Jeanette K. Gundel, Nancy Ann Hedberg, R. Zacharski
In this chapter a case is made for six implicationally related cognitive statuses relevant for explicating the use of referring expressions in natural language discourse. These statuses are the conventional meanings signaled by determiners and pronouns, and interaction of the statuses with general conversational principles such as Grice’s Maxim of Quantity accounts for the actual distribution and interpretation of forms when necessary conditions for the use of more than one form are met. This proposal is supported by an empirical study of the distribution of referring expressions in naturally occurring discourse in five languages: English, Japanese, Mandarin Chinese, Russian, and Spanish.
The present study explores the relationship between Foreign Language Enjoyment (FLE) and Foreign Language Classroom Anxiety (FLCA) and a number of teacher-centered variables within the Spanish classroom context. Participants were 210 former and current learners of English as a foreign language (EFL) from all over Spain who filled out an online questionnaire with Likert scale items. A moderate negative relationship emerged between FLE and FLCA. Participants who had an L1 English speaker as a teacher reported more FLE and less FLCA than those with a foreign language user of English. Teacher characteristics predicted close to 20% of variance in FLE but only 8% of variance in FLCA. The strongest positive predictor of FLE was a teacher's friendliness while a teacher's foreign accent was a weaker negative predictor. Teacher-centered variables predicted much less variance for FLCA. Participants experienced more FLCA with younger teachers, very strict teachers, and teachers who did not use the foreign language much in class. The findings confirm earlier research that FLE seems to be more dependent on the teachers? pedagogical skills than FLCA.
Large Language Models (LLMs) are increasingly used to ``professionalize'' workplace communication, often at the cost of linguistic identity. We introduce "Cultural Ghosting", the systematic erasure of linguistic markers unique to non-native English varieties during text processing. Through analysis of 22,350 LLM outputs generated from 1,490 culturally marked texts (Indian, Singaporean,& Nigerian English) processed by five models under three prompt conditions, we quantify this phenomenon using two novel metrics: Identity Erasure Rate (IER) & Semantic Preservation Score (SPS). Across all prompts, we find an overall IER of 10.26%, with model-level variation from 3.5% to 20.5% (5.9x range). Crucially, we identify a Semantic Preservation Paradox: models maintain high semantic similarity (mean SPS = 0.748) while systematically erasing cultural markers. Pragmatic markers (politeness conventions) are 1.9x more vulnerable than lexical markers (71.5% vs. 37.1% erasure). Our experiments demonstrate that explicit cultural-preservation prompts reduce erasure by 29% without sacrificing semantic quality.
Mohammad Shojaeinia, Azamossadat Hosseini, Mostafa Naderi
et al.
Abstract Purpose To synthesize evidence on deep learning applications for diagnosing central serous chorioretinopathy (CSCR), a macular disorder associated with vision loss, this systematic review categorized studies by diagnostic task and imaging modality. The study evaluates advances in deep learning performance, clinical integration potential, dataset limitations, and the contributions of multimodal imaging and Explainable AI (XAI) to diagnostic accuracy and clinical decision-making. Methods We conducted a PRISMA-compliant systematic review of PubMed, Scopus, and IEEE Xplore, including peer-reviewed English-language studies published from January 1990 to February 2024 that reported quantitative deep learning metrics for CSCR diagnosis. A two-stage selection process was applied (Cohen’s κ = 0.84), resulting in 96 studies for analysis. Risk of bias was evaluated using the QUADAS-2 tool, and data were synthesized by imaging modality, model architecture, and diagnostic task. Results Deep learning models demonstrate exceptional performance in CSCR diagnosis. DenseNet architectures applied to optical coherence tomography (OCT) images achieved peak Metrics, including 99.78% accuracy, 99.68% sensitivity, and 100% specificity. Segmentation models for subretinal fluid (SRF) reported Dice scores of up to 0.965, while multimodal models for differential diagnosis achieved an area under the curve (AUC) of 0.999. Despite these advances, clinical adoption remains limited by several challenges: scarce and imbalanced datasets (e.g., SRF/non-SRF ratio of 1:8), lack of open-access datasets and models, risks of overfitting, and insufficient external validation. Emerging approaches, such as few-shot learning and diffusion models, are promising for mitigating data constraints; however, improvements in dataset quality and the implementation of rigorous cross-institutional validation are essential for real-world deployment. Conclusions By leveraging OCT and multimodal imaging data, deep learning has the potential to transform CSCR diagnosis through enhanced accuracy and automation. However, translating these advances into routine clinical practice necessitates overcoming key challenges, including limited and heterogeneous datasets and models with restricted generalizability. Future research should prioritize standardized reporting frameworks, transparent model interpretability through XAI, and rigorous large-scale validation. Essential strategies include employing federated learning to leverage distributed data, implementing effective multimodal fusion techniques, and fostering collaborative frameworks to improve diagnostic accuracy, ensure algorithmic fairness, and enable real-world clinical applicability.
