According to Christen et al. (2013, pp. 28–30), four main dialect areas exist in German-speaking Switzerland, three of which belong to the ALM dialect family. In Basel, Low ALM is spoken, otherwise north of the Alps, High ALM is used. The dialects spoken in the Alpine southern region of German-speaking Switzerland belong to Highest ALM, except in Samnaun, where a Bavarian dialect is used (Haas, 2000, p. 71). As Low ALM and Bavarian make up for only a little part of the dialect diversity in German-speaking Switzerland, they are not part of the dialects scrutinised in this article.

The High and Highest ALM area has four identifiable broad regions (see Figure 1) that originate in the division into a northern and southern part and an eastern and western one (Haas, 2000, p. 67). These divisions are based on averaging different isoglosses that approximately run along the same geographic areas. Although much variation can still be observed within each quadrant, assuming these four regions can help order the vast amount of dialectal diversity. This study focusses on one representative of each of the four regions, i.e., the dialects of Bern (BE), Chur (GR),¹ Brig (VS),² and Zurich (ZH).

Before dialectal influences on SSG are discussed, it is important to note that no generally accepted SSG pronunciation norm exists as it is the case for Germany (e.g., Siebs, 1969; Duden, 2015). Rather, there is a lot of articulatory leeway when Swiss people speak SSG. As Hove (2002, p. 6) claims, it is a popular idea in German-speaking Switzerland that Swiss people try to speak German ‘correctly’ but fail to do so, which is why they employ ALM sounds. From the set of these ALM sounds, the general public seems to have an unwritten agreement over which ones are more and which ones are less acceptable (Guntern, 2012, p. 103), a situation which Hove calls ‘language convention’ (2002, p. 6). Obviously, it is difficult to predict how SSG is spoken exactly, as this ‘language convention’ does not categorically exclude the emergence of certain stigmatised dialectal features. There is nevertheless a greater likelihood for some variants to occur, i.e., those that are not stigmatised, and for others not to occur, i.e., those that are stigmatised. This SSG variety based on likelihood of occurrence within the framework of the ‘SSG language convention’ (Hove, 2002, p. 6) has been referred to as typical SSG (Zihlmann, 2020a, p. 8), a term that will also be used throughout this study.

Given the three studies presented have found interdialectal differences, the question arises whether a SSG speaker’s ALM dialect origin can be auditorily identified. In fact, empirical evidence from a perception experiment (Guntern, 2011, p. 177) suggests that listeners are indeed able to identify the ALM origin of a SSG speaker. However, the accuracy rate differs depending on the dialect. Specifically, it was 90% correct recognition for VS, 75% for BE, 40% for GR, and 25% for ZH. Guntern claims that cues for this auditory dialect localisation in spoken SSG lie primarily on the level of vowel quality (and consonant quality). More precisely, she explains that listeners relied on the quality of the short vowels /ɪ ʊ ʏ ɔ œ ɛ/, the long vowels /iː uː yː oː øː eː/, the <a>-sounds, the diphthongs as well as the quality of /r l k/ when successfully judging a SSG speaker’s ALM dialect origin (Guntern, 2011, p. 181). However, she adds that those ALM dialects that were identified the most seem to contain phonetic features that are rather rare within German-speaking Switzerland, as it is the case for VS ALM. Thus, if ALM interferences in SSG take place, it is easier for listeners to infer the dialectal origin of a SSG speaker if said interferences are caused by phonetic features that are unique to a given ALM dialect.

Next to the field of dialectology, the insights of this study could potentially be of interest to forensic caseworkers. Given that dialects are an essential element of a speaker’s identity (Rose, 2002, pp. 44–48; Leemann et al., 2018, p. 81), the way in which ALM dialects colour the same speaker’s SSG, and the possible stability of certain features could prove beneficial, e.g., in the context of speaker identification, verification or elimination. Of course, this will have to be taken with a grain of salt as the International Association for Forensic Phonetics and Acoustics (IAFPA) states in its Code of Practice, rightly so, that ‘[m]embers should exercise particular caution with cross-language comparisons’ when carrying out forensic speaker identification/elimination work. Nevertheless, the results might help forensic caseworkers by adding an additional piece of evidence to a given case, especially if dialectologically rare features are involved.

The following question guided the research: Keeping in mind a speaker’s ALM dialect background, how does interspeaker variability regarding SSG consonant quantity, vowel quality, and temporal features manifest itself?

