2.1 The terms big accent and small accent

The frequently used pair of terms focus accent and word accent brings with it a number of problems which will become very prominent in the present discussion. Therefore, these terms are replaced here, with the pair of terms big accent and small accent, following Myrberg & Riad (2015). Thus, what has previously been referred to as focus accent is termed big accent here, and what has been termed word accent is termed small accent.

The primary problem with the term pair focus accent and word accent is that it fails to separate the phonological form of an accent from its function. The term focus accent implies that this accent signals focus, and the term word accent implies that this accent expresses the lexical tone accent distinction. However, there is no one-to-one correlation between focus and big accents (a word may carry a big accent even though it is not focused). There is also not a one-to-one correlation between the lexical tone accent distinction and the distribution of small accents (a word may be specified for tone accent 2 but still be realized without a small accent). The lack of a one-to-one correlation between big accents and information-structural focus creates a need to speak of the accent as an object separate from the information-structural status of the constituent on which the accent appears. Likewise, there is sometimes a need to speak separately of the lexical tone accent marking of a morpheme and the pitch accent that appears on a word in a specific realization.

Further complicating the picture, different dialects of Swedish and Norwegian have different tonal representations for the big and small accent, meaning that the terms should also not be too tightly tied to the specific shape of the contour. The terms small and big accent will allow cross-dialectal discussion of the accent types.

2.2 The prosodic phrase φ and the intonation phrase ι in Stockholm Swedish

Following Itô & Mester (2012), three universal categories of the prosodic hierarchy will be assumed here: the prosodic word ω, the prosodic phrase φ, and the intonation phrase ι.⁶ A fuller discussion and application of this model to Stockholm Swedish can be found in Myrberg & Riad (2015). For the discussion in this article, the most relevant domains will be the φ and the ι. Sections 4 and 5 will present evidence that these two categories need to be distinguished for Stockholm Swedish. While such a distinction was also assumed for Swedish by Myrberg (2010; 2013) and Myrberg & Riad (2014; 2015), the empirical evidence presented in these publications was not extensive.

An important part of the argument for a separation between the φ and the ι regards the existence of two separate prominence types, each of which serves as the head of one prosodic category. I will argue here that such a separation is necessary within the group of big accents, meaning that the prenuclear area may contain big as well as small accents (i.e., it is not the case that all prenuclear accents are small and all nuclear accents are big).

Nuclear accents are heads of ι (the rightmost big accent in ι), and have a similar distribution in Swedish to that described for West Germanic languages like English, German and Dutch (Pierrehumbert 1980; Selkirk 1984; Féry 1993; Selkirk 1995; Gussenhoven 2004; Féry & Samek-Lodovici 2006; Ladd 2008). As in other Germanic languages, information-structural foci obligatorily contain the nuclear accent, which is the rightmost and strongest prominence in Swedish sentences (Myrberg 2010; Myrberg & Riad 2015; 2016).⁷ In Swedish, like in other Germanic languages, the requirement that a focus must contain the nuclear accent results in a ban on strong prominences in the postfocal area (e.g., Selkirk 1984; Ladd 2008; Féry 2013). In Swedish, this means that there are no big accents in postfocal areas (note, however, that small accents are realized postfocally in the Stockholm variety of Swedish, cf., discussion in Hansson 2003; Bruce 2007; Myrberg 2010).

