Previous studies have shown duration, f0, and intensity to be the primary acoustic correlates of focus marking in English (Weismer & Ingrisano, 1979; Folkins et al., 1976; Atkinson, 1973; Breen et al., 2010; Cooper et al., 1985; Eady et al., 1986; Eady & Cooper, 1986; Kochanski et al., 2005; O’Shaughnessy, 1979; Pell, 2001; Roettger et al., 2019). However, there is a notable variety in the reported findings, which is likely at least partially due to studies differing with regard to investigated dialects and populations, as well as methods. In particular, it is likely that the use of different subsets of acoustic correlates with varied definitions examined in these studies likely have contributed to the mixed findings, making it difficult to evaluate how focus effects systematically manifest in different acoustic correlates. Also, the effects of lexical stress and focus were not clearly delineated or controlled in many of the earlier studies, partly because lexical stress and focus placement are closely related (Turk & White, 1999). Therefore, the current study aimed to examine the effects of focus on a variety of acoustic correlates and build on the previous findings in different varieties of English to gain a more precise understanding of how focus is produced in a single dialect, Canadian English.

Focus-induced lengthening has been consistently observed in previous studies, but there are substantial inconsistencies regarding temporal modulation of words or segments preceding or following focus. For instance, Cooper et al. (1985) and Eady & Cooper (1986) reported lengthening of the focused item without shortening of adjacent words, while Weismer & Ingrisano (1979) and Folkins et al. (1975) found post-focal shortening in addition to focus-induced lengthening on the focused item. Eady et al. (1986) replicated this post-focal shortening effect, but only in shorter sentences (5–7 syllables) and not in longer sentences (10–12 syllables) used in the experiment. Pell (2001) found the post-focal shortening effect in both short (6 syllables) and long (10 syllables) sentences. In terms of pre-focal shortening, Pell (2001) observed pre-focal shortening in long sentences, contrary to studies that did not find shortening preceding focus (Weismer & Ingrisano, 1979; Cooper et al., 1985; Eady & Cooper, 1986).

Intensity has been argued to be another primary acoustic correlate of focus, but this is partly based on earlier findings that greater intensity is associated with lexical stress in English: Fry (1955), Lieberman (1960), and Beckman (1986) observed an increase in the amplitude of the stressed syllable compared to unstressed syllables, and Turk & Sawusch (1996) showed that the contribution of intensity to predicting lexical stress in perception was significant, although its independent contribution in the absence of duration information was small. Kochanski et al. (2005) provided evidence for loudness as a better predictor for detecting prominent syllables than f0, using a corpus of spontaneous speech in seven dialects of British and Irish English. This finding is in contrast to previous studies that found f0 to be a major correlate for prominence. For instance, Xu & Xu (2005) found an expansion of f0 range to be associated with focus and a compression of f0 range with post-focal material, suggesting that f0 makes an independent contribution to the production of focus.

In addition, there is no clear agreement concerning the phonetic details of focus effects on f0. Many previous studies observed higher f0 to be associated with focus, followed by a sharp drop in f0 (Atkinson, 1973; O’Shaughnessy, 1979; Cooper et al., 1985; Eady & Cooper, 1986; Eady et al., 1986; Pell, 2001). However, how and where f0 was measured (e.g. mean or peak, syllable or whole word) and what constitutes a higher f0 also differed between studies, with some reporting a paradigmatic change, e.g., comparing the realization of a word in narrow focus to the realization of the same word in a broad focus sentence, and others a syntagmatic change, e.g. comparing a focused word to the pre- and post-focal words in the same sentence. For example, Atkinson (1973) described a lowering of f0 after a syllable in an emphasized word and discussed how the f0 fluctuation makes the syllable stand out from the rest of the utterance. He also found that this lowering is sometimes accompanied by a higher f0 (peak and mean) of the prominent syllable compared to the same syllable in the same utterance produced without an emphasis on the word. On the other hand, O’Shaughnessy (1979) only stated that emphasis results in larger f0 obtrusions in the emphasized syllables compared to the other syllables in the utterance. Later studies such as Cooper et al. (1985), Eady et al. (1986) and Pell (2001) compared the mean f0 of the focused item with the mean f0 of the corresponding item produced in the neutral condition.

