Search

This research explores the potential of machine learning (ML) in identifying vocal markers for schizophrenia. While previous research showed that voice-based ML models can accurately predict schizophrenia diagnosis and symptoms, it is unclear to what extent such ML markers generalize to different clinical subpopulations and languages: the assessment of generalization performance is however crucial for testing their clinical applicability.We systematically examined voice-based ML model performance on a large cross-linguistic dataset (3 languages: Danish, German, Chinese). Employing a rigorous pipeline to minimize overfitting, including cross-validated training sets and multilingual models, we assessed generalization on participants with schizophrenia and controls speaking the same or different languages. Model performance was comparable to state-of-the art findings (F1-score ~ 0.75) within the same language; however, models did not generalize well - showing a substantial decrease - when tested on new languages, and the performance of multilingual models was also generally low (F1-score ~ 0.50).

The current study investigated whether the difficulty in finding group differences in prosody between speakers with autism spectrum disorder (ASD) and neurotypical (NT) speakers might be explained by identifying different acoustic profiles of speakers which, while still perceived as atypical, might be characterized by different acoustic qualities. We modelled the speech from a selection of speakers (N = 26), with and without ASD, as a network of nodes defined by acoustic features. We used a community-detection algorithm to identify clusters of speakers who were acoustically similar and compared these clusters with atypicality ratings by naïve and expert human raters. Results identified three clusters: one primarily composed of speakers with ASD, one of mostly NT speakers, and one comprised of an even mixture of ASD and NT speakers. The human raters were highly reliable at distinguishing speakers with and without ASD, regardless of which cluster the speaker was in. These results suggest that community-detection methods using a network approach may complement commonly-employed human ratings to improve our understanding of the intonation profiles in ASD.

Capacities for abstract thinking and problem-solving are central to human cognition. Processes of abstraction allow the transfer of experiences and knowledge between contexts helping us make informed decisions in new or changing contexts. While we are often inclined to relate such reasoning capacities to individual minds and brains, they may in fact be contingent on human-specific modes of collaboration, dialogue, and shared attention. In an experimental study, we test the hypothesis that social interaction enhances cognitive processes of rule-induction, which in turn improves problem-solving performance. Through three sessions of increasing complexity, individuals and groups were presented with a problem-solving task requiring them to categorize a set of visual stimuli. To assess the character of participants' problem representations, after each training session, they were presented with a transfer task involving stimuli that differed in appearance, but shared relations among features with the training set. Besides, we compared participants' categorization behaviors to simulated agents relying on exemplar learning. We found that groups performed superior to individuals and agents in the training sessions and were more likely to correctly generalize their observations in the transfer phase, especially in the high complexity session, suggesting that groups more effectively induced underlying categorization rules from the stimuli than individuals and agents. Crucially, variation in performance among groups was predicted by semantic diversity in members' dialogical contributions, suggesting a link between social interaction, cognitive diversity, and abstraction.

Background To date, evidence on late-onset immune-related adverse events (irAEs) with immune checkpoint inhibitors (ICIs) is limited to a small number of clinical cases. This study aimed to identify drug- and patient-related characteristics potentially associated with the reporting of late-onset irAEs with ICIs in VigiBase, the WHO global database of individual case safety reports (ICSRs).Methods Observational study comparing deduplicated ICSRs with ICIs reporting late-onset irAEs (occurred >90 days after ICI discontinuation) versus ICSRs with ICIs not reporting late-onset irAEs, collected in VigiBase from 2011 to December 31, 2022. Logistic regression was used to model the relationship between drug-related and patient-related characteristics of ICSRs and the reporting of late-onset irAEs. Significance was determined for variables with the lower bound of the 95% CI of the reporting OR (ROR) higher than 1 and a p value

When speaking to infants, adults often produce speech that differs systematically from that directed to other adults. To quantify the acoustic properties of this speech style across a wide variety of languages and cultures, we extracted results from empirical studies on the acoustic features of infant-directed speech. We analysed data from 88 unique studies (734 effect sizes) on the following five acoustic parameters that have been systematically examined in the literature: fundamental frequency (f0), f0 variability, vowel space area, articulation rate and vowel duration. Moderator analyses were conducted in hierarchical Bayesian robust regression models to examine how these features change with infant age and differ across languages, experimental tasks and recording environments. The moderator analyses indicated that f0, articulation rate and vowel duration became more similar to adult-directed speech over time, whereas f0 variability and vowel space area exhibited stability throughout development. These results point the way for future research to disentangle different accounts of the functions and learnability of infant-directed speech by conducting theory-driven comparisons among different languages and using computational models to formulate testable predictions.

