Abstracts | Cognitive and Language Sciences
Commenced in January 2007
Frequency: Monthly
Edition: International
Paper Count: 2473

World Academy of Science, Engineering and Technology

[Cognitive and Language Sciences]

Online ISSN : 1307-6892

2473 Haunted Streets And Moral Maps: Reconstructing Urban Memory In Dickens’s London

Authors: Aku Youse

Abstract:

This paper examines the spectral and ethical configuration of urban space in Charles Dickens’s literary depictions of London, arguing that the Victorian city functions not as a neutral backdrop but as a palimpsest of unresolved histories, affective traces, and moral contradictions. Drawing upon spectral theory (Derrida; Gordon), urban memory studies, and moral geography, the study investigates how Dickens spatialises social trauma and ethical complexity through his narrative architecture. Focusing on Bleak House, Oliver Twist, and the Mutual Friend, the analysis explores how spectral residues—abandoned passageways, fog-laden streets, and decaying interiors—imprint the city with invisible yet forceful histories. These “haunted streets” operate as dynamic narrative sites where forgotten lives re-emerge, and moral disquiet disrupts spatial and historical continuity. The concept of “moral maps” is employed to reveal how Dickens aligns character movement, spatial disorientation, and confinement with broader ethical tensions, allowing geography to articulate critique. Haunting, in this context, is not a gothic ornament but a method of remembering and resisting: a mode of literary ethics embedded in space. Ultimately, the paper argues that Dickens constructs a “spectral ethics of space” that unsettles Victorian ideals of linear progress and rational order. His imagined London becomes a contested archive of marginalised voices, institutional failures, and unresolved injustices. This study contributes to literary urbanism by framing the city not merely as a setting, but as an active agent of ethical engagement and historical reckoning.

Keywords: charles dickens, moral geography, spectrality, urban memory, victorian london

Procedia PDF Downloads 0
2472 Does ChatGPT Bring ESL/EFL Student Writing Closer to Academic Standards: A Cross-Disciplinary Study of Linguistic Complexity

Authors: Ying Chai, Yingqi Huang

Abstract:

In recent years, Artificial Intelligence (AI) tools such as ChatGPT-4o have gained increasing popularity in supporting English as a Second/Foreign Language (ESL/EFL, henceforth FL) students in academic writing tasks, including paraphrasing to reduce grammatical errors and enhance fluency and academic style. Given the global role of English as a lingua franca, the use of ChatGPT in assisting FL students’ academic writing has attracted growing scholarly attention. However, it remains unclear whether ChatGPT effectively improves the linguistic complexity of FL students’ academic abstracts, whether such improvements make their writing more comparable to that of academic scholars, and whether these effects differ across academic disciplines.This study investigates whether ChatGPT-paraphrased abstracts approximate professional writing more closely than original student writing, and whether disciplinary background (hard vs. soft sciences) affects this approximation. To explore these questions, we constructed three million-token corpora consisting of three text types across both hard and soft disciplines: (1) FL students’ original English abstracts, (2) ChatGPT-paraphrased versions of these abstracts, and (3) abstracts authored by academic scholars published in high-impact journals.Using the entropy weight method, this study analyzed lexical, syntactic, and overall linguistic complexity across these texts.The results indicate that ten relatively significant linguistic complexity indicators were identified. Generally, ChatGPT-paraphrased abstracts demonstrated higher overall linguistic complexity scores compared to FL students’ original texts but remained significantly different from published abstracts in renowned journals. Specifically, ChatGPT-paraphrased abstracts showed superior performance in syntactic complexity measures such as dependent clause ratios (DC/C, DC/T) and complex T-unit ratios (CT/T), while L2-authored abstracts exhibited greater lexical diversity in certain measures. Disciplinary variations revealed that hard science abstracts (Biology, Chemistry, Physics, Engineering) displayed more consistent complexity patterns across the three text types, whereas soft science abstracts (Art, Sociology, Linguistics, Economics) showed greater variability. These findings suggest that while ChatGPT can effectively enhance certain aspects of academic writing complexity, significant gaps remain between AI-generated and expert-level academic discourse, with implications for L2 academic writing pedagogy and AI-assisted writing tool development.

Keywords: academic writing, artificial intelligence, ESL/EFL, disciplinary variation, linguistic complexity

Procedia PDF Downloads 0
2471 Semiotics of Subjective Well-Being in Balkan Cinema

Authors: Jan Levchenko, Alena Maslova

Abstract:

Today, studies of the subjective well-being of characters in cinema seem particularly relevant in the context of the search for a decent way of life and sustainable development, since they provide an opportunity to obtain data on the behavioral models of the character in the socio-cultural context and their perception by the modern viewer. In this regard, in our article, we will analyze from the point of view of cinema semiotics how the image of the internal and external well-being of the characters or their deficiency is constructed in Balkan cinema using the material of four contrasting Yugoslav films: Who's Singing There? (1980) by Slobodan Šijan, When I Will Be Dead and White (1967) by Živojin Pavlović, Love and Fashion (1960) by Ljubomir Radičević, Underground (1995) by Emir Kusturica. Psycholinguocultural methods of modern analysis of linguistic personality (V.N. Telia, V.V. Krasnykh), as well as semiotic analysis and critical discourse analysis of film texts, allow us to develop a protocol for analyzing film texts for the subjective well-being of characters and present the corresponding results in the article.