Prenuclear big accents are heads of φ and there may be several prenuclear big accents inside one ι. From the perspective of the literature on West Germanic languages, where prenuclear pitch accents are relatively well described, and where there are frequently multiple pitch accents inside a sentence, the need for a distinction between nuclear and prenuclear big accents might seem somewhat trivial. In the Swedish literature, however, it is far from obvious that big accents can appear alongside small accents in prenuclear position. Earlier accounts of Swedish prosodic structure belonging to the so-called Lund-model of Swedish intonation made no distinction between nuclear and pre-nuclear prominences, beyond the separation between small and big accents (cf., e.g., Bruce 1982; 1998; Heldner 2001; Hansson 2003). Hansson (2003) makes a thorough review of the known facts about the domains in the prosodic hierarchy of Swedish, and reaches the conclusion that only one category above the ω can be supported. In the Swedish literature, the big accent has been strongly associated with the expression of focus, and in the seminal work by Bruce (1977) it was referred to as a sentence accent, capturing the observation that (almost) all utterances must have at least one big accent. In most subsequent literature, the big accent has been referred to as focus accent, indicating that signalling of focus is its primary function (e.g., Bruce 1987; 1998; Gussenhoven & Bruce 1999; Heldner 2001; Hansson 2003; Gussenhoven 2004; Bruce 2007; Ambrazaitis 2009; Roll, Horne & Lindgren 2009; Myrberg 2013b). The conception of big accents as focus accents or sentence accents implies that the big accent should bear some correlation with what in the West Germanic literature has been referred to as the “nuclear accent”, i.e., the rightmost and most prominent accent in a sentence, which also correlates with the focus of the sentence. Under this conception, there should, in most cases, be one big accent per sentence (or per focus domain, which would correlate with the sentence in simple cases like those presented here, cf., discussion by e.g., Truckenbrodt 1995; Féry & Samek-Lodovici 2006). The results from the experiment presented here, however, will give ample evidence that there are frequently more than one big accent per sentence in Stockholm Swedish.

2.3 Initiality accent

In addition to appearing on focused items as described in the previous section, big accents have been observed to appear at the left edges of phrases, i.e., phrase-initially. In this position, they function as a phrasing cue (Horne 1994; Roll 2006; Roll, Horne & Lindgren 2009; Myrberg 2010; 2013b; Söderström 2017). These big accents will be referred to here as initiality accents, following Myrberg (2010; 2013b).

The initiality accent appears on the leftmost ω in (the leftmost φ of) an ι.⁸ The requirement on ω-status means that almost all content words in the relevant position obligatorily receive initiality accent, whereas function words, such as pronouns and articles, are often not eligible for initiality accent (see Section 5.2, Figure 20).

An example of an initiality accent is provided in Figure 1 where the initial word många ‘many’ carries the initiality accent. Note that the right H tone of this accent is realized very late, on the initial syllable of rostiga ‘rusty’. This late timing is typical of initiality accents, and separates the tonal behaviour of the initiality accent from other big accents (Roll, Horne & Lindgren 2009; Myrberg 2010). The late timing is, however, not obligatory, as illustrated by the initiality accent on dåliga ‘bad’ in Figure 2. When appearing on a compound, the second rise begins on the secondary stress, in parallel with other big accents on compounds, (2c) and Figure 3.

Figure 1

Initiality accent with late timing. Adapted from Myrberg (2010: 57).

Figure 2

Initiality accent with early timing. Adapted from Myrberg (2010: 54).

Figure 3

Initiality accent on compound. Adapted from Myrberg (2010: 56).

Initiality accents frequently appear in the results of the experiment presented in this article. In (1), big accents which appear on the words bruna or många are initiality accents, because they appear on the leftmost ω in their φ, cf., (1a, d, f). As mentioned in Section 1, initiality accents are analyzed as left-aligned φ-heads throughout the current article. We return to this issue in Section 5.2.

2.4 The syntax-prosody interface

It will be tacitly, and sometimes explicitly, assumed throughout this article that the syntax-prosody correspondence is governed by a set of competing Optimality Theoretic constraints (OT, Prince & Smolensky 1993). These constraints govern the relation between prosodic and syntactic edges, while simultaneously taking into account factors of information structure, prosodic well-formedness, and phrase length. This view on the syntax-prosody interface is based on a long line of previous publications (among many others Selkirk 1996; 2011; Truckenbrodt 1999; Féry & Samek-Lodovici 2006; Elordieta 2007; 2015; Itô & Mester 2007; 2012; 2019a; 2019b; Feldhausen 2010; 2016; Féry 2013; Myrberg 2013b; Ishihara 2014; 2015; Elfner 2015; Bennett, Elfner & McCloskey 2016; 2019; Kalivoda 2018). Laying out the full workings of the interactions between OT-constraints that generate the prosodic structures described here lies beyond the scope of this article and will be left for future research (but cf., Myrberg 2019). Nonetheless, this section reviews the most important constraints that are assumed to be responsible for the prosodic structures observed in Section 4.

An important aspect of the syntax-prosody interface is that there can be optionality in the syntax-prosody correspondence. This means that one and the same syntactic structure can result in several different prosodic output structures (cf., Myrberg 2013b). Factors that govern the variation in the syntax-prosody correspondence include information structure and phrase length, as well as speech style and individual variation.