Engraved ochres and ostrich eggshells from the South African Blombos Cave and Diepkloof Rock Shelter are among the earliest expressions of human symbolic behavior. Furthermore, they appear to document a continuous practice of abstract mark-making across ~40.000 years. During this time, the engraved patterns change from simpler unstructured patterns to complex, ordered and symmetric cross-hatchings. To inform discussions of the possible function of the engravings, we conducted a two-part experimental study. Based on the assumption that the pragmatic use of an artifact will motivate incremental adaptive refinements, we used transmission chain experiments to reconstruct the original trajectory of changes. We then conducted five experiments to assess the cognitive implications of changes to the patterns and compared these to the original engravings. Although we observe interesting similarities, our findings suggest the Blombos and Diepkloof engravings are not only a product of human cognitive biases and constraints on working memory.

This study compared the acoustic properties of 26 (100% female, 100% monolingual) Danish caregivers' spontaneous speech addressed to their 11- to 24-month-old infants (infant-directed speech, IDS) and an adult experimenter (adult-directed speech, ADS). The data were collected between 2016 and 2018 in Aarhus, Denmark. Prosodic properties of Danish IDS conformed to cross-linguistic patterns, with a higher pitch, greater pitch variability, and slower articulation rate than ADS. However, an acoustic analysis of vocalic properties revealed that Danish IDS had a reduced or similar vowel space, higher within-vowel variability, raised formants, and lower degree of vowel discriminability compared to ADS. None of the measures, except articulation rate, showed age-related differences. These results push for future research to conduct theory-driven comparisons across languages with distinct phonological systems.

Linguistic repetitions in children are conceptualized as negative in children with autism -- echolalia, without communicative purpose -- and positive in typically developing (TD) children -- linguistic alignment involved in shared engagement, common ground and language acquisition. To investigate this apparent contradiction we analyzed spontaneous speech in 67 parent-child dyads from a longitudinal corpus (30 minutes of play activities at 6 visits over 2 years). We included 32 children with autism and 35 linguistically matched TD children (mean age at recruitment 32.76 and 20.27 months). We found a small number of exact repetitions in both groups (roughly 1% of utterances across visits), which increased over time in children with autism and decreased in the TD group. Partial repetitions were much more frequent: children reused caregivers' words at high rates regardless of diagnostic group (24% of utterances at first visit), and this increased in frequency (but not level) over time, faster for TD children (at final visit: 33% for autism, 40% for TD). The same happened for partial repetition of syntax and semantic alignment. However, chance alignment (as measured by surrogate pairs) also increased and findings for developmental changes were reliable only for syntactic and semantic alignment. Children with richer linguistic abilities also displayed a higher tendency to partially re-use their caregivers' language (alignment rates and semantic alignment). This highlights that all children commonly re-used the words, syntax, and topics of their caregivers, albeit with some quantitative differences, and that most repetition was at least potentially productive, with repeated language being re-contextualized and integrated with non-repeated language. The salience of echolalia in ASD might be partially explained by slight differences in frequency, amplified by lower semantic alignment, persistence over time, and expectations of echolalia. More in-depth qualitative and quantitative analyses of how repetitions are used and received in context are needed.

Voice atypicalities are a characteristic feature of schizophrenia, often associated with core negative symptoms. A recent meta-analysis identified atypicalities in pitch, speech rate, and pauses. However, heterogeneity across studies was large and replications almost nonexistent. Further, it is not clear whether vocal patterns are directly related to the mechanisms underlying the disorder and could therefore be found across languages, or not.In this study we implemented a more rigorously cumulative scientific approach by collecting and analyzing a large cross-linguistic corpus of voice recordings. We critically employed meta-analytic priors to systematically assess the replicability of previous findings, and modeled between-participants variability and cross-linguistic differences.We replicate previous meta-analytic findings across all languages for reduced pitch variability, while increased pause duration and lower speech rate results were replicated only in some languages. Most atypical voice patterns, thus, seem not to be distinctive of schizophrenia in general, but more specifically situated in linguistic/cultural differences.

Fluent conversation requires temporal organization between conversational exchanges. By performing a systematic review and Bayesian multi-level meta-analysis, we map the trajectory of infants' turn-taking abilities over the course of early development (0 to 70 months). We synthesize the evidence from 26 studies (78 estimates from 429 unique infants, of which at least 152 are female) reporting response latencies in infant-adult dyadic interactions. The data were collected between 1975 and 2019, exclusively in North America and Europe. Infants took on average circa 1 s to respond, and the evidence of changes in response over time was inconclusive. Infants' response latencies are related to those of their adult conversational partners: an increase of 1 s in adult response latency (e.g., 400 to 1400 ms) would be related to an increase of over 1 s in infant response latency (from 600 to 1857 ms). These results highlight the dynamic reciprocity involved in the temporal organization of turn-taking. Based on these results, we provide recommendations for future avenues of enquiry: studies should analyze how turn-by-turn exchanges develop on a longitudinal timescale, with rich assessment of infants' linguistic and social development.