Keywords: Balkan cinema, semiotics of cinema, concept of subjective well-being, symbolism, everyday life and meaning, visual metaphors, cinema as a mirror of society, inner world of the hero, local color

Procedia PDF Downloads 8
2470 Empathic Capacity with Patients of Brazilian Dental Students of Different Genders: Evaluating the Humanization of Dental Care

Authors: Junia Maria Serra-Negra, Ivana Meyer Prado, Gabrielle Carvalho, Saul Martins Paiva, Isabela Almeida Pordeus

Abstract:

Objectives: The aim of the study was to analyze empathetic ability with patients of dental medicine students associated with gender and sociodemographic factors. Methods: A total of 441 Brazilian dental medicine students participated in this cross-sectional study using Google Forms and snowball strategy to collect data. The Checklist for Reporting of Survey Studies CROSS was followed. The study used a structured questionnaire with questions related to sociodemographic status and Higher Education Institution characteristics. The Brazilian version for students of the Jefferson Scale of Medical Empathy (JSE-BR) was administered to assess the level of empathy with patients of students. The JSE-BR score ranges from 20 to 140. The higher the score, the higher the level of empathy. Statistical analyses were performed with the software SPSS-25 using the Kolmogorov-Smirnov test and bivariate analyses with the Mann-Whitney U and Kruskall-Wallis tests (p<0.05). Results: The average age of the students was 24.1 years (±5.4), and the majority were female (77.8%). Most self-declared to be white skin color (54.9%) and 45.1% non-white. A total of 54.3% of the students belonged to public Higher Education Institutions and 45.7% to private ones. Most dental medicine students were exclusively engaged in study activities (70%), and 30% combined study activities with paid work. JSE-BR scores ranged from 46 to 140 (median 125.2). In the bivariate analyses, women had higher empathy scores compared with men (p=0.005). Conclusion: The students´ gender influenced their level of empathy with the patients, with women showing higher empathic ability when compared with men.

Keywords: empathy, epidemiology, dentistry, gender, university students

Procedia PDF Downloads 6
2469 About 2.3 million Ghanaians are Suffering from Some Form of Mental Illness: Exploring Mental Health Frames in Ghanaian Parliamentary Discourse

Authors: Hannah Deloris De-Heer

Abstract:

Mental health has become pervasive in today’s world. This study examines the discursive framing of mental health in Ghanaian parliamentary debates. It used the Frame Theory and corpus linguistic methods. Analysing 555 parliamentary Hansards (14,453,568 tokens), the research identifies five dominant frames: biomedical, social, cultural, rights-based, and institutional. Findings reveal that Members of Parliament (MPs) predominantly employ a biomedical frame, emphasizing clinical treatment and policy integration, while social framing links mental health to poverty, unemployment, and substance abuse. Cultural framing acknowledges traditional and spiritual perspectives, whereas rights-based discourse positions mental health as a constitutional and human rights issue. Institutional discussions highlight governance gaps, including funding disparities and challenges related to service decentralization. The study concludes that while parliamentary discourse aligns mental health with global health standards and local sociocultural contexts, implementation remains hindered by structural and resource constraints. Recommendations include further research on Ghanaian definitions of mental health and their representation in digital spaces.

Keywords: mental health, parliamentary discourse, framing, Ghana, corpus linguistics, policy

Procedia PDF Downloads 3
2468 A Design of Shang Mi 4 Optimization with Chen Chaos Key Expansion

Authors: Kai Shen, Lei Li, Wanting Zhou, Guanjun Yao

Abstract:

As China's national commercial encryption standard, SM4 is widely used in data encryption and secure communication. Its key expansion module generates round keys through 32 rounds of iterations and participates in encryption and decryption operations. However, the core parameter of this module, the round constant (CK), is completely fixed and public, which may be a weak link for some malicious attacks. To improve security, this paper proposed an SM4 key expansion optimization scheme based on the Chen chaos algorithm(SM4-Chen). The key expansion was associated with the chaotic system, and 32 sets of pseudo-random sequences were generated by the four-dimensional Chen chaos algorithm to replace the traditional fixed CK, which improved the randomness and security of the round key. In terms of hardware implementation, the Euler method was used to discretize the Chen chaos algorithm, and the throughput was improved by combining the optimized SM4 eight-stage pipeline architecture. Experimental results show that, under the same initial key, the Shannon entropy value of the optimized round key can be improved from 3.9431 of the original scheme to 3.9769. While ensuring the security of the algorithm, this solution can improve throughput through pipeline design, with the latency of 8 clock cycles , providing a safe and efficient solution for the application of the SM4 in IoT encryption scenarios.