The constraints that govern the relation between syntactic and prosodic edges are assumed here to be the Match constraints developed by Selkirk (2011), (3).

An important assumption behind these constraints is that the syntax-prosody interface is separately governed on three different levels. The morphosyntactic word has a correspondence with the ω, the syntactic phrase has a correspondence with the φ, and the syntactic clause has a correspondence with the ι. How the syntactic word, phrase, and clause should be formally defined in terms of nodes in the generative model of syntax is yet to be fully understood. For the purposes of the current paper, it will suffice to assume that any XP is matched to φ-edges (see also e.g., Truckenbrodt 1999), and any CP is matched to ι-edges (especially the CP to ι correlation is, however, a simplification, cf., e.g., Selkirk 2011; Myrberg 2013b; Hamlaoui & Szendrői 2017).

The Match constraints interact with a constraint STRESSFOCUS, formulated as in (4) from Truckenbrodt (1995: 160, though Truckenbrodt refers to this constraint as FOCUS).

In addition, the Match constraints interact with constraints that govern the relation between phonological constituents and their heads. HEADEDNESS (5) is inviolable, which means that every phonological constituent in the output structure will have one head (Selkirk 1996).

ALIGNHEAD (6)–(7) is a family of constraints that govern the location of heads inside prosodic constituents. The formulations of ALIGNHEAD in (6)–(7) are from Féry & Samek-Lodovici (2006), though they refer to it as HP and HI, respectively (see also McCarthy & Prince 1993; Féry 2013). ALIGNHEAD as formulated in (6)–(7) will place a phonological head at the right constituent edge. In addition, a constraint that aligns φ-heads with their left edges must be active in Stockholm Swedish, to account for the initiality accent.

STRESSFOCUS is higher ranked than ALIGNHEAD, which means that under e.g., narrow focus, the head appears at the right edge of the focused constituent rather than at the right edge of the phonological constituent (Myrberg 2010).

In addition to the above mentioned constraints, some constraints which control phrase length must be assumed to affect the prosodic phrasing. These account for the phrasing differences between long and short constituents. Constraints that may play a role in this are binarity constraints (e.g., Itô & Mester 2007; 2019b).

3.1 Annotation procedure

The sentences were annotated using Praat (Boersma & Weenink 2016). The left and right edges of each word in the subject were manually inserted. Thereafter, a script was used to automatically insert three tonal target points (points A, B, C in Figures 4 and 5) in the initial and the final ω of the subject. After the tonal points had been automatically inserted, the author controlled all inserted points, manually correcting erroneous measurements due to e.g., creaky voice or consonantal effects in the f0 contour. A detailed account of the procedure through which each point was defined is provided in Sections 3.1.1 (tone accent 1) and 3.1.2 (tone accent 2).¹¹

Figure 4

Result of the annotation procedure for tone accent 1 words with big accents (left panel) and small accents (right panel). See text for explanation.

Figure 5

Result of the annotation procedure for tone accent 2 words with big accents (left and middle panels) and small accents (right panel). See text for explanation.

Crucially, the annotation procedure resulting in tonal target points did not rely on any subjective judgment or analysis regarding which tonal contour the target word carried, but operated independently of the author’s (or any other speaker’s) judgement regarding the phonological representation of the contour. This means that the script was designed so as to apply independently of any assumption about whether the target word was assigned a small accent contour (tone accent 1: HL*, tone accent 2: H*L, cf., Table 1), or a big accent contour (tone accent 1: L*H, tone accent 2: H*LH, cf., Table 1). One necessary consequence of this is that the final annotation contains points that do not correspond to phonological H or L targets (specifically, point C in the tone accent 1 big accent, point A in the tone accent 1 small accent, and point C in the tone accent 2 small accent, cf., Figures 4 and 5).

The second step of the annotation procedure took place after the completion of the semi-automatic annotation, and consisted in the author making an auditory and visual judgment of whether there was a big accent or a small accent contour in each word in the subject. This procedure is described in Section 3.1.3.

