

However, we do not adopt Raffelsiefen's view, because the assumption that Parse(seg) is undominated is a stipulation without empirical support. Rather, there is evidence that may support the view that the suffix -s in English is actually extrasyllabic. Walsh & Parker (Reference Walsh and Parker1983) found that the suffix -s, as in (6b), was longer than word-final s in monomorphemic forms. On the assumption that segment duration relates to prosodic structure, their observation suggests that the suffix -s forms part of a different prosodic structure than corresponding segments in monomorphemic forms. It therefore appears possible to represent the suffix -s as phonologically extrasyllabic.

Hall (Reference Hall2001) and Raffelsiefen (Reference Raffelsiefen, Downing, Hall and Raffelsiefen2005) further claim that there is no PWd boundary between a stem and a vowel-initial Level II suffix such as -ing, and propose that a PWd boundary should be present only before Level II suffixes such as -ness and -less, which begin with a consonant. However, as noted in §2.1.1, Level II suffixes that begin with vowels induce similar phonological behaviour to consonant-initial Level II suffixes, i.e. final consonant cluster simplification, as shown again in (7a), and stress neutrality, as in (7b).

1. (7)

   

These phenomena provided the original motivation for a word-level boundary before Level II suffixes, regardless of whether they are vowel- or consonant-initial, contra Raffelsiefen and Hall.

It is nevertheless worthwhile considering why Raffelsiefen and Hall argue against a PWd boundary before vowel-initial suffixes. According to Hall (Reference Hall2001), these suffixes absorb the final consonant of the immediately preceding stem as their syllable onset, as shown in (8) for sunning, and this configuration is claimed to be incompatible with a PWd boundary between the stem-final consonant and a Level II suffix.

1. (8)

   

However, flapping phenomena in American English suggest that stem-final consonants immediately preceding these vowel-initial Level II suffixes are not completely integrated into the syllable of the suffix. For example, the stem-final consonant /t/ in date surfaces as a flap when followed by -ing and -er, as in da[ɾ]ing and da[ɾ]er, and is therefore very different from the realisation of syllable-initial /t/ in word-initial and pretonic positions. The flapping of an intervocalic alveolar stop between a stressed vowel and an immediately following unstressed syllable nucleus has been proposed to be due to its being ambisyllabic, i.e. being simultaneously the coda of the preceding syllable and the onset of the following syllable (Kahn Reference Kahn1976). Selkirk (Reference Selkirk, Hulst and Smith1982b), however, claims that it is unambiguously syllabified into the preceding stressed syllable. Turk (Reference Turk1993, Reference Turk and Keating1994) presents experimental evidence that suggests that intervocalic consonants in this context are more similar in their articulatory characteristics to syllable-final than to syllable-initial stops. Turk (Reference Turk and Keating1994), in particular, lends support to Selkirk's (Reference Selkirk, Hulst and Smith1982b) view. She compared the upper lip gestures of clearly syllable-initial labial stops (e.g. reˈpair), those of clearly syllable-final stops (e.g. ˈcaptor) and those of stops immediately preceded by stressed vowel and followed by unstressed nucleus (e.g. ˈleper), whose syllabic affiliation is not known. She found that the upper lip gestures of the unknown stops patterned with those of the clearly syllable-final ones.Footnote ⁴

Contrary to Hall's (Reference Hall2001) proposal in (8), Kahn's (Reference Kahn1976) and Selkirk's (Reference Selkirk, Hulst and Smith1982b) views of the syllabification of post-tonic intervocalic consonants into the preceding tonic syllable allow for a PWd boundary between the stem and the suffix, as shown in (9).

1. (9)

   

In the representation in (9a), the ambisyllabic affiliation of the stem-final consonant violates the CrispEdge(σ) constraint in (10), which prohibits double linking of segments at syllable edges (Itô & Mester Reference Itô, Mester, Merchant, Padgett and Walker1994).

1. (10) CrispEdge(σ)

   Segments that coincide with the edges of σ must be exhaustively contained in σ.

Nonetheless, the representation in (9a) may still surface as optimal, provided that Onset, i.e. a constraint that calls for an onset consonant in every syllable, dominates CrispEdge(σ), as suggested by Itô & Mester (Reference Itô, Mester, Merchant, Padgett and Walker1994), who claim that the flapping of word-final /t/ immediately followed by a word-initial vowel, as in sough[ɾ] Ed is due to its being ambisyllabic.

We are not in a position to choose between the representations in (9a) and (9b). The significant point here is that both of these approaches allow for a PWd boundary between the stem and the following Level II suffix. One of the goals of the current study is to test the possibility that a PWd boundary occurs between a stem and a Level II suffix, even when the suffixes are monoconsonantal (e.g. -s) or begin with a vowel (e.g. -ing).

2.1.4 Approaches without a PWd boundary: paradigm uniformity and serialism

In Raffelsiefen's (Reference Raffelsiefen, Downing, Hall and Raffelsiefen2005) approach, where a PWd boundary does not occur before unstressed vowel-initial Level II suffixes, the coda-cluster simplification observed in e.g. conde[m]ing should be due to Paradigm Uniformity or output–output (OO) correspondence.

The principal tenet of the OO-correspondence approach is that the phonological similarity between two related forms in the same paradigm (e.g. a simplex word and a suffixed word) comes from a mapping relationship between the output forms of suffixed words and those of their base (e.g. simplex words). According to Benua (Reference Benua1997), ‘word-boundary’ phenomena, e.g. stem-final consonant-cluster simplification before Level II suffixes, are formalised by adopting ‘OO-Identity’ constraints which require the optimal output of the stem in the affixed word (e.g. condemn in condemning) to be identical to the surface representation of its base form (the stem forming an independent word on its own without affixes, e.g. (conde[m])_PWd, accompanied by word-final consonant-cluster simplification).

