Diffusion Large Language Models (DLLMs) are inherently ill-suited for variable-length generation, as their inference is defined on a fixed-length canvas and implicitly assumes a known target length. When the length is unknown, as in realistic completion and infilling, naively comparing confidence across mask lengths becomes systematically biased, leading to under-generation or redundant continuations. In this paper, we show that this failure arises from an intrinsic lengthinduced bias in generation confidence estimates, leaving existing DLLMs without a robust way to determine generation length and making variablelength inference unreliable. To address this issue, we propose LR-DLLM, a length-regularized inference framework for DLLMs that treats generation length as an explicit variable and achieves reliable length determination at inference time. It decouples semantic compatibility from lengthinduced uncertainty through an explicit length regularization that corrects biased confidence estimates. Based on this, LR-DLLM enables dynamic expansion or contraction of the generation span without modifying the underlying DLLM or its training procedure. Experiments show that LRDLLM achieves 51.3% Pass@1 on HumanEvalInfilling under fully unknown lengths (+13.4% vs. DreamOn) and 51.5% average Pass@1 on four-language McEval (+14.3% vs. DreamOn).
We introduce the Mandarin-English Language Interview (MELI) Corpus, an open-source resource of 29.8 hours of speech from 51 Mandarin-English bilingual speakers. MELI combines matched sessions in Mandarin and English with two speaking styles: read sentences and spontaneous interviews about language varieties, standardness, and learning experiences. Audio was recorded at 44.1 kHz (16-bit, stereo). Interviews were fully transcribed, force-aligned at word and phone levels, and anonymized. Descriptively, the Mandarin component totals ~14.7 hours (mean duration 17.3 minutes) and the English component ~15.1 hours (mean duration 17.8 minutes). We report token/type statistics for each language and document code-switching patterns (frequent in Mandarin sessions; more limited in English sessions). The corpus design supports within-/cross-speaker, within/cross-language acoustic comparison and links acoustics to speakers' stated language attitudes, enabling both quantitative and qualitative analyses. The MELI Corpus will be released with transcriptions, alignments, metadata, scans of labelled maps and documentation under a CC BY-NC 4.0 license.
In recent years, the use of technology has become an important part of language learning and the development and improvement of learner autonomy and motivation. Moreover, it has been found that learning a third language is easier for those who are already proficient in a second language, as they have developed linguistic and cognitive skills that facilitate the acquisition of new languages. Furthermore, if the student has theoretical training in language acquisition, their learning process is expected to be even more effective. This study explores how individuals with a background in language acquisition and teaching make use of technology, specifically the language learning app Duolingo, to learn a third language. Following a qualitative approach, the perceptions and practices of 19 participants with a Bachelor’s, Master’s, or Doctorate degrees in language acquisition, teaching, or similar areas were explored. Data were collected through questionnaires and interviews and were analyzed using the Constant Comparative Method. Findings suggest that, while Duolingo promotes autonomy, those who complemented their learning with real-life interactions made more progress than those who relied solely on the app. Thus, technology is a valuable tool, but it should be combined with other strategies to improve the acquisition of a third language.
Adisai Na-Thalang, Chanakan Wittayasakpan, Kritsadha Phatcharoen
et al.
This paper introduces the development of the first open conversational speech dataset for the Isan language, the most widely spoken regional dialect in Thailand. Unlike existing speech corpora that are primarily based on read or scripted speech, this dataset consists of natural speech, thereby capturing authentic linguistic phenomena such as colloquials, spontaneous prosody, disfluencies, and frequent code-switching with central Thai. A key challenge in building this resource lies in the lack of a standardized orthography for Isan. Current writing practices vary considerably, due to the different lexical tones between Thai and Isan. This variability complicates the design of transcription guidelines and poses questions regarding consistency, usability, and linguistic authenticity. To address these issues, we establish practical transcription protocols that balance the need for representational accuracy with the requirements of computational processing. By releasing this dataset as an open resource, we aim to contribute to inclusive AI development, support research on underrepresented languages, and provide a basis for addressing the linguistic and technical challenges inherent in modeling conversational speech.
Richard G. McGee, Stuart Wark, Felista Mwangi
et al.