Safety data on MET inhibitors in patients with advanced NSCLC harboring MET exon 14 mutation and treated with frontline immune checkpoint inhibitors (ICIs) are still limited. Here, we describe clinical characteristics, liver biopsy features, and management of liver injury of two patients with a diagnosis of MET exon 14-mutant NSCLC receiving capmatinib after ICI failure. On the basis of histologic findings and exclusion of other potential causes, a diagnosis of drug-induced liver injury (DILI) associated with portal fibrosis was made in both cases. The use of hepatoprotective drugs, in addition to oral ursodeoxycholic acid, resulted in liver blood tests normalization. To provide a global safety perspective, we queried the Food and Drug Administration Adverse Event Reporting System and detected a robust disproportionality signal. Out of the 918 total reports with capmatinib from the Food and Drug Administration Adverse Event Reporting System database, DILI was recorded in 43 cases (4.7%), mostly serious (93.0%) with hospitalization and death recorded in 25.6% and 16.3% of the cases, respectively. The median time to onset was 42 days, with discontinuation and positive dechallenge documented in 41.9% and 39.5% of the cases, respectively. Anti–programmed cell death protein-1 agents were coreported in 11 DILI cases. Only two cases of DILI out of 105 reports were found for tepotinib. Our data support a potential association between capmatinib and DILI in patients who have also been previously exposed to immunotherapy. Considering the potential implications for sequence strategy and timing of ICI and MET inhibitor, further investigation is warranted.

Theories of vocal signalling in humans typically only consider communication within the interactive group and ignore intergroup dynamics. Recent work has found that colaughter generated between pairs of people in conversation can afford accurate judgements of affiliation across widely disparate cultures, and the acoustic features that listeners use to make these judgements are linked to speaker arousal. But to what extent does colaughter inform third party listeners beyond other dynamic information between interlocutors such as overlapping talk? We presented listeners with short segments (1-3 s) of colaughter and simultaneous speech (i.e. cospeech) taken from natural conversations between established friends and newly acquainted strangers. Participants judged whether the pairs of interactants in the segments were friends or strangers. Colaughter afforded more accurate judgements of affiliation than did cospeech, despite cospeech being over twice in duration relative to colaughter on average. Sped-up versions of colaughter and cospeech (proxies of speaker arousal) did not improve accuracy for either identifying friends or strangers, but faster versions of both modes increased the likelihood of tokens being judged as being between friends. Overall, results are consistent with research showing that laughter is well suited to transmit rich information about social relationships to third party overhearers-a signal that works between, and not just within conversational groups.

Background: Although endoscopic ultrasound-guided gallbladder drainage (EUS-GBD) using lumen-apposing metal stents (LAMS) has become one of the treatments of choice for acute cholecystitis (AC) in fragile patients, scant data are available on real-life settings and long-term outcomes. Methods: We performed a multicenter retrospective study including EUS-guided GBD using LAMS for AC in 19 Italian centers from June 2014 to July 2020. The primary outcomes were technical and clinical success, and the secondary outcomes were the rate of adverse events (AE) and long-term follow-up. Results: In total, 116 patients (48.3% female) were included, with a mean age of 82.7 ± 11 years. LAMS were placed, transgastric in 44.8% of cases, transduodenal in 53.3% and transjejunal in 1.7%, in patients with altered anatomy. Technical success was achieved in 94% and clinical success in 87.1% of cases. The mean follow-up was 309 days. AEs occurred in 12/116 pts (10.3%); 8/12 were intraprocedural, while 1 was classified as early (15 days). According to the ASGE lexicon, two (16.7%) were mild, three (25%) were moderate, and seven (58.3%) were severe. No fatal AEs occurred. In subgroup analysis of 40 patients with a follow-up longer than one year, no recurrence of AC was observed. Conclusions: EUS-GBD had high technical and clinical success rates, despite the non-negligible rate of AEs, thus representing an effective treatment option for fragile patients.