In Benua's model, there are two kinds of OO-Identity constraints: one referring to Level I affixation and the other referring to Level II affixation. Level I OO-Identity constraints and Level II OO-Identity constraints are ranked differently with respect to other types of constraints, such as input–output (IO) constraints, i.e. constraints requiring the output string to have identical representations to its input. Since Level I OO-Identity constraints are dominated by IO constraints, the surface representation of the Level I suffixed stem retains all the segments from its input representation, e.g. conde[mn]ation. In contrast, Level II OO-Identity constraints outrank the IO constraints, and therefore result in consonant-cluster simplification at the end of the Level II suffixed stem. Similarly, the highly ranked Level II OO-Identity constraints force the stress location of Level II suffix forms and base forms to be identical. In this way, the OO-Identity approach captures the difference in the behaviour of Level II and Level I suffixed stems, without resorting to postulating differences in their morphological and phonological constituent structure.

Another approach that does not require a PWd boundary before Level II suffixes is a serial derivational approach (e.g. Siegel Reference Siegel1974, Kiparsky Reference Kiparsky, Hulst and Smith1982, Halle & Mohanan Reference Halle and Mohanan1985, Borowsky Reference Borowsky, Hargus and Kaisse1993).Footnote ⁵ According to Borowsky (Reference Borowsky, Hargus and Kaisse1993), for example, Level I suffixes attach to their stems earlier in the derivation, and the combination of stem+Level I suffix undergoes word-level phonological operations as a single unit, while Level II suffixes attach to their stems after the stems have independently undergone word-level phonological operations such as stress assignment and final coda-cluster simplification. It has also been proposed that internal constituent boundaries are erased after completion of phonological rule application at each lexical level, i.e. ‘bracket erasure’ (Kiparsky Reference Kiparsky, Hulst and Smith1982: 140), so that phonological processes at later levels are inaccessible to any constituent structure constructed during earlier levels of derivation. In this approach, the phonological differences between Level I and Level II suffixed forms such as condemn-ing and condemn-ation are explained by ordered rules that apply to different morphological levels, rather than by the necessary occurrence of a PWd boundary before the Level II suffix.

In summary, there is no clear agreement as to the existence of word-internal PWd boundaries in Level II suffixed forms. The main goal of the experiments presented in what follows is to test for durational evidence for such boundaries. Segment duration is investigated, since this is known to vary systematically according to the presence of a prosodic boundary, and its level within the prosodic hierarchy (e.g. Wightman et al. Reference Wightman, Shattuck-Hufnagel, Ostendorf and Price1992). Experiment 1 tests for durational differences between monomorphemic words and Level II suffixed forms, and Experiment 2 tests for differences between Level I suffixed forms and Level II suffixed forms. Word-internal PWd boundary proposals predict that there should be durational differences in both cases.

2.4.1 Speakers

Five native female speakers of Scottish Standard English (SSE) from Edinburgh served as paid subjects in the experiment. None reported any hearing or speaking difficulties. They were all University of Edinburgh undergraduate students (aged between 18 and 40) at the time of the recording.

The phonemic inventory of SSE provided in Table I. The most notable durational features of SSE are the patterns arising from the Scottish Vowel Length Rule (SVLR), which applies primarily to the vowels /i ʉ ʌi/ (Scobbie et al. Reference Scobbie, Hewlett, Turk, Foulkes and Docherty1999, Scobbie et al. Reference Scobbie, Gordeeva and Matthews2006), lengthening them immediately before lexical word boundaries, Level II morpheme boundaries, voiced fricatives and /r/. They are short elsewhere. Therefore, both brute and brood are pronounced with a short vowel [ʉ], while brewed (brew#ed) has long [ʉː]. According to Scobbie et al. (Reference Scobbie, Hewlett, Turk, Foulkes and Docherty1999), the difference in duration between vowels in the short and long contexts is 40 to 80 ms.

Table I

The phonemes and major allophones of Scottish Standard English, adapted from Scobbie et al. (2006: 7). The arrows indicate ‘a particularly notable range of productions’.

In our experiment, we avoid words that would undergo the SVLR. Potential durational differences between Level II suffixed stems and their monomorphemic counterparts in our experiments are therefore expected to be much more subtle than those of the SVLR.

2.4.2 Stimulus design

The stimulus sets consisted of (i) bimorphemic items consisting of closed syllable stems followed by a Level II suffix and (ii) their monomorphemic counterparts. Level II suffixes included the voiceless plural or 3rd person singular suffix -s [s], the voiceless past tense suffix -ed [t] and two syllabic suffixes: the gerundive suffix -ing [ɪŋ] and the plural inflectional suffix -es [ɪz]. Test pairs therefore included monosyllables such as tack-s vs. tax and miss-ed vs. mist, as well as disyllables such as bake-ing vs. bacon, puff-ing vs. puffin and ax-es vs. axis.

We emphasise here that the test sequences in disyllabic pairs crucially contain the same number of letters, and are therefore not confounded by potential spelling effects on duration: pairs like puff-ing and puffin have the same number of letters for their target VC sequences. Non-homophonous pairs like baking vs. bacon were included to ensure that any durational differences between the two morphological conditions would not be due to speakers artificially contrasting members of homophonous pairs within the experimental task.

Three pairs were chosen for each of the monomorphemic word comparison sets, e.g. tack-s vs. tax (Set 1) and miss-ed vs. mist (Set 2), seven pairs for one of the disyllabic word comparison sets, e.g. bake-ing vs. bacon (Set 3) and two pairs for the other disyllabic word comparison set, e.g. ax-es vs. axis (Set 4). The four sets of test pairs (15 pairs in total) are shown in Table II. No proper names were included in the stimulus sets, because an earlier pilot study suggested that proper names tend to be longer than words spelled without capitals.

Table II

Monosyllabic (Sets 1 and 2) and disyllabic (Sets 3 and 4) item pairs used in Experiment 1.

We provide in Fig. 3 the log word frequencies of the stimulus pairs, which are obtained from the CELEX database (Baayen et al. Reference Baayen, Piepenbrock and van Rijn1993). Lexical frequency is known to contribute to segment duration (e.g. Munson Reference Munson2001, Bell et al. Reference Bell, Gregory, Brenier, Jurafsky, Ikeno and Girand2002, Aylett & Turk Reference Aylett and Turk2004, Pluymaekers et al. Reference Pluymaekers, Ernestus and Baayen2005). For example, Aylett & Turk (Reference Aylett and Turk2004) show that words whose log frequency exceeds 4 to 4·5 are realised with shorter durations than less frequent ones. Since we limited the log frequency count of each item to below 2·5, we expected little effect of frequency on the durations of our items.