Abstract Background The constraints imposed by the COVID-19 pandemic has led to the rapid development and implementation of digital methods for teaching clinical skills in medical education. This systematic review presents both the benefits, challenges, and effectiveness of this transition. Methods A systematic search of six electronic databases (SCOPUS, Medline, CINAHL, PsycINFO, ERIC & Informit) was conducted on 1st October 2023 and updated on 1st April 2024 to identify peer- reviewed articles, from 2019 onwards, which used any type of digital tool (online or otherwise) to teach clinical skills to medical trainees (undergraduate or postgraduate) and were published in English language. The primary outcome synthesised was the reported effectiveness of these digital tools in the development of clinical skills. Risk of bias of included studies was assessed using the Quality Assessment With Diverse Studies (QuADS) tool. Results Twenty-seven studies involving 3,895 participants were eligible for inclusion in this review. The QuADS quality assessment scores ranged from 22 to 35, indicating medium quality and thirteen of the studies were randomized trials. Overall, digital teaching of clinical skills demonstrated improved or comparable outcomes to in-person teaching. There was a beneficial effect of digital learning tools on assessment outcomes, with meta-analysis showing a mean difference increase of 1.93 (95% CI 1.22 to 2.64), albeit with a high amount of statistical heterogeneity I2 97%, P < 0.001. Digital clinical skills teaching also resulted in improved student satisfaction scores in many situations and was also shown in one study to be cost effective. Conclusion Teaching of clinical skills using digital tools is an important alternative to the traditional format of face-to-face delivery, which is resource intensive and difficult to implement during a pandemic. This review demonstrates their potential efficacy in improving education outcomes, student satisfaction and potentially reducing costs. However, the integration of traditional and innovative digital teaching methods appeared to provide the most comprehensive learning experience. Future research could focus on longitudinal studies to assess the long-term impact and efficacy of different digital and blended learning modalities on the acquisition of clinical skills and professional competencies.
Emerging applications such as blockchain, autonomous vehicles, healthcare, federated learning, self-consistent large language models (LLMs), and multi-agent LLMs increasingly rely on the reliable acquisition and provision of data from external sources. Multi-component networks, which supply data to the applications, are defined as data provision networks (DPNs) and prioritize accuracy and reliability over delivery efficiency. However, the effectiveness of the security mechanisms of DPNs, such as self-correction, is limited without a fine-grained log of node activities. This paper presents <span style="font-variant: small-caps;">DPTracer</span>: a novel logging system designed for DPNs that uses tamper-evident logging to address the challenges of maintaining a reliable log in an untrusted environment of DPNs. By integrating logging and validation into the data provisioning process, <span style="font-variant: small-caps;">DPTracer</span> ensures comprehensive logs and continuous auditing. Our system uses Process Tree as a data structure to store log records and generate proofs. This structure permits validating node activities and reconstructing historical data provision processes, which are crucial for self-correction and verifying data sufficiency before results are finalized. We evaluate the overheads introduced by <span style="font-variant: small-caps;">DPTracer</span> regarding computation, memory, storage, and communication. The results demonstrate that <span style="font-variant: small-caps;">DPTracer</span> incurs reasonable overheads, making it practical for real-world applications. Despite these overheads, <span style="font-variant: small-caps;">DPTracer</span> enhances security by protecting DPNs from post-process and in-process tampering.
Abstract This study aimed to (1) elaborate on microblogging-based instruction, which was built on social media platform X and designed based on the four phases of the design-based research (DBR) framework (Amiel and Reeves in J Educ Technol Soc 11(4):29–40, https://www.jstor.org/stable/jeductechsoci.11.4.29 , 2008); (2) put microblogging-based instruction into practice in a medical English for Specific Purposes (ESP) course; (3) examine students’ perceptions of the pedagogical value of X platform usage on their learning experiences; and (4) investigate how well microblogging-based instruction enhances the students’ specialised-language performance. On the basis of the four phases of the DBR framework, a study was conducted with students studying a medical ESP course to identify the problems in the teaching practices used in ESP courses. To improve the identified problems, following the second phase, social media platform X was used to supplement ESP face-to-face instruction. Three tasks were designed for the students to do on X. These tasks were implemented and assessed on 19 EFL first-year premedical undergraduate students over 16 weeks at a university in the Kingdom of Saudi Arabia. Data collected from the students included a copy version of the ESP final test, a questionnaire, and semi- structured interviews. The results revealed that X-based instruction improved students’ ability to use medicine-specialised language, offered opportunities for students to become more familiar with medicine-specific terms, perceived ability in using writing and reading skills, collaborative learning, and generated thoughtful discussions outside the confines of the classroom. However, the students’ inadequate knowledge of the specialised subject, the stress of making X mandatory, and log- in overload remain key challenges against the effective appropriation of X use in an ESP context.
Special aspects of education, Language acquisition
Rod Sheaff, Angela Ellis Paine, Mark Exworthy
et al.
