Dialectal Variation in European Portuguese Central Vowel Perception

Dialectal Variation in European Portuguese Central Vowel Perception Valerie Horn1, Esther Rinke2 and Cristina Flores3 1 Department of Psycholinguistics and Didactics of German, Goethe University Frankfurt am Main, DE 2 Department of Romance Languages and Literatures, Goethe University Frankfurt am Main, DE 3 Departamento de Estudos Germanísticos e Eslavos/Centro de Estudos Humanísticos, Universidade do Minho, Braga, PT Corresponding author: Valerie Horn (v.horn@em.uni-frankfurt.de)


Introduction
It has been mentioned in a number of studies that the European Portuguese (EP) dialect spoken in the region of Braga in northern Portugal differs from the EP variety spoken in the Lisbon area with respect to the realization of the central vowels. More precisely, it has been reported in the literature that in stressed syllables, the Littoral/Center dialect shows allophonic variation of [a] and [ɐ], whereas the open vowel [a] is predominant in the north (Barbosa 1965 and1994;Gonçalves 2008;Martinet 1985).
This observation has been corroborated in acoustic studies focussing on language production which have shown that in the speech of Braga, the vowels [ɐ] and [a] are acoustically and articulatorily equivalents in the tonic syllable (cf. Varanda 2015). As a result, the contrast [a]-[ɐ] is less prominent, or even absent, compared to central and southern Portuguese speech in these contexts. For example, where the Lisbon dialect distinguishes between tomamos [tu'mɐmuʃ] (Engl. 'take', present indicative 1 st person plural) and tomámos [tu'mamuʃ] (Engl. 'take', preterite indicative 1 st person plural), according to Varanda, Barroso & Rato (2016) [ɐ] exists mainly in non-stressed syllables, even though it is not identical to the stressed [ɐ] of the Standard variety due to the process of vowel reduction that generally affects non-stressed vowels in EP (Veloso 2013). Hence, the two dialects do not differ with respect to the realization of [ɐ] in the unstressed syllable but in the stressed syllable.
The present paper takes the findings of Varanda, Barroso & Rato (2016) as a starting point and investigates whether the dialectal difference regarding the production of [a] and [ɐ] is reflected in the speakers' general ability to perceive a contrast between the two vowels. The central research question of this study is the following: Do speakers of the northern variety of Portuguese differ in discriminating between [ɐ] and [a], compared to speakers of the Littoral Center variety, because there are no contexts where the two sounds are contrastive in their dialect?

The EP vowel system
According to Mateus, Falé & Freitas (2005: 79-80), EP has nine phonetic oral vowels. Depending on the opening of the jaw (open, midopen, midclosed, closed), and the position of the tongue in the oral cavity (front, central, back), they can be distinguished in the following way. (