Figure 3

Log word frequency count of items used in Experiment 1.

2.4.3 Elicitation and recording

Each set of sentences was printed on a set of cards and presented to speakers in random order (by shuffling). Our strategy was to elicit target words in contexts where the subtle durational effects expected here would be most likely to occur. Because the durational effects of word-constituent structure are known to be larger in pitch-accented contexts than non-pitch-accented contexts (Beckman & Edwards Reference Beckman, Edwards, Kingston and Beckman1990, Turk & Shattuck-Hufnagel 2000), background and carrier sentences were designed to elicit nuclear pitch accents on target words (by contrastive focus for Sets 1 and 2 and presentational focus for Sets 3 and 4), as illustrated in the examples in (11) (the full set of test materials is given in the Appendix).

1. (11)

   1. a. A special event at the cooking institute. Say ‘canning event’ for me.

   2. b. The 1 o'clock event at the Edinburgh Castle. Say ‘cannon event’ for me.

Target words are italicised in (11), but were not highlighted in any way when presented to subjects. The topic sentence in (11) states that the following carrier sentence refers to a particular event, and the target word in the carrier sentence further specifies the type of event. The target word is therefore interpreted as new (i.e. in presentational focus), and speakers put a nuclear pitch accent on the target word, while the words following the target word, i.e. event for me, are unaccented. Within each carrier sentence, the target word was immediately preceded by Type (when the target word began with a vowel or a rhotic) or Say (when the target word began with another consonant). This first word in the carrier sentence will henceforth be called the head. The target word was also followed by one or more words (the post-target). The head and the post-target were kept constant between the two utterances in each comparison pair.

We elicited our materials at different rates, on the assumption that boundary effects would be magnified at slow speech rates. Beckman & Edwards (Reference Beckman, Edwards, Kingston and Beckman1990) found that the duration difference of the [ɑ] in pop opposed vs. poppa posed was magnified by about 50~100% at a slower speech rate.

Our recording consisted of multiple blocks, where each block was read at either a normal or a slow rate. Only the blocks consisting of pairs from Sets 3 and 4 were read at an extra-slow rate in addition to the normal and slow speech rates. When reading at the normal rate, speakers were instructed not to put any pause between words in a single sentence; none of the speakers had difficulty in doing this, i.e. they read the sentences fluently. If they stumbled over the pronunciation of a sentence, they were instructed to read the whole sentence again from the beginning, and the disfluent speech was discarded. When reading at the slow rate, speakers were instructed to read not only slowly but also articulately (‘please say these phrases as if you talk to someone having difficulty in hearing or to small children’), and for the extra-slow rate they were told to read even more slowly than for the slow rate speech produced in the previous block. The use of articulate speech helped speakers to slow down their speech rate. At the slow rates, speakers were allowed to put pauses between words.

The two stimuli in each comparison pair (e.g. the sentences in (11a) and (b)) always occurred together in the same block, but were never presented to speakers consecutively. For each of the speech rates, six repetitions were elicited for stimulus pairs in Sets 1 and 2 for four of the five speakers; four repetitions were elicited in all other cases. Note that not all pairs were read by all of the five speakers. Two of the nine pairs in Set 3 (i.e. puffing vs. puffin and bobbing vs. bobbin) and the two pairs in Set 4 (i.e. axes vs. axis and bushes vs. bushel) were read by four of the five speakers.

Recordings were made in a sound-treated recording studio at the Department of Theoretical and Applied Linguistics, University of Edinburgh, using a hypercardioid microphone and a SONY PCM-2700A DAT recorder. Recordings were amplified, low-pass filtered at 48 kHz and downsampled to 16 kHz for further analysis.

2.4.4 Measurements and analyses

We measured the rhyme duration of the stem of each target word, as well as the corresponding sequence in the monomorphemic counterpart. For example, the target sequence for the monosyllabic words tacks and tax [tʰaks] was the sequence [ak], i.e. the nucleus vowel and the final coda consonant of the stem. In the same way, the target sequence for baking and bacon (i.e. [bekɪŋ] and [bek]) was the sequence [ek].

We defined the target sequence acoustically as starting at the onset of the word-initial consonant-constriction release and ending at the onset of stem-final coda consonant-constriction release (or the onset of the corresponding coda consonant in the monomorphemic counterparts).Footnote ⁸ Therefore, the period of aspiration immediately following the constriction release of a word-initial voiceless stop was also included in the target sequence. We chose this criterion because the release of each word-initial consonant can be interpreted as the onset of supralaryngeal articulation of the following nucleus vowel (see Turk et al. Reference Turk, Nakai, Sugahara, Sudhoff, Lenertová, Meyer, Pappert, Augurzky, Mleinek, Richter and Schließer2006 for discussion).

Our speakers pronounced the initial rhotic of the raisin and raising pair as an approximant [ɹ]. The segmentation of prevocalic approximant variants is not straightforward, because voicing continues from the approximant to the following vowel, formant trajectories are continuous throughout the approximant–vowel sequence and amplitude does not change abruptly from the approximant to the vowel. [ɹ] was therefore included in the target-measurement sequence.

Part of the initial word, i.e. the head, of each carrier sentence (Say or Type) was also measured, as a check for overall speech-rate differences between test conditions. The interval measured for Say began at the frication offset of the onset /s/ and ended at the F2 offset of the vowel. For Type the interval began at the onset of constriction release for /t/ and ended at the onset of constriction release for /p/. These control intervals were therefore comparable to the test intervals in size.Footnote ⁹

Because each test sentence in the suffixed condition was paired with a corresponding test sequence in the monomorphemic condition in each recording block, we analysed our data using paired t-tests. One-tailed tests were adopted instead of two-tailed tests, because we were testing the hypothesis that the duration of one condition is longer than the other (rather than assessing a more general hypothesis of a difference in either direction).