Background As a matter of policy, voluntary, community and social enterprises contribute substantially to the English health and care system. Few studies explain how the National Health Service and local authorities commission them, what outputs result, what contexts influence these outcomes and what differentiates this kind of commissioning. Objectives To explain how voluntary, community and social enterprises are commissioned, the consequences, what barriers both parties face and what absorptive capacities they need. Design Observational mixed-methods realist analysis: exploratory scoping, cross-sectional analysis of National Health Service Clinical Commissioning Group spending on voluntary, community and social enterprises, systematic comparison of case studies, action learning. Social prescribing, learning disability support and end-of-life care were tracers. Setting Maximum-variety sample of six English local health and care economies, 2019–23. Participants Commissioning staff; voluntary, community and social enterprise members. Interventions None; observational study. Main outcome measures How the consequences of commissioning compared with the original aims of the commissioners and the voluntary, community and social enterprises: predominantly qualitative (non-measurable) outcomes. Data sources Data sources were: 189 interviews, 58 policy and position papers, 37 items of rapportage, 692,659 Clinical Commissioning Group invoices, 102 Freedom of Information enquiries, 131 survey responses, 18 local project group meetings, 4 national action learning set meetings. Data collected in England during 2019–23. Results Two modes of commissioning operated in parallel. Commodified commissioning relied on creating a principal–agent relationship between commissioner and the voluntary, community and social enterprises, on formal competitive selection (‘procurement’) of providers. Collaborative commissioning relied on ‘embedded’ interorganisational relationships, mutual recognition of resource dependencies, a negotiated division of labour between organisations, and control through persuasion. Commissioners and voluntary, community and social enterprises often worked around the procurement regulations. Both modes were present everywhere but the balance depended inter alia on the number and size of voluntary, community and social enterprises in each locality, their past commissioning experience, the character of the tracer activity, and the level of deprivation and the geographic dispersal of the populations served. The COVID-19 pandemic produced a shift towards collaborative commissioning. Voluntary, community and social enterprises were not always funded at the full cost of their activity. Integrated Care System formation temporarily disrupted local co-commissioning networks but offered a longer-term prospect of greater voluntary, community and social enterprise influence on co-commissioning. To develop absorptive capacity, commissioners needed stronger managerial and communication capabilities, and voluntary, community and social enterprises needed greater capability to evidence what outcomes their proposals would deliver. Limitations Published data quality limited the spending profile accuracy, which did not include local authority commissioning. Case studies did not cover London, and focused on three tracer activities. Absorptive capacity survey was not a random sample. Conclusions The two modes of commissioning sometimes conflicted. Workarounds arose from organisations’ embeddedness and collaboration, which the procurement regulations often disrupted. Commissioning activity at below its full cost appears unsustainable. Future work Spending profiles of local authority commissioning; analysis of commissioning in London and of activities besides the present tracers. Analysis of absorptive capacity and its consequences, adjusting the concept for application to voluntary, community and social enterprises. Comparison with other health systems’ commissioning of voluntary, community and social enterprises. Funding This award was funded by the National Institute for Health and Care Research (NIHR) Health and Social Care Delivery Research programme (NIHR award ref: NIHR128107) and is published in full in Health and Social Care Delivery Research; Vol. 12, No. 39. See the NIHR Funding and Awards website for further award information.
Plain language summary National Health Service commissioners and local councils often buy health and care services from voluntary, community and social enterprises. This study aimed to explore how commissioners and voluntary, community and social enterprises worked together and where improvements could be made. We talked to commissioners and voluntary, community and social enterprises in six areas across England and focused on services for learning disabilities, social prescribing and end-of-life care. We analysed National Health Service financial accounts to see how much the National Health Service was spending on services provided by voluntary, community and social enterprises. We surveyed how commissioners and voluntary, community and social enterprises were using information and knowledge to make decisions. We organised events bringing together commissioners and voluntary, community and social enterprises to share knowledge and experience. We found there were two ways commissioners bought services from voluntary, community and social enterprises. One was commodified (a buying-and-selling model), the other collaborative (based on working together). Both were happening at the same time in all the areas of the study, but usually one of them was more present than the other. We saw a general move towards collaboration, but some areas were further along with this than others. Various things helped commissioners and voluntary, community and social enterprises collaborate, such as: paying voluntary, community and social enterprises enough for their services; having people and networks that encouraged others to work with voluntary, community and social enterprises; and including voluntary, community and social enterprises in making decisions about health and care. Commissioners and voluntary, community and social enterprises might therefore lean further towards working together to see how collaborative commissioning can be further developed; how to make contract prices cover voluntary, community and social enterprises’ costs; how to enable longer-term contracts; and how to enable less hurried, more considered ways of renewing contracts (e.g. by putting draft contract specifications out to pre-tender consultation). The new Integrated Care Systems in local areas could encourage all this to happen, but pressures elsewhere in the health and care sector might make it more difficult.