Nevertheless, a post-focal f0 decrease has been observed relatively consistently compared to an f0 increase on a focused item. In particular, Cooper et al. (1985) argued that this post-focal f0 drop, rather than the increased f0 on the focused words, is the major effect of focus in sentence-initial and sentence-medial positions (also see Sánchez-Alvarado, 2020; Arnhold, 2021), whereas Eady et al. (1986) observed a lack of post-focal f0 drop when there was an additional focused word later in the utterance (i.e., in a double-focus condition), which appears to undermine the claim. Eady et al. (1986) also reported a higher f0 peak in the sentence-initial focused word, which was not found consistently in the previous studies (Cooper et al., 1985 with f0 peak; Eady & Cooper, 1986 with f0 peak; Sánchez-Alvarado, 2020 with f0 range; Arnhold, 2021 with f0 peak). Moreover, pre-focal f0 mean lowering shown in Atkinson (1973) and Eady & Cooper (1986) was not found in Pell (2001).

More recently, Breen et al. (2010) examined a relatively large number of acoustic correlates to investigate whether and how they are employed to systematically distinguish between focus on different constituents (subject, verb, object), focus breadths (narrow object focus vs. broad focus), and types of focus (contrastive vs. non-contrastive) in American English. The results of a series of production experiments showed that the focused constituent was signalled with longer duration, higher f0 mean and maximum, and higher maximum intensity, compared to pre- and post-focal words. In addition, objects with narrow focus were marked by longer duration, higher f0 mean and maximum, and higher maximum intensity compared to objects with broad focus. Interestingly, subjects and verbs with contrastive focus were produced with lower f0 mean and maximum than those with noncontrastive narrow focus, contrary to previous studies that found higher f0 to be associated with contrastive focus (Ladd & Morton, 1997 for British English; Ito & Speer, 2008 for American English).

It is crucial to acknowledge that these previous studies examined different varieties of English using different tasks to elicit the speech data, which has likely contributed to the mixed findings on prosodic focus marking. In an attempt to address the resulting contradictions and gaps in the literature, the present study examined the prosodic marking of focus in Canadian English, taking into account a broad range of acoustic correlates and directly comparing realizations of broad focus, narrow contrastive focus and given (or background) information in the sentence-initial, -medial and -final positions (see 1.5 for the hypotheses).

Information structure is also phonologically signaled by pitch accents. Studies on pitch accents or focus in American English suggest that a H* (or L+H*) accent is associated with new information (or contrastive focus) and a L* or absence of pitch accent (or de-accenting) is associated with given information. For example, Pierrehumbert & Hirschberg (1990) argued that new information is rendered salient by a H* while given information does not bear a pitch accent or receives a L*. They also suggested that a L+H accent marks the salience of a scale: a L*+H accent signals a lack of speaker commitment, while a L+H* signals that the speaker asserts that the accented item should be believed over a possible alternative. In other words, under this account, a L+H* can be characterized as a contrastive accent. Similarly, Steedman (2008) characterized H* accents as signalling uncontentious rhemes (a notion that can be roughly equated with focus or new information) and L+H* as uncontentious themes (a notion corresponding to topic or givenness). Welby (2003) showed that listeners are indeed sensitive to the presence or absence of a pitch accent when perceiving the information structure of an utterance. Theoretical accounts also generally agree that the representation of information structure is closely related to the location and type of pitch accent (Büring, 2007; Selkirk, 1995; Truckenbrodt, 1995).

However, empirical studies provide evidence that the relationship between information structure and the pitch accent types is less straightforward than previously considered (Terken & Hirschberg, 1994; Ito et al., 2004; Hedberg & Sosa, 2007; Chodroff & Cole, 2019). Ito et al. (2004) found that while the L+H* pitch accent was more likely to appear in contrastive contexts than non-contrastive contexts, it did appear in non-contrastive contexts in both new and given conditions of information status. De-accenting was also not exclusively found in the given condition. Chodroff & Cole (2019) argued for a probabilistic relationship between information structure and the type of pitch accent on the target word: Given information was more likely to appear with a L* or no pitch accent, whereas new and contrastive information were more likely to appear with H* or L+H*. Their results also showed that, similarly to what Ito et al. (2004) reported, (L+)H* pitch accents were observed on target words in the given condition, and new and contrastive target words with L* or no pitch accent were also observed (also see Grice et al., 2017, for a similar conclusion on German).