2.4.5 Results 1: speech-rate index

The mean duration of the control segment (i.e. the rhyme of the initial word of each carrier sentence: Say and Type) of the stem–suffix cases and that of the monomorphemic cases were compared, using a two-tailed paired t-test, for each comparison pair at each speech rate. Results suggested that speech rate was comparable across experimental conditions, i.e. the control-segment duration mean of the stem–suffix cases was neither significantly longer nor significantly shorter than that of the monomorphemic cases.

The only exception was Set 4 at the normal speech rate: the control-segment duration mean for the stem–suffix cases was 2·3% (i.e. 5 ms, SD=18) longer than that of the monomorphemic case, which was marginally significant in a two-tailed paired t-test (α=0·05): (control-segment mean for the stem–suffix cases=174 ms (SD=63), control-segment mean for the monomorphemic cases=169 ms (SD=60); t=1·68, df=47, p=0·1).

We therefore consider it unlikely that any observed systematic differences between our test sequences across experimental conditions are due to differences in overall speech rate, with the possible exception of the differences found in Set 4 (at the normal rate only).

2.4.6 Results 2: test sequences

The differences observed between stem–suffix and monomorphemic cases were in the direction predicted by the word-internal PWd-boundary hypothesis for some sets at the normal rate, and for all sets at the slow speech rate.

2.4.6.2 The slow rate

At the slow rate, all four sets showed statistically significant differences in the direction predicted by the word-internal PWd hypothesis, i.e. suffixed rhymes were reliably longer than the corresponding stretches in monomorphemic words. For Set 1, the mean of the stem–suffix cases was 336 ms (SD=54) and that of the monomorphemic cases was 321 ms (SD=43), a difference of 4·7% (15 ms, SD=35; t=3·281, df=59, p=0·001). For Set 2, the mean of the stem–suffix cases was 379 ms (SD=88) and that of the monomorphemic cases was 362 ms (SD=75), a difference of 4·7% (17 ms, SD=43; t=2·989, df=59, p=0·002). For Set 3, the mean of the stem–suffix cases was 303 ms (SD=118) and that of the monomorphemic cases was 289 ms (SD=109), a difference of 4·8% (14 ms, SD=46; t=3·432, df=130, p=0·0005). For Set 4, the stem–suffix mean was 409 ms (SD=145) and the monomorphemic mean was 381 (SD=116), a difference of 7·3% (28 ms, SD=78; t=2·025, df=31, p=0·026).

Additional analyses were conducted for pairs that satisfied our word-frequency criteria. The mean of the stem–suffix cases was significantly longer than that of the monomorphemic cases in three of the four sets: Set 1 (stem–suffix mean=337 ms (SD=43), monomorphemic mean=317 (SD=53), difference=20 ms (SD=46, stem–suffix 6·3% longer than monomorph); t=1·91, df=19, p=0·036), Set 2 (stem–suffix mean=422 ms (SD=71), monomorphemic mean=397 (SD=51), difference=25 ms (SD=43, 6·3%); t=3·58, df=39, p=0·0005), Set 3 (stem–suffix mean=370 ms (SD=127), monomorphemic mean=347 (SD=108), difference=23 ms (SD=50, 6·6%); t=3·30, df=51, p=0·001). Although Set 4, i.e. the bushes vs. bushel pair, also showed the same direction of difference between the two morphological conditions: (stem–suffix mean=342 ms (SD=137); monomorph mean=311 ms (SD=77)), the difference of 31 ms (SD=91, 10%) was not statistically significant: t=1·35, df=15, p=0·10. This may be due to the fact that there was more inter-speaker variation for the bushes vs. bushel pair in Set 4 than for the pairs in other sets, as shown in Fig. 5: only Speaker A showed a substantial difference in the expected direction, 32·4%, while the other three speakers did not.Footnote ¹¹ That is, the relatively large difference of 10% in Set 4 comes from Speaker A only, and as a result the difference is non-significant. These results are shown in Fig. 4.

Figure 4

(a) Mean rhyme durations (ms) at the slow speech rate, from pairs satisfying the frequency criteria. P-values preceded by an asterisk indicate statistical significance in a one-tailed t-test comparison of the two morphological conditions (α=0·05). (b) Differences in mean rhyme duration of stem–suffix cases as compared to monomorphemic cases, expressed as a percentage, for pairs satisfying the frequency criteria: (suffixed rhyme duration/monomorphemic rhyme duration – 1)×100.

The inter-speaker variation of the other three sets was far smaller, as shown in Fig. 5. For Set 1, four of the five speakers showed the expected difference. For Set 2, the majority of the speakers also showed the expected difference, although Speaker C showed difference in neither of the two item pairs. For Set 3, all speakers except Speaker R showed the expected difference of more than 3% in the comparison of raising vs. raisin. For the other pairs (bobbing vs. bobbin, puffing vs. puffin), there were always two speakers who showed the expected difference of more than 5%. Furthermore, there is one crucial aspect shared by all pairs in the four sets: when there was a difference of more than 1% between the two morphological conditions, the difference was always in the expected direction.

Figure 5

Differences in first syllable mean rhyme durations at the slow rate for stem–suffix vs. monomorphemic cases, expressed as a percentage, for pairs satisfying the frequency criteria. Speaker C did not record the bobbing vs. bobbin, puffing vs. puffin and bushes vs. bushel pairs.

2.4.6.3 The extra-slow rate

The stimulus pairs in Sets 3 and 4 were also recorded at an extra-slow speech rate. Both sets showed a mean difference in the direction predicted by the word-internal PWd-boundary hypothesis. However, the difference was statistically significant only for Set 3 (stem–suffix mean=403 ms (SD=179), monomorphemic mean=385 ms (SD=169), difference=18 ms (SD=109, 4·7%); t=1·892, df=126, p=0·03). For Set 4, there was no significant difference between the two morphological conditions.

Similar results were obtained for the pairs satisfying our word-frequency criteria in Sets 3 and 4, as shown in Fig. 6. For Set 3, the mean of the stem–suffix cases was 481 ms (SD=195) and that of the monomorphemic cases was 439 ms (SD=182), with a mean difference of 42 ms (SD=67, 9·6%; t=4·47, df=50, p=0·00). For Set 4, there was no significant difference between the two morphological conditions.