Scientific summary Background This study examines how the NHS and local authorities commissioned voluntary, community and social enterprises (VCSEs); some outcomes for commissioners, VCSEs and the health system; and which contexts affected these outcomes. Existing studies describe how the public funding of independent providers occurs through a commissioning cycle of service specification, provider selection, contract-letting and monitoring of the activities actually delivered. The original policy (and theoretical) formulations of this cycle presupposed a clear separation of commissioners from providers. In practice, however, the cycle is partly implemented, and attenuated, by co-commissioning activities, collaborations in which potential providers, citizens and the public contribute at each stage. Commissioning has been explained as, inter alia, a means by which governments continued to exercise governance (not direct control) over independent providers of public-funded services through six main media of power: managerial techniques, negotiated order, discursive control, resource dependencies, provider competition, and juridical control. Each particular combination constitutes a ‘mode of commissioning’. Previous studies have examined the modes of commissioning applied elsewhere, but not to VCSEs, and then usually from the standpoint of policy-makers’ aims and intended service outcomes of commissioning. Fewer studies have explored commissioning from the VCSE standpoint. Some that did reported largely negative consequences for the commissioned VCSEs: a loss of freedom to criticise policy and a ‘degeneration’ of the VCSEs’ democratic internal regimes. Still fewer studies have closely examined the interactions between commissioners and VCSEs as they engage in the commissioning cycle and in co-commissioning. The practical import of these questions is whether the commissioning of health-related VCSEs enables them to supplement the reach of NHS activities and strengthen users’ voice in the health system, or whether it undermines the characteristics of VCSEs which first motivated the commissioning of them. Because commissioning involves two main groups of agents, their interactions involve not just one but two context–mechanism–outcome configurations (CMOCs). In one, the commissioners are the focal actor (as in most realist evaluations); in the other, VCSEs are. The two configurations intersect at the commissioning mechanisms, where the parties interact. Earlier studies suggested that the parties’ absorptive capacity (ACAP) to acquire, assimilate, transform and exploit externally sourced knowledge was an important context affecting what interactions occurred, and to what effect. Objectives This study aimed to produce knowledge about which factors strengthen (or weaken) collaboration between healthcare commissioners and VSCEs, and make commissioning relationships between the commissioners and VCSEs more productive for all. Research questions were: RQ1. How do healthcare commissioners address the task of commissioning VCSEs as service providers, and what barriers do they face? RQ2. What are the consequences for VCSEs of the public bodies commissioning services from them? RQ3. How are VCSEs involved in Clinical Commissioning Group (CCG), local authority and other [e.g. Integrated Care System (ICS), NHS England] commissioning decisions? RQ4. What ACAPs do healthcare commissioners and VCSEs respectively need for enabling VCSEs to be commissioned, and for co-commissioning? Methods This study was a mixed-methods realist analysis of the mechanisms by which English health-related VCSEs are commissioned. To investigate the intersecting CMOCs, we used five main methods: Preliminary scoping work with national-level NHS and VCSE organisations to identify important current developments in this domain and likely data sources, based on interviews and content analysis of policy documents. A cross-sectional profile of CCG spending on VCSEs, which provided data about patterns of VCSE commissioning and was a sampling frame for the three following work packages. We content-analysed 226,138 CCG invoices (for sums over £25k) from 2018 to 2019 to discover the distribution of CCG spending on VCSEs and how it compared with CCG spending on non-VCSE providers. Using findings from the preceding work packages, we drew a sample of six places contrasted by their proportion of spending on VCSEs. We systematically compared case studies of VCSE–commissioner collaboration in formulating local commissioning strategies (‘co-commissioning’) in them (2020–2). Commissioners’ and VCSEs’ aims when engaging in commissioning were examined separately for either side, as were the outcomes relevant to each, but not the structures and activities through which they interacted. A systematic comparison of case studies of the commissioning of VCSEs, using the same study sites, methods and unit of analysis as the preceding work package (late 2020 to summer 2022). Action learning activities (2020–3) in the same sites: Local project reference groups supported the project in each study site and combined as: national action learning workshops a preliminary, exploratory (not randomised sample) survey of ACAP in the study sites local co-researchers who conducted local research projects into the commissioning of VCSEs and whom the research team mentored As tracer studies we examined social prescribing, end-of-life care and support for learning disabilities, on the assumption that this selection gave variety in the scale and number of VCSEs involved, and type of activity (clinical vs. preventive). Framework analyses were used to synthesise the five sets of findings and map them onto the research questions. Results Two modes of commissioning VCSEs existed in parallel: A commodified mode centred on the commissioning cycle, financial dependencies, formalised procurement [a regulatory (i.e. juridically based) practice] and provider competition. Consulted VCSEs were often excluded at either service specification or provider selection stage in the cycle. The financial dependency of VCSEs was a central medium of power for commissioners. Juridical power was in the background but seldom used in practice. The paradigm form of this mechanism was a direct bilateral principal–agent relationship between one commissioner and one VCSE, with the latter wielding less power. This was a commodified mode of commissioning. To reduce their transaction work (costs), commissioners were introducing larger contracts (e.g. with one large ‘lead’ provider subcontracting many smaller VCSEs). Networks for negotiating what activities were commissioned and how they were