Keywords: SM4, chen chaos algorithm, pipeline, key expansion

Procedia PDF Downloads 11
2467 Algorithm-Based Detection and Correction for Soft Errors in Convolutional Neural Networks Accelerators

Authors: Guanjun Yao, Lei Li, Wanting Zhou, Kai Shen

Abstract:

With the widespread deployment of convolutional neural networks (CNNs) in safety-critical applications such as autonomous driving and industrial control, the reliability of dedicated CNN accelerators becomes increasingly important. Due to external factors such as high-energy particle radiation and temperature fluctuations, soft errors are prone to occur during the operation of the accelerator, resulting in abnormal calculation results, which in turn affect the accuracy and safety of inference. In order to effectively detect and correct the soft errors that could occur during the operation of the CNN accelerator, this paper studied and implemented an algorithm-based detection and correction (ABDC) mechanism based on checksums. In order to verify the effectiveness of the mechanism on an actual hardware platform, this paper designed and implemented a CNN accelerator that supported LeNet-5 network inference, and used it as an integration and evaluation platform for the ABDC mechanism. Hardware synthesis results show that the proposed ABDC mechanism can achieve a high error detection rate with an accepted area overhead. To further evaluate its fault tolerance, this paper built a soft error injection platform and conducted 10,000 experiments to randomly inject single-bit errors during the convolution calculation process to verify the robustness and accuracy of the proposed method. It can be concluded that the ABDC mechanism can enhance the fault tolerance of CNN accelerators under low-overhead conditions, offering valuable insights and references for the future design of high-reliability CNN accelerators.

Keywords: algorithm-based detection and correction, convolutional neural network accelerator, fault tolerance, soft errors

Procedia PDF Downloads 8
2466 Exploring Chinese Educational Planning: How the Neigao Ban Program Influences on Language Attitudes and Identity Construction

Authors: GuoFeng Zheng, Ruxianguli Wulayin

Abstract:

Established in 2000, the Neigao Ban Program (“Outside-Xinjiang Boarding Schools”), is being implemented in developed regions of eastern coastal China. The program offers three to four years of education, including a preparatory year, in boarding schools for Xinjiang prospective high school students in grades 10 to 12. This study investigates its impact on different ethnic groups regarding language attitudes and identity construction, drawing on the education in Bourdieu’s cultural capital. To achieve this, the study follows several steps. First, it provides a comprehensive analysis of the existing literature on the Neigao Ban program, including its history and current status. Next, questionnaires were administered to 200 students who participated in the program in 2009, regardless of their program location or duration, to provide a macro-level evaluation. Then, to explore the program’s impact on different ethnic groups in more detail, semi-structured online interviews were conducted with six students who graduated from Shanghai Chongming High School in 2013 after four years of study. These students represent a variety of ethnic backgrounds, including Han Chinese, Uyghurs, Kazakhs, Hui, Xibe, and Manchu. Preliminary results indicate that the program has largely achieved its goals of cultivating talent and promoting the development of Xinjiang, as evidenced by the entry of these students into various sectors of the workforce in and outside of Xinjiang. Additionally, it fosters a shared national Chinese identity while enhancing students’ educated elite ethnic identity. Furthermore, the program has a positive influence on students’ language attitudes: they embrace multiple languages (Chinese, English, their mother tongue and those spoken by their friends from other ethnic minority groups) along with the rich cultures associated with these languages.

Keywords: educational planning, identity construction, language attitude, Neigao Ban

Procedia PDF Downloads 4
2465 Media and the Construction of Linguistic Ideologies: An Analysis of Le Nouvelliste’s Positioning within the Context of Linguistic Tensions in Haiti

Authors: Nazaire Joinville

Abstract:

The coexistence of French and Creole in Haiti, the result of over two centuries of history, constitutes a major sociolinguistic issue that continues to provoke intense debate within the national media. In this context, the present study aims to analyze the linguistic ideologies conveyed by Le Nouvelliste, Haiti’s leading daily newspaper, over the period from 2000 to 2023, in relation to the enduring tensions between the two languages. To this end, the research draws on a corpus of fifty press articles. The study thus highlights a media discourse marked by a notable ambivalence: on the one hand, a symbolic recognition of Creole as a central component of national identity; on the other, a functional valorization of French, justified by its perceived social and economic utility. This tension reveals the ideological complexities surrounding Haitian multilingualism and calls for a broader reflection on linguistic dynamics in postcolonial societies