The current study investigates how different pitch accents are associated with information structure in Canadian English by analyzing the presence/absence, as well as the types of pitch accents occurring in broad focus, narrow contrastive focus and on pre- and post-focal given constituents. We hypothesize that H* and L+H* are more likely to appear in the items in contrastive condition than given condition, while L* and the absence of an accent are more likely to be observed in items in given than contrastive condition. In addition, the possibility remains that the division between the types of pitch accents occurring in contrastive/given condition is not clear-cut, such that some contrastive constituents do not appear with (L+)H* or some given items appear with pitch accents that are not L*.

Dialectal variation has been argued to be one of the factors that systematically modulate acoustic correlates of focus, which can lead to phonological and/or phonetic differences. For example, differences pertaining to pitch accent inventories and tonal alignment in the varieties of Indian English (Maxwell & Payne, 2018), different pitch accent types (i.e., rising vs. falling) associated with focused elements between Indian and British English (Féry et al., 2016), and differences in the pitch accent alignment between Scottish Standard English and Southern British English (Ladd et al., 2009) have been reported. In addition, O’Reilly et al. (2010) examined f0 values and contours to investigate how broad and narrow contrastive focus are realized in Donegal varieties of Irish (Gaelic) and Irish English. The results suggest that contrastive focus is produced with a large f0 excursion and post-focal f0 lowering compared to broad focus in Irish English, as has been reported previously for other varieties of English (cf. section 1.1). However, there are some characteristics that do not align with some of the earlier results reported for different varieties of English: f0 lowering before contrastive focus was observed (contra. Xu & Xu, 2005 for American English) and broad focus was associated with a rising (L*H) pitch accent (contra Chen et al., 2007 for Southern British English).

Turning to the regional dialects of American English, Arvaniti & Garding (2007) showed evidence that phonological and phonetic differences in focus production may exist between Minnesotan English and Southern Californian English. The use of H* and L+H* in their Southern Californian speakers was modulated by the number of iterations of the target word, whereas the Minnesotan speakers used L+H* irrespective of the number of repetitions, suggesting that H* may not be used in Minnesotan English. In addition, the alignment of pitch accents investigated in this study was generally later in Southern Californian speakers than Minnesotan speakers. Arvaniti & Garding (2007) also noted other differences in phonetic details of pitch accents between their findings and some earlier studies, which the authors argued might be due to the different varieties of American English. For instance, the L tone target of the L+H* was found to be aligned within the accented syllable in their data, whereas it was observed to be aligned with the preceding syllable in earlier studies (Pierrehumbert, 1980; Beckman & Pierrehumbert, 1986). A similar cross-dialectal phonological difference in pitch accent inventories was reported in Clopper & Smiljanic (2011) as tentative findings: A relatively stronger preference for L* pitch accents was observed for Southern American English speakers than Midlands American English speakers.

Previous studies on the relationship between different pitch accents and information status also showed cross-dialectal differences in the interpretation of pitch accents. Chen et al. (2007) investigated the role of four nuclear pitch accents in the processing of information status in Southern British English, building on previous findings in German (Baumann et al., 2006) that different pitch accent types are associated with different types of information status. The authors of this eye-tracking study found that speakers of Southern British English associated fall (H*L) and rise-fall (L*HL) pitch accents with new information, while rise (L*) pitch accents were associated with given information. This differs from how listeners of American English associate different pitch accents with information status. In an eye-tracking experiment, Watson et al. (2008) found that L+H* created a strong bias towards contrastive (and given) information, whereas H* created bias towards both contrastive and new information, suggesting that interpretations of H* and L+H* pitch accents in American English are neither mutually exclusive nor variations of a single pitch accent category.