implemented. Commissioners and VCSEs were both embedded in these networks, which constituted an interorganisational negotiated order among commissioners, among VCSEs, and between commissioners and VCSEs. Persuasion and legitimation were the main discursive media of power, supplemented by relationality (mutual trust) and mutual ‘real-side’ (as opposed to financial) resource dependency. Workarounds such as alliance and lead provider commissioning were important media of managerial power, adopted in order to diminish the expected, and in some cases observed, adverse consequences of commodified commissioning. VCSEs’ commissioning and co-commissioning activities overlapped considerably. The networking mechanisms supplemented and attenuated the quasi-market mechanisms, and indeed were partly intended to. This was a collaborative mode of commissioning. It was not that where one mode of commissioning existed, the other did not; both coexisted and interacted everywhere but the balance between them varied. Certain contexts affected how these mechanisms worked in practice: Local health and care system characteristics: Fiscal constraints upon commissioners constrained how much VCSE activity could be commissioned, compelled commissioners to prioritise VCSE activities that appeared to offer immediate cost savings elsewhere in the health system, and destabilised VCSE income. Diverse ownership mix of provider organisations (e.g. in learning disability services) motivated more cautious, commodified commissioning insofar as commissioners anticipated challenges to their decisions from unsuccessful bidders. Having a small number of stable VCSEs (e.g. in end-of-life care) enabled and necessitated collaborative commissioning. Geographical and historical characteristics: Population deprivation motivated commissioners and VCSEs to respond through collaborative commissioning. Spatial dispersion increased the practical difficulty and costs of, and the number of networks involved in, collaborative commissioning. Co-terminosity of commissioners minimised the number of interorganisational interfaces that the networks had to span. Local policies, ethos and history strongly influenced whether collaboration and trust between commissioners and VCSEs was long-established, and thus how collaborative commissioning could be. The presence and nature of networking spaces, in particular which VCSEs participated, and how much voice they had when they did participate, affected how collaborative co-commissioning could be. Organisational characteristics: ACAP affected commissioners’ ability to know what resources VCSEs could offer, and what VCSEs were aiming for, in becoming involved in commissioning, and vice versa. Commissioners relied on formal sources of information and evidence more than VCSEs tended to. Discursive (translation) gaps between VCSEs and commissioners were widespread. Organisational systems, culture and bureaucracy: the more narrowly procurement regulations were interpreted and implemented, the more commissioning was commodified rather than collaborative. Individuals’ role, discretion and influence: individuals who lobbied for VCSEs to engage, or be engaged, in commissioning and who were boundary-spanners with knowledge of both commissioning and VCSEs facilitated the development of collaborative commissioning, as did ‘maverick’ individuals who devised workarounds when procurement regulations appeared to obstruct the commissioning of VCSEs, especially collaborative commissioning. The nature of VCSE activity: Statutory or non-statutory status: statutory requirements for service provision exposed commissioners’ resource dependence on VCSEs, which led to more collaborative, flexible commissioning. Tracer group characteristics: tracer groups’ characteristics affected commissioning partly through the above contexts; that is, whether VCSEs were few and large (e.g. hospices) or the opposite (social prescribing); undertaking statutorily mandated activity (e.g. hospices, some learning disability support). The degree of specialisation of VCSE work affected the number and mix of providers. Stability of demand for a VCSE’s activity favoured more collaborative commissioning. A temporal context: the COVID-19 pandemic shifted the commissioning of VCSEs substantially towards a more collaborative mode, and this change had not fully reverted. The outcomes of these mechanisms included some that approximated to commissioners’ and VCSEs’ respective aims in using the above commissioning mechanisms. For commissioners the main outcomes were to obtain the use of VCSE resources, sometimes at below the full cost of provision. It was often claimed that VCSE activities had relieved, or would relieve, pressure on NHS services, but firm evidence was scarce. Commissioning VCSEs also offered a way for commissioners to pilot test possible innovations (e.g. for self-help in maintaining health) and a ready-made route to access patient, carer and public opinions. For VCSEs the main outcome was income, although the income flow was often unstable, which made it hard to retain paid staff, and below the cost of their activities. Some VCSEs also increased their voice in the local health system, but we also found a large periphery of VCSEs that did not participate in the networks described above and were not commissioned. We found a more nuanced picture than previous studies’ account of the internal ‘degeneration’ of VCSEs as a result of being commissioned. Most VCSEs reported little change to their aims or ‘mission’, and some had increased the professionalism of their management. We also found emergent, unforeseen outcomes. There was a general shift towards collaborative commissioning away from commodified commissioning, but it made the discursive gap between commissioners and VCSEs more apparent. The undercosting of some VCSE activities amounted to a hidden subsidy to commissioners from VCSEs. VCSE activities added a new category of public health activities, those of preventive self-care at a personal and family level, to the longer-established models of clinical prevention (vaccination, etc.) and the ‘new’ public health (legislative and policy changes, e.g. food labelling controls). ICS formation tended to disrupt co-commissioning networks in the short term, but offered a longer-term prospect of greater VCSE input into co-commissioning. Conclusions By research question: RQ1. Commissioners used two commissioning mechanisms in parallel. Each embodied a different kind of governance structure and a different mode of commissioning. Centred on the commissioning cycle, the quasi-market mechanisms implemented a heavily commodified mode of