Due to their aspecific macroscopic appearance, uncommon pancreatic cystic lesions (PCLs) are often misdiagnosed as mucinous lesions and improperly resected. We aimed to evaluate the endoscopic ultrasound (EUS)-guided through-the-needle biopsy (TTNB) capacity of the preoperative diagnosis of uncommon PCLs. Overall, 136 patients with PCLs who underwent EUS-TTNB between 2016 and 2022 were retrospectively identified. Common histotypes (e.g., IPMN, serous cystadenoma, and mucinous cystadenoma) were excluded and 26 (19.1%) patients (15 female, mean age 52.9 ± 10.4) were analyzed. The EUS findings, adverse events (AEs), and TTNB outcomes in uncommon PCLs were evaluated. The cysts histotype was accurately diagnosed by TTNB in 24/26 (92.3%) cases (seven cystic neuroendocrine tumors, four squamoid cysts, three acinar cells cystadenomas, two lymphoepithelial cysts, two mucinous non-neoplastic cysts, two bronchogenic cysts, two cystic lymphangiomas, one solid-pseudopapillary neoplasm, and one schwannoma). In the remaining two cases, lymphangioma was eventually diagnosed after resection. Surgery was performed in 15/26 (57.7%) patients. The mean follow-up of non-surgical patients was 32.5 months. One severe acute case of pancreatitis (3.8%) that required surgery occurred after EUS-TTNB. Uncommon pancreatic/peripancreatic lesions represent the 19.1% of PCLs in our series, with mainly benign histotypes. TTNB demonstrated a high diagnostic performance with a low rate of AEs in this setting, representing a reliable tool with which to avoid useless surgery.

Speech input is often noisy and ambiguous. Yet listenersusually do not have difficulties understanding it. A keyhypothesis is that in speech processing acoustic-phoneticbottom-up processing is complemented by top-downcontextual information. This context effect is larger when theambiguous word is only separated from a disambiguating wordby a few syllables compared to many syllables, suggesting thatthere is a limited time window for processing acoustic-phoneticinformation with the help of context. Here, we argue that therelative weight of bottom-up and top-down processes may bedifferent for languages that have different phonologicalproperties. We report an experiment comparing two closelyrelated languages, Danish and Norwegian. We show thatDanish speakers do indeed rely on context more thanNorwegian speakers do. These results highlight the importanceof investigating cross-linguistic differences in speechprocessing, suggesting that speakers of different languagesmay develop different language processing strategies.

Language processing depends on the integration of bottom-upinformation with top-down cues from several differentsources—primarily our knowledge of the real world, ofdiscourse contexts, and of how language works. Previousstudies have shown that factors pertaining to both the senderand the receiver of the message affect the relative weighting ofsuch information. Here, we suggest another factor that maychange our processing strategies: perceptual noise. Wehypothesize that listeners weight different sources of top-downinformation more in situations of perceptual noise than innoise-free situations. Using a sentence-picture matchingexperiment with four forced-choice alternatives, we show thatdegrading the speech input with noise compels the listeners torely more on top-down information in processing. We discussour results in light of previous findings in the literature,highlighting the need for a unified model of spoken languagecomprehension in different ecologically valid situations,including under noisy conditions.

Do people adjust their conversational strategies to the specificcontextual demands of a given situation? Prior studies haveyielded conflicting results, making it unclear how strategiesvary with demands. We combine insights from qualitative andquantitative approaches in a within-participant experimentaldesign involving two different contexts: spontaneouslyoccurring conversations (SOC) and task-oriented conversations(TOC). We systematically assess backchanneling, other-initiated repair and linguistic alignment. We find that SOCexhibit a higher number of backchannels, a reduced and moregeneric repair format and higher rates of lexical and syntacticalignment. TOC are characterized by a high number of specificrepairs and a lower rate of lexical and syntactic alignment.However, when alignment occurs, more linguistic forms arealigned. The findings show that conversational strategies adaptto contextual demands.