Empirical findings thus established that different varieties of English realize focus differently. However, Canadian English has been assumed to be a subgroup of American English (Bloomfield, 1970) or intermediate between American and British English (Canepari, 2010), and little attention has been given to whether and how Canadian English may differ from the varieties of American English in terms of prosodic aspects (e.g., Wagner & McAuliffe, 2019, treat their Canadian English data as broadly representative of English). Moreover, the prosodic characteristics of Canadian English are usually not considered in the phonetic or phonological literature describing Canadian English (Chambers, 1973, 1991, 2006; Labov et al., 2005; Sundara, 2005; Boberg, 2008). A few exceptions gave minimal and/or impressionistic descriptions (Halford, 2007; Canepari, 2010), described a distinct dialect (Clarke, 2010 for Newfoundland English), or did not aim to address prosodic aspects specific to Canadian English (DePape et al., 2012). Some studies have investigated a particular phenomenon concerning phrase-level intonation called High Rising Terminal (HRT), also known as uptalk (Lacey et al., 1997; Shokeir, 2008; Sando, 2009; Ouafeu, 2009; also see e.g., Fletcher et al., 2005 on HRTs in other varieties of English). While the differences in Canadian and American English in vowel production have been relatively well studied (Joos, 1942; Clarke et al., 1995; Hagiwara, 2006), the prosodic differences between these two broadly defined varieties of English have been rarely investigated.

Only a few studies provide information on how focus is produced or perceived in Canadian English, although focus marking was not always the primary topic of the study. The primary goal of DePape et al. (2012) was not to investigate focus production in Canadian English, but rather to show how adults with autism spectrum disorder and varying levels of language functioning use prosodic features to mark information status. Nevertheless, the six speakers in the control group provided insight into how focus is marked in Canadian English: larger f0 falls were observed in the focused word regardless of the location of the word in an utterance (i.e., subject or object), and larger f0 rises in the focused word in sentence-final positions compared to given topics in the same position. Patience et al. (2018) examined the first pitch accents of statements and questions using a sentence repetition task. They found that H* is the dominant pitch accent type in short SVO sentences (e.g., Bobby looked for his football), taking up nearly 80% of all first pitch accents, while L*+H is also observed (approximately 15%).

Arnhold et al. (2020) is the only study to our knowledge that directly investigated how regional accent influences the perception of information structure in Canadian English. Experiment 1 was an eye-tracking study: Listeners of Canadian English were given auditory instructions spoken in Standard Southern British English and their eye movements were monitored while they moved a computer mouse to click and move objects presented on the monitor. The target words in the instruction sentences varied in pitch accent type (fall, rise, no accent) and information status (given, new). The results showed that Canadian listeners associated falling accents with new information, and interpreted unaccented words as given, just as British listeners responding to the same items (Chen et al., 2007). Crucially, the rising accents were not clearly associated with either given or new information. This finding, which is in contrast with the results of Chen et al. (2007) for British listeners, suggests that the way Canadian English uses pitch accents to mark information status differs from Standard Southern British English. Given the phonological differences observed between Canadian English and British English, one may assume that there are also phonetic differences in how the acoustic correlates of focus are employed in these varieties of English.

We investigated the prosodic marking of information structure in Western Canadian English by analyzing a scripted speech corpus of 37 native speakers of Alberta and British Columbia English collected through a speech production experiment, which we compared to previous research on other varieties of English. Our main point of comparison was Mainstream American English (MAE), as it has been the subject of most previous research and due to the common assumption that Canadian English is vastly similar to MAE. Given the findings that suggest information structure is marked differently in different dialects of English, we hypothesized that the phonetic and phonological details of focus marking in Western Canadian English are different from those previously observed for MAE. By examining a variety of acoustic correlates to broad focus, narrow focus and given information, the current study aims to show whether and how they are modulated in a systematic way to mark information structure. By annotating and analyzing the types of pitch accents, another goal of the current study is to establish a core pitch accent inventory (limited to declarative sentences) in Canadian English and to test the relationship between information structure and the presence and types of pitch accents. Based on previous findings for other varieties of English, particularly MAE, the following hypotheses can be formulated:

Sixty-one speakers of English participated in a sentence production experiment. They were undergraduate students recruited through the University of Alberta linguistics subject pool. Out of the 61 participants, the recordings of 37 participants who identified themselves as native speakers of Canadian English were used in the analysis, after excluding 18 non-native speakers, 4 native speakers who were balanced bilinguals, 1 native speaker of American English, and 1 native speaker whose audio was not recorded. Of the 37 participants, 26 were female and 11 were male. All were from Western Canada, with 35 identifying various locations in Alberta as the place they primarily lived growing up and 2 identifying locations in British Columbia. Their median age was 19.5, ranging from 17 to 34. The study was approved by the Research Ethics Board 2 of the University of Alberta (study ID: Pro00066772).