It is important to understand that the variables used to explore speaker individuality were chosen because of statistically significant average interdialectal differences in Zihlmann’s previous two studies (2020a, 2021) and, in the case of temporal variables, because previous research (Wiget et al., 2010; Leemann et al., 2014b; Dellwo et al., 2015; Asadi et al., 2018) has reported that AR and %V are valuable variable to explore interspeaker differences. No claims are made that other variables not scrutinised in the current study are not useable to address the research question or that the variables scrutinised in this study are the best ones to explore speaker individuality.

32 speakers from four regions were recorded (8 each from BE, GR, VS, and ZH; 50% female; age range: 17–32 years; mean = 22.5; standard deviation (SD) = 3.42). All subjects held a Matura degree (higher secondary education) except one, and only three did not hold a university degree. One (or both) of each speakers’ parents had to have grown up with the same dialect, and if only one parent spoke the same dialect, the other parent could not have grown up in a country where another German variety is spoken. To reduce the amount of possible dialect contact, the participants either had to have still been residing in the city in which they had grown up or they could not have lived elsewhere for more than three years. 2 GR and 7 VS speakers used to study in Bern or Zurich but they reported that their primary social group consisted of people speaking the same dialect. Generally, it was made sure that the speakers in the four groups were as homogeneous as possible to reduce potential effects of social factors as, e.g., education level and age.

Two wordlists were used, (1) for the assessment of consonant (and vowel) quantity as well as temporal variables, and (2) for the assessment of vowel quality.

Wordlist 1 consisted of disyllabic words with one of the vowels /i a u/ as the nucleus of the first syllable plus the consonants /p b̥ t d̥ k g̊ l n s z̥/ as the onset of the second syllable in the four phonotactically permissible vowel-consonant sequences VC, VCː, VːC, and VːCː. Due to variety-specific phonotactic constraints (see Zihlmann, 2020a, pp. 13–14), 61 words were used for BE, 65 for GR, 59 for VS, 64 for ZH, and 62 for SSG.

Wordlist 2 consisted of monosyllabic words representing the variety-specific long-vowel inventories. Due to each variety’s historical development (see Zihlmann, 2021, pp. 3–4), different numbers of vowels exist (specified in Table 1).

The words from both lists were put into variety-specific generic carrier phrases whose sound immediately before the target word was a vowel (see Table 2). All ALM words were written in Dieth’s (1938) spelling system, and all SSG words were written in SSG orthography.

Subsequently, for each variety blocks were created containing all variety-specific words embedded in the carrier phrases in randomised order. These blocks were arranged as ALM-SSG-ALM-SSG-ALM-SSG, resulting in three repetitions of each word. To familiarise the participants with the recording situation and elicit more naturalistic speech, prior to the recording session an interview was conducted in which metadata on the participants was collected. While most of the insights of the interviews were irrelevant to the purpose of the study, it provided an overview of frequency of SSG use and self-evaluated SSG proficiency. However, the analysis of frequency of use and the self-evaluation yielded no statistically significant correlations, which is why they are not mentioned in the result section.

For the recording of the stimuli, SpeechRecorder version 3.28.0 (Draxler & Jänsch, 2004) was used, so the participants could read the words from a screen sentence by sentence. They were instructed to speak as naturally as possible without artificial hyperarticulation. If possible, the recordings took place in a sound-attenuated booth at the University of Zurich with the interface USBPre^® 2 by Sound Devices and the microphone NT2-A by RØDE (at 16-bit/44.1 kHz in mono, stored as .WAV). If a participant could not come to Zurich (which was the case for 19 subjects), the recording was conducted in a quiet furnished room either at the University of Bern or at their homes with portable recording equipment consisting of an identical interface model and the microphone Opus 54.16/3 by BeyerDynamic (at 16-bit/44.1 kHz in mono, stored as .WAV). The collected corpora contained about 12,000 tokens for wordlist 1, and about 2,000 tokens for wordlist 2.

The recordings were automatically segmented via an R script (2019; courtesy of Markus Jochim, Ludwig Maximilian University of Munich) using the Munich AUtomatic Segmentation (MAUS) System (Schiel, 1999; Kisler et al., 2017) with the language setting General Swiss German for ALM and standard German for SSG. Subsequently, the sentences were uploaded to the EMU Speech Database Management System (Winkelmann et al., 2017), where they were manually corrected by Zihlmann (78%) and, due to reasons of time, two additional researchers (22%).