Keywords: Linguistic ideologies, press discourse, French, Creole, bilingualism, identity

Procedia PDF Downloads 10
2464 On Descriptions and Characterization: A New Analytical Framework

Authors: Boyan Xu

Abstract:

This paper revisits the issue of definite descriptions, not from the perspective of reference theory or truth-conditions, but from the angle of speaker subjectivity. Philosophers such as Russell, Donnellan, Strawson, and Grice have laid out various functions of descriptions—quantificational, referential, attributive, presuppositional, and implicational. But what remains insufficiently addressed is the fact that speakers do not always describe for the sake of reference. The way a description is constructed often reflects the speaker’s emotional stance, cognitive orientation, or evaluative bias. In this sense, description is not only a way of picking out a referent but also a means of expressing the describer’s relation to what is described. To clarify this expressive aspect, I introduce the notion of characterization as a triadic relation involving a speaker (P), a target (T), and a descriptive way (W). This structure allows us to treat descriptions as pragmatic acts rather than solely semantic constructions. Further, I propose the concept of reflective characterization to capture cases where the speaker’s choice of W does not only describes T, but also reflects P’s internal state. For example, calling someone “the crazy man” or “the saviour” is not simply referential—it simultaneously expresses a stance, an emotion, or a belief. The proposal is not meant to replace existing theories of description, but to account for a range of descriptive practices that they leave unexplained. I argue that reflective characterization occupies a space between semantic determination and pragmatic implication. It allows us to analyze how descriptive language can function as a projection of the speaker’s own position within discourse. This paper aims to make this expressive dimension visible and analytically accessible

Keywords: Characterisation, Reflective Characterisation, Descriptive Pragmatics, Speaker Subjectivity, Expressive Meaning, Philosophy of Language

Procedia PDF Downloads 10
2463 Mapping Single Word Processing Onto Brain Dynamics

Authors: Ali Al-Azem, Emmanuel Chemla, Minye Zhan, Christophe Pallier, Yair Lakretz

Abstract:

The human brain can process single words during different tasks and via different modalities. it can hear the word ‘apple’ and repeat it out loud, or we see the word written on a page and read it, or we think of its concept, or see an image of it, and name it out loud. These three tasks (word repetition, naming and reading) require different computations at early stages, but share others at higher, amodal, levels.Decades of research in neuropsychology and cognitive sciences have produced a detailed description of the various information-processing stages during each of these tasks, based on examination of patients with brain damage, often highly localized. Each of these processing steps is considered a separate module, which can be selectively impaired following a focal brain lesion, which would lead to specific error patterns in one or more ofthese tasks.Neuroimaging studies have isolated cortical regions putatively associated with several of these processing stages, however, they often target a specific component of the model, in a specific modality, and thus could not provide a comprehensive view on single-word processing in the human brain. it suggest a new experimental design to study the neural basis of these three tasks jointly. it present results from a pilot study, which reveal cortical regions sensitive to specific factors of the experimental design, such as modality selectivity.Overall, the study provides first steps towards producing a global map of the cortical regions involved in modality-specific and amodal processing of words, as well as of the various information-processing pathways. These results could provide important insights into both typical and abnormal single-word processing, such as dyslexia or anomia.

Keywords: fMRI, naming, reading, word repetition

Procedia PDF Downloads 12
2462 An Evaluation of Arabic News Translation on the Tradukka Website: A Critique of Translation Techniques and Quality

Authors: Nur Akidah Khalimatussa Diyah

Abstract:

This study investigates the translation quality of Arabic news texts translated by the online machine translation platform Tradukka. The research evaluates the translation techniques applied and the overall quality of translations produced when Arabic news from Al-Jazeera is rendered into Indonesian. Employing a descriptive qualitative method, this study uses Molina and Albir’s theory to analyze translation techniques and Nababan’s model to assess translation quality through the parameters of accuracy, acceptability, and readability. The data source includes three selected news articles from the Al-Jazeera Arabic website, translated using Tradukka and assessed by 36 respondents with specific linguistic qualifications. The findings reveal the presence of 12 out of 18 possible translation techniques, with literal translation being the most frequently used. The highest acceptability scores are associated with the use of established equivalence, while the lowest readability scores were observed in texts employing the addition technique. This research underlines both the advantages and limitations of machine translation platforms like Tradukka, particularly regarding their capability to produce culturally and semantically accurate translations. It contributes to the broader discourse on machine translation quality in the Arabic-Indonesian language pair and offers insights for both academic and practical applications in translation studies.