Dialectal variation in EP vowel production
According to Frota et al. (2015: 236) based on Cintra (1971), Portuguese dialects divide into two main groups: the northern and the central-southern varieties, which are characterized by a number of differential phonetic features. The northern dialects are considered to be more conservative and are characterized by "the absence of the distinction between /v/ and /b/, the presence of /tʃ/, the apical realization of the dental-alveolar fricative (i.e., [ʂ], [ʐ]), or the preservation of /ow/ (Cintra 1971;Segura & Saramago 2001)." (Frota et al. 2015: 236). The central and southern varieties are considered to be the result of innovations based on territorial expansions from the 13 th century onwards (Cintra 1971;Frota et al. 2015).
The northern varieties of EP further divide into the Transmontano and Alto-Minhoto dialect and the Baixo-Minhoto, Duriense and Beirão dialect; the central-southern varieties consist of the Littoral Center dialect and the Interior Center and South variety. Figure 1, taken from Frota et al. (2015: 237), gives an overview of the dialectal division of European Portuguese. The present study concentrates on the comparison of the variety of EP spoken in the Braga area and the variety of EP spoken in the Lisbon area with respect to the perception of the contrast [a] and [ɐ]. The Lisbon dialect is part of the Littoral Center varieties and considered to be the Standard Variety of EP (SEP). The dialect of Braga belongs to the northern dialects and is part of the Baixo Minhoto variety.
As already mentioned, the Braga dialect has been described in many studies to differ phonologically from SEP with respect to the realization of [a] and [ɐ]. In the Standard dialect, /a/ is realized as [ɐ] in unstressed position or in stressed open syllables before a nasal consonant (cf. Barbosa 1994: 176). For the dialect of Braga, Barbosa (1965) reports that there is no vowel alternation in the context of stressed /a/ in an open syllable followed by a nasal consonant in the north and that [a] is predominant in this dialectal variety. Table 1 gives an overview of the different realizations of /a/ in different contexts in the two dialects.
Rodrigues & Martins (1999) examine vowel realization in six low-educated speakers of the Braga region. The authors observe that the data from Braga show F1 and F2 values that are different from the ones reported in the literature for the Lisbon variety. Amongst other things, they observe that the informants realize the vowel /a/ in different ways.
Rodrigues, Rato & Silva (2014) compare the realization of vowels in speech corpora from Braga to the results of earlier studies on vowel realization in the Lisbon dialect in Delgado-Martins (2002[1973) and Escudero, Boersma, Rauber & Bion (2009). Their investigation also revealed dialectal differences. One interesting observation for the present study concerns the finding that the F1 values in the Braga corpus of Rato (2013) are higher than in the Lisbon dialect speakers in Escudero et al. (2009) and that overall /a/ is realized more openly in the Braga dialect.
Based on these previous findings, Varanda (2015) investigated the acoustic realization of [a] and [ɐ] in tonic open syllables followed by nasal in the speech of inhabitants of the city of Braga. The data was taken from the spontaneous speech corpus Perfil Sociolinguístico da Fala Bracarense 2 (PSFB) and analysed phonetically by examining the F1 and F2 values of the vowels in question. This study revealed that in the Braga region, the production of [a] prevails in all studied contexts independent of the education level and the gender of the speakers.
The comparison with studies by Escudero et al. (2009) and Santos (2013), who examined the F1 and F2 values of vowels by speakers from the Lisbon region, corroborates the dialectal differences between the Lisbon and the Braga dialects. Varanda, Barroso & Rato 2 Sociolinguistic Profile of Braga's speech, cf. Barbosa (2009 nós tomámos (past) 'we took' nós tomamos (pres.) 'we take' (2016) compare the results obtained by Escudero et al. (2009)

Research question and hypothesis
Based on what we know so far about dialectal differences concerning the production of [ɐ] and [a] in the stressed syllable, the present study focuses on the perception of the two vowels. The main aim of this study is to investigate whether speakers of the Braga dialect perform differently in discriminating between [ɐ] and [a] in a vowel discrimination task, compared to speakers of the Littoral Center variety of Lisbon.
Our main research question is therefore the following: • This does, however, not mean that speakers living in the city of Braga will generally be unable to distinguish between the two sounds. Given the fact that the Lisbon variety represents the Standard European Portuguese dialect, which is present in the media and in formal settings, we expect that speakers of the Braga variety will not be completely ignorant of the vowel contrast. This means that they may show some perception abilities of the vowel contrast but they will differentiate less between the two vowels and show a higher degree of intra-and inter-individual variation than the speakers from Lisbon.