Participants gave written consent at the beginning of the experiment. Research assistants verbally explained the task to the participants and gave them time to read the written instructions. Participants were seated in a sound-attenuated booth with a monitor displaying the stimuli and were recorded with a Countryman headset microphone (H6 Omni) connected to a Fostex recorder (model FR-2LE) at a sampling frequency of 48,000 Hz. Participants completed two practice trials before proceeding to the main experiment. After the experiment, participants were asked to fill out an exit questionnaire containing questions about their language background and other demographic information.

During each trial, participants were given a short paragraph describing a situation, followed by a question and an answer about the situation. The questions were designed to elicit broad or narrow focus on different constituents (Subject, Verb, Object) in the answers. An example of a context paragraph and the Q&A pairs in different focus conditions is given in Table 1. The context and question were presented aurally and in writing. Participants were asked to read the answers out loud as if they were in that situation. All answers in the narrow focus conditions began with “no”, which was not included in the acoustic analysis.

Twenty-four answers each occurred in the four focus conditions – broad focus (BF), subject focus (SF), verb focus (VF), object focus (OF) – balanced across four lists with a Latin square design. Each list also contained fifteen filler trials. All three constituents are labeled as BF in the broad focus condition. However, for the three narrow focus (NF) conditions, only one constituent in each answer received narrow focus and the other two constituents without focus were always given. Therefore, Subjects received narrow focus in SF, but Subjects were given in VF and OF, where either the verb (in the VF condition) or object (in the OF condition) received narrow focus.

Word sequences containing an article or possessive pronoun in the case of subjects and objects (e.g., the woman, her photos) or consisting of an auxiliary and a main verb (e.g., was moving) were considered as single constituents in the analysis. All subject, verb and object constituents consisted of three syllables, the second of which was always the lexically stressed syllable (e.g., [mɪ.ˈɹæn.də] Miranda). All materials were also chosen to consist of voiced sounds as much as possible. Read speech was used in this study in order to exclude non-prosodic strategies for marking information structure such as pronominalization and ellipsis, and to ensure that we obtained comparable utterances suitable for phonetic measurements in all focus conditions.

All target utterances were manually segmented in Praat (Boersma & Weenink, 2022). Segmentation used auditory and the following visual cues, in order of importance where available: silent intervals, fricative noise, and the third and second formant. At the same time, f0 contours were inspected in pitch objects: The f0 maximum and minimum of each constituent were automatically identified and manually inspected in order to correct or exclude f0 maxima and minima that do not accurately represent the highest or lowest points in the pitch excursion due to microprosodic f0 movements. Octave jumps and other f0 measurement errors due to noise, etc. were removed from the pitch objects using the ‘unvoice’ button on the affected stretches. Based on these annotations, seven acoustic measurements were taken for each constituent with a Praat script (available at Arnhold, 2021): duration of the constituent (in ms), mean intensity (in dB; extracted in the middle 50% of the nucleus of the three syllables and scaled to 70dB), f0 range (f0 maximum – f0 minimum), f0 maximum, f0 mean, and f0 minimum of the constituent (all f0 measurements in semitones; st henceforth).

Of a total of 888 utterances (37 participants x 24 items), 69 were sorted out due to disfluency. The remaining 819 utterances contained 2,457 constituents (three each), of which 12 constituents were further removed due to internal pauses. Thus, 2,445 constituents underwent acoustic analysis for the duration and intensity measures. A total of 225 constituents were further excluded from the f0 analyses after manual inspection because of creaky voice and other issues making reliable f0 measurements impossible.