Keywords: machine translation, tradukka, translation techniques, translation quality, Arabic, Indonesian, Nababan, Molina, Albir

Procedia PDF Downloads 10
2461 Cognition in Early Mathematic Learning

Authors: José I. Navarro

Abstract:

Piaget’s studies proposed that logical thinking forms the foundation for the development of number and arithmetic skills in early childhood. While the role of Piagetian operations remains relevant, the concept of Early Mathematical Competence (EMC) provides a broader perspective on early math learning. EMC refers to the ability to understand and apply mathematics in a variety of contexts, depending on children's individual capabilities and the demands of specific situations.The Early Numeracy Test–Revised (ENT-R), grounded in this interactionist model, assesses core components of mathematical competence in children aged 4 to 7. It helps identify those at risk of mathematical learning difficulties by evaluating underlying cognitive and relational factors. The Spanish adaptation has been digitally implemented, and an educational app has also been developed based on this version.This study presents preliminary findings from the app’s development process, with a particular focus on the challenges of integrating voice recognition technology.

Keywords: early math, voice recognition, math cognition, APP

Procedia PDF Downloads 12
2460 Mapping Single-Word Processing onto Brain Dynamics

Authors: Ali Al-Azem, Emmanuel Chemla, Minye Zhan, Christophe Pallier, Yair Lakretz

Abstract:

The human brain can process single words during different tasks and via different modalities. We can hear the word ‘apple’ and repeat it out loud, or we can see the word written on a page and read it, or we can think of its concept, or see an image of it, and name it out loud. These three tasks (word repetition, naming and reading) require different computations at early stages, but share others at a higher, amodal level. Decades of research in neuropsychology and cognitive sciences have produced a detailed description of the various information-processing stages during each of these tasks, based on examination of patients with brain damage, often highly localized. Each of these processing steps is considered a separate module, which can be selectively impaired following a focal brain lesion, which would lead to specific error patterns in one or more of these tasks. Neuroimaging studies have isolated cortical regions putatively associated with several of these processing stages, however, they often target a specific component of the model, in a specific modality, and thus could not provide a comprehensive view of single-word processing in the human brain. We suggest an experimental design to study the neural basis of these three tasks jointly. We present results from a pilot study, which reveals cortical regions sensitive to specific factors of the experimental design, such as modality selectivity. Overall, our study provides first steps towards producing a global map of the cortical regions involved in modality-specific and amodal processing of words, as well as of the various information-processing pathways. These results could provide important insights into both typical and abnormal single-word processing, such as dyslexia or anomia.

Keywords: reading, word repetition, naming, fMRI

Procedia PDF Downloads 12
2459 Disentanglement and Compositionality of Letter Identity and Letter Position in Variational Auto-Encoder Vision Models

Authors: Bruno Bianchi1, Aakash Agrawal, Stanislas Dehaene, Emmanuel Chemla, Yair Lakretz

Abstract:

Human readers can accurately count how many letters are in a word (e.g., 7 in "buffalo''), remove a letter from a given position (e.g., "bufflo'') or add a new one. The human brain of readers must have therefore learned to disentangle information related to the position of a letter and its identity. Such disentanglement is necessary for humans' compositional, unbounded ability to create and parse new strings, with any combination of letters appearing in any position. An important consideration is whether modern deep neural models possess this crucial compositional ability. Here, we tested whether neural models that achieve state-of-the-art on disentanglement of features in visual input can also disentangle letter position and letter identity when trained on images of written words. Specifically, we trained beta variational autoencoder (β-VAE) to reconstruct images of letter strings and evaluated their disentanglement performance using CompOrth - a new benchmark that we created for studying compositional learning and zero-shot generalization in visual models for orthography. The benchmark suggests a set of tests, of increasing complexity, to evaluate the degree of disentanglement between orthographic features of written words in deep neural models. Using CompOrth, we conducted a set of experiments to analyze the generalization ability of these models, in particular, to unseen word length and to unseen combinations of letter identities and letter positions. We found that while models effectively disentangle surface features, such as horizontal and vertical `retinal' locations of words within an image, they dramatically fail to disentangle letter position and letter identity and lack any notion of word length. Together, this study demonstrates the shortcomings of state-of-the-art β-VAE models compared to humans and proposes a new challenge and a corresponding benchmark to evaluate neural models.

Keywords: reading, deep learning, compositionality, auto-encoders

Procedia PDF Downloads 13
2458 A Model Interpretation Method Based On Feature Dependency

Authors: Luo Chen

Abstract:

Explainable Artificial Intelligence (XAI) aims to clarify the reasoning behind model predictions in ways that are understandable to humans, thereby helping users comprehend the internal mechanisms of models. However, most current interpretation methods are based on the assumption of feature independence. Existing feature-dependent interpretation approaches often fail to fully account for the dependencies among sample features when fitting the explanatory model, resulting in low interpretation accuracy in scenarios where feature dependencies exist. To address this issue, this paper proposes an Interpretation Method with Feature Dependency (IMFD). IMFD groups features based on their correlations and utilizes local perturbations and techniques such as Group Lasso to generate linear models that approximate the local decision boundaries of deep learning models, thereby enhancing interpretation accuracy. For each data sample, IMFD provides the weight of each feature in the linear model to explain how different features contribute to the sample’s classification outcome. Furthermore, IMFD is applied to enhance widely-used black-box interpretation methods such as LIME and SHAP. Comparative experiments on a network intrusion detection model demonstrate that IMFD achieves superior interpretation accuracy and feature selection capability. The improved methods, LIME-FD and SHAP-FD, also show varying degrees of performance improvement over their original counterparts.

