The timing of nuclear falls: Evidence from Dutch, West Frisian, Dutch Low Saxon, German Low Saxon, and High German

1.1 Background

One issue that has arisen from two decades of research into the time alignment of tonal targets is the behaviour of post-nuclear tones. A number of cases have been presented of such tones being aligned with stressed syllables, while in other cases the targets of similar tones have been shown to have a timing relation either with the targets of other tones or with the upcoming phrasal boundary. The ‘stress-seeking’ kind was labelled a ‘phrase accent’ by Grice et al. (2000), meaning a non-starred tone that associates with an unaccented stressed syllable. In at least three cases they present, the effect of the location of the post-nuclear stress on the alignment of the tone has been found to be considerable. First, their Eastern European Question Tune, a L* H-L% melody which occurs in a number of languages in southeastern Europe, shows H aligning with the stressed syllable of the last unaccented word, in Athenian Greek creating peaks on the antepenult, penult, or final syllable of the phrase, depending on the location of the word stress of the final post-nuclear word (Arvaniti 2002). Similarly, the second target of the H_ι in the Roermond Dutch interrogative nuclear melody L* H_ιL_ι associates with the last post-nuclear stressed syllable, which again is final, penultimate, or antepenultimate (Gussenhoven 2000; H_ι and L_ι are right-hand boundary tones of the intonational phrase). ^[1] Third, the English vocative chant has a mid-pitched trailing H-tone which associates with a post-nuclear stressed syllable (Liberman 1975; Ladd 1978; Hayes and Lahiri 1992); a Dutch counterpart spreads the H-tone to all post-nuclear stressed syllables (Gussenhoven 1993).

The status of ‘phrase accent’ has also been attributed to tones whose alignment varies less obviously with context. Barnes et al. (2006) found that the alignment of the first target of L- in a H* L-L% contour varied with the distance to the first post-nuclear stress in American English, suggesting that L- is stress-seeking. An extended study showed, however, that this target was in fact aligned at a more or less fixed distance from the target of H* (Barnes et al. 2010). This latter finding is in line with Caspers and van Heuven (1993) for Dutch, who found that increasing speech rate compressed the rising, but not the falling, movement of nuclear rising-falling accents. Because the determination of the alignment of the turning point in a concave pitch movement, a pitch ‘elbow’, may be ambiguous, Benzmüller and Grice (1998) examined the timing of two alignments in German. The first was defined as “the point where the slope of the fall turns from convex to concave, usually well above the baseline”, while the second was “the elbow, or point where the baseline level is reached [….]” (81). They observed that L1 occurred at a fixed distance after the preceding F₀ peak, while L2 was preferably aligned with the first postnuclear stressed syllable. The ambiguity in the timing of the low turning point is also illustrated by a preliminary conclusion drawn by Pierrehumbert (1980: 85), who suggested that the right edge of the accented word was a likely alignment point for her phrase accent L-. Van de Ven and Gussenhoven (2011) investigated the alignment of the end of the low plateau in nuclear rising-falling-rising contours in Dutch, varying both the interval from the last stress to (a) the phrase end and (b) the location of second-occurrence focus. They found that the alignment was largely determined by the distance to the final boundary, showing that the L-tone responsible for the beginning of the final rise is not stress-seeking. The evidence for the status of a post-H* ‘phrase accent’ in these less obvious cases therefore is ambiguous at best.

An understanding of the timing of the post-H* low target will require a contextualization within the entire rising-falling pitch accent. The beginning of rising pitch movements of rising and rising-falling pitch accents was found to be aligned with the beginning of the accented syllable or rime by Caspers and van Heuven (1993) and Ladd et al. (2000) for Dutch, Niebuhr and Ambrazaitis (2006) for German, and Ladd et al. (2009) for English. As for the F₀ peak, Silverman and Pierrehumbert (1990) report earlier prenuclear peaks in American English with smaller distances from the end of the accented word as well as from the first stressed syllable after the accented word. This finding was replicated in Greek for some speakers by Arvaniti et al. (1998). Earlier, Steele (1986) had found variation of the timing of the nuclear peak as a function of the distance to the upcoming IP boundary. For Dutch, Schepman et al. (2006) found no effect of word length on the timing of H* in nuclear rising-falling accents, while there was only a small effect of the distance to the IP-end.