Jalal Safari Bazargani, Abolghasem Sadeghi-Niaraki, Xinyu Shi
et al.
Spelling is a crucial language skill, yet traditional instruction often relies on rote memorization rather than meaningful learning. Despite various approaches to spelling instruction, the potential of virtual reality (VR) along with the integration of gamification and LLMs, remains underexplored. This study explores a VR-based, gamified approach using LLM-driven adaptive learning to improve spelling acquisition of English color words among young learners. The study employed a quasi-experimental, pre-test-post-test design with a control group. The participants were 50 male students aged 10, divided into an experimental group (N=25) that used the LLM-enhanced VR game and a control group (N=25) that received traditional instruction. The VR intervention consisted of a three-stage game built on the whole-word approach, featuring gamified elements and adaptive feedback from an LLM. Data were collected via spelling tests (pre-test, immediate post-test, and delayed post-test), user experience surveys, and semi-structured interviews. Results showed that the VR-based approach, significantly improved spelling performance and engagement. Specifically, the experimental group demonstrated substantially higher scores in both immediate vocabulary uptake and long-term retention after one week compared to the control group. Furthermore, qualitative and survey data indicated the VR experience was perceived as significantly more interesting, effective, and motivating. These findings highlight the potential of immersive, gamified learning environments to enhance spelling education, offering an effective alternative to conventional methods.
Vera P. Frolova, Elena N. Miroshnichenko, Irina S. Voronkova
Background. The relevance of the research theme is due to the necessity to improve the quality of foreign language teaching to Master Degree students at a technical university by searching and improving pedagogical methods in order to meet the requirements of new generation federal educational standards. In this regard, the use of edutainment technology in combination with artificial intelligence (AI) appears to be an important condition for the effective organization of the educational process with a focus on high motivation of the students. Thanks to the functionality of AI, lecturers have the opportunity to develop exciting interactive tasks as the basis of the technology analyzed both for the classroom activities and independent work of the learners. This will help to increase the language proficiency level of the students and prepare them for successful professional activity in an international environment.
Purpose. The purpose is to analyze the peculiarities and potential of applying edutainment technology in combination with elements of artificial intelligence in teaching English to Master Degree students at a technical university.
Materials and methods. The study is based on a set of theoretical and methodological approaches that has allowed us to conduct an analysis and effectiveness assessment of the applying edutainment technology in combination with AI in the educational process. It has been used such methods as observation, description, data collection and analysis, generalization, and systematization.
Results. Within the framework of integrating edutainment technology into the educational process, it has been studied and tested GigaChat tools to determine their functionality and potential for their application in foreign language teaching. The results of the study allow us to conclude that the application of this technology combined with AI elements has made classes more interesting and contributed to the intensification of the knowledge acquiring process. It has been identified scientific conferences, communicative trainings, and discussions as the most productive forms of intellectual work, which stimulated student active participation during classroom activity and independent study. The article provides examples of artificial intelligence tools application in the practice of teaching English to Master Degree students of technological faculty. The materials of the study can be used when organizing foreign language classes in higher schools.
EDN: FMZJBH
Wiwat Chancharoenthana, Asada Leelahavanichkul, Claudio Ronco
et al.