Keywords: explainable artificial intelligence, feature dependency, deep learning models, group lasso

Procedia PDF Downloads 12
2457 A CRM (Customer Relationship Management) Plugin That Enhances Customer Satisfaction and CRM Staff Performance Through Sentiment Analysis

Authors: Buğrahan Kiziltaş, Kadirhan Sağlam

Abstract:

This study aims to perform sentiment analysis on customer messages within CRM (Customer Relationship Management) systems. The goal is to develop a CRM plugin that enhances customer satisfaction and enables employees to provide more effective feedback. The system is implemented on an e-commerce platform, where customers communicate with employees via voice messages regarding the products they have purchased. These recordings are analyzed and delivered to employees as notifications. In the modeling process, a pre-trained model such as Wav2Vec2 is utilized, while the RAVDESS, CREMA-D and Emotions datasets supported the training and testing phases. The project is developed using .NET technologies; Identity is used for security, Entity Framework for database operations, SignalR for real-time communication, and Stripe for payment integration.

Keywords: CRM, .NET, speech emotion recognition, text emotion recognition

Procedia PDF Downloads 12
2456 Reasoning over Knowledge Graphs with Large Language Models via Relation-Path Planning and Rewriting

Authors: Tang Yichen

Abstract:

Large Language Models (LLMs) often suffer from hallucinations and factual incompleteness due to outdated or missing knowledge. To address this, we propose ROK (Reasoning over Knowledge), a reasoning framework that integrates Knowledge Graph (KG) structure into LLMs through a relation-path planning module, KG-based retrieval, and natural language rewriting. This structured augmentation enhances both the accuracy and explainability of model predictions. Experiments on WebQSP and CWQ show that ROK achieves state-of-the-art performance across various LLMs while providing interpretable reasoning chains, bridging the gap between symbolic knowledge and generative models.

Keywords: large language models, knowledge graphs, knowledge-augmented reasoning, explainable generation, retrieval-augmented generation

Procedia PDF Downloads 11
2455 The Impact of Spatial Design Parameters on Adult Workers’ Attention and Focus in Individual Home Office Environments

Authors: Or Regev, Dafna Fisher-Gewirtzman

Abstract:

This research investigates the influence of spatial design parameters on the attention and focus of adult workers in individual home office environments, a topic of increasing relevance due to the rise of remote work following the COVID-19 pandemic. With a significant shift toward working from home, the design of home workspaces has become a critical factor in employee wellbeing and productivity. This research aims to explore how architectural and interior design elements, particularly window size, window position, and workstation location, affect cognitive functioning, focusing on attention, concentration, and overall subjective wellbeing. The experimental methodology utilized a controlled virtual reality (VR) setup in which participants experienced nine distinct immersive home office environments. Each environment featured one of three window options: a curtain wall, a standard-sized window (120W x 100H), or no window at all. These variables were paired with three different workstation placements: facing the window, positioned between the window and the door, and with the back to the window (facing the door). After exploring each environment, participants completed a detailed virtual questionnaire assessing their subjective perceptions of the space, which facilitated the creation of a wellbeing index. To objectively measure attention and focus, participants also completed a virtual puzzle-assembling task, with task completion times recorded as indicators of cognitive performance across the different spatial setups. In total 53 participants(n=53) completed the experiment, primarily architecture students aged 20 to 40, all of whom engaged in remote work or study at least one day a week. Findings reveal the relations between environmental design, task efficiency, and subjective wellbeing. The combination of a participant facing a standard window emerged as the most efficient configuration, producing both high wellbeing scores and the quickest task times for a significant portion of participants. However, comfort and subjective wellbeing did not always directly correlate with immediate task speed. Some settings with high wellbeing scores did not correspond with fast task performance, indicating that while comfort is essential for overall welfare, it may not always enhance efficiency. Notably, windowless configurations with the workstation facing away from the windowless wall (NW_B) showed high efficiency and wellbeing ratings. Overall Ratings of privacy and security emerged as significant contributors to perceived wellbeing across several configurations, emphasizing the importance of these elements in home office design. The findings highlight the importance of balancing visual exposure, environmental comfort, and psychological factors in workspace design. These insights are valuable for architects, interior designers, and remote workers looking to enhance focus and overall wellbeing in home office environments. We are currently conducting a follow-up experiment to enhance realism by blending real-life and digital environments through projections on a large screen. We replace the current task with the Trail Making Test (TMT). These changes aim to better reflect real-world scenarios and improve the assessment of attentional processes. By conducting this experiment alongside the previous one, we can compare attentional performance between the VR and integrated real-life conditions.

