Idiolectal variation: is migrants’ speech more variable than that of non-migrants?

Resumo

Research on migrants’ speech has documented substantial inter- and intra-speaker variation (Nycz 2018; Silveira 2022; Walker 2019), as well as weak lectal coherence (Beaman; Guy 2022). These findings suggest that migrants may exhibit greater linguistic variance than non-mobile speakers due to dialect contact, but few studies have explicitly tested this hypothesis (see Xie; Jaeger 2020 in L2 context). This paper analyzes whether internal migrants display higher statistical variance in their linguistic patterns compared to non-migrants. To this end, we acoustically analyze the production of Brazilian Portuguese pretonic mid-vowels /e/ and /o/ (e.g., relógio ‘clock,’ romã ‘pomegranate’) in phonological contexts favoring vowel lowering ɛ, ɔ, a well-attested pattern in Northeastern dialects (Nascentes 1922) but largely absent in São Paulo speech. The dataset comprises 60 sociolinguistic interviews, evenly distributed across gender and region/status: 20 non-mobile Alagoans and Paraibans (Northeasterners), 20 non-mobile Paulistas (Southeasterners), and 20 Alagoan and Paraiban migrants residing in the state of São Paulo. All participants are adults with no more than a high school education, and migrants are further balanced for Age of Arrival and Length of Residence. Recordings were automatically aligned using Montreal Forced Aligner, and F1/F2 measurements were Lobanov-normalized and analyzed in R using Levene’s tests (1960) for homogeneity of variance and mixed-effects linear regression models. Model comparisons included marginal R², conditional R², and intraclass correlation coefficients (ICCs) for random effects of Speaker and Lexical Item. Figure 1 shows that, for vowel /e/ (N = 5,085), migrants have an intermediate mean F1 value (µ) compared to both non-migrant groups, indicating convergence towards the host community, and an intermediate variance (σ) in vowel height, with no significant variance difference from both control groups. For vowel /o/ (N = 3,837), however, migrants show no general convergence towards Paulista norms and display significantly higher variance than both non-migrant Northeasterners (F = 4.37, p = 0.04) and Paulistas (F = 61.7, p < 0.001). This higher variance among migrants is driven by males (σ=0.67), as migrant Northeastern women (σ=0.57) exhibit less variance than their non-migrant female counterparts (σ=0.61). Mixed-effects multivariate models (Table 1) indicate that, across all groups, most variance is accounted for by random effects, as conditional R² values (fixed + random effects) are considerably higher than marginal R² values (fixed effects alone). Figure 2 shows that Speaker explains more variance in the non-migrant Northeasterners’ data than in migrants’ and Southeasterners’ speech. Variance due to Lexical Item also decreases substantially from non-migrant Northeasterners to migrants, who approach native Paulista’s variance. Contrary to expectations, these results challenge the hypothesis that dialect contact necessarily increases individual variation. Most cross-comparisons reveal greater variability among non-migrant Northeasterners than among migrants. The greater vowel height variability among non-mobile Northeasterners likely reflects their dialect’s tendency for pretonic mid-vowel lowering, while migrants’ reduced variation suggests accommodation toward Paulistas’ non-lowering pattern both in mean F1 values and in F1 dispersion. The asymmetry between /e/ and /o/ also suggests selective dialect leveling, which underscores the importance of dialect-specific features as sources of variation.