Linear Mixed-Effects Regression models were fit separately to the six measured acoustic correlates – duration, relative mean intensity (see section 3.1.2 for details of calculation), f0 range, f0 maximum, f0 mean, and f0 minimum – using the lme4 package (Bates et al., 2015) in R (R core team, 2021). The base model included the focus condition of the utterance (Focus: BF, SF, VF, OF), the constituent on which the measurement was obtained (Constituent: S, V, O), and an interaction between them as fixed effects. Participant and Item were included as random intercepts. We compared the fitness of the base model against more complex models that included Focus, Constituent, or their interaction as random slopes. Model comparisons to determine the random effects structure, as well as whether the interaction term should be retained in the fixed effects, were done using the ANOVA function to determine if there was a significant difference in fit. A more complex model was only selected when it provided a significantly better fit based on the AIC scores (Matuschek et al., 2017). All final models except those for f0 range, f0 maximum, f0 mean, and f0 minimum included Constituent as random slope for both Participant and Item. Based on the resulting models, pairwise comparisons were computed via Tukey test using the emmeans package (Lenth, 2018). The figures in the results section showing the estimates and the results of the pairwise comparisons in the form of compact letter display (Piepho, 2004) were produced using the multcomp package (Hothorn et al., 2008; Bretz et al., 2010). All differences reported below were significant and associated with a p-value smaller than 0.05.

A group of native speakers of Canadian English who did not participate in the experiment annotated the evaluated 819 utterances using the Tone and Break Indices (ToBI) transcription system originally developed for MAE (Beckman & Ayers, 1997). The MAE ToBI system was used to identify pitch accents of Canadian English, given the widespread assumption that Canadian English is similar to MAE in terms of speech prosody, which the current study aims to put to the test. The annotators were seven undergraduate students who received basic training for the ToBI transcription using existing training materials (Veilleux et al., 2006) and instruction from the second author. They annotated a pitch accent for each constituent unless clearly absent, in addition to the boundary tones at the end of each utterance. For categorizing different pitch accents, annotators were instructed to pay attention to both the shape and meaning of the pitch accents, e.g., whether they perceived a constituent as prominent or as implying a different pragmatic meaning from the example recordings of different pitch accents provided by Veilleux et al. (2006). For an illustration of average f0 contours for the most frequent accent types and a discussion of the correspondence between contours and classification, see section 3.2.

The inter-annotator reliability was assessed by calculating Cohen’s Kappa value (Cohen, 1960) in which 0 indicates chance agreement (i.e., annotators selected the same categorical value by chance) and a positive value suggests that the agreement was not due to chance. A trained phonetician (first author) transcribed approximately 45% of the data (370 utterances) using the same MAE ToBI annotation conventions, blind to the original annotation. The inter-annotator reliability between the original annotation and the subset annotation amounted to the kappa value of 0.24, which is considered fair (Cohen, 1960). The rate of agreement, calculated as a percentage of matched annotation out of all 370 utterances, is 68%. When the inter-annotator reliability was calculated between the presence or absence of pitch accent, which formed the basis of the first GLMM that examined the likelihood of pitch accent annotation (accent vs. no accent), the kappa value increased to 0.49 or moderate agreement, and the rate of agreement increased to 79%. The kappa value for the comparison of H* vs. non-H* after excluding the 343 ‘no accent’ constituents from either of the annotators was 0.43, which is considered fair according to Cohen (1960). In this comparison, the rate of agreement was 84%.

To examine the relationship between the information structural condition of a constituent and the pitch accent annotation in terms of likelihood of annotation for any pitch accent or specific pitch accent types, a series of binomial Generalized Linear Mixed-Effects Models (GLMM) was fit using the lme4 package (Bates et al., 2015) in R. The dependent variable for the first GLMM was the presence of pitch accent annotated for each constituent, which is coded as 1 if annotated and 0 if no pitch accent was annotated. The second GLMM, motivated by the frequent occurrence of H* accents across the focus conditions (see section 3.2), was fit for accented constituents only (i.e., excluding unaccented constituents) to examine the probability of the pitch accent annotated for a constituent being a H* variation (i.e., inclusive of the downstepped H* and the H* with a delayed peak) versus any other pitch accent. The dependent variable was coded as 1 if the pitch accent was a H* variation and 0 for any other pitch accent. The base model for both GLMMs included Focus (recoded as BF, NF, GV)1 and Constituent (S, V, O) as fixed effects. The first GLMM failed to converge when the interaction between focus and constituent was added, and therefore it was not included in the final model. The second GLMM included the interaction term. Participant and Item were included as random intercepts for both models. Any more complex random structure resulted in convergence failure and therefore was not used in the final models. The model selection process was the same as the linear mixed-effects models used in the acoustic analysis.