Keywords: work from home, attention and focus, wellbeing, architecture and spatial design, experiments in VR

Procedia PDF Downloads 12
2454 An Examination of Code-Switching and Linguistic Coinages as Linguistic Tools in Yoruba Child Acquisition of English

Authors: Oluwafemi Emmanuel Bamigbade

Abstract:

Second language acquisition and second language learning are two linguistic phenomena studied within the ambiance of applied linguistics that is interwoven. One of the most widely used hypotheses to describe language acquisition is the innateness theory, amongst others. It is in light of the innateness hypothesis that this paper intends to investigate the pattern of code-switching and the level of interference and transfer of linguistic features in the language developmental stages of Yoruba children while learning to use the English language, especially in an informal setting. This is significant because an average Yoruba child is exposed to using the English language at a very early stage at home and in the neighborhood, thus making the child an early coordinate bilingual. The data for this work include recorded conversations of ten randomly selected Yoruba children ages four to eight. To collect the data, the researcher engaged in participant observation with the children in their school during the break period and also collected some other data by interacting with a few of the children in their homes. The data were subjected to descriptive analysis. The findings reveal that an average Yoruba child either employs code-switching, code-mixing, and lexical coinage or accommodates language interference and/or language transfer while attempting to use the English language during his/her early exposure to English.

Keywords: acquisition, coinage, innateness, interference, transfer

Procedia PDF Downloads 14
2453 An Enhanced YOLOv5-Based Algorithm for Ship Detection

Authors: Yuan Tan

Abstract:

To address the challenge of reduced ship detection accuracy in complex maritime environments and across vast sea surfaces, this paper proposes an improved YOLOv5-based ship detection algorithm. The proposed method integrates the self-organizing competitive attention (SOCA) mechanism, which leverages second-order statistical characteristics of the image to enhance the network’s feature representation capability. In addition, a dedicated small-object detection layer is incorporated to improve the detection performance for small-scale targets. The loss function is replaced with WIoU, which not only emphasizes the overlap area between bounding boxes but also incorporates their directional relationship, thereby offering a more precise regression metric and improving localization accuracy. Experimental results demonstrate that the proposed method increases detection accuracy by 3.8%, with [email protected] improved by 1.3%, along with enhancements across other performance metrics. Furthermore, due to its lightweight design, the algorithm is suitable for deployment on mobile terminals and can be effectively applied in real-time maritime target detection scenarios.

Keywords: target detection, YOLOv5, SOCA attention mechanism, WIoU

Procedia PDF Downloads 17
2452 Balancing the Specificity and Generalisability of Learning: Roles of Sleep and Wakefulness

Authors: Zhishan Liu, Chen Song

Abstract:

Visual perceptual learning (VPL), the improved ability to extract visual information through training, occurs not only during active task engagement (the "online" phase) but also during subsequent rest or sleep (the "offline" phase). However, it remains unclear how these two phases interact to optimize learning outcomes. In this study, we examined whether improvements in visual performance following training are specific to the trained visual feature and location or whether they generalize to untrained features and locations, and how such changes unfold during wakefulness and sleep. Participants were trained on a backward masking task involving either orientation or luminance discrimination in a specific visual field quadrant. Across five test sessions—including immediate, 12-hour, 24-hour, and delayed follow-up—we assessed performance on tasks that varied by visual feature, spatial location, and task rule. By comparing performance in morning and evening training groups, we isolated learning effects occurring during wakefulness and those emerging after sleep. Our findings revealed a functional division of effort between the sleep and wake phases. Improvements in visual performance during wakefulness were highly specific, restricted to the trained visual feature and location, suggesting the involvement of local, task-dependent plasticity within stimulus-relevant circuits. In contrast, post-sleep improvements generalized to untrained features and locations, suggesting that sleep supports global, homeostatic plasticity and widespread reactivation of neural circuits. Both local and generalized gains were sustained over several weeks, indicating long-term retention. These findings suggest that wakefulness and sleep jointly shape perceptual learning by balancing precision and flexibility. Wake promotes specificity for targeted improvements, while sleep fosters adaptability by enabling generalization beyond the trained context. This mechanism may reflect a broader neural strategy for optimizing learning across different domains. Our study highlights a central challenge in learning: maintaining specific, high-level performance without losing the ability to generalize. We show that this balance is naturally achieved by assigning specificity to wakeful learning and generalization to sleep-driven processes. These findings offer practical insights for optimizing learning strategies in both clinical and educational settings.