Method and participants
This study consists of an online vowel discrimination task, which was conducted with the perception tool Percy, developed by the LMU Munich. 4 This tool is specifically designed for perception experiments. As it is not suited for the elicitation of detailed sociobiographical data, SosciSurvey was chosen to serve for these matters. Both parts of the test were linked by the insertion of a personalized pseudonym. The whole test with both parts had a duration of approximately 15 minutes.
A total of 46 speakers living in two different regions of Portugal participated in this experiment. Their consent to participate in the study was obtained before they started. According to the sociobiographical questionnaire, 23 participants were born and have always lived in the Lisbon area and 23 participants are from the northern Portuguese area around Braga and have always lived in this region. The participants' age ranges from 18 to 50 years (mean: 25; SD: 8.67), 33 identify themselves as female, 12 as male and one as other. 31 out of 46 participants are university students, the remaining 15 already obtained an academic degree. The majority of the test subjects studied or studies humanities (43), only three have a different academic background. All participants were raised in a monolingual context in Portugal and learned one or more second languages later in life, mostly English. It is important to stress that the northern variety is generally present in all social contexts, i.e., there is no diaglossic distribution of dialectal use and no division into high and low variety in the sense of Ferguson (1959).
The vowel identification task was based on Darcy & Krüger's (2012) oddity vowel categorization task: The participants listened to 60 stimulus sequences in total, each consisting of three monosyllabic CVC pseudo-words and had to determine which of the pseudo-words had a different vowel. For the present study, focused on the oral central vowels, we will analyse the speakers' perception of 28 vowel sequences, which contain the target vowels, either in contrast to each other or, as control condition, in contrast to the vowel [i]. 5 In each trial, three items were presented as a stimulus sequence. All tested vowels were embedded in the consonantal contexts /bVʃ/ and /zVʃ/. The plosive and the fricative as the consonants preceding the target vowels were chosen as they were the only possible combinations where the insertion of all four vowels did not lead to an existing European Portuguese word. Because only stressed syllables were tested, the monosyllabic strings are closer to stressed lexical forms than to unstressed clitic forms.
Out of the 28 sequences, 12 were the test items that only contained the vowel contrast

between [a] and [ɐ]. Another 12 sequences served as control items and included a contrast between the target vowels and [i] (i.e., [a]-[i] and [ɐ]-[i]). Given that [i] is acoustically and articulatorily very different from both [a] and [ɐ]
, discrimination should be easy if the participants correctly understood the task. The remaining four sequences were used as distractors as they contained three times the same target vowel, i.e., the participants heard sequences with three identical pseudo-words (catch trials).
The stimuli were produced by three female native speakers of EP from the region of Braga.  [ziʃ]) were implemented in the carrier sentence "Eu digo…" (Engl. 'I say…') and read by the speakers with a normal voice and intonation. The pseudo-words were then extracted and normalized by using Audacity.
For each stimulus, the interstimulus interval (ISI) between the three CVC words was determined as 1.5 seconds, following Colantoni, Steele & Escudero's explanation: "It has been argued that an ISI of 500 ms promotes acoustic rather than phonological comparisons between sound, while an ISI of 1.5 seconds or more ensures phonological processing" (2015: 97). The task was to identify the pseudo-word containing a different vowel (see Figure 2). Previous to the actual perception task the participants completed a training task consisting of three test items with vowels not used in the actual task in order to adapt to the test scheme. The items in the main task were presented in a randomized order.
A mixed-effect model was applied for statistical analysis, conducted in SPSS version 26. In section 3.3 we will present the descriptive results and the statistical analysis.

Results and statistical analysis
We will start by presenting the accuracy scores for all 28 test conditions. Subsequently, we will have a closer look on the target vowels [a] and [ɐ] and present the results for those conditions where they are directly contrasted to each other. We will then compare these results to the accuracy rate in contrast to the control item [i]. Finally, the accuracy scores for the distractor items (catch trials) will be shown. The overall accuracy rates in Table 2 show that the two groups differ with respect to their overall performance. Participants from Lisbon differentiated the vowel contrasts in 92.1% of the cases and participants from Braga in 80.1%. The range of the Lisbon group  , i.e., the contexts where the target vowels have to be identified against each other. In these contexts, the results show that the identification rate is considerably lower for the Braga group (64.9%) compared to the Lisbon group (89.1%). Here we highlight the range between minimal and maximal scores in both groups. The minimal accuracy score in the Braga group is 8.3%, whereas the maximum score reaches 100%. Accordingly, the Standard Deviation reaches a score of 22.9 in the Braga group. In contrast, for the Lisbon group, the range and the Standard Deviation are between 66.7 and 100%; SD = 10.5. The difference between the two speaker groups is illustrated in the boxplot in Figure 3.
Finally, both groups show similar results for the catch trial items (see Table 2, IV), which served as distractors. The Braga group scores 81.5% of accuracy (SD: 20.3), the Lisbon group 80.4% (SD: 28.2). Both groups show high variation for these trials, which is typical for catch trial items.
For the statistical analysis, a binary logistic regression model was applied. The dependent variable was answer value, meaning whether the given answer was accurate or inaccurate.
The factor given answer indicates which position was chosen as the one to contain the different vowel (the first, second or third position, or whether all three vowels sounded the same) and therefore takes into consideration whether the speakers have a general tendency for choosing the odd sound of a certain position. Position of the target vowel refers to the actual position where the different vowel was placed in the sequencing of the stimuli. The question was whether its position mattered for the accuracy of discrimination. Target vowel specifies whether [a] or [ɐ] is asked for, consonantal context denotes whether the target vowel is preceded by /b/ or /z/ and EP variety contains the information whether the test person answering the item came from the Braga or the Lisbon region. Because the contingency coefficent in the test for collinearity (Cramér's V) showed that given answer and position of the target vowel are highly associated (see Table 3), given answer was excluded from the mixed-effects model.
Hence, we included position of the target vowel, target vowel, consonantal context and EP variety as fixed effects in the statistical analysis. Participant and item were entered as random effects. A backward selection of the binary logistic regression models led to the model that had the best overall fit. The winning model consisted of the independent variables position of the target vowel, target vowel, consonantal context, and EP variety. Table 4 shows the results of the selected model. There is no effect of the variables position of the target vowel, target vowel, and consonantal context. The factor European Portuguese variety shows a significant negative effect in relation to the reference value, which in this case is Lisbon, encoded as 0. This means that the group from Braga, encoded as 1, is significantly more likely not to differentiate between the two vowels. Hence, the statistical model underscores that the (only) significant factor determining the identification rate of the vowel contrast [a]-[ɐ] is the speaker's belonging to the northern or central EP dialect.