commissioning whose paradigm was a principal–agent relationship between commissioner and VCSE. The networking mechanisms of collaborative commissioning consisted of a triple negotiated order: networking among commissioners, networking among VCSEs and networking between commissioners and VCSEs. A barrier to using this hybrid mechanism was that at certain points in the commissioning cycle, especially provider selection, the mechanisms conflicted. Unfavourable contexts were other barriers: more specialised VCSE activities, dependence on a few individuals, non-coterminous commissioners, possible competition from corporate and public providers, population dispersal and fiscal austerity. RQ2. For VCSEs, consequences of being commissioned depended on the mode of commissioning but were mainly some extension of VCSE activities, including advocacy; marginal adjustments (not drift) in mission; and additional funding, but often unstable and below the full cost of these activities. RQ3. VCSE involvement in co-commissioning occurred through networking rather than quasi-market mechanisms. VCSEs with long-established working relationships with commissioners were often the ones involved, and were often larger VCSEs. Small local VCSEs were involved vicariously through their participation in VCSE infrastructure bodies or networks which were more directly involved, but a large periphery of small VCSEs remained outside these networks. RQ4. The ACAPs that commissioners needed were a combination of managerial and communication skills: to routinise ‘acquiring’ and ‘assimilating’ information from VCSEs; to specify calls for practical proposals or bids from VCSEs so as to elicit responses from suitable VCSEs; and then to apply (‘exploit’) information and ideas from VCSEs in ways that did not then penalise the VCSEs which provided them (e.g. by excluding them from tendering). VCSEs needed to develop the capacities to supply evidence, both hard and soft intelligence, about why they wished to supplement or amend commissioners’ assumptions and proposals, and to express what outcomes their proposals would offer, what metrics could be used to verify that, and more generally how the VCSE’s proposals would add social value. In ACAP terms, the commissioners appeared to need to develop their knowledge acquisition and exploitation capacities especially, while VCSEs needed to develop their knowledge acquisition and transformation capacities. Implications for health and care Below-cost funding for VCSE activity is not sustainable in the long term because VCSE activities require paid-for inputs, even when volunteer labour is used. The instability of VCSE income from commissioners contributed to VCSE staff turnover. If it continues, commissioners’ move towards longer contracts implies that a stratification and concentration of the VCSE side of the quasi-market may occur. Research recommendations are, in descending priority order, for research to: Quantify the impacts of VCSE activity on demand on NHS services. Extend our analysis of NHS commissioner spending on health-related VCSE activity to include local authority spending, and longitudinally. Test whether our findings apply to the commissioning of large (cross-England) VCSEs engaged in health-related activity. Assess the nature, effects and development of ACAP in commissioners and VCSEs by means that combine representative sample surveys (e.g. using the survey instrument developed for this study) and deeper analysis of the specific mechanisms by which ACAP affects commissioning practice. Funding This award was funded by the National Institute for Health and Care Research (NIHR) Health and Social Care Delivery Research programme (NIHR award ref: NIHR128107) and is published in full in Health and Social Care Delivery Research; Vol. 12, No. 39. See the NIHR Funding and Awards website for further award information.
While numerous studies have established the roles of metacognitive awareness and self-efficacy in foreign and second language teaching and learning, research specifically focusing on Arabic language acquisition in this context remains limited. This study aims to examine the metacognitive awareness and self-efficacy of Arabic language learners at Universitas Islam Tribakti Lirboyo Kediri. The participants in this study consisted of 72 Arabic learners. To collect data, two types of instruments were used: the Junior Metacognitive Inventory and the Self-Efficacy Questionnaire for Language Learning Strategies. The collected data were analyzed statistically by calculating the percentages of students based on their levels of metacognitive awareness and self-efficacy. The Kolmogorov-Smirnov test was employed to assess the normality of the data, while the Levene test was used to check for homogeneity. A t-test was conducted to analyze the significance of the differences in metacognitive awareness and self-efficacy between male and female learners. The results indicated that the metacognitive awareness of the Arabic learners varied across different levels, both in terms of cognitive knowledge (very good, good, fair, poor) and in the regulation of cognition (very good, good, fair). There was a significant difference between male and female learners in the regulation of cognition, while no significant difference was found regarding knowledge of cognition. Additionally, the Arabic learners exhibited varying levels of self-efficacy, including competence, confidence, and comfort; however, the difference between male and female learners in this area was not significant.
In Emergent Communication (EC) agents learn to communicate with one another, but the protocols that they develop are specialised to their training community. This observation led to research into Zero-Shot Coordination (ZSC) for learning communication strategies that are robust to agents not encountered during training. However, ZSC typically assumes that no prior data is available about the agents that will be encountered in the zero-shot setting. In many cases, this presents an unnecessarily hard problem and rules out communication via preestablished conventions. We propose a novel AI challenge called a Cooperative Language Acquisition Problem (CLAP) in which the ZSC assumptions are relaxed by allowing a 'joiner' agent to learn from a dataset of interactions between agents in a target community. We propose and compare two methods for solving CLAPs: Imitation Learning (IL), and Emergent Communication pretraining and Translation Learning (ECTL), in which an agent is trained in self-play with EC and then learns from the data to translate between the emergent protocol and the target community's protocol.