For other languages, Dalton and Ní Chasaide (2007) report that the timing of the prenuclear and nuclear peak in Cois Fharraige Irish was unaffected by the number of adjacent unstressed syllables, while Prieto et al. (1995) report effects of the end of the accented word, the distance to the upcoming prosodic boundary, and the strength of that boundary on the timing of nuclear F₀ peaks in Mexican Spanish, with closer distances leading to earlier peaks. As for the end of the fall, Frota (2002) shows that the trailing L of the European Portuguese focal H*+L is timed with reference to H*, in contrast to the leading H of the H+L* pitch accent, which is rather timed independently of L*. For Pisa Italian, Gili Fivela (2002) found that the post-peak low target was later in words containing two or three postnuclear syllables than in words containing one postnuclear syllable, at least in sentences with broad focus, which in her data can be interpreted either as an effect of the distance to the final word boundary or the final IP boundary. Though methodologies vary in these investigations, the variation in the findings may in part reflect cross-linguistic differences.

1.2 Cross-linguistic variation

In the British School (e.g., O’Connor and Arnold 1973), the rising-falling contour is classified as the High Fall nuclear tune. The determination of any systematicity in the dependence of the alignment of the falling slope of the High Fall on landmarks to its left and right will require a large database, because effects may be small. Also, there is likely to be regional variation, as suggested by various findings that cross-linguistic variation in West Germanic contributes to variation in the timing of nuclear and prenuclear pitch accents, even in closely related languages and dialects. Atterer and Ladd (2004) found small differences of the timing of both the low and the high targets of prenuclear rising accents in Northern and Southern German, with later peaks attested for the Southern speakers (see also Gilles 2005). In addition, both Northern and Southern speakers of German were found to align the prenuclear rises later than suggested by data for English and Dutch. Mücke et al. (2008, 2009) found a similar difference between speakers of (Northern) Düsseldorf German and (South Eastern) Vienna German, the latter aligning the beginning and end of prenuclear rises and the peaks of rising-falling accents later. Kleber and Rathcke (2008) supplemented these studies with data from East Central German, which showed earlier alignments of the beginnings of prenuclear rises than both Northern and Southern German. Cross-linguistic variation has also been found for nuclear F₀ peaks. Ladd et al. (2009) found both prenuclear and nuclear peaks to be aligned later in Scottish Standard English than in RP English and nuclear peaks to be aligned earlier in RP English than in Dutch. Van Leyden and van Heuven (2006) observed a similar variation of peak timing in rising accents of the Lowland Scots dialects of Orkney and Shetland, with later peaks being found in the Orkney dialect. ^[2]Grabe (1998) reports nuclear peaks aligning at 66% of the accented syllable in Southern British English, but at 98% by speakers of Braunschweig German. Peters (1999) found similar differences between northern German varieties, with speakers of Hamburg German aligning the nuclear peak at 57% of the accented syllable and Berlin speakers at 79%. This difference was increased when alignment is expressed as a proportion of the duration of the vowel (Hamburg 35%, Berlin 71%). Interestingly, in both varieties the timing of F₀ was found to be affected by the distance to the upcoming IP boundary, but only in Berlin German was there an effect of the end of the accented word, which may indicate that the timing of earlier peaks may be less affected by the end of the accented word. As for the alignment of the target of the L-tone after H*, no cross-linguistic variation has to our knowledge been reported for West Germanic.

In the investigation reported here, we were particularly interested in the extent to which the dependence of the timing of the post-peak low target on nearby prosodic boundaries varies across varieties and languages that are geographically close. We report a reading experiment that was designed to examine the variation in the timing of the rising-falling accent in seven varieties spoken along the coast of the Netherlands and North-Western Germany. While representing a section of a traditional dialect continuum, this geographic cline includes three standard language areas, Dutch, Frisian, and German. A further opportunity to show that effects of geographic proximity may be independent of the sociolinguistic allegiance of the dialects was provided by the bidialectal speakers from North-West Germany, who read comparable materials twice, once in their dialect and once in their regiolectal standard German.

There is no unique autosegmental representation of the High Fall. The ToBI framework originally developed for English (Beckman and Ayers Elam 1997; Beckman et al. 2005) represents it as (L+)H* L-L%, where the end of the fall is accounted for by the ‘phrase accent’. In the analysis of Gussenhoven (2005), the High Fall is represented as H*L L%, in which the end of the falling movement is represented as a trailing tone. The same analysis was applied to English (Gussenhoven 1991, 2004) and Standard German (Peters 2009, 2014). In our investigation, we will put the two hypotheses of the alignment of L-, i.e., an alignment with the upcoming post-nuclear stressed syllable and an alignment with the end of the nuclear word, to the test. In Section 4, we will suggest that the assumptions underlying the existence of a trailing tone in those analyses do not in fact imply the phonetic alignment characteristics that have been assumed in the ToBI framework.