Tropical nephrology refers to kidney diseases commonly found in tropical and subtropical regions. These conditions, such as malaria-associated acute kidney injury, leptospirosis with renal involvement, schistosomiasis-related nephropathy, HIV-associated nephropathy, and dengue-associated kidney injury, are becoming increasingly relevant to clinicians worldwide due to global travel, climate change, and migration. However, their coverage in foundational English-language medical education resources may be inadequate, potentially impairing clinicians’ ability to manage these conditions effectively. To assess the extent of this gap, a structured content review was conducted across 12 widely used English-language educational materials, including general internal medicine and nephrology textbooks, tropical medicine references, and digital platforms like UpToDate. Each resource was evaluated for its coverage of five conditions across six educational domains (epidemiology, pathophysiology, clinical presentation, diagnosis, management, and prevention) using a modified DISCERN tool with a 5-point scale. The review found that overall coverage was limited, with a mean DISCERN score of 2.2 out of 5. Tropical medicine textbooks (mean 3.2) and digital platforms (mean 2.8) scored higher than general internal medicine texts (mean 1.7). Diagnosis and prevention were the least covered domains, while HIV-associated nephropathy received the most attention. These findings highlight significant gaps in core English-language educational materials that may contribute to challenges in how clinician manage these diseases. There is a clear need for improved and updated medical curricula to support better recognition, diagnosis, and treatment of tropical kidney diseases in an increasingly interconnected world.
Large Language Models (LLMs) have demonstrated remarkable capabilities across many domains, yet their application to specialized fields remains constrained by the scarcity and complexity of domain-specific training data. We present a novel approach that achieves a 261x cost reduction for maritime intelligence by using LLMs as one-time teachers rather than using them directly for inference. Our method transforms 3.2 billion Automatic Identification System (AIS) vessel tracking records into 21,543 synthetic question and answer pairs through multi-model generation (GPT-4o and o3-mini), preventing overfitting and ensuring accurate reasoning. The resulting fine-tuned Qwen2.5-7B model achieves 75% accuracy on maritime tasks, while being substantially cheaper than using a larger model for inference. We show that smaller, cheaper models -- when fine tuned properly -- can provide similar accuracy compared to larger models that are prohibitively expensive. Our work contributes to the growing field of synthetic dataset generation for specialized AI applications and presents a highly reproducible framework for domains where manual annotation is infeasible. Beyond expanding research in the growing field of specialized small language models, our approach has immediate applications in maritime safety, security operations, and vessel traffic management systems in various industries.
Scholars of English Literature unanimously say that J.R.R. Tolkien influenced C.S. Lewis’s writings. For the first time, we have investigated this issue mathematically by using an original multi-dimensional analysis of linguistic parameters, based on surface deep language variables and linguistic channels. To set our investigation in the framework of English Literature, we have considered some novels written by earlier authors, such as C. Dickens, G. MacDonald and others. The deep language variables and the linguistic channels, discussed in the paper, are likely due to writers’ unconscious design and reveal connections between texts far beyond the writers’ awareness. In summary, the capacity of the extended short-term memory required to readers, the universal readability index of texts, the geometrical representation of texts and the fine tuning of linguistic channels within texts—all tools largely discussed in the paper—revealed strong connections between <i>The Lord of the Rings</i> (Tolkien), <i>The Chronicles of Narnia</i>, <i>The Space Trilogy</i> (Lewis) and novels by MacDonald, therefore agreeing with what the scholars of English Literature say.
The purpose of this paper is to explore the role of kennings’ use in Old English (OE) poetry beyond their rhetorical power, more specifically, their role as mnemonic devices. Generally, kennings are used to refer to a certain entity using a more complex and descriptive way, more than one individual tag. This way of encoding referents seems to carry more than aesthetic value for poets and bards. Seeing as Old English poetry is oral in nature, I believe there is an argument to be made for the use of specific structures that can aid word and context retrieval, especially in longer-form content. As such, kennings would function as anchors, and I argue that they function this way because they contain semantic information that supports word retrieval. The framework for analysing this type of word formation is based on semantic feature analysis, which is a protocol used in the therapy of aphasia and anomia to improve word retrieval in post-stroke patients. Beyond this analysis, this paper will argue for the importance of considering alternate, novel techniques and methodologies for the study of Old English and for the diachronic study of language altogether, hoping to help bridge the gap between different areas of research.