Large Vision-Language Models (LVLMs) evolve rapidly as Large Language Models (LLMs) was equipped with vision modules to create more human-like models. However, we should carefully evaluate their applications in different domains, as they may possess undesired biases. Our work studies the potential behavioral biases of LVLMs from a behavioral finance perspective, an interdisciplinary subject that jointly considers finance and psychology. We propose an end-to-end framework, from data collection to new evaluation metrics, to assess LVLMs' reasoning capabilities and the dynamic behaviors manifested in two established human financial behavioral biases: recency bias and authority bias. Our evaluations find that recent open-source LVLMs such as LLaVA-NeXT, MobileVLM-V2, Mini-Gemini, MiniCPM-Llama3-V 2.5 and Phi-3-vision-128k suffer significantly from these two biases, while the proprietary model GPT-4o is negligibly impacted. Our observations highlight directions in which open-source models can improve. The code is available at https://github.com/mydcxiao/vlm_behavioral_fin.
Recently, much work has concerned itself with the enigma of what exactly pretrained language models~(PLMs) learn about different aspects of language, and how they learn it. One stream of this type of research investigates the knowledge that PLMs have about semantic relations. However, many aspects of semantic relations were left unexplored. Generally, only one relation has been considered, namely hypernymy. Furthermore, previous work did not measure humans' performance on the same task as that performed by the PLMs. This means that at this point in time, there is only an incomplete view of the extent of these models' semantic relation knowledge. To address this gap, we introduce a comprehensive evaluation framework covering five relations beyond hypernymy, namely hyponymy, holonymy, meronymy, antonymy, and synonymy. We use five metrics (two newly introduced here) for recently untreated aspects of semantic relation knowledge, namely soundness, completeness, symmetry, prototypicality, and distinguishability. Using these, we can fairly compare humans and models on the same task. Our extensive experiments involve six PLMs, four masked and two causal language models. The results reveal a significant knowledge gap between humans and models for all semantic relations. In general, causal language models, despite their wide use, do not always perform significantly better than masked language models. Antonymy is the outlier relation where all models perform reasonably well. The evaluation materials can be found at https://github.com/hancules/ProbeResponses.
Wondimagegnhue Tsegaye Tufa, Ilia Markov, Piek Vossen
Social media conversations frequently suffer from toxicity, creating significant issues for users, moderators, and entire communities. Events in the real world, like elections or conflicts, can initiate and escalate toxic behavior online. Our study investigates how real-world events influence the origin and spread of toxicity in online discussions across various languages and regions. We gathered Reddit data comprising 4.5 million comments from 31 thousand posts in six different languages (Dutch, English, German, Arabic, Turkish and Spanish). We target fifteen major social and political world events that occurred between 2020 and 2023. We observe significant variations in toxicity, negative sentiment, and emotion expressions across different events and language communities, showing that toxicity is a complex phenomenon in which many different factors interact and still need to be investigated. We will release the data for further research along with our code.
Stefan Elmer, Mireille Besson, Antoni Rodriguez-Fornells
et al.
Learning new words in an unfamiliar language is a complex endeavor that requires the orchestration of multiple perceptual and cognitive functions. Although the neural mechanisms governing word learning are becoming better understood, little is known about the predictive value of resting-state (RS) metrics for foreign word discrimination and word learning attainment. In addition, it is still unknown which of the multistep processes involved in word learning have the potential to rapidly reconfigure RS networks. To address these research questions, we used electroencephalography (EEG), measured forty participants, and examined scalp-based power spectra, source-based spectral density maps and functional connectivity metrics before (RS1), in between (RS2) and after (RS3) a series of tasks which are known to facilitate the acquisition of new words in a foreign language, namely word discrimination, word-referent mapping and semantic generalization. Power spectra at the scalp level consistently revealed a reconfiguration of RS networks as a function of foreign word discrimination (RS1 vs. RS2) and word learning (RS1 vs. RS3) tasks in the delta, lower and upper alpha, and upper beta frequency ranges. Otherwise, functional reconfigurations at the source level were restricted to the theta (spectral density maps) and to the lower and upper alpha frequency bands (spectral density maps and functional connectivity). Notably, scalp RS changes related to the word discrimination tasks (difference between RS2 and RS1) correlated with word discrimination abilities (upper alpha band) and semantic generalization performance (theta and upper alpha bands), whereas functional changes related to the word learning tasks (difference between RS3 and RS1) correlated with word discrimination scores (lower alpha band). Taken together, these results highlight that foreign speech sound discrimination and word learning have the potential to rapidly reconfigure RS networks at multiple functional scales.
We present an empirical evaluation of various outputs generated by nine of the most widely-available large language models (LLMs). Our analysis is done with off-the-shelf, readily-available tools. We find a correlation between percentage of memorized text, percentage of unique text, and overall output quality, when measured with respect to output pathologies such as counterfactual and logically-flawed statements, and general failures like not staying on topic. Overall, 80.0% of the outputs evaluated contained memorized data, but outputs containing the most memorized content were also more likely to be considered of high quality. We discuss and evaluate mitigation strategies, showing that, in the models evaluated, the rate of memorized text being output is reduced. We conclude with a discussion on potential implications around what it means to learn, to memorize, and to evaluate quality text.