Figure 6

(a) Mean rhyme durations (ms) at the extra slow speech rate, from pairs satisfying the frequency criteria. P-values preceded by an asterisk indicate statistical significance in a t-test comparison of the two morphological conditions (α=0·05). (b) Differences in mean rhyme duration of stem–suffix cases as compared to monomorphemic cases, expressed as a percentage, for pairs satisfying the frequency criteria.

Although one speaker showed a substantial difference between the two morphological conditions for Set 4 at the slow speech rate, none of the speakers showed a difference at this extra-slow rate. In contrast, four of the five speakers showed a difference of more than 7% in Set 3, as shown in Fig. 7.

Figure 7

Differences in first syllable mean rhyme durations at the extra slow rate for stem–suffix vs. monomorphemic cases, expressed as a percentage, for pairs satisfying the frequency criteria. Speaker C did not record the bobbing vs. bobbin, puffing vs. puffin and bushes vs. bushel pairs.

2.4.7 Summary: Experiment 1

Overall results were consistent with the word-internal PWd hypothesis for most of the word comparisons at the slow speech rate. Analyses for sets of pairs that satisfied the frequency criteria and those that did not satisfy the criteria also showed similar results. Target sequences were longer in suffixed forms for two types of monosyllabic word pairs (e.g. tacks vs. tax and missed vs. mist): these showed similar behaviour to the -s suffixed forms in Schwarzlose & Bradlow (Reference Schwarzlose and Bradlow2001). Target sequences were also longer in suffixed forms of disyllabic words controlled for homophony and the number of letters used in the spelling of the target sequences (e.g. baking vs. bacon, puffing vs. puffin). These results are consistent with the view that the Level II suffixed words are different from their monomorphemic counterparts in their prosodic constituent representation. On this view, the difference in prosodic structure is reflected in surface segment durations through the application of phonetic implementation processes which refer the structure.

It should be noted that there was a difference among sets: although three of the four sets (Sets 1, 2 and 3) showed a difference between Level II suffixed and monomorphemic forms at the slow and extra-slow rates, Set 4 did not. Unfortunately, Set 4 contained just two pairs, and only one of them satisfied our frequency criteria. It is therefore unclear whether the exceptional behaviour of Set 4 is due to the small number of pairs or to the intrinsic nature of -es suffixation. This should be further investigated with pairs such as lex-es vs. lexis (both of them equally infrequent), miss-es (more frequent) vs. missus (less frequent), etc.

Given that we did not obtain the expected result in Set 4 for the suffix -es, the suffixes that induced the expected difference in our experiment are -s, -ed and -ing. The sets with the two consonantal suffixes, however, contained pairs whose stem–suffix and monomorphemic members were different in the number of letters with which they are spelled, which might have affected the durational differences between these two morphological forms. However, we also obtained longer target durations in stem+-ing cases as compared to their monomorphemic counterparts, in spite of the fact that stem+-ing cases did not have spelling confounds.

Notice also that the durational differences across the three speech rates in Set 3 are consistent with Beckman & Edwards (Reference Beckman, Edwards, Kingston and Beckman1990), who showed that boundary effects are more exaggerated at slower speech rates. The pairs in Set 3 that satisfy our frequency criteria showed enhancement from the normal to the slow rate, and from the slow to the extra-slow rate. Figure 8 shows how the magnitude of the durational difference between the two morphological conditions of Set 3 varied across the three rates of speech (shown as a percentage), was obtained from the pairs that satisfy the frequency criteria (i.e. bobbing vs. bobbin, puffing vs. puffin, raising vs. raisin). There was no such enhancement from the slow rate to the extra-slow rate in Set 4, which was exceptional because only Speaker A showed a substantial difference between the two morphological conditions in the slow rate of speech. The speaker showed an unusually large difference (i.e. 32·4%; see Fig. 5) at the slow rate of speech, while the difference was reduced to −2·2% (see Fig. 7) in the unpredicted direction at the extra-slow rate.

Figure 8

Percentage differences in mean rhyme duration for Set 3 stem–suffix words as compared to monomorphemic words, at different speaking rates. These differences come from pairs satisfying the frequency criteria.

A further finding is that there were inter-speaker differences in the magnitude of the difference between the two morphological conditions. One of our speakers, Speaker A, consistently showed the greatest differences between the two morphological conditions; she was the only speaker that showed the expected difference for Set 4.

2.5.1 Stimulus design

Stimulus sets consisted of the eight pairs of Level I and Level II suffixed words in Table III.

Table III

Item pairs used in Experiment 2.

The Level II suffix tested in this experiment was always -ing, while the Level I suffixes were -ance, -ence, -ent and -al. The suffixes -ance and -ence belong to the Level I class because they can trigger stress alternation of the stem (e.g. igˈnore vs. ˈignorance, preˈfer vs. ˈpreference) and also attach to bound root morphemes which never stand alone as words (e.g. brilli-ance and experi-ence). The suffix -al is also a Level I morpheme, because it can trigger segmental alternation of the stem (e.g. n[eɪ]tion vs. n[a]tional) and syllabification of a stem-final sonorant consonant into the following suffix-initial vowel (e.g. centre [sʵn.t] vs. central [sʵn.tɹəl]). All these phenomena support the view that the stems to which these suffixes attach lack independent word status in the lexicon. In this experiment, we chose pairs in which the stem with the Level I suffix (e.g. attend in attendance) and that with the Level II suffix (e.g. attend in attending) are homophonous.

The frequency data for these words are shown in Fig. 9. For all pairs except for assisting vs. assistance, the log frequency of the Level II suffixation cases was either greater than or the same as that of the Level I suffixation cases. These cases therefore satisfy our frequency criteria.

Figure 9

Log word frequency count of items used in Experiment 2.

2.5.2 Speakers

Three of the native female speakers of Scottish Standard English who participated in Experiment 1 served as paid subjects.

2.5.3 Elicitation and recording

Only slow speech was elicited in Experiment 2, given that the results from Experiment 1 suggested that the durational effects of the stem–Level II suffix boundaries were greatest and most reliable at the slow rate. Four repetitions of each pair were recorded, giving a total of 168 items (i.e. 3 speakers×4 repetitions×2 morphological conditions×7 pairs). The rest of the recording procedure was the same as that of Experiment 1. The carrier sentences in which each target word was embedded are shown in the Appendix.

