Describing stylistic and prosodic variation in Italian poetic speech: spaciousness, wiggliness, and reader-specific patterns.

Background: Recent work in experimental phonetics has revived interest in poetry performance [1–7], yet intonation – central to poetic enunciation – remains comparatively underexamined. Literary and philological traditions have long stressed melodic organisation, but systematic experimental evidence is limited. Early studies in English reported recurrent patterns, including a so-called monotonous style [8–9]; Romance-language research, and Italian in particular, has complicated this view, identifying register-specific contours such as the poetic declarative (a medium–low terminal pattern) and a huge stylistic variability in interrogative, continuative, and enumerative shapes [6–7;10–11]. To move beyond impressionistic labels (see monotonous patterns), in this study, we quantify intonational dynamics using Wiggliness (rate of f0 slope changes) and Spaciousness (size of f0 excursions), complemented by Interpausal (IPU) Duration and pause distribution [12–16]. Our goal is descriptive: to characterise these indices in Italian poetic speech, to estimate between-reader variability, and to assess how segment length structures intonational realisation; era (historical period) and sex are included as contextual controls rather than primary targets of inference. Corpus and Methods: The corpus comprised 767 IPUs from 30 author-performed recordings (20 male poets and 10 female poets, spanning 1960 to 2022) selected into the Voices of Italian Poets (VIP) archive [5]. f0 tracks were annotated and hand-corrected, following [12–16]. Spaciousness and Wiggliness were modelled with z-scored log-Duration (zlogDur) as a covariate. We fitted Bayesian hierarchical Student-t regressions with random intercepts for Speaker (and File where warranted), reporting posterior means, 90% credible intervals, and P(β>0). Between-reader dispersion was quantified via ICCs and reader baselines; duration–prosody coupling via population effects of zlogDur and reader-specific slopes. Our new intonation visualization model [Fig. 1; 2] – by representing f0 movements, vowel intensity (curve thickness), and relative articulation rate (color- coded) – complements these statistics by enabling qualitative and stylistic analysis. Results: Speaker identity accounts for a substantial share of variance [Fig. 3]. ICCs from random- intercept models attribute a large proportion of variance in Spaciousness to Speaker (ICC = 0.67, 90% CI 0.55–0.79) and a moderate proportion in Wiggliness (0.35, 0.22–0.52) and log(Duration) (0.29, 0.15–0.52). These values indicate robust, reader-specific fingerprints, particularly for the range of f0 movement. Duration strongly organises intonation. At the population level, longer IPUs are associated with larger excursions (Spaciousness ~ zlogDur: β = 0.63, 90% CI 0.52–0.75, P(β>0)=1.00) and with fewer slope changes per unit time (Wiggliness ~ zlogDur: β = −0.28, 90% CI −0.35–−0.21, P(β>0)=0.00). Random-slopes models show heterogeneity around these averages – some readers amplify Spaciousness steeply with duration, while most show a negative duration slope for Wiggliness (yet the direction of the population effects is stable). Overall, results support a range–density trade-off: as segments lengthen, pitch movement expands in amplitude while becoming sparser in directional reversals. Once Speaker and Duration are modelled, era and sex have comparatively small effects. Descriptive era contrasts are small and uncertain for Spaciousness (δ = 0.65, 90% CI −0.67–1.86) and Wiggliness (δ = 0.30, −0.17–0.77), and negligible for log(Duration) (δ = −0.05, −0.19–0.09). Per-reader era shifts are rare and modest. Discussion and Conclusion. This study, replicating the modelling framework of [13–16], offers a compact quantitative account of melodic organisation in Italian poetry performance. Two findings are central: (I) reader identity is the primary source of variation, especially for Spaciousness, indicating stable, recognisable stylistic profiles; and (II) Duration has a robust, interpretable effect: longer segments permit broader f0 excursions while reducing the density of slope reversals, evidencing a range–density trade-off. Although reader-specific slopes vary in magnitude, the population pattern is consistent. Two caveats apply: unequal IPUs counts with a partial sex–era confound, and possible residual text/setting effects despite modelling File where warranted. Overall, poetic speech in this corpus is best characterised as reader-driven intonational style structured by segment length, with Spaciousness and Wiggliness capturing complementary dimensions of that organisation.