The analysis of phonological quantity included PVD, amongst other variables not discussed here (see Zihlmann, 2020a, pp. 15–16). PVD has the advantage of being a value that disposes of articulation-rate differences, which makes varieties comparable. As mentioned in section 1.2.1, given that the average SSG PVD values of Bullen ‘cops’ were only statistically significantly different for BE and GR speakers (Zihlmann, 2020a, p. 30), VS and ZH SSG speakers are excluded from the analysis, and only the subset of the 16 BE and GR SSG speakers is analysed.

As there was a great amount of intradialectal interspeaker variation observed with regard to vowel quality for the BE SSG realisations of <ä>, the vowel-quality analysis will focus on the 8 BE SSG speakers. The assessment of their realisations of SSG <ä> was done by Zihlmann auditorily (grouped categorically by phoneme) and visually (by means of vowel plots) due to the relatively small amount of data per speaker. The vowel plots were made in R (2019) with phonR (McCloy, 2016) using frequencies normalised with Lobanov’s (1971) procedure with NORM (Thomas & Kendall, 2007).

Regarding temporal variables, the current study will analyse the data of all 32 speakers given that the factors AR and %V were reported to be good indicators for interspeaker variation (see, e.g., Leemann, 2017; Asadi et al., 2018).

All statistical analyses were conducted in R (2019) and involved linear mixed-effects models (LMM) with lme4 (Bates et al., 2015). No obvious deviations from homoscedasticity or normality were observed for the residual plots. For post-hoc pairwise comparisons (Tukey method), lsmeans (Lenth, 2016) was used. Regarding the LMMs for temporal variables, I opted for a model with a double interaction rather than a triple interaction as dialect turned out not to be a statistically significant fixed factor for AR and SR (Zihlmann, 2020b, p. 622). Concerning %V, the fixed factor dialect was only statistically significant for ALM but not for SSG in Zihlmann’s (2020b) study. However, due to convergence issues, a double interaction was opted for. For correlation analyses, ggpubr (Kassambara, 2019) was employed using the Pearson method.

Figure 2 depicts each BE and GR speaker’s PVD of the word Bulle/Bullen ‘cops’ in ALM/SSG.

For the consonant-quantity results, the speakers’ behaviour will be categorised in three patterns. Pattern 1 refers to SSG speakers behaving how it would be expected in typical SSG. Pattern 2 refers to SSG speakers behaving how it would be expected in their ALM dialects, i.e., showing ALM interferences. Pattern 3 is used for anything that is not covered by patterns 1 and 2.

In the context of the word Bulle/Bullen ‘cops’, pattern 1 corresponds to SSG speakers using long realisations of /l/, resulting in smaller PVD values. This is, incidentally, also pattern 2 for BE SSG speakers, given that BE ALM produces the word Bulle ‘cops’ with a geminated /l/. Pattern 2 for GR speakers corresponds to shorter SSG /l/, i.e., higher PVD values, due to their ALM dialect equivalent having a singleton /l/.

Zihlmann (2020a, p. 29) reports that on average, speakers behaved according to pattern 2. This, however, does not completely hold true on the speaker level. Most BE SSG speakers, i.e., BE01, BE03, BE06, BE07, and BE09, behave as expected (though in this case, patterns 1 and 2 are congruent). However, BE02, BE05, and to some degree also BE04 show a tendency to produce shorter /l/, in a way that resembles, e.g., GR04 or GR05. Thus, it can be argued that these five speakers behave according to pattern 3. As for the remaining GR speakers, most of them show pattern 2, i.e., GR01, GR02, GR03, GR07, and GR08. However, GR06 has a PVD value that is commensurate with typical SSG, i.e., comparable to the BE speakers, thus showing pattern 1.

The LLM with PVD as the dependent variable, and speaker and variety as independent variables (with interaction term) shows that the interaction (F(1,15)=3.45, p<.001) is statistically significant. The pairwise comparison reveals that there are intra- and interspeaker as well as intra- and interdialectal differences and similarities. Regarding intraspeaker differences, only one speaker showed statistically significantly different ALM and SSG PVDs, namely GR08 (p=.017). BE06, although in the opposite direction, came very close as well and narrowly missed the 5% threshold (p=.068). Moreover, GR07 (p=.151) shows a weak tendency to produce a shorter, i.e., nongeminated, consonant in SSG.

The statistically significant interspeaker differences in SSG PVDs are summarised in Table 3. The values are to be interpreted in a way that, e.g., GR02’s PVD was statistically significantly different from BE01’s, BE03’s, and GR06’s PVDs.