2.5.4 Measurements and analyses

We measured the rhyme duration of the stem-final syllable of each target word, e.g. /ʵnd/ in attendance and attending. Measurement criteria were the same as those for the previous experiment: the target sequence was measured as the interval between the onset of constriction release of the first consonant and that of the coda consonant of the stem-final syllable (see again note 11). For the target sequence with a sonorant onset [ɹ], i.e. [ɹez] in phrasal and phrasing, it was impossible to determine a boundary between the sonorant consonant and the following vowel. We therefore included the sonorant onset consonant in the target measurement sequence. Part of the initial word of each carrier sentence (Say or Type) was measured as a check for overall speech-rate differences (see §2.4.5 for more detailed discussion). Due to segmentation difficulties, 17 item pairs were discarded, giving 67 pairs out of 84 for analysis. 14 pairs discarded out of 17 were Speaker K's, due to the unclear onset of the stem-final consonant release. Data were analysed using one-tailed paired t-tests; each repetition of a test sentence in one morphological condition was always paired with that of the other morphological condition within a single block.

2.5.6 Summary: experiment 2

Our results are consistent with the view that the stem–Level II suffix words differ from their stem–Level I suffix counterparts in their prosodic constituent representation. That is, Level II suffixes adjoin to a word-level constituent, and the presence of the word boundary can result in a longer stem duration at a slow rate of speech. In contrast, the shorter stem-final rhyme durations before Level I suffixes suggest that no such word boundary is present between the stem and the following Level I suffix.

3.1.1 Stress placement and syllable weight in English

One of the motivations for the word-internal foot constituent in English comes from the interaction between the location of lexical stress and syllable weight (Selkirk Reference Selkirk1980a). In American English, open syllables with a stressed lax (i.e. short monomoraic) vowel are rare or non-existent (a) in word-final position: *[… Cˈ#], and (b) in word-medial position immediately followed by another stressed syllable when at most one consonant intervenes between the lax vowel and the following stressed vowel: *[… Cˈ(C)ˈVC …].Footnote ¹⁴

This fact can be accounted for by foot theory in the following way. Since every stressed vowel, whether short (monomoraic) or long (bimoraic), must be the head of a left-headed foot, all word-final and word-medial stressed syllables must be foot-initial. Because they are not followed by any unstressed syllables, they must form a foot on their own. A problem arises when these syllables are monomoraic, because a foot consisting of only a monomoraic syllable is illegal if each foot should dominate at least two syllables or moras (Mester Reference Mester1994). American English therefore has neither a word-final nor a word-medial lax stressed vowel immediately followed by another stressed syllable (with at most one intervening consonant). In this way, the foot hypothesis gives a simple account of the interaction between the location of a stressed syllable and its weight.

Foot theory, however, is not the only possible account for these facts. First, the lack of word-final lax vowels is accounted for by the familiar requirement that a stressed syllable should be bimoraic. Secondly, the lack of a lax vowel immediately followed by only one consonant and a stressed vowel is also accounted for by the bimoraicity requirement of stress, and a requirement for an onset consonant in every stressed syllable.Footnote ¹⁵ The *[… CˈCˈVC …] configuration violates one of the requirements no matter what the syllable affiliation of the intervening consonant is. If it is affiliated to the initial syllable, then the following syllable violates the onset requirement. If it is affiliated to the second syllable, then the preceding syllable with a lax vowel violates the bimoraicity. As a result, the configuration fails to surface.

Thus the lack of syllables ending with a lax vowel in word-final and pre-tonic positions can be explained by syllable-related constraints or principles which do not refer to foot structure.

3.1.2 Unstressed syllable truncation

Some additional pieces of evidence for the foot hypothesis in English come from unstressed syllable omissions in infant speech: speakers often truncate unstressed syllables that are not parsed into a trochaic stress (SW) foot. Gerken (Reference Gerken1994a, Reference Gerkenb), Wijnen et al. (Reference Wijnen, Krikhaar and den Os1994) and Demuth (Reference Demuth, Morgan and Demuth1996), among others, claim that infants make use of a trochaic rhythmic unit, i.e. the stress foot, as a template for word production. For example, children often produce ˈraffe for giˈraffe, while keeping both syllables in a strong–weak sequence like ˈtiger, and produce ˈnana for baˈnana, with omission of the first syllable. Gerken (Reference Gerken1994b) further investigates infants' imitation tasks for four-syllable words, and found that children often truncated even non-initial unstressed syllables when these syllables did not fit in the foot template, i.e. unstressed syllables not immediately following a stressed syllable. Since the foot template allows at most two syllables (i.e. a stressed syllable and an immediately following unstressed one), such an unstressed syllable is unparsed and undergoes truncation.

The foot-based account, however, is not the only way to capture the truncation phenomena. First, the tendency for word-initial unstressed syllable truncation may be due to the fact that the majority of English words start with a stressed syllable (Cutler & Butterfield Reference Cutler and Butterfield1992). That is, infants' unstressed syllable truncation reflects the distribution of stress patterns in the words they hear, and they adjust their output to be similar to more familiar initially stressed words by omitting the word-initial unstressed syllable. In addition, word-initial unstressed syllables may be less perceptually salient than word-final unstressed syllables in disyllabic words in some contexts, due to the higher probability of phrase-final lengthening, and accentual lengthening of the preceding accented syllable carries over to word-final syllables (Wightman et al. Reference Wightman, Shattuck-Hufnagel, Ostendorf and Price1992, Turk & White Reference Turk and White1999). Word-medial unstressed vowel truncation may reflect a preference for stress patterns that have a strict alternation between strong and weak syllables (i.e. strong–weak–strong–weak) to more anomalous patterns such as two consecutive weak syllables followed by a strong syllable, etc. In addition, children's tendency to truncate a word-medial unstressed syllable not immediately preceded by a stressed syllable may also be due to such a syllable not being as perceptually salient as an unstressed syllable immediately preceded by a stressed syllable (in phrasally stressed words), because of the rightward spread of phrasal stress-related (accentual) lengthening (Turk & Sawusch Reference Turk and Sawusch1997, Turk & White Reference Turk and White1999). It is therefore possible to explain most unstressed syllable omissions in English without recourse to the word-internal foot.