AbstractThis paper concerns the acquisition of the sign lexicon in L2 learners of Swedish Sign Language. Sampled data (conversation and narrative retelling) from a longitudinal learner corpus with 16 adult L2 signers was analyzed and compared with data from nine L1 signers. The use of three broad types of signs was analyzed: lexical signs, partly-lexical signs (i.e. depicting signs) and non-lexical signs. The results revealed some differences between L1 and L2 signers, especially with regard to depicting signs. The number of depicting signs used by L2 learners increased over time, approaching the target language use. Qualitatively, we observed differences between L1 and L2 signers in their use of depicting signs, related to handshape choice and sign constructions. We discuss these findings in light of previous research linked to L2 vocabulary as well as the role of gestural knowledge in sign L2 acquisition.
Abstract This qualitative study explores Korean high school students’ exercising of agency in processing and producing L2 writing. Data were collected from off-line and online interviews, field notes, and other written materials over the course of two years and analyzed from a social view of agency (Ahearn, in: Jaspers, Östman, Verschueren (eds) Society and language use, John Benjamin Publishing Company, Amsterdam, 2010; van Lier in Sociocult Theory Teach Second Lang 163:186–193, 2008). The students’ engagement in varied L2 writing projects and their writing artifacts consistently showed their enhanced awareness of linguistic and other semiotic resources which resulted in their frequent and continuous use of multiple languages and other placed resources. Meanwhile, they developed their strategies and reshaped their L2 writing practices considering the given context, placed resources, and their funds of knowledge. Findings from this study provide valuable insights into the open possibilities of EFL students’ exercise and development of agency, which is an increasingly necessary feature of life-long learners in the post-pandemic era.
Special aspects of education, Language acquisition
Introduction
Insufficient or faulty cooperation between the various child protection professions can have an extremly negative impact on the well-being of the concerned children. Communication problems that were revealed when dealing with cases of child abuse show the importance of adequate cooperation and common language of the involved professions in child protection.
Objectives
An online-course adressing medical-therapeutic professionals, youth welfare as well as judiciary and police was developed to impart skills and knowledge in child protection to create interdisciplinary understanding and improve cooperation between the involved professions.
Methods
The acquisition of competencies, the transfer of learning content into everyday work and the quality of the online-course are determined using an online-survey before starting (t1) and after completing (t2) the course. T1-assessment is currently being evaluated with 1034 datasets, t2-assessment will take place 03/2022.
Results
Intended target groups could be accessed and participated in the online-course, although the ratio of medical-therapeutic participants was greater than of judiciary professionals. Specific results of T1- and T2- assesssment and comparing analyses are expected in March 2022 and will be presented.
Conclusions
Based on existing online-courses developed by the Universityhospital Ulm, the suitability of online-education for training professionals in the field of child protection could be proven. If comparable effects can be shown for this online-course, there is an increase in evaluated offers of high quality. These enable comprehensive and low-threshold access to the subject of interdisciplinary communication and cooperation in child protection for involved professionals.
Disclosure
No significant relationships.
<p><em>The role of parents' beliefs or ideology is one of the most important factors in promoting the strategies for encouraging children's bilingual development. This case study examined how the parents' beliefs influenced an effort to practice the languages at home. The data were obtained through observation, in-depth interviews, digital artefacts, and language portrait silhouettes with English-Indonesian bilingual parents, and three children at the age of 6 to 11. Nvivo12 Plus application was used to organize and code the data. The results revealed that the parents believed in the importance of laying language foundations at the golden age as this could lead to shaping confident bilingual children. Findings also highlight the influential factors in parental ideology, i.e. the changing era, parents’ experience, and the support from the extended family. Although parents chose a non-bilingual school for their children, they could practice English dominantly with children at home with a wide range of strategies.</em> </p><p>Kepercayaan atau ideologi orang tua merupakan salah satu faktor yang paling penting untuk mendukung strategi dan mendorong perkembangan bilingual anak-anak. Studi kasus ini bertujuan untuk meneliti bagaimana kepercayaan orang tua mempengaruhi upaya untuk mempraktekkan bahasa di rumah. Data diperoleh melalui observasi, wawancara mendalam, artefak digital, dan siluet potret bahasa dengan orang tua bilingual Inggris-Indonesia, dan tiga orang anak berusia 6 hingga 11 tahun. Selanjutnya, data tersebut disusun dan dikodekan dengan menggunakan aplikasi Nvivo12 Plus. Hasil penelitian ini menunjukkan bahwa orang tua mempercayai pentingnya meletakkan dasar bahasa pada usia emas karena hal ini dapat membentuk anak bilingual menjadi percaya diri. Hasil ini juga menunjukkan bahwa ada beberapa faktor yang memengaruhi ideologi orang tua, yaitu perubahan zaman, pengalaman orang tua, dan dukungan dari keluarga besar. Meskipun orang tua memilih sekolah non-dwibahasa untuk anak-anak mereka, mereka dapat berlatih bahasa Inggris secara dominan dengan anak-anak di rumah dengan berbagai strategi.</p>