3.2 Validity of binary judgments

The author judged each word in the subject with respect to whether the word carried a big or a small accent. This was a binary judgment, looking only at the tonal contour, i.e., if there was a rise or a fall associated with the stressed syllable in accent 1 words (L*H or HL*), and whether there was one or two peaks in accent 2 words (H*L or H*LH). A question mark was used to indicate any case where it was difficult to decide between the two options. In general, the binary choice between big and small accent was uncomplicated to make. This is reflected in the small number of question marks used in the annotation. In total, 4192 words were labelled.¹⁵ Of these, 1886 were labelled big accent, 2276 small accent, and 30 had a ‘?’ after their label. Words marked with ‘?’ are excluded in the analyses below.

In order to verify the validity of the subjective judgments, they were compared with the results of the f0 measure from the tonal annotation. Recall that the annotation procedure that extracted the tonal targets operated independently of the binary judgment. Figure 6 shows the f0 measure in words judged to have big accent versus words judged to have small accent. The left panel shows the f0 of annotation point B in accent 1 words. The right panel shows the f0 measure of annotation point C in accent 2 words. Point B for accent 1 and point C for accent 2 are the points where the crucial difference between big and small accent is expected given the tonal representation of big versus small accents (cf., Figures 4 and 5). In Figure 6, data from all five speakers and from the initial as well as the final words of the subject is plotted together. In spite of this, there is a clearly significant difference between the f0 among words labelled big accent and words labelled small accent. This difference is present for words with tone accent 1 as well as tone accent 2.

Figure 6

F0 measures of words rated as big versus small accents. The left panel shows the f0 measure of annotation point B (see Figure 4 for explanation), in words with accent 1, which are rated as big accent and small accent, respectively. The right panel shows the f0 measure of annotation point C (see Figure 5 for explanation), in words with accent 2, which are rated as big accent and small accent, respectively.

Importantly, no judgment was made with respect to perceived degree of prominence. Though the shape of the f0 contour is strongly related to the perceived degree of prominence, other factors including scaling and duration also play a role for the perceived degree of prominence. Perceived degree of prominence is a gradient phenomenon, which cannot be adequately described on a binary scale, and possibly cannot be adequately annotated by a single speaker, cf., e.g., Strangert & Heldner (1995a; 1995b). See Heldner (2001) and Hansson (2003) for discussion and studies regarding perceived degree of prominence in Swedish. Thus, judging the degree of prominence of a word is more complex than judging the tonal contour on that word, at least for Swedish, where the number of possible tonal contours on any given word is small.

4.1 Subjects with one big accent, one φ

A central observation is that the subject always contains at least one big accent, cf., the ungrammaticality of (12c).¹⁶ Note that the big accent is obligatory even when the subject is information-structurally given. Under the assumption that every big accent is head of one φ (cf., Sections 1 and 5.1), the preverbal constituent obligatorily forms a φ, as indicated with round brackets in (12a) (Figure 7) and (12b) (Figure 8).

Figure 7

The f0 contour for (12a). ‘The brown hare with many cute kids lives in the park’.

Figure 8

The f0 contour for (12b). ‘The wild llama from Lima was moved to the national park’. Note that the f0 contour here is illustrated with a different experiment sentence than in (12). All prosodic options cannot be illustrated with the same sentence throughout the article. However, no significant differences could be found between the different items in terms of prosodic structure.

One of the most interesting aspects of the obligatory big accent regards its position within the preverbal subject. As seen from the contrast between (12a) and (12b), the big accent may appear either on the leftmost or on the rightmost ω of the subject. However, it may not appear on a subject-medial ω, cf., the ungrammaticality of (12d). Assuming that every big accent is head of one φ, we can refer to (12a) as left-headed, and (12b) as right-headed. The big accent that appears in the left-headed φ is an initiality accent (cf., Sections 2.3 and 5.2, see also Myrberg 2010; 2013b; Myrberg & Riad 2015).

In total, 557 of the 1200 recorded subjects have a single big accent (46,4%). Figure 9 shows the frequency of subjects that contain a single big accent for subjects of different lengths.

Figure 9

Number of given and focused subjects phrased in a single φ, in each length condition. Short subjects contain a single big accent more often than long subjects and therefore have higher bars. The total for each bar is 150. Given subjects are shown to the left and focused subjects to the right. Light bars indicate left-headed subjects (12a) and dark bars indicate right-headed subjects (12b).