1.3 Timing strategies

Assuming an accented word with initial stress, there are theoretically three strategies speakers may adopt in the alignment of an accentual peak when accommodating to variation in word length. First, increasing the number of word-internal postnuclear syllables may affect the distance of a given target in the LHL-contour to the end of the nuclear word, but not its distance to the beginning of the word. In this strategy, the tonal target stays at a fixed distance from the beginning of the nuclear word irrespective of its duration (Strategy 1). Second, increasing the number of word-internal postnuclear syllables may affect the distance of the tonal target to the beginning of the nuclear word, but not its distance to the word end. This strategy represents Pierrehumbert’s (1980) word-based hypothesis for the alignment of L-. In this case, the tonal target occurs at a fixed distance from the end of the nuclear word irrespective of its duration (Strategy 2). Third, increasing the size of the nuclear word may affect the target’s distance from both the beginning of the nuclear word and from its end. In this case, increasing word size results in a proportional adjustment of the alignment (Strategy 3). The same three options will apply to the alignment of the rise-fall with respect to the interval between the beginning of the accented syllable and the first post-nuclear word-stressed syllable. We will refer to this interval as the ‘nuclear foot’, in the spirit of Abercrombie (1964). Note that the nuclear foot may be longer than the nuclear word. This is the case if the first post-nuclear word begins with an unstressed syllable, which is included in the nuclear foot. As will be clear, Strategy 2 now represents the widely entertained hypothesis that the upcoming stress provides the alignment point for L-.

Figure 1 illustrates the three strategies for the alignment of H* relative to either the nuclear word or nuclear foot. C1 marks the beginning of the nuclear word or nuclear foot, where the initial syllable is accented. Wb and St mark the final boundary of the nuclear word and foot, respectively, which may or may not coincide. t1 and t2 mark the intervals from H* to the beginning of the nuclear word or foot and from H* to the end of the nuclear word or nuclear foot, respectively. The interval t1+t2 corresponds to the overall duration of the nuclear word or foot. The three panels of Figure 1 illustrate the three strategies for accommodating an increased length of the nuclear word or foot on t1 and t2. A correlation analysis will be used to detect these possible effects of increasing length. Specifically, a positive correlation of t1+t2 with t2 and no correlation with t1 will indicate that Strategy 1 is adopted (left panel), while a positive correlation of t1+t2 with t1 and no correlation with t2 suggests that Strategy 2 is adopted (central panel). Finally, positive correlations between t1+t2 and both t1 and t2 suggest that Strategy 3 is adopted (right panel).

Figure 2 illustrates the three strategies for the timing of the tonal low target at the end of the nuclear falling movement, here referred to as L1. We defined t1 as the distance of L1 from the target of H*, because the location of H* more so than that of C1 determines the time available for it. t2 is the distance of L1 from the end of the nuclear word or nuclear foot (Wb/St).

The aim of the present paper is to answer the following questions:

1. The timing strategy for H. Does H occur at a fixed distance from the beginning of the nuclear word/foot? (Strategy 1) Or is it affected by the location of the final boundary of the nuclear word or foot (Wb or St)? And if so, does H stay at a fixed distance from Wb or St (Strategy 2), or does it move proportionally to the duration of the nuclear word or foot (Strategy 3)? ^[3]

2. The timing strategy for L1. Does L1 occur at a fixed distance from the preceding H? (Strategy 1) Or is it affected by the location of Wb or St? And if so, does H stay at a fixed distance from Wb or St (Strategy 2), or does it move proportionally to the duration of the nuclear word or foot (Strategy 3)?

3. Are there overall differences in the timing of H and L1 in the varieties examined? And if so, do these differences have any relevance for the strategies adopted for the timing of H and L1?

2.1 Speech materials

We used ordinary statements as carrier sentences. In order to elicit High Falls, these carrier sentences were presented as answers to a preceding question, as in (1).

Each carrier sentence contained a fictitious proper name that was intended to bear the nuclear accent, Mol in (1b), followed by an infinitive verb form in postnuclear position, beweren in (1b). To prevent stress clash effects on the timing of the nuclear pitch accent gesture, we made sure that in each carrier sentence the proper name was preceded by an unstressed syllable. As proper names we used three oxytones, three paroxytones, and three proparoxytones, all of which start with the lexical stressed syllable. As verb forms we used three pairs of paroxytones which differed by the presence or absence of an initial unstressed syllable corresponding to the prefix be- in all our languages. Table 1 lists all proper names and verb forms used in the Standard Dutch version of the test sentences classified according to their position in the carrier sentence (nuclear vs. postnuclear) and their stress patterns.