3.1.3 Expletive infixation

Another argument provided by previous studies for the sublexical foot constituent hypothesis in English comes from ‘expletive infixation’, discussed in detail by McCarthy (Reference McCarthy1982) and Hammond (Reference Hammond1999). Although this phenomenon has also been treated as evidence for the English sublexical foot, other possible accounts are also available: these invoke native speakers' preference for alternating strong–weak rhythmic patterns.

English infixation is a productive morphological process in which expletives such as fuckin', goddam (American English) and bloody (British English) are inserted in a word-medial position to give an emphatic connotation to words, e.g. fan-fuckin-tastic (derived from fantastic). McCarthy and Hammond claim that infixes are never able to break up a foot. Therefore, one never finds infixed words like *a(ˈmal-ˌfuckin-ga) (ˌmated), where the infixation breaks up two syllables that are parsed into the same trochaic foot, whereas a stressed syllable and an immediately preceding unstressed syllable are easily separated, as in a(ˈmalga-ˌfuckin-) (ˌmated), (ˌKala)ma-ˌfuckin-(ˈzoo), (ˌKala)-ˌfuckin-ma(ˈzoo) and ad-ˌfuckin- (ˈvance), because the infixation does not break up two syllables within the same foot.

Although this foot-based formalisation of English infixation is simple and straightforward, it is also possible to formalise the phenomenon by stating that native speakers of English prefer the rhythmic pattern to alternate strictly between strong and weak syllables. The strong–weak alternation pattern by itself is a simple and primitive notion, which does not necessarily require a foot-level constituent in its definition. The English infixes all consist of an initial strong syllable followed by a weak syllable as in fuc (S) kin (W), and infixation immediately after a stressed (strong) syllable is therefore undesirable because such infixation would yield two consecutive strong syllables, thus violating the favoured alternation pattern. Therefore *aˈmal-ˌfuckin-gaˌmated (WS-SW-WSW) is bad because the infixation creates a stress clash between ˈmal (S) and ˌfuc (S). In contrast, aˈmalga-ˌfuckin-ˌmated (WSW-SW-SW) is preferred, because it avoids both stress clash and two consecutive unstressed syllables. However, stress-clash cases like ˌfan-ˌfuckin-ˈtastic (S-SW-SW) are allowed because no better option is available: if the infix is inserted between the penultimate stressed syllable and the final unstressed syllable as in *ˌfanˈtas-ˌfuckin-tic (SS-SW-W), the outcome representation contains two stress clashes (i.e. ˌfanˈtas and ˈtas-ˌfuc…), while the well-formed representation ˌfan-ˌfuckin-ˈtastic (S-SW-SW) has only one. Thus the foot-based account is not the only possible explanation for infixation in English.

3.1.4 Durational evidence for word-internal feet?

Phonetic evidence for word-internal feet in English is ambiguous at best. Although some results reported in the literature, e.g. Fowler (Reference Fowler1981) and Rakerd et al. (Reference Rakerd, Sennett and Fowler1987), are consistent with the view that word-internal feet affect segmental duration, White (Reference White2002) suggests that these results are more likely to be word-sized constituent-related phenomena, rather than phenomena relating to word-internal feet. Van Lancker et al. (Reference Van Lancker, Kreiman and Bolinger1988) also presents effects less easily interpreted without recourse to the foot, but the observed effects have a variety of experimental flaws which need to be corrected in subsequent studies.

Fowler (Reference Fowler1981) found asymmetric polysyllabic shortening between trochaic (strong–weak) words and iambic (weak–strong) words in a metronome-paced speech-production experiment of nonsense words in English. Stressed syllables immediately followed by unstressed syllables in trochaic words such /ˈsisa/ were shorter on the average by 101 ms than stressed syllables produced in isolation in the word /ˈsi/. However, stressed syllables immediately preceded by unstressed syllables in iambic words such /sʌˈsi/ were only 9 ms shorter than those produced in monosyllabic words. Rakerd et al. (Reference Rakerd, Sennett and Fowler1987) interpret this asymmetric polysyllabic shortening as evidence for the trochaic foot hypothesis. They consider the domain of polysyllabic shortening to be the trochaic foot constituent, while the stressed syllable of iambic words forms a monosyllabic foot of its own, and therefore does not undergo within-foot polysyllabic shortening. However, White's (Reference White2002) research on the durational patterns of phrases containing identical foot structure, e.g. [(ˈshake)_Ft (ˌdown)_Ft]_Wd [(ˈstairs)_Ft]_Wd ‘shakedown stairs’ vs. [(ˈshake)_Ft]_Wd [(ˌdown)_Ft (ˈstairs)_Ft]_Wd ‘shake downstairs’ – each foot in those words is non-branching, consisting of a stressed syllable only – suggests that this asymmetric behaviour is not due to foot structure. The duration of shake was shorter in shakedown stairs than in shake downstairs, indicating that the asymmetry had to do with word structure rather than foot structure, while word-final syllables were consistently longer than non-final syllables. White's results suggest that asymmetric shortening of syllables has nothing to do with the number of syllables or the position of syllables in a foot, but instead relates to processes involving word-sized constituents, such as word-final lengthening or polysyllabic shortening (see Turk & Shattuck-Hufnagel Reference Turk and Shattuck-Hufnagel2000 for a discussion of these mechanisms).

Van Lancker et al. (Reference Van Lancker, Kreiman and Bolinger1988) also provide durational observations that support the foot hypothesis. However, we show below that their study has several flaws, and as a result their observations do not offer unambiguous support for the word-internal foot hypothesis. They found that syllable duration in English varied with the quality of the immediately following syllable: it was about 20 ms longer when immediately followed by a stressed syllable, e.g. [an] in ˈAnˌdes, than when followed by an unstressed syllable, e.g. [an] in ˈAndy's. These results can be accounted for by foot-level polysyllabic shortening, if one assumes that a foot boundary separates the two heavy syllables in the strong–strong word, (An)(des), but that the two syllables in the strong–weak word belong to the same foot, (Andy's). However, several factors confounded their results. For example, their comparison pairs differed in syllable structure (e.g. to.ma.to vs. tor.ture), syllable count within each test word (e.g. ˈcompˌlex vs. ˈcomplicated, ˈvorˌtex vs. ˈvortices), the level of stress prominence (e.g. ˌbanˈdana (secondary stress) vs. ˈbandit (primary stress)), position within carrier sentences (e.g. Nimrod in utterance-initial position vs. nimble in utterance-medial position) and the mixing of proper names and common nouns (e.g. Sunday vs. sundown), etc. Further controlled experiments are therefore required to provide durational evidence for the word-internal foot hypothesis.