We see in Figure 9 that short subjects have a single accent more frequently than long subjects (cf., Sections 4.2 and 4.3). Short subjects are almost always phrased as a single φ.

We also see in Figure 9 that the location of focus is an important factor for whether a subject is left-headed (12a) or right-headed (12b). When containing only one big accent, given subjects tend to have an initiality accent (left-headed, (12a)), whereas focused subjects in principle obligatorily have right-aligned prenuclear accents (right-headed, (12b)). This is entirely expected given what we know about the interface between prosody and information structure in Germanic languages. It is well observed that languages like English, German and Swedish have a strong requirement that the rightmost word in a focused constituent carry a so-called nuclear accent, i.e., the strongest prominence in the intonation phrase (among others Bruce 1977; Selkirk 1984; Féry 1993; Selkirk 1995; Féry & Samek-Lodovici 2006; Myrberg 2010; Féry 2013). The left-aligned head in (12a) fails to satisfy this requirement, and thus cannot express narrow focus on the subject. We may note that the left-aligned head is the most common realisation of given subjects with a single big accent, but that right-aligned subjects are also an available option.¹⁷

4.2 Subjects with two big accents, two φ’s

We have seen that the subject obligatorily contains at least one big accent, which is the head of a φ. It is, however, also common for the subject to contain two big accents. Under the assumption that each big accent is the head of one φ, subjects with two big accents are divided into two φ’s. In terms of left- versus right-headedness of the two φ’s, there are three available options, shown in (13). In (13a) and Figure 10, both φ’s are right-headed. In (13b) and Figure 11, the first φ is left-headed whereas the second is right-headed. In (13c) and Figure 12, both φ’s are left-headed. The logically possible option in (13d), where the first φ is right-headed and the second is left-headed, is unattested.

Figure 10

F0 contour of (13a). ‘The one hundred lambs on the rightmost meadow which is farthest away will be sold.’

Figure 11

F0 contour of (13b). ‘The brown hare with many cute kids lives in the park.’

Figure 12

F0 contour of (13c). ‘The brown hare with many cute kids lives in the park.’

Figure 13 shows the number of subjects with two big accents for different subject lengths. The three shades of grey inside each bar indicate the three structures in (13a)–(13c).¹⁸

Figure 13

Number of given and focused subjects phrased in two φ’s, in each length condition. The light part of each bar represents subjects with two right-aligned big accents (13a). The mid-grey represents subjects with one initiality accent and one left-aligned big accent followed by one right-aligned big accent (13b). The dark part of each bar represents two left-aligned big accents (13c). The total for each bar is 150.

4.3 Subjects with three big accents, three ϕ’s

In a comparatively small number of subjects, the subject contains three big accents and is thus divided into three φ’s. Specifically, this is the case in 9% of subjects with 4ω, and 16% of subjects with 5ω. The patterns with three big accents are shown in (14).¹⁹ Contours for each pattern are provided in Figures 14 and 15. Figure 16 shows the frequency of each pattern.

Figure 14

F0 contour of (14a). ‘The kind fairy with long thin wings lives in a mushroom’. Note that the subject is focused in this structure, because this structure is considerably more common among focused subjects than given subjects.

Figure 15

F0 contour of (14b). ‘The kind fairy with long thin wings lives in a mushroom’.

Figure 16

Number of given and focused subjects phrased in three ϕ’s, in each length condition. Only the two longest subject types exhibit three big accents.

4.4 Scaling of nuclear versus prenuclear big accents

As discussed in Sections 1, 2.2 and 5.1, this article argues that a distinction should be made between nuclear big accents (ι-heads) and prenuclear big accents (φ-heads). In relation to this discussion it is interesting to look at the scaling of nuclear versus prenuclear accents. The relevant comparison is between the rightmost word in focused versus given subjects, i.e., the nuclear big accent on information-structurally focused ungar in (15a), and the prenuclear big accent on information-structurally given ungar in (15b).²⁰

Figures 17 and 18 show the scaling of the right H in big accents on the rightmost word (ungar in (15)) of information-structurally focused versus information-structurally given subjects. Figure 17 shows tone accent 1 subjects and Figure 18 shows tone accent 2 subjects. It can be observed that big accents have a significantly higher f0 value in the nuclear accents than in the prenuclear accents. This is true for all length conditions in tone accent 1 and tone accent 2 words, except the short accent 2 subjects. Importantly, only subjects which were realized with a big accent are included in the plots. This means that subjects which did not have a big accent on their last word were excluded from the analysis. The difference between nuclear and prenuclear big accents observed here is previously undocumented and to my knowledge previously unstudied for Swedish.