Seven SSG speakers, i.e., BE02, BE04, BE05, BE06, GR04, GR05, and GR08, did not show any statistically significant interspeaker PVD differences.

The speaker-specific vowel plots are visualised in Figure 3.

When auditorily (by grouping the realisations categorically) and visually (based on Figure 3) inspecting the production of the BE speakers’ SSG /eː/ (See ‘lake’) and /ɛː/ (Bär ‘bear’), several phenomena can be identified. Firstly, some SSG vowels are pronounced according to the typical SSG language convention. Secondly, some SSG vowels are produced with ALM interference. Thirdly, some speakers pronounced their SSG vowels inconsistently, resulting in intraspeaker variation, which is evident when the SSG mean values are between two ALM mean values. Lastly, although it was only observed once, hypercorrection occurred, i.e., BE09 producing SSG /ɛː/ as [eː], which, however, is also accepted in German Standard German (König, 1989, pp. 44–46).

To bring some order in these results, the speakers were grouped by shown SSG behaviour. Behaviour 1, i.e., typical SSG pronunciation, was only shown by BE02. Behaviour 2, i.e., showing ALM interferences in both SSG vowels, could be observed for BE03 and BE05 as they produced either vowel in a statistically significantly lower manner. Producing one SSG vowel consistently according to the SSG language convention and the other one consistently with ALM interference was regarded as behaviour 3. Specifically, this was observed for BE06 and BE07. However, while BE06 showed ALM interferences for SSG /eː/, BE07 produced SSG /ɛː/ in a lowered fashion. Therefore, two manifestations exist for this behaviour, i.e., producing SSG /eː/ as [ɛː] and producing SSG /ɛː/ as [æː], while pronouncing the other vowel compliant with typical SSG. Similarly, behaviour 4 describes speakers who show ALM interferences in a vowel, yet they do it inconsistently, i.e., they show intraspeaker variation. This was the case for BE01 and BE04, who sometimes produced SSG /ɛː/ as [ɛː] and sometimes as [æː], thus showing a varying pronunciation of the word Bär ‘bear’. However, the two speakers differed in their production of SSG /eː/. While BE01 realised it as [eː], which is typical with regard to Hove’s (2002) language convention, BE04 showed ALM interferences and produced it lower as [ɛː]. Therefore, regarding the intraspeaker-variation behaviour, speakers may either pronounce the other vowel according to the SSG language convention, or with ALM interference. Lastly, behaviour 5 was observed for speaker BE09, who hypercorrected SSG /ɛː/ to [eː], which resulted in an overlap with the SSG phoneme /eː/, which the speaker produced typically.

The LMM regarding vowel height with the first formant (F1) as dependent variable, variety, speaker and vowel quality as fixed factors (with interaction term), and random intercepts for word shows that the triple interaction variety * speaker * vowel quality is statistically significant (F(7,119)=14.71, p<.001). With regard to vowel backness, the LMM with the second formant (F2) as dependent variable, variety, speaker and vowel quality as fixed factors (with interaction term), and random intercepts for word shows the triple interaction variety * speaker * vowel quality to be statistically significant (F(7,119)=15.32, p<.001).

Figure 4 shows each speaker’s AR in ALM and SSG in syl/s. The LMM with AR as dependent variable, variety and speaker as fixed factors (with interaction term), and random intercepts for word shows the interaction (F(31,8830.70)=77.10, p<.001) to be statistically significant.

Due to the large amount of data, pairwise comparisons were only conducted for each speaker’s ALM and SSG differences. Only two speakers showed no statistically significant differences in AR between their ALM and SSG varieties, namely BE02 and GR03. All other speakers had a statistically significantly higher AR in SSG.

As for %V, Figure 5 depicts the ALM and SSG results by speaker. The LMM with %V as dependent variable, variety and speaker as fixed factors (with interaction term), and random intercepts for word shows the interaction (F(31,1146)=7.99, p<.001) to be statistically significant.

Here as well, pairwise comparisons were limited to intraspeaker differences between the two varieties. The results suggest that five speakers show no statistically significant differences between their ALM and SSG %V values, i.e., VS01, VS02, ZH02, ZH05, and ZH08. All other speakers had a statistically significantly lower %V value in SSG.

The correlation analysis between AR and %V shows a statistically significantly negative correlation between the two variables (R=-.37; p=.038). In other words, the quicker the speech, the more vowel reduction occurs.