Proper names and verb forms were combined such that six types of carrier sentences were created, differing by the distance between the lexically stressed syllable of the nuclear word and the final word boundary of the nuclear word (=WbDist) and by the distance between the lexically stressed syllable of the nuclear word and the first postnuclear lexical stress (=StDist), both counted in terms of the number of intervening syllables. In Mol beweren in (1), for example, WbDist = 0, as no syllable intervenes between the accented syllable and the final word boundary of Mol, while StDist = 1, as one syllable intervenes between the nuclear and the first postnuclear stress. Table 2 lists one sample dialogue for each experimental condition of the Standard Dutch version. WbDist was varied from 0 to 2, whereas StDist was varied from 0 to 3.

Note that WbDist and StDist are not independent. StDist corresponds to WbDist or WbDist +1, but it is never smaller than WbDist. Different levels of WbDist can be compared while keeping StDist constant if we check condition 2 against 3 for StDist = 1 and condition 4 against 5 for StDist = 2. Likewise, different levels of StDist can be compared if we check condition 1 against 2 for WbDist = 0, condition 3 against 4 for WbDist = 1, and condition 5 against 6 for WbDist = 2

We used three carrier sentences per condition, which amounts to a total of 18 sentences, each of which was prompted by a question. The Standard Dutch version of the sentences was used for speakers from Zuid-Beveland, Rotterdam, and Amsterdam. For the other speakers, we used translations into the local language, which was West Frisian, Dutch Low Saxon, German Low Saxon, or High German (for more information on our speakers see Section 2.3). In the translations, the proper names were kept constant, whereas the postnuclear words were adapted according to the respective language. In some cases, particularly in the Low Saxon and High German language versions, modified versions rather than translations of the Standard Dutch version had to be used in order to keep the metrical structure constant across languages and dialects. An overview of the sentence in all language versions is given in the Appendix.

2.2 Recording procedure

The mini-dialogues were presented in a booklet, one dialogue per page. To prevent order effects, the dialogues were presented in pseudo-randomized order, which was reversed for half of the subjects per variety, and 87 mini-dialogues from other experiments were added as fillers. To reduce effects of the experimenter’s presence on the speakers’ dialect level, our speakers were recorded in pairs, with one speaker producing the context sentence and the other the carrier sentence. The participants switched roles at the end of the task after they had repeated any mispronounced sentences. The German Low Saxon and High German test sentences were recorded by the same speakers on two visits which were at least four weeks apart. Recordings were made in a quiet room either in the homes of our speakers or in a public building. We used a portable digital recorder (Zoom H4) with a 48 kHz sampling rate, 16 bit resolution, and stereo format. The participants wore head-mounted Shure WH30XLR wired condenser microphones. ^[4]

2.3 Speakers

Recordings were made in six places along the coastal line stretching from Zeeland in the South-West to Weener in the North-East (see Figure 3). These recordings covered six local varieties, Zeelandic Dutch in Zuid-Beveland (ZB), Rotterdam Dutch (RO), Amsterdam Dutch (AM), West Frisian in Grou (GR), Dutch Low Saxon in Winschoten (WI), and German Low Saxon in Weener (WL). The Weener speakers were recorded on a second visit to obtain comparable data of the local standard variety, which is Weener High German (WH).

We recorded a total of 125 speakers, aged between 16 and 45. The speakers from Zuid-Beveland, Grou, and Winschoten were bilingual with Standard Dutch and their local language. All regional speakers and at least one of their parents spoke the indigenous variety fluently, and they were raised in the location concerned. The speakers from Weener were bilingual with German Low Saxon and High German. Except for the speakers of West Frisian, our speakers were less familiar with their local language as a written language, which may have had a negative influence on the fluency of the speech in the reading task of some speakers. This is particularly true of our Weener speakers when speaking German Low Saxon.

Eleven speakers were excluded because their speech was disfluent or because their linguistic background later turned out not to be representative of the local variety. Table 3 gives an overview of the speakers recorded per variety. The participants were naïve as to the purpose of the experiment and were paid for their participation.

2.4 Acoustic analysis

All recorded carrier sentences were converted to monaural files and stored on computer disk as separate wave files. Utterances were excluded from further analysis if they showed deviant pitch patterns due to accent position, choice of pitch accent, or choice of final boundary tone. In particular, we excluded utterances with a downstepped nuclear accent. As it happens, a relatively small number of the utterances in our experiment had downstep. We also excluded utterances with hesitation pauses on and around the target word. After exclusion of all irregular items, about 81% of the recorded utterances were left for acoustic analysis, which were evenly distributed across speaker groups.