Figure 17

Subjects with tone accent 1. A comparison of nuclear (15a) versus prenuclear (15b) big accents on the rightmost words of 2, 3, 4, and 5 ω subjects. The boxplots show the height, in semitones, of the H in the big accent 1 contour L*H. (Observe that (15) illustrates sentences with tone accent 2 words. See Appendix A for the set of sentences with tone accent 1). Only words labeled big accent in the subjective judgment are plotted. The four panels show subjects of different lengths. Inside each panel, the dark grey box shows the f0 for focused subjects (nuclear big accent). The light grey box shows the f0 for given subjects (prenuclear accent).

Figure 18

Subjects with tone accent 2. A comparison of nuclear (15a) versus prenuclear (15b) big accents on the final words of 2, 3, 4, and 5 ω subjects. The boxplots show the height, in semitones, of the right H in the big accent 2 contour H*LH. Only words labeled big accent in the subjective judgment are plotted. The four panels show subjects of different lengths. Inside each panel, the dark grey box shows the f0 for focused subjects (nuclear big accent). The light grey box shows the f0 for given subjects (prenuclear big accent).

It is worth noting that the nuclear accents plotted in Figures 17 and 18 are induced by information-structural narrow focus. Nuclear accents also appear in all-new sentences, usually on the rightmost ω (cf., Myrberg 2010 for detailed discussion and exceptions). Myrberg (2013a) showed that nuclear accents in all-new sentences are scaled lower than nuclear accents induced by narrow or contrastive focus. Further research is necessary to understand the scaling of nuclear accents in all-new sentences.

5.1 Arguments for the distinction between φ and ι

The data presented in Section 4 has been claimed to support a distinction between φ and ι. The first part of the argument for such a division relates to the sequencing of φ’s within the ι. Without a domain type governing the φ, it will not be possible to formulate a rule, which can account for the distribution of right- versus left-headed φ’s. Assuming a category ι, however, the distribution of left-aligned heads can be stated as follows: If the leftmost φ in the ι is left-headed, the following φ in the ι may also be left-headed. If the leftmost φ in the ι is right-headed, the following φ in the ι must be right-headed. These distributional patterns of left-headed φ’s were illustrated in (12)–(13), and are repeated in (16)–(17).

The second part of the argument for a φ versus ι distinction relates to the expression of focus. A clause-initial φ may be left-headed only if it does not contain an information-structural focus. Prosodic constituents that do contain focus may not be left-headed in Swedish (nor in other Germanic languages, see references in Section 2.2). Thus, among focused subjects with a single big accent, left-headedness is ungrammatical independent of subject length (see Figure 9).

The obligatory right-headedness of constituents that contain focus is accounted for here by the assumptions that a) the focus is obligatorily expressed as an ι-head, and b) the ι is obligatorily right-headed.²¹ It is very generally accepted in the literature on West Germanic that the nuclear accent is right-aligned inside a focused constituent, and Swedish behaves very much like other Germanic languages in terms of the distribution of nuclear accents (cf., Myrberg 2010; Myrberg & Riad 2016 for a longer discussion on the distribution of accents in relation to focus in Swedish and other Germanic languages).

The third part of the argument for distinguishing the categories φ and ι consists of the scaling difference between big accents that are the head of an ι (nuclear accents) and big accents that are heads of a φ (prenuclear accents), presented in Section 4.4.

The fourth part of the argument relates to the existence of an elaborate set of boundary tones at the right edge of the ι, which is not present at the right edge of φ. A thorough review of these boundary tones is beyond the scope of the present article. The interested reader is referred to Myrberg (2010: 99ff) and Myrberg and Riad (2015).