In sum, no unambiguous phonetic or phonological evidence has yet been provided in support of the word-internal foot. However, the potential for durational support exists, if Van Lancker et al.'s observation of foot-level polysyllabic shortening holds even when confounding factors are removed. In the experiment presented below, we control as many of these factors as possible, in order to see if foot-level polysyllabic shortening persists, as predicted by the word-internal foot hypothesis.

3.2.1 Materials

Materials consisted of seven pairs of (a) disyllabic words consisting of a primary stressed syllable followed by a secondary stressed syllable, e.g. ˈArˌgos, where the first syllable constitutes a monosyllabic foot, i.e. (ˈAr)_Ft (ˌgos)_Ft, and (b) disyllabic words consisting of a primary stressed syllable followed by an unstressed syllable, e.g. ˈArgus, where the first syllable belongs to a disyllabic foot, i.e. (ˈAr.gus)_Ft (see Table IV).

Table IV

Item pairs used in Experiment 3.

The segmental organisation of the initial target syllable, i.e. the primary stressed syllable, was kept constant across stress conditions. The vowel quality of the second syllable was different in the two stress conditions, because stressed syllables in English are by definition realised with full vowel nuclei, while unstressed vowels are not (see Footnote note 13). In our materials, as shown in the list of item pairs in Table IV, unstressed second syllables always had either schwa, as in climate, or a syllabic consonant, as in amber or common, as their nucleus. The log word frequencies of the stimulus pairs are given in Fig. 12.

Figure 12

Log word frequency count of items used in Experiment 3.

These items were embedded in carrier sentences and preceded by background sentences designed to elicit primary phrasal stress on the target word, as in (12) (see the Appendix for the full list).Footnote ¹⁶

1. (12) To overcome and take control. Say ‘conquer’ for me.

3.2.2 Speakers and recordings

The same five speakers who participated in Experiment 1 read the stimulus set in this experiment. Recording procedures were the same as those in Experiment 1.

3.2.3 Measurements

The target sequence in this experiment consisted of the rhyme of the first syllable in disyllabic words containing two stresses, as well as the corresponding stretch in the words containing single stress. The onset of the target stretch was the onset-consonant release, when an onset was present, thereby including any onset VOT in the duration of the target sequence. For the pair climax vs. climate, where there is an obstruent–sonorant consonant cluster [kl] in onset position, we took the obstruent-consonant release to be the onset of the rhyme: the rhyme therefore included VOT as well as the sonorant onset consonant [l]. The offset of the rhyme differed for different stimulus pairs. When only a single consonant was present between the two syllables (e.g. acorn vs. acre), the offset of the initial syllable nucleus F2 energy was taken as the offset of the rhyme. When there was a consonant cluster between the first and second syllable nuclei ([ŋk] in Concord vs. conquer, [mb] in ambush vs. amber, [nt] in centaur vs. centre, [ɹg] in Argos vs. Argus), the rhyme included acoustic evidence of the first member of the consonant cluster. For homorganic clusters, the offset of the rhyme was taken to be the offset of nasal formant energy. Cases of unclear boundaries between the two consonants were excluded from our analyses (normal rate: four pairs of Concord vs. conquer, ten pairs of ambush vs. amber, eight pairs of centaur vs. centre; slow rate: five pairs of ambush vs. amber, four pairs of centaur vs. centre).

3.2.5 Partial effects

A possible criticism of this set of comparisons would be that some pairs differed in their surface segment count. For example, the second syllable of the monosyllabic foot case (ˈcli)(ˌmax) consists of four segments [maks], while that of the disyllabic foot case (ˈclimate) consists of three segments [mət]. Six of our seven pairs actually came with such second syllable segment-count differences: the monosyllabic feet cases contained more second syllable segments than the disyllabic foot cases. Crude measurements of the second syllables of those six word pairs (including final coda consonant release intervals) at the slow rate of speech also show that the second syllable duration of the monosyllabic foot cases was 80 ms to 150 ms longer than that of the disyllabic foot cases, as shown in Fig. 14.

Figure 14

Mean difference (ms) in second syllable duration for monosyllabic and disyllabic foot cases at the slow rate, obtained by subtracting the latter from the former.

Given this, one might speculate that the shorter rhyme duration of the initial syllable in the monosyllabic feet cases observed above would be due to word-level polysegmental shortening, i.e. compensatory shortening of the initial syllable under pressure from the heavier second syllable, and that this effect could have obscured the foot-boundary effect.

It is therefore necessary to analyse separately the only pair without such second syllable segment-count differences, i.e. (ˈAr)(ˌgos) vs. (ˈArgus).Footnote ¹⁷ The second syllable rhyme-duration differences between the two foot conditions in this pair turned out to be relatively small in comparison to those in the six pairs above: only 13 ms to 22 ms (at the slow rate of speech).

However, the results of the separate analysis of the rhyme durations of the first syllables of this pair did not support the foot hypothesis either. The mean duration of the initial rhyme of (ˈAr)(ˌgos) was 164 ms (SD=19) at the normal rate of speech and 249 ms (SD=41) at the slow rate of speech, while that of (ˈArgus) was 165 ms (SD=17·8) at the normal rate of speech and 255 ms (SD=39) at the slow rate of speech. That is, the mean of the initial rhyme duration of (ˈAr)(ˌgos) was again shorter than that of (ˈArgus), contrary to our prediction. The difference was, however, statistically insignificant under a one-tailed paired t-test (normal rate: t=−1·354, df=14, p=0·1; slow rate: t=−0·479, df=14, p=0·32).