All models showed a significant interaction between Focus and Constituent (see the fixed effects summary tables in section 1 of the supplementary materials). Therefore, in this section, we focus on the results of the pairwise comparisons between individual factor level combinations, which are based on the Linear Mixed Effects Models for each acoustic correlate. Results of these comparisons are presented with compact letter display. The output of the pairwise comparisons are given in section 2 of the supplementary materials.

The acoustic correlates examined in the current study were systematically modulated by the focus conditions. The presence of narrow focus resulted in increased magnitudes of various correlates, most notably constituent duration, intensity and f0 range of the focused constituent relative to their realization in given conditions. The other f0 measures did not show enhancement effects due to narrow focus across the board: narrow-focused subjects were actually lower in f0 mean and f0 minimum than given subjects and comparable to given in f0 maximum. The f0 maximum, mean, and minimum of verbs were higher in the narrow focus condition than those in the post-focal given condition (i.e., subject focus) but not different from those in the pre-focal given condition (i.e., object focus). As for objects, the presence of narrow focus manifested as enhancement in some contexts (f0 mean compared to post-focal condition and f0 maximum) but not in the others (f0 mean compared to pre-focal condition and f0 minimum).

The comparison between narrow and broad focus revealed a complex pattern: while the durations of all constituents were longer under narrow focus than broad focus, only verbs showed enhancement of relative intensity under narrow focus compared to broad focus, and none of the constituents showed significant differences in f0 range between the narrow and broad focus conditions. With regards to the other f0 measures, objects did not show significantly differences between broad focus and narrow (object) focus, while narrow-focused subjects and verbs were even lower in f0 maximum, mean and minimum than subjects and verbs in broad focus, respectively.

Finally, comparing broad focus and given conditions showed that subjects showed lower f0 mean and f0 maxima in given conditions than in broad focus, while given objects had shorter durations, lower relative intensity, f0 ranges and f0 maxima than in broad focus. For verbs, durations were shorter in both given conditions (SF and OF) than in broad focus, while f0 maximum, f0 mean and f0 minimum were lower in SF compared to broad focus, suggesting clearer marking post-focally than pre-focally.

Overall, the hypotheses in H1 (see 1.5) were partially supported: narrow focus did not induce enhancement of all acoustic correlates in an across-the-board manner, although enhancement of duration, intensity, and f0 range in the narrow-focused constituents compared to given constituents were consistently observed. Given constituents differed not only from narrow focus, but also from broad focus realizations, though less extensively. As hypothesized, several measures showed consistent marking of givenness only post-focally.

Figure 7 displays all constituents submitted to the pitch accent analysis, divided by the presence or absence as well as the types of pitch accents. The most frequent label was “no accent”, i.e., unaccented constituents (35% of the data; 860 constituents). Among the accented constituents, the dominant pitch accent type was H* – a total of 625 constituents were annotated with a H*. The number of the H* variations (i.e., including the constituents in which a delayed peak and/or downstepping was additionally identified) added up to 1,160 constituents, which took up approximately 73% of all accented constituents and 47% of all constituents overall. Variations of the L+H* pitch accent accounted for approximately 14% (228 constituents) and variations of the L* accent accounted for 12% (183 constituents) out of all accented constituents. The least common pitch accents were L*+H (16 constituents) and H+!H* (1 constituent).