Acoustic and auditory analyses of the data were done with the help of the speech processing software package Praat (Boersma and Weenink 2008). We inserted the labels listed in Table 4.

In general, syllable boundaries were determined on the basis of visual inspection of the waveform and the broadband spectrogram, aided by auditory information. We placed all labels at negative-to-positive zero-crossings of the sound wave. L0 and H were determined semi-automatically using a Praat function to locate the F₀ minimum or maximum in a selected region. Semi-automatic determination of the elbow after the nuclear peak (L1) was found to yield less inter-rater agreement for our data set. Therefore, we determined L1 visually by looking for the location of the highest rate of F₀ change near the bottom line of the nuclear contour. For a comparison of manual with automatic detection methods, see del Giudice et al. (2007). Pitch tracking errors such as octave jumps were corrected by hand. Using the labels in Table 4, we calculated the acoustic variables given in Table 5.

From the above measures, all variables can be derived that are necessary to identify the timing intervals t1 and t2 that we used to define the three timing strategies in Section 1. Table 6 gives an overview of variants of t1 and t2 for the timing of H and L with respect to the nuclear word and foot.

To check the reliability of measurements, one sentence per experimental condition was quasi-randomly selected from one male and one female speaker per variety and labelled independently by the first author and a trained phonetician (6 × 2 × 7 = 84 sentences). Table 7 gives the averaged absolute differences between the two measurements for each of the acoustic variables listed in Table 5. For most variables inter-rater agreement was acceptable. The largest mean absolute difference was found for L1 Delay, which can mainly be attributed to a lower agreement about the location of L1.

3.1 Duration of nuclear word and nuclear foot

To ensure that the experimental manipulation of the number of syllables of the nuclear word and the nuclear foot was not compensated for by a change in speech rate, we first checked whether adding syllables indeed increased the duration of the nuclear word (WordDur) and foot (FootDur).

Figure 4 reveals that increasing the number of syllables between the accented syllable and the right word boundary (=WbDist) from 0 to 2 had the desired effect. Every additional syllable increased WordDur. For statistical analysis, we built LME Models including Speaker and Sentence as random factors and WbDist (0, 1, 2) and Dialect (ZB, RO, AM, GR, WI, WL, WH) as fixed factors. Significant main effects on WordDur were found for both WbDist, F(2,1863) = 3353.58, p < 0.001, and Dialect, F(6,139) = 7.59, p < 0.001. In addition, there was a significant interaction between WbDist and Dialect, F(12,1862) = 10.85, p < 0.001.

Pairwise comparisons of levels of WbDist revealed significant differences between all levels of WbDist at the 0.1% level. Levels of Dialect revealed that WL differed significantly from ZB (p < 0.01), RO (p < 0.05), and WH (p < 0.001).

Figure 5 shows analogous measurements of the effect of varying the distance between the nuclear syllable and the first postnuclear stress (StDist) from 0 to 3 syllables on the duration of the nuclear foot (FootDur). Again, increasing the number of intervening syllables had the desired effect. For statistical analysis we included Speaker and Sentence as random factors and StDist (0, 1, 2, 3) and Dialect as fixed factors. Significant main effects on FootDur were found for StDist, F(3,1859) = 2813.01, p < 0.001, and Dialect, F(6,141) = 6.37, p < 0.001. There was also a significant interaction between StDist and Dialect, F(18,1858) = 4.28, p < 0.001.

Pairwise comparisons of levels of StDist revealed significant differences between all levels of WbDist at the 0.1% level. Comparisons of levels of Dialect revealed that WL differed significantly from WH (p < 0.001). We conclude that all experimental manipulations had the desired effects on the location of the final word and foot boundary in all varieties.

In Sections 3.2 and 3.3 we examine the effects of WbDist and StDist on the timing of H within the accented word and the nuclear foot and of L between the preceding tonal target, H, and the end of the nuclear word and foot. As we were particularly interested in differences in timing strategies as a function of language, we fitted separate regression models for each variety.