5.2 The initiality accent and the head hypothesis

As already mentioned in Sections 1 and 2.3, an important implication of the analysis pursued here is that initiality accents are left-aligned heads of φ’s. We may refer to this analysis of the initiality accent as the head hypothesis. While left-headedness at the level of the φ is exotic among the Germanic languages, this is arguably the most convincing analysis of the complex patterns of big accent distribution observed here and in previous literature (cf., the previously proposed prominence hypothesis and boundary tone hypothesis summarized below).

First, the head hypothesis allows the substantial variation in terms of initiality accent distribution that has been observed in this experiment to be captured via the syntax-prosody interface, i.e., via the alignment of φ’s with syntactic XP’s. If all big accents are heads of φ, and if the φ aligns with the syntactic XP, big accents should exhibit a basic pattern of one big accent per syntactic XP, with exceptions relating e.g., to phrase length and information-structural focus. These predictions regarding the correlation between syntactic XP’s and big accents have been shown in Section 4 to be borne out. It is predicted that CP structures should be divided into (at least) two φ’s: one containing the preverbal XP, and one containing the rest of the clause (cf., Footnote 16). This is in line with the observation that the preverbal constituent must contain one big accent (head of the φ that contains the preverbal constituent) and that there is one more (nuclear) big accent in the VP (head of the constituent that contains the rest of the clause).²² Further partitioning of the subject into φ’s is predicted to occur e.g., due to complex XP-structures and phrase length, but is not predicted to be obligatory. This is also in accordance with the results.

The prediction that CP-structures are parsed as two φ’s seems to be borne out even in very short two-ω clauses like that in (18) and Figure 19. In Figure 19, a big accent is obligatory at the beginning of the CP (Myrberg 2010: 54–56). The obligatoriness of the double accents in the short (two ω’s) sentence of Figure 19 is curious from the point of view that subjects of similar length rarely contain two big accents (cf., Figure 9, where it is shown that the two-ω subjects almost always contain a single big accent). However, the head hypothesis offers the intuitive explanation that the prosodic difference between the two-ω subject and the two-ω clause originates in syntax. A two-ω NP is not parsed as two φ’s, whereas a two-ω clause is parsed as two φ’s. It should be noted that in a sentence like that in Figure 19, there will be no way of determining whether the big accent on the first ω is a left-aligned initiality accent or a right-aligned prenuclear accent, as there is only one possible position for the head to appear in within the φ. The only sure way of deciding the alignment of a φ-head, is to look at a constituent that contains at least two ω. We can get some indication but no definite answer to the question by looking at the alignment of the second peak, which regularly but not obligatorily exhibits a later alignment in left-aligned initiality accents than in right-aligned big accents (cf., Figures 1, 2 and 3).

Figure 19

Initiality accent and nuclear accent in the two-ω sentence from (18).

The head analysis also offers a way of understanding why unstressed elements in Spec-CP fail to get initiality accents (Myrberg 2010: 56–57). Specifically, the generalization that the constituent in Spec-CP forms a φ holds only for constituents that are able to project their own ω (cf., the constraint HEADEDNESS from (5), Selkirk 1996). Because unstressed elements like pronouns (e.g., hon ‘she’, han ‘he’, den ‘it’) fail to project a ω, they can also not become the head of any φ. Instead of forming φ’s on their own when in Spec-CP position, they incorporate into a ω to their right. Thus, in (19b) and Figure 20, vi ‘we’ incorporates into one ω together with the verb lämnade ‘left’. It remains to be understood why the verb cannot receive an initiality accent in this type of sentence, cf., the ungrammaticality of (19c).

Figure 20

Lack of initiality accent in the sentence from (19b), with an unstressed pronoun in Spec-CP.

The head hypothesis, further, circumvents several problems associated with the two competing analyses. According to the prominence hypothesis (Myrberg 2010), the initiality accent is a type of prominence or pitch accent. However, this “prominence accent” has an unsatisfactorily unclear status in relation to prosodic phrasing, being analysed as a type of pitch accent, yet lacking status as the head of any prosodic category (Myrberg 2010; 2013b).

Under the competing account, the boundary tone hypothesis (Roll 2006; Roll, Horne & Lindgren 2009), the initiality accent also has a somewhat unclear status, being analysed as a H% boundary tone directly associated with the CP node of main clauses. The H% does not associate with a prosodic edge, however, as would be default for boundary tones. Instead, it appears after the leftmost small accent (tone accent 2: H*L, tone accent 1: (H)L*), thus rendering a contour similar to that of the big accent (tone accent 2: H*L+H%, tone accent 1: (H)L*+H%). The boundary tone hypothesis, therefore, makes the large similarities between the big accent and the initiality accent look accidental.