Discussion and Conclusion
The main finding of the present study is the corroboration of existing dialectal differences between speakers of the Braga region and speakers of the Lisbon region when it comes to the discrimination between the two vowels [a] and [ɐ]. The speakers of the northern Portuguese variety differ from the speakers of the Littoral Center variety in identifying the vowel contrast under consideration, while all speakers perform similarly in the control conditions. Previous acoustic studies (e.g., Varanda, Barroso & Rato 2016) showed that in the northern variety of Braga, the realization of /a/ as [a] prevails in stressed contexts and that there is no systematic alternation between [a] and [ɐ] as in the Lisbon (Standard EP) variety. The present study adds to this observation by showing that the missing distinction between [a] and [ɐ] is reflected in perception differences. In addition to not producing the difference between both central vowels in open and stressed syllables before a nasal consonant, the speakers of the northern dialect also show a general tendency to not consistently perceive this difference in stressed syllables.
As already discussed in section 2.2, the absence of the vowel contrast [a]- [ɐ] in the northern variety is restricted to stressed syllables, particularly to stressed open syllables  . This means that the more restricted occurrence of [ɐ] and, mainly, the absence from stressed contexts and, particularly from contexts where it has a discriminating function appears to affect the speakers' overall ability to perceive the difference between [a]- [ɐ]. The trials included in the present experiment are monosyllabic CVC-items; still, the speakers of the Braga dialect show significant lower discrimination abilities than the Lisbon speakers. We therefore conclude that the low ability to perceive the contrast is not restricted to the contexts described in the literature, but also found in monosyllabic stressed contexts followed by a fricative. It is also noteworthy to mention that the speakers of the Braga dialect show a high degree of intra-and interindividual variation. The ability of the northern dialect speakers to perceive the central vowel(s) is shaped by their exposure to their native dialect where the contrast [a] and [ɐ] is less prominent than in Standard Portuguese speech. However, they are also exposed to some extent to the vowel contrast [a] and [ɐ] when listening to speakers of the central variety (i.e., Standard EP). SEP is of course available to the speakers of the Braga variety as it is ubiquitous for instance in the media. This explains the high variation observed in the group of the northern dialect speakers: some speakers seem more familiar with the Standard dialect than others.
In sum, this study shows that speakers of two different EP varieties perform differently in discriminating the vowel contrast [a]-[ɐ] in a perception task. We presume that the underlying factor might be a difference in the status of [ɐ] in both varieties resulting in a different language acquisition situation. To support our assumptions concerning the status of [ɐ] in European Portuguese, further inquiry needs to be executed to compare the vowel inventories of the northern to the central and southern Portuguese variety. More studies on the contrast [a]-[ɐ] are required, including perception and production tests with different syllable structured test items, including unstressed syllables.