3.2 Alignment of H

3.3 Alignment of L

3.4 Cross-linguistic variation

In Sections 3.2 and 3.3 we carried out separate analyses for each level of Dialect, as we were interested in the timing strategies of individual varieties. In this section, we add some information on this cross-linguistic variation based on direct comparisons of the varieties included in our corpus. The comparisons of the nuclear words and feet produced in different experimental conditions in Section 3.1 have shown that word and foot duration differed only marginally across varieties. The picture changes when we look at the timing of the accentual gesture. Figure 12 shows the alignment of L0 and H relative to the beginning of the nuclear word and foot (L0 Delay and H Delay) and of L1 relative to H (L1 Delay). For all variables we observe inverted U-shaped patterns, with a later occurrence of the three tonal targets in the more ‘central’ varieties, AM, GR, RO, and, to a lesser extent, WI than in the more ‘peripheral’ varieties, ZB, WL, and WH. These data suggest that the whole rising-falling F₀-gesture occurs later when we approach the geographical midpoint of our research area. ^[5]

To asses the effect of the linguistic background of our speakers on the alignment of L0, H, and L1, we fitted LME models for the three dependent variables L0 Delay, H Delay, and L1 Delay, including Speaker and Sentence as random factors and Dialect as a fixed factor in each case. Dialect had a significant effect on L0 Delay, F(6,163) = 14.67, p < 0.001; H Delay, F(6,135) = 21.34, p < 0.001; and L1 Delay, F(6,138) = 15.70, p < 0.001. Table 16 shows pairwise comparisons of levels of Dialect. The distribution of significant differences between levels is in line with the overall inverted U-shaped patterns observed in Figure 12. Most significant differences were found between the more ‘peripheral’ varieties, ZB, WL, and WH, on the one hand and the more ‘central’ varieties on the other. For L0 Delay only those differences between levels that involved either WL or WH ^[6] reached significance.

4.1 Alignment of H

The nuclear F₀ peak, H, kept a fairly constant latency from the beginning of the nuclear word and foot when the distance between the nuclear syllable and the final word boundary (WbDist) or the following postnuclear stress (StDist) was varied. Similar results were obtained when continuous variables such as the overall word duration (WordDur) or duration of the nuclear foot (FootDur) were used as predictors of the location of H.

Statistical analyses revealed that speakers of all varieties either used Strategy 1 or Strategy 3. Strategy 1 was defined as exclusively increasing the H-distance from the final word or foot boundary when word or foot size is increased, respectively. Strategy 3 was defined as increasing both the H-distance from the final word or foot boundary and the initial word or foot boundary when word or foot size is increased, respectively. Under Strategy 3, delaying the final word or foot boundary is accompanied by a proportional delay of H in the same direction. Estimates of the variance of residuals explained, Ω², however, showed that even in varieties using Strategy 3, this effect is fairly small. A comparison of the Ω² estimates for t1_H/w and t2_H/w in Tables 8 and 9 and of t1_H/f and t2_H/f in Tables 10 and 11 shows that the amount of variance of residuals explained in predicting t2_H/w or t2_H/f was at least 5 times larger than when predicting t1_H/w or t1_H/f, respectively. We conclude that in the varieties examined here, the size of the nuclear word or foot has little or no effect on the timing of the nuclear F₀ peak relative to the beginning of the nuclear word and foot.

This finding is largely in line with the findings of Schepman et al. (2006) on Standard Dutch, who found no general effect of the size of the nuclear word for WbDist = 0 and WbDist = 1, and only a small effect for StDist = 1 when compared with StDist = 0. Ladd et al. (2009) found an effect of stress distance on nuclear accentual peaks in speakers of English (RP and Scottish Standard English). In this case, however, the target words occurred in final or prefinal position of the utterance, such that tonal crowding effects arising from the presence of IP-final boundary tones may have come into play.

4.2 Alignment of L1

L1 was found to be aligned at a fairly fixed distance from the preceding H when WbDist or StDist was increased stepwise. Again, statistical analyses suggested that either Strategy 1 or Strategy 3 was adopted, and again the outcome in adopting these strategies was very similar. Even when Strategy 3 was adopted, increasing the size of the word or foot had only a small effect on the distance of L1 from the preceding high target. One exception was AM, which was found to adopt Strategy 2 when WbDist was increased from 1 to 2 syllables (for StDist = 2; cf. Table 12) and when StDist was increased from 1 to 2 syllables (for WbDist = 0; cf. Table 14). In these cases, L1 was kept at a rather fixed distance from the end of the nuclear word or foot rather than from the beginning of the nuclear word or foot, as illustrated by the upper half of the panel for AM in Figure 6 and the lower third of the panel for AM in Figure 8. Using the absolute durations of the nuclear word and foot as predictor variables and including data of all levels of WbDist and StDist, all varieties except AM and GR for WordDur and AM for FootDur were found to adopt Strategy 3 (see Tables 13 and 15). Again, however, the effects on the distance of L1 from the beginning of the word and foot were small. Estimates of Ω² for t2_L/w and t2_L/f were more than five times larger than for t1_L/w and t1_L/f, respectively. Moreover, with foot duration as the dependent variable, speakers were no longer found to adopt Strategy 2.