Importantly, the boundary tone hypothesis also fails to account for the complex distribution of initiality accents evidenced by the results of the current experiment. The clearest example of this is that the boundary tone hypothesis predicts that the intiality accent should be obligatory at the left edge of CP structures. This prediction is not borne out. A relatively large number of Spec-CP positions contain right-aligned prenuclear accent instead of the left-aligned initiality accent (Figure 9), and the initiality accent also fails to appear in clauses where the Spec-CP position contains an unstressed element, (19b). In addition, the boundary tone hypothesis violates the indirect reference hypothesis (Inkelas 1991), as it assumes a direct correlation between syntactic structure (the CP) and prosody (the H%). Since the advent of the prosodic hierarchy, the indirect reference hypothesis is regarded an important part of phonological theory (cf., e.g., discussions in Nespor & Vogel 1986; Truckenbrodt 1999) and any violations of it should be well motivated.

Before moving on, a few predictions of the head hypothesis will be outlined. The head hypothesis predicts that only syntactic structures composed of more than one XP are eligible for initiality accent. Because information-structural focus must have an ι-head at its right edge (see 5.1), an initiality accent can never project to ι-head in any focused constituent or clause. This, in turn, means that the initiality accent can never be the sole big accent in an ι, if the ι contains focused information, as most ι’s do. A left-headed φ, then, must always be followed by another right-headed φ contained in the same focused constituent and by the same ι, which can project to ι-head.

This prediction potentially has implications for an issue yet to be empirically studied and theoretically explained. The initiality accent has, by now, been rather well-studied in utterances consisting of clauses, but when it comes to utterances that consist of a noun phrase, like that in (20b), we have much less knowledge. In order to say something about this, an additional recording was made with two speakers of Stockholm Swedish, reading three of the five-ω subjects from the current experiment in the subject focus context, as in (20). Each subject was repeated twice, rendering a set of twelve sentences. Ten of the twelve noun phrases were read with a single big accent on the last word, as in (20b) and Figure 21. One sentence (both repetitions) was read with two right-aligned accents as in (20c).

Figure 21

Lack of initiality accent in noun phrase utterance, cf., (20b).

This small data set suggests that initiality accents are not common, and possibly not allowed, in utterances consisting of a noun phrase. If true, this distinguishes noun phrase utterances from clause utterances. One prediction of the head analysis is that only noun phrases that are phrased as more than one φ should be eligible for initiality accent (to the extent that NP utterances ever receive an initiality accent). Future research must determine whether noun phrases may have a left-aligned φ-head, and if so, under what conditions.

5.3 Final accenting: Additional evidence for ϕ-break after Spec-CP

As previously discussed, the analysis pursued in this article implies that it is ungrammatical for a non-pronominal preverbal subject to be contained in the same prosodic phrase as the rest of the clause, i.e., that the constituent in Spec-CP obligatorily forms a φ. The main argument for this is the unacceptability of (16c). In this section, I discuss some data, yet to be experimentally tested, which provide additional indications that there is a prosodic edge at the right edge of Spec-CP. It can be noted that the patterns described in this section are reminiscent of patterns reported from English (e.g., Selkirk 1996; Itô & Mester 2019a).

Some words are deaccented in non-final positions. This applies especially to some groups of verbs, as illustrated in (21a) with auxiliary verbs and in (21b) with the verb går ‘goes’.²³ Verbs and other words that fail to receive a small or big accent do not form maximal ω on their own, but incorporate across the ω-edge into the ω to their right, as in (21).²⁴ There is, to date, no exact description of which words or verbs may be deaccented or why, but see e.g., Myrberg & Riad (2015) for discussion.

When otherwise deaccented words as in (21) appear rightmost in an ι, they are obligatorily accented, as illustrated in (22).

Interestingly, the verbs in (21)–(22) are also obligatorily accented (big or small accent) when they appear rightmost in Spec-CP. This is illustrated in (23a) and (23b). This fact constitutes additional evidence for an obligatory prosodic edge at the right edge of Spec-CP.