Introduction Machine learning (ML) approaches are a promising venue for identifying vocal markers of neuropsychiatric disorders, such as schizophrenia. While recent studies have shown that voice-based ML models can reliably predict diagnosis and clinical symptoms of schizophrenia, it is unclear to what extent such ML markers generalize to new speech samples collected using a different task or in a different language: the assessment of generalization performance is however crucial for testing their clinical applicability. Objectives In this research, we systematically assessed the generalizability of ML models across contexts and languages relying on a large cross-linguistic dataset of audio recordings of patients with schizophrenia and controls. Methods We trained ML models of vocal markers of schizophrenia on a large cross-linguistic dataset of audio recordings of 231 patients with schizophrenia and 238 matched controls (>4.000 recordings in Danish, German, Mandarin and Japanese). We developed a rigorous pipeline to minimize overfitting, including cross-validated training set and Mixture of Experts (MoE) models. We tested the generalizability of the ML models on: (i) different participants, speaking the same language (hold-out test set); (ii) different participants, speaking a different language. Finally, we compared the predictive performance of: (i) models trained on a single language (e.g., Danish) (ii) MoE models, i.e., ensemble of models (experts) trained on a single language whose predictions are combined using a weighted sum (iii) multi-language models trained on multiple languages (e.g., Danish and German). Results Model performance was comparable to state-of-the art findings (F1: 70%-80%) when trained and tested on participants speaking the same language (out-of-sample performance). Crucially, however, the ML models did not generalize well - showing a substantial decrease of performance (close to chance) - when trained in a language and tested on new languages (e.g., trained on Danish and tested on German). MoE and multi-language models showed a better increase of performance (F1: 55%-60%), but still far from those requested for achieving clinical applicability. Conclusions Our results show that the cross-linguistic generalizability of ML models of vocal markers of schizophrenia is very limited. This is an issue if our first goal is to translate these vocal markers into effective clinical applications. We argue that more emphasis needs to be placed on collecting large open datasets to test the generalizability of voice-based ML models, for example, across different speech tasks or across the heterogeneous clinical profiles that characterize schizophrenia spectrum disorder. Disclosure of Interest None Declared

Purpose: The aim of the study was to use systematic review and meta-analysis to quantitatively assess the currently available acoustic evidence for prosodic production impairments as a result of right-hemisphere damage (RHD), as well as to develop methodological recommendations for future studies. Method: We systematically reviewed papers reporting acoustic features of prosodic production in RHD in order to identify shortcomings in the literature and make recommendations for future studies. We estimated the meta-analytic effect size of the acoustic features. We extracted standardized mean differences from 16 papers and estimated aggregated effect sizes using hierarchical Bayesian regression models. Results: RHD did present reduced fundamental frequency variation, but the trait was shared with left-hemisphere damage. RHD also presented evidence for increased pause duration. No meta-analytic evidence for an effect of prosody type (emotional vs. linguistic) was found. Conclusions: Taken together, the currently available acoustic data show only a weak specific effect of RHD on prosody production. However, the results are not definitive, as more reliable analyses are hindered by small sample sizes, lack of detail on lesion location, and divergent measuring techniques. We propose recommendations to overcome these issues: Cumulative science practices (e.g., open data and code sharing), more nuanced speech signal processing techniques, and the integration of acoustic measures and perceptual judgments are recommended to more effectively investigate prosody in RHD.

Laughter is a nonverbal vocalization occurring in every known culture, ubiquitous across all forms of human social interaction. Here, we examined whether listeners around the world, irrespective of their own native language and culture, can distinguish between spontaneous laughter and volitional laughter-laugh types likely generated by different vocal-production systems. Using a set of 36 recorded laughs produced by female English speakers in tests involving 884 participants from 21 societies across six regions of the world, we asked listeners to determine whether each laugh was real or fake, and listeners differentiated between the two laugh types with an accuracy of 56% to 69%. Acoustic analysis revealed that sound features associated with arousal in vocal production predicted listeners' judgments fairly uniformly across societies. These results demonstrate high consistency across cultures in laughter judgments, underscoring the potential importance of nonverbal vocal communicative phenomena in human affiliation and cooperation.

How do solution strategies spread in teams? Inspired by the 80s sci-fi movie Close encounters of the third kind, weset up a networked multiplayer game where participants had to signal peace to invading aliens from space by playingmusic. In each round of the game, participants were matched in dyads and through chat had to jointly construct a four-tone melody. Melodies translated to points according to a rugged landscape score system. We compared three networktopologies: a lattice network (participants only play with immediate neighbours), a fully connected network, and dyads.Furthermore, we manipulated the nature of the problem being either open-ended or with fixed solutions by making themaximum possible score known or not. With known maximum score, lattice networks show an advantage with increasedpropensity to explore and diffusion of good solutions, compared to fully connected networks. No effects are observedwhen the maximum score is unknown.

Abstraction lies at the heart of human cognition. While most approaches to abstraction implicitly take the individual as astarting point, we hypothesize abstraction to be contingent on the interactive sharing of diverse perspectives. Interactivealignment, however, can reduce diversity making group members’ contributions more similar and redundant, especially ifthey have not had time to form their own impressions and opinions. We report an experiment investigating the conditionsunder which participants arrive at a superior, abstract rule-based solution to a problem: inferring the direction of the lastgear from the rotation of the first in a series of connected gears. Participants were assigned to three different conditions:1) individual, 2) dyadic, 3) combined: dyad members start individually but are joint mid through the experiment. We findthat performance is significantly higher in the dyadic than in the individual condition, but highest in the combined.