Verbs form the backbone of language, providing the structure and meaning to sentences. Yet, their intricate semantic nuances pose a longstanding challenge. Understanding verb relations through the concept of lexical entailment is crucial for comprehending sentence meanings and grasping verb dynamics. This work investigates the capabilities of eight Large Language Models in recognizing lexical entailment relations among verbs through differently devised prompting strategies and zero-/few-shot settings over verb pairs from two lexical databases, namely WordNet and HyperLex. Our findings unveil that the models can tackle the lexical entailment recognition task with moderately good performance, although at varying degree of effectiveness and under different conditions. Also, utilizing few-shot prompting can enhance the models' performance. However, perfectly solving the task arises as an unmet challenge for all examined LLMs, which raises an emergence for further research developments on this topic.
Paramita Das, Amartya Roy, Ritabrata Chakraborty
et al.
Although Wikipedia is the largest multilingual encyclopedia, it remains inherently incomplete. There is a significant disparity in the quality of content between high-resource languages (HRLs, e.g., English) and low-resource languages (LRLs, e.g., Hindi), with many LRL articles lacking adequate information. To bridge these content gaps, we propose a lightweight framework to enhance knowledge equity between English and Hindi. In case the English Wikipedia page is not up-to-date, our framework extracts relevant information from external resources readily available (such as English books) and adapts it to align with Wikipedia's distinctive style, including its \textit{neutral point of view} (NPOV) policy, using in-context learning capabilities of large language models. The adapted content is then machine-translated into Hindi for integration into the corresponding Wikipedia articles. On the other hand, if the English version is comprehensive and up-to-date, the framework directly transfers knowledge from English to Hindi. Our framework effectively generates new content for Hindi Wikipedia sections, enhancing Hindi Wikipedia articles respectively by 65% and 62% according to automatic and human judgment-based evaluations.
A pivotal advancement in the progress of large language models (LLMs) is the emergence of the Mixture-of-Experts (MoE) LLMs. Compared to traditional LLMs, MoE LLMs can achieve higher performance with fewer parameters, but it is still hard to deploy them due to their immense parameter sizes. Different from previous weight pruning methods that rely on specifically designed hardware, this paper mainly aims to enhance the deployment efficiency of MoE LLMs by introducing plug-and-play expert-level sparsification techniques. Specifically, we propose, for the first time to our best knowledge, post-training approaches for task-agnostic and task-specific expert pruning and skipping of MoE LLMs, tailored to improve deployment efficiency while maintaining model performance across a wide range of tasks. Extensive experiments show that our proposed methods can simultaneously reduce model sizes and increase the inference speed, while maintaining satisfactory performance. Data and code will be available at https://github.com/Lucky-Lance/Expert_Sparsity.
Ishita Dasgupta, Christine Kaeser-Chen, Kenneth Marino
et al.
Reasoning in a complex and ambiguous environment is a key goal for Reinforcement Learning (RL) agents. While some sophisticated RL agents can successfully solve difficult tasks, they require a large amount of training data and often struggle to generalize to new unseen environments and new tasks. On the other hand, Large Scale Language Models (LSLMs) have exhibited strong reasoning ability and the ability to to adapt to new tasks through in-context learning. However, LSLMs do not inherently have the ability to interrogate or intervene on the environment. In this work, we investigate how to combine these complementary abilities in a single system consisting of three parts: a Planner, an Actor, and a Reporter. The Planner is a pre-trained language model that can issue commands to a simple embodied agent (the Actor), while the Reporter communicates with the Planner to inform its next command. We present a set of tasks that require reasoning, test this system's ability to generalize zero-shot and investigate failure cases, and demonstrate how components of this system can be trained with reinforcement-learning to improve performance.