Cong Zhang

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.

Calculates and visualises speech rhythm and timing metrics. The 'rhythm.metrics' package provides a standardised workflow to compute common metrics including Delta C, Delta V, VarcoC, VarcoV, the percentage of vocalic intervals (%V), and both raw and normalised Pairwise Variability Indices (rPVI, nPVI). It includes functions for calculating and visualising these measures to facilitate cross-linguistic and developmental rhythm research. Delta C, Delta V, and %V measures are based on Ramus et al. (1999) ; VarcoC and VarcoV measures are based on Dellwo (2006, ISBN: 9783631554777); and rPVI-C and nPVI-V are based on Grabe & Low (2002) .

This paper explores the integration of human linguistic insights into multilingual text-to-speech (TTS) systems by evaluating the Featurally Underspecified Lexicon (FUL) as a theory-driven input representation. Unlike data-intensive end-to end models, FUL offers a compact, interpretable feature set grounded in phonological principles, enabling scalable and equitable TTS development for low-resource languages. We provide a mapping from language-specific phones to FUL feature vectors via a SAMPA intermediate and incorporate these features into a modified FastSpeech architecture. Experiments were conducted to evaluate their ability to generate native, non-native, and code-mixed speech in English and Mandarin. We ran an experiment with a small dataset and one with a larger dataset, which showed that TTS with FUL features as input could produce intelligible native speech with as little as 8 h of training data; with 100 h of training data, intelligible speech could be generated for a language not present in the training data. The approach further supports code-mixed synthesis while preserving consistent timbre and interpretable phonetic control. These results highlight the potential of theory-driven representations for building efficient, scalable, and linguistically informed TTS systems, demonstrating that phonological features can function as both analytical tools and practical inputs for speech technology.

Gamification has been increasingly adopted across a range of domains, including academic research, thanks to its inclusive and scalable nature. Gamified tasks involve the incorporation of game elements such as points, levels, and rewards within non-game contexts to enhance participant’s engagement and motivation. Despite its online accessibility and motivating nature, however, there is no agreed protocol regarding which gamified elements to include in what type of tasks, especially in linguistic research which is mostly text-based. Rather than proposing a one-size-fits-all solution, this paper demonstrates the viability of audio-based gamified tasks, testing whether listeners from the North East of England, a region in which many residents believe there are fine-grained dialectal distinctions, can accurately distinguish speakers by location. In comparison to traditional identification tasks, our tasks were gamified through the inclusion of elements such as a points system, a post-task leaderboard, and rewards for high-scoring participants. Results indicate that within–North East speaker identification was a more difficult task for listeners than folk belief would indicate. We suggest that despite its limitations, gamification successfully replaced traditional per-participant incentives as a data collection strategy in a phonetic perception study.

Purpose: We explored the prosody and voice quality of people with latent aphasia, a mild form of aphasia, against a neurotypical control group. In contrast to people with moderate to severe aphasia, people with latent aphasia can have similar speech production compared to neurotypical controls. Method: We analysed three prosodic measures (fundamental frequency [f0], intensity, and duration) and three voice quality measures (local jitter, local shimmer, and harmonic-to-noise ratio) of utterance-initial and utterance-final words from ten speakers with latent aphasia and ten matched controls. Three relevant linguistic features (syllable number, part of speech, and word type [content or function words]) were also considered. Linear mixed-effects models assessed the effects of word position and speaker group on prosodic features.Results: Both groups showed parallel patterns in f0 and intensity: Utterance-final words were produced with higher f0 and lower intensity than utterance-initial words. Significant group-by-position interactions indicated that these measures were reduced in the latent aphasia group, suggesting subtle but systematic prosodic distinctions. Word duration further differentiated the two groups, with the latent group consistently showing longer duration than controls. Voice quality measures showed minimal group differences, consistent with the latent group being clinically classified as “normal.” Nonetheless, these measures provide baseline values for spontaneous speech, which diverge from cut-offs based on sustained vowels.Conclusions: Prosodic features are sensitive indicators of latent aphasia, offering an explanation for why predictive models using prosody alone can successfully distinguish affected individuals from controls. Voice quality measures, while less informative for group separation, contribute novel data for this population. Together, the results underscore the importance of prosodic research for identifying subtle pathological differences and support future clinical and computational approaches to detecting clinical language disorders.

