Amalia Arvaniti

The present paper advocates for incorporating Big Team Science practices into the language sciences. Big Team Science is a promising mode of collaborative research in which top-down coordination directs efforts by a large number of researchers from different institutions towards a common scientific goal. The redistribution of resources and division of labor enabled by Big Team Science may enable researchers to increase methodological rigor, systematically explore claims across a wider sample of languages, and to diversify both the languages studied and the researchers engaged. However, implementing Big Team Science practices is not without challenges, some of which are unique to the language sciences. We suggest strategies and best practices for adapting to these challenges, including theoretical and methodological pluralism, variation in conditions among data collection sites, and navigating a trade-off between maximizing methodological rigor and ability to engage with the most inclusive, typologically broad language samples.

This paper examines the scripted vs. unscripted speech debate from the perspective of statistical power. Using a case study on Greek pitch accents and simulation-based power analysis, we find that unscripted speech offers greater power for main effects, while scripted speech is more powerful for interactions. The optimal strategy for increasing power — adding speakers vs. adding items — varies depending on the effect being tested. We argue that neither method is inherently superior; the choice should be guided by the research question, the effect size, the results of the power analysis and available resources.

Purpose: When studying how factors influence multiple outcomes (e.g., acoustic measures in phonetics), researchers often analyze each outcome separately using a univariate approach. However, this approach ignores relationships between outcomes, which can reduce estimate accuracy and make it difficult to examine how effects are related across outcomes. A multivariate approach addresses these issues by modeling all outcomes jointly. This tutorial illustrates how to fit Bayesian multivariate linear mixed-effects models using the R package brms . Method: We present three example applications in phonetic research using a corpus of Greek utterances. We focus on a rising accent, that is, a deliberate fundamental frequency movement temporally aligned with a word's stressed syllable and used to highlight that word in speech. Specifically, we examine whether the phonetic characteristics of the rising accent depend on two factors: (a) the presence of a preceding accent within the same utterance and (b) the location of the stress in the accented word relative to its final syllable. Results: The multivariate approach reduced uncertainty in population-level effect estimates compared to univariate models and provided a convenient way to examine correlations among effects across outcomes. Conclusion: This tutorial provides guidance on implementing Bayesian multivariate linear mixed-effects models and demonstrates their potential.

Purpose: When studying how factors influence multiple outcomes (e.g., acoustic measures in phonetics), researchers often analyze each outcome separately using a univariate approach. However, this approach ignores relationships between outcomes, which can reduce estimate accuracy and make it difficult to examine how effects are related across outcomes. A multivariate approach addresses these issues by modelling all outcomes jointly. This tutorial illustrates how to fit Bayesian multivariate linear mixed effects models using the R package brms.Method: We present three example applications in phonetic research, using a corpus of Greek utterances. We focus on a rising accent, that is, a deliberate f_0 movement temporally aligned with a word’s stressed syllable and used to highlight that word in speech. Specifically, we examine whether the phonetic characteristics of the rising accent depend on two factors: (a) the presence of a preceding accent within the same utterance, and (b) the location of the stress in the accented word relative to its final syllable.Results: The multivariate approach reduced uncertainty in population-level effect estimates compared to univariate models and provided a convenient way to examine correlations among effects across outcomes.Conclusions: This tutorial provides guidance on implementing Bayesian multivariate linear mixed effects models and demonstrates their potential.

Greek uses H*, L + H*, and H* + L, all followed by L-L% edge tones, as nuclear pitch accents in statements. A previous analysis demonstrated that these accents are distinguished by F0 scaling and contour shape. This study expands the earlier investigation by exploring additional cues, namely, voice quality, amplitude, and duration, in distinguishing the pitch accents, and investigating individual variability in the selection of both F0 and non-F0 cues. Bayesian multivariate analysis and hierarchical clustering demonstrate that the accents are distinguished not only by F0 but also by additional cues at the group level, with individual variability in cue selection.

Previous analysis of Greek H*, L + H*, and H* + L used FPCA on the f0 contours of these accents (N = 844; 13 speakers), and showed that they are distinguished by their PC1 and PC2 scores: PC1 captures differences in pitch height (H* < L + H* < H* + L) and PC2 differences in accent shape. We extend this analysis by investigating the use of voice quality (H1-H2) and duration in distinguishing the accents, with a focus on individual variation in the use of these cues. Results obtained from Bayesian multivariate analysis demonstrate considerable variability in the use of voice quality and duration in distinguishing between L + H* and H*, as compared to H* + L and H*. Specifically, all speakers used F0 to distinguish H* + L and L + H* from H*. In addition, 7/13 speakers employed a breathier voice to distinguish H* + L from H*, while the use of duration as a cue was relatively infrequent. In contrast, 8/13 speakers employed voicing and 7/13 employed duration to distinguish L + H* from H*, with most speakers using at least one parameter in addition to F0 to make this accentual distinction. These results highlight the importance of investigating non-F0 cues used in encoding intonation categories, as well as individual variation in such use.

The autosegmental-metrical (AM) theory of intonational phonology posits that tunes are composed of independent elements, pitch accents and edge tones. However, configurational approaches remain popular, while AM relies on but cannot prove compositionality. We addressed this issue with a corpus of Greek wh-questions (N = 2135), elicited from 18 speakers using a discourse completion task that involved communicative situations resulting in rise-fall-rises or rise-falls. The rise-fall-rises are autosegmentally represented as a L* + H pitch accent on the utterance-initial wh-word, followed by a L- phrase accent and a H% boundary tone; the rise-falls are represented as L + H* L − L%. We applied functional principal component analysis (FPCA), a data-driven method that breaks down curves into components that capture independent modes of curve variation, and followed FPCA with LMEMs on the principal component coefficients. The pitch movement associated with each of the posited tonal elements was captured by a different PC: PC2 captured differences in the extent of the rise and the peak alignment of the pitch accent, PC1 differences in fall steepness, and PC4 the difference between a final rise and low, flat pitch. These results provide prima facie evidence for tune compositionality, since the differences are driven by the analysis not human annotation.

Early Spanish-English bilinguals and English controls were tested on the production and perception of negative, short-lag, and long-lag Voice Onset Time (VOT), VOT types spanning the Spanish and English phonetic categories: phonologically, negative and short-lag VOT stops are distinct phonemes in Spanish, but realizations of voiced stops in English. Dominance was critical: more English-dominant bilinguals produced more short-lag VOT stops in response to negative VOT stimuli, and were also less accurate than more balanced bilinguals at discriminating negative from short-lag VOT. Bilinguals performed similarly to monolinguals overall, but they produced more negative VOT tokens and shorter short-lag VOT in response to negative VOT. Their productions were also less well correlated with perception and showed more variation between individuals. These results highlight the variable nature of bilingual production and perception, and demonstrate the need to consider language dominance, individual variation, as well as modalities and tasks when studying bilinguals.