Overall, the timing patterns of L1 were very similar to those of H in most varieties. L1, like H, was fairly stably aligned with the beginning of the preceding landmark, with a tendency to proportional timing in some varieties. These findings are fully in line with those by Barnes et al. (2010) for American English, which, unlike Barnes et al. 2006, found the alignment of L1 relative to H to be unaffected by the distance of the word boundary and first postnuclear stress. A comparison of estimates of Ω² for t2_H/w, t2_H/f, t2_L/w, and t2_L/f for H and L1 shows that the amount of variance explained when predicting the alignment of L1 was smaller than when predicting the alignment of H. One possible explanation may be that in the case of L1 a tonal target rather than a segmental landmark was used as point of reference. Therefore, we carried out additional analyses not reported here, in which we used segmental landmarks like the beginning of the nuclear word and foot or the end of the accented syllable as points of reference for the alignment of L1. These analyses provided the same overall patterns. ^[7]

4.3 Cross-linguistic variation

Whereas comparisons of the overall length of the nuclear word and foot in Section 3.1 did not yield substantial cross-linguistic variation, except for WL (see below), comparisons of the timing of L0, H, and L1 in Section 3.4 have shown that in the more ‘central’ varieties, RO, AM, GR, and WI, the overall accentual gesture is delayed when compared to the ‘peripheral’ varieties, ZB, WL, and WH. Accordingly, we observed inverted U-shaped patterns in Figure 12, with the largest values for L0 Delay, H Delay, and L1 Delay being found in AM and GR. Cross-linguistic differences in the timing of the accentual gesture have been found for a number of languages and dialects for both nuclear and prenuclear accents (e.g., Atterer and Ladd 2004; Ladd et al. 2009), but we did not expect gradual variation as it is manifested by the observed inverted U-shaped pattern, which suggests that geographical distance matters more in this area than historical-grammatical origin of the languages involved (for similar findings see Peters et al. 2014).

We also observed that AM and GR most often adopted Strategy 3 for the timing of both H and L1, and AM was the only variety that adopted Strategy 2 for the timing of L1 (see Table 12). Strategy 2 and Strategy 3 have in common that increasing word or foot size increases H Delay or L1 Delay. The word and foot boundaries in these varieties might conceivably have acted as stronger attractors for H and L1 because of smaller absolute distances between the prosodic boundaries and the tonal targets when compared to other varieties (cf. Schepman et al. 2006). In that case, however, we should expect stronger effects of WbDist and StDist for smaller values of WbDist and StDist, where H and L1 occur closer to the prosodic boundaries, which was not the case.

We also observed that the timing of L1 was generally more variable in AM than in all other varieties, which may be explained by the higher number of F₀ contours that could not easily be classified as either ordinary high falls or as ‘late peak’ modification of the rising-falling accent (Ladd 1983), the late C-fall instead of the early A-fall of ‘t Hart et al. (1990). The latter is typically realized in AM with a high plateau, ranging from the nuclear syllable to the last stress in the intonational phrase. In Figure 13, we illustrate the ordinary fall, the ‘late peak’ modification with a high plateau, and an F₀ contour that could not easily be classified as either one of these. Note that the classification of these contours as instances of ordinary falls yielded very large values for L1 Delay.

Finally, the question arises of why the nuclear word was found to be longer in WL than in ZB and RO in Section 3.1, and why both the nuclear word and foot were found to be longer in WL than in WH, which data were recorded by the same speakers. As for WL and WH, there are two factors which may have contributed to these differences. First, Weener speakers were less familiar with Low Saxon as a written language than with High German. As a result, they may have read the Low Saxon sentences less fluently than the High German ones, which in turn may have led to overall differences in speech rate. Second, closer inspection of Figure 4 reveals that the longer overall durations of the nuclear word in Weener Low German may be particularly due to longer durations of target words with WbDist = 2, that is Molberen, Lumberen, and Melberen. In unpublished research the same speakers were found to realize target words of the Molberen type as Mol.be:rn ([mɔl.bɛːɐn]) rather than as Mol.be.ren ([mɔl.bə.ʀṇ]) in nearly 50% of the cases when they read the Low Saxon versions of the test sentences, but in less than 25% of the cases when they read the High German versions. Even if word forms like Mol.be:rn contain one syllable less than words like Mol.be.ren, the longer duration of the full-voweled second syllable increases the ovesall duration of both the nuclear word and the nuclear foot.