The precise nature of the prosodic contribution to disambiguating open and polar questions with indefinite content pro-forms in Korean (e.g., nwukwu “who/someone”) remains a matter of debate. One possible relevant prosodic feature is expanded F0 range at the site of question focus. We report the results of a pilot experiment followed by a large-scale online speech perception gating study ( n = 124 ) that manipulated the natural prosody of identical strings read as statements or open or polar questions. We found a tendency for all questions to be perceived as open questions, regardless of prosody. Open questions were reliably disambiguated from other utterance types, and there was no effect of the prosodic manipulation. Polar questions did show a significant effect of the prosodic manipulation ( p < . 001 ) , but even with full natural prosody, these stimuli were never correctly identified above chance levels. We suggest that in the absence of context, there is a strong preference for hearers to interpret content pro-forms as question words, and that subsequent prosodic information may be discounted if a hearer has already committed to their interpretation. The present findings have implications for formal accounts of the Korean syntax-prosody interface.

The study of citation tones, lexical tones produced in isolation, is one of the first steps towards understanding speech prosody in tone languages. However, methodologies for investigating citation tones vary significantly, often leading to limited comparability of tone inventories, both within and across languages. This paper presents a systematic review of research methods and practices in 136 citation tone studies on 129 tonal language varieties in China, including 99 studies published in Chinese, which are therefore not easily available to an international scientific readership. The review provides an overview of possible analytical decisions along the research pipeline, and unveils considerable variation in data collection, analysis, and reporting conventions, particularly in how f0, the primary acoustic correlate for tone, is operationalised and reported across studies. Key methodological issues are identified, including small sample sizes and inadequate transparency in communicating methodological decisions and procedure. This paper offers a clear road map for citation tone production research and proposes a range of recommendations on speaker sampling, experimental design, acoustic processing techniques, f0 analysis, and result reporting, with the goal of facilitating future tonal research and enhancing resources for underrepresented tonal varieties.

This study addresses the TAUKADIAL challenge, focusing on the classification of speech from people with Mild Cognitive Impairment (MCI) and neurotypical controls. We conducted three experiments comparing five machine-learning methods: Random Forests, Sparse Logistic Regression, k-Nearest Neighbors, Sparse Support Vector Machine, and Decision Tree, utilizing 1076 acoustic features automatically extracted using openSMILE. In Experiment 1, the entire dataset was used to train a language-agnostic model. Experiment 2 introduced a language detection step, leading to separate model training for each language. Experiment 3 further enhanced the language-agnostic model from Experiment 1, with a specific focus on evaluating the robustness of the models using out-of-sample test data. Across all three experiments, results consistently favored models capable of handling high-dimensional data, such as Random Forest and Sparse Logistic Regression, in classifying speech from MCI and controls.

Increasingly, phonetic research uses data collected from participants who record themselves on readily available devices. Though such recordings are convenient, their suitability for acoustic analysis remains an open question, especially regarding how recording methods affect acoustic measures over time. We used Quantile Generalized Additive Mixed Models (QGAMMs) to analyze measures of F0, intensity, and the first and second formants, comparing files recorded using a laboratory-standard recording method (Zoom H6 recorder with an external microphone), to three remote recording methods: (1) the Awesome Voice Recorder application on a smartphone (AVR), (2) the Zoom meeting application with default settings (Zoom-default), and (3) the Zoom meeting application with the “Turn on Original Sound” setting (Zoom-raw). A linear temporal alignment issue was observed for the Zoom methods over the course of the long, recording session files; however, the difference was not significant for utterance-length files. F0 was reliably measured using all methods. Intensity and formants presented non-linear differences across methods that could not be corrected for simply. Overall, the AVR files were most similar to the H6’s, and so AVR is deemed to be a more reliable recording method than either Zoom-default or Zoom-raw.

Recent empirical studies have highlighted the large degree of analytic flexibility in data analysis that can lead to substantially different conclusions based on the same data set. Thus, researchers have expressed their concerns that these researcher degrees of freedom might facilitate bias and can lead to claims that do not stand the test of time. Even greater flexibility is to be expected in fields in which the primary data lend themselves to a variety of possible operationalizations. The multidimensional, temporally extended nature of speech constitutes an ideal testing ground for assessing the variability in analytic approaches, which derives not only from aspects of statistical modeling but also from decisions regarding the quantification of the measured behavior. In this study, we gave the same speech-production data set to 46 teams of researchers and asked them to answer the same research question, resulting in substantial variability in reported effect sizes and their interpretation. Using Bayesian meta-analytic tools, we further found little to no evidence that the observed variability can be explained by analysts’ prior beliefs, expertise, or the perceived quality of their analyses. In light of this idiosyncratic variability, we recommend that researchers more transparently share details of their analysis, strengthen the link between theoretical construct and quantitative system, and calibrate their (un)certainty in their conclusions.