Figure 8 illustrates time-normalized average f0 contours for the three most common main accent types, i.e., variations due to a delayed peak and downstepping are combined into a single category. These were obtained from f0 measurements at 10 equidistant points on each constituent. Averages of these measurements, as well as error bars representing standard errors, appear in Figure 8 for each constituent and focus condition by main accent type. Average contours are clearly distinct for the three accent types H* (and its variations), L* (and its variations) and L+H* (and its variations), as well as consistent across constituents, with H*-accents showing a small peak followed by a slight fall, L* accents displaying a dip in f0 and L+H* accents showing a low turning point (L+) followed by a rise to a peak (H*) followed by a small fall. The difference between the H* contours and the “no-accent” contours was less than prominent. Here, differences in perceived prominence leading to the respective classification (cf. section 2.5) may not be reflected in average f0 contours. Finally, Figure 8 illustrates that average contours for all constituents and accent types were consistent across focus conditions, further suggesting that annotations reflect prosodic realizations rather than experimental conditions as such. Instead, some accents were never observed in certain experimental conditions – for example, no L+H* accent was observed for objects in the given conditions – i.e., SF and VF.

To directly illustrate the correlation between the frequency of accent types and experimental conditions, Figure 9 shows the occurrences of each main pitch accent type divided by focus condition and constituent. The most noticeable pattern was the high number of unaccented constituents (labeled “No Accent”) in the given conditions across the constituents. It is also noteworthy that, while the H* pitch accent is the dominant type across the conditions, verbs with BF and VF are annotated with L* more frequently than subjects and objects in any condition. In addition, the L+H* accent was rare on verbs and objects, particularly in given conditions, but appeared relatively frequently on subject constituents in all conditions.

The first GLMM examined the likelihood of pitch accent annotation (vs. no accent) depending on location and focus condition (see Tables 15 and 16 in the Supplementary Materials). Recall that focus was re-coded for the analyses of pitch accents to reflect the information status of the individual constituent as being in broad focus (BF), narrow focus (NF) or given (GV) rather than reflecting the information structure of the whole utterance (BF, SF, VF, OF). The results showed significant main effects of focus and constituents: Constituents with NF were significantly more likely to be annotated as accented than constituents with BF, which in turn were more likely to be accented than given constituents. The low probability of pitch accent annotation for the given constituents corresponds to the high frequency of unaccented given constituents shown in Figure 9. Comparing the different constituents, subjects were significantly more likely to be accented than verbs, which were more likely to be accented than objects.

Focusing on the accented constituents in Figure 9, it appears that the majority of them was annotated with a H* variation, while far fewer constituents were annotated with a different type of pitch accent, regardless of the focus conditions and for all three constituents. The second GLMM therefore examined the probability of H* annotation compared to all other pitch accent types, excluding the unaccented constituents (see Tables 17 and 18 in the Supplementary Materials). Figure 10 shows the results of the inverse calculation of the estimates of the fixed effects in the model (specified in the logit scale), with higher values indicating more frequent/probable H* accents than lower values. The interaction between Focus and Constituent was significant, and the results of the multiple comparisons using the Tukey test are indicated by the compact letter display. The results of the second GLMM showed that, overall, accented constituents showed a probability of H* annotation higher than 50% across the focus conditions and locations. The results of the multiple comparisons between the levels of Focus and Constituent showed that narrow-focused verbs were significantly less likely to be annotated with a H* (‘a’) than given verbs (“bcd”) or objects in any focus condition (‘d’ or ‘cd’). Moreover, objects in any focus condition were more likely to be annotated with H* accents than broad-focused subjects (‘ab’) or given subjects (‘ab’). All other combinations of focus and location did not differ significantly from each other, suggesting an overall high probability of H* accents.

The pitch accent analysis showed that the presence of narrow focus led to higher probability of the focused constituents being annotated as accented, compared to broad focus and given constituents. It also showed the prevalence of H* accent under the presence of broad or narrow focus in the variety of Canadian English examined in the current study. The average f0 contours in different focus conditions and constituents suggest that the f0 shapes differ between the broad pitch accent categories such as H* and L+H*, although further research is needed to evaluate how the different types of focus map onto the different f0 shapes. Overall, the hypotheses in H2 (see 1.5) were partially supported: narrow-focused constituents were most frequently annotated as H*, which was far more frequent than L+H* across all conditions. In addition, while given constituents were less likely to be annotated as accented than broad-focused constituents, L* accents were not consistently associated with the given, accented constituents.