4.4 Implications for phonological theory

According to classical ToBI, trailing tones are tones that occur at a fixed distance from the preceding starred tone, whereas ‘phrase accents’ are timed independently. Grice et al. (2000) defined the ‘phrase accent’ as a boundary tone that seeks an association with a postnuclear stressed syllable. If L1 were in fact attracted to a postnuclear stress in languages like English, German, or Dutch, it would be reasonable to adopt the ToBI representation of the nuclear fall as H* L-L%, originating from Pierrehumbert (1980).

The attraction of L1 to a postnuclear stress was apparently supported by smaller-scale studies like Benzmüller and Grice (1998) for German and Barnes et al. (2006) for American English. In the data reported here, the position of the first postnuclear lexical stress had no appreciable effect on the timing of either H or the end of the fall in seven varieties of coastal continental West Germanic. ^[8] Together with the findings by Barnes et al. (2010), our results provide evidence that the timing of L1 is not substantially affected by the position of the following lexical stress, which from the view of classical ToBI and its further developments in Grice et al. (2000) argues in favour of representing L1 as a trailing tone rather than a ‘phrase accent’. While this representation is consistent with that proposed in Gussenhoven (1991, 2005), which is a bitonal pitch accent H*L, his analysis of a tone as either trailing or phrasal was not primarily guided by constant latencies of tonal targets within the pitch accent, but by contextual and language-specific implementation rules. Specifically, for Dutch, English, and German (cf. Peters 2014), L in a nuclear H*L is timed with reference to H*, but in a prenuclear H*L, it may be timed with reference to the first tone in a following pitch accent. Rather, L1 is regarded as part of the nuclear accent for structural and semantic reasons. For instance, in Dutch, the nuclear and pre-nuclear realizations of H*L appear functionally identical, both being typical of citation pronunciation. Moreover, the allophonic right-alignment of the trailing tone of prenuclear pitch accents is a general feature of the analysis and equally applies to trailing H. The decision to analyse the H-tone that associates with the last stressed syllable in the tonal dialect of Roermond Dutch as a boundary tone by Gussenhoven (2000) is based on the fact that the tone complex H_ɩL_ɩ consistently expresses either non-finality or interrogativity as well as on the fact that they appear without the pitch accent L* in unaccented syllables. Additionally, in this case their status as a complex boundary melody is indicated by the failure of the two tones making up the complex to be separated by a lexical tone on the final mora of the IP.

An alternative interpretation of these findings is presented by Barnes et al. (2010). While observing that the target of L is stably timed with the preceding H* target in American English, they suggest that this is due to the need for low F₀ to occur immediately after the nuclear peak, in order for the contour to be distinct from one where prenuclear H* is followed by L*, where the slope down from H* may be more gradual. They suggest that, given their findings, L may still represent a phrase accent L- that finds an association somewhere in the post-nuclear stretch, but that the preservation of the contrast between H* L- and H* L* requires speakers to produce a sharp fall. There are two considerations that may be relevant here. First, there is no evidence for any syllabic association of L at all (Pierrehumbert 1980: 57); rather, the constant timing of L after H* suggests the left-alignment of an unassociated tone, where ‘alignment’ has the meaning of Prince and Smolensky (1993; cf. Gussenhoven 2000). In that meaning, a tone’s alignment specifies its location with respect to some other element in the phonological structure, independent of whether the tone associates in that location. Second, the existence of contrastive patterns cannot generally be used to argue for a non-phonological interpretation of the phonetic characteristics that are unique to one or the other member of a contrast. A perhaps somewhat crude analogy may illustrate this second point. If we were to argue that the short duration of the tongue glide in roil serves to distinguish it from raw eel, this would in no way support any independent, earlier position that roil was a disyllabic structure. At a minimum, we can maintain, therefore, that an interpretation of the post-H* low target as a phrase accent is not supported by the timing data.

One of the advantages of defining tone class without reference to detailed timing patterns becomes evident when comparing the phonetic realisation of closely related varieties that differ with respect to the timing of L1. Reference to timing patterns in distinguishing between H*+L L% and H* L-L% contours would imply that the nuclear falls produced by the same speaker are represented differently. As Table 14 shows, AM speakers adopted Strategy 2 for the timing of L1 in sentences containing monosyllabic nuclear words (WbDist = 0), but Strategy 1 in sentences containing disyllabic nuclear words (WbDist = 1). According to the ToBI-Grice et al. view, we would have to conclude that in the first case they used H* L-L% and in the second H*+L L%. There is, however, no other motivation for assuming that AM speakers used categorically different contours depending on whether WbDist was 0 or 1. They clearly did not express different meanings in the two contexts. The simpler assumption is that H*L L% was used in all cases, and that phonetic implementation rules show some sensitivity for the timing of L1 to an upcoming stress in words with final main stress.