Keywords: perceptual learning, generalisation, sleep, memory, learning, features, tasks

Procedia PDF Downloads 12
2451 Multimodal Fake News Detection Method With Semantic Consistency And Image Manipulation

Authors: Hailong Gui, Dan Liu

Abstract:

Nowadays, the internet plays an increasingly indispensable role in people's lives, and fake information, which is pervasive across various social media and online spaces, has become a serious factor affecting people's lives. As a result, detection methods for different types of fake information have also been developed. Compared to single-modal fake information, multimodal fake information detection methods are more complex, but their practical significance is also greater. In this paper, in addition to utilizing traditional BERT and ResNet34 models for extracting text and image features from multimodal data, it also design two modules to investigate the presence of image forgery and semantic consistency issues in the input data. This paper proposes a model combining Semantic Consistency and Image Manipulation features (SCIM model). Besides using the multi-head attention mechanism to fuse the extracted features from different modalities, it also uses the fused features for authenticity recognition. Additionally, the model constructs two sub-tasks for classification of semantic consistency and image forgery, employing three loss functions to form the overall task optimization objective. The SCIM model has been compared with baseline models on three widely used real-world datasets—Twitter, Weibo, and GossipCop—and outperforms the baseline models in terms of fake information detection accuracy. The experimental results show that the proposed model can effectively identify the authenticity of multimodal fake information.

Keywords: fake information detection, multimodal, multi-head attention mechanism, image manipulation features, semantic consistency

Procedia PDF Downloads 12
2450 Research on Container Scheduling Techniques for Deep Learning Training Tasks

Authors: Guanping Xie

Abstract:

Abstract——With the widespread application of machine learning across various fields, current AI models possess powerful capabilities in data processing and analysis, enabling them to effectively meet the functional requirements of complex environments. Meanwhile, containerization technology, due to its lightweight nature, isolation, and ease of deployment, has become an effective technical vehicle for task scheduling and resource management in training clusters. However, in machine learning training clusters, the diverse container resource demands brought by complex model structures, as well as heterogeneous resource contention, pose significant challenges to the orchestration and management of compute-intensive machine learning tasks. Furthermore, due to the limited processing capabilities of a single node, distributed training techniques have been introduced to accelerate the handling of massive datasets and complex model training, which in turn leads to more complex topological relationships among task containers and increases the difficulty of container scheduling. In this context, this paper focuses on containerized AI model training tasks and conducts research on task orchestration from the perspectives of interference mitigation and load optimization in containerized machine learning systems. The research is carried out for both centralized and distributed training scenarios, and the main contributions are summarized as follows: (1) Deployment strategy for containers in centralized training tasks based on model characteristics. To address the issues of insufficient awareness of resource demands during containerized deep learning deployments and the performance interference caused by resource contention among co-located tasks, this paper proposes a container deployment scheme for deep learning applications based on model characteristics. The strategy incorporates awareness of characteristics such as computational workload, parameter count, and model architecture to predict container resource requirements. Based on this, it evaluates the interference level by measuring the increased task completion time under different model co-location combinations. The scheduling problem is then modeled with consideration of system load balancing, and a heuristic scheduling solution based on a genetic algorithm is employed to complete task container placement. Experimental results show that the proposed method can accurately estimate resource demands, effectively balance task interference and system load, and reduce the completion time of deep learning training tasks. (2) Scheduling scheme for distributed training task containers based on hybrid optimization of interference and communication.To address the performance interference and communication bottlenecks between distributed units that arise when deep learning workloads are co-located in a distributed environment, this paper proposes a container scheduling scheme based on hybrid optimization of interference and communication. Based on a proposed affinity-based container deployment framework for scheduling units, a resource model for distributed units is established to sense interference levels, and a cross-device communication model is constructed according to the structural characteristics of distributed training. Finally, a scheduling policy based on reinforcement learning is designed to perform hybrid optimization scheduling of containers in terms of interference and communication. Simulation results demonstrate that the proposed scheme can optimize data transmission between containers in distributed training tasks while avoiding task performance degradation due to excessive resource contention. This effectively compensates for the shortcomings of current cluster schedulers in addressing communication and contention problems in distributed training, and provides a viable solution for performance optimization in containerized distributed deep learning systems.

Keywords: Container Technology, Machine Learning, Distributed Training, Resourcce Contention.

Procedia PDF Downloads 16
2449 Intercultural Competence Development Through the Translation of Chinese Emotional Vocabulary

Authors: Jingsong Ma

Abstract:

The interpretation of culturally-specific emotional terminology in classical Chinese texts offers valuable potential for advancing intercultural learning. This investigation explores pedagogical approaches for teaching complex affect-laden concepts like yuàn (怨) and chóu (愁) - terms that defy simple translation due to their deep cultural and philosophical roots - to facilitate meaningful cross-cultural engagement. Analysis of representative texts from the Tang-Song period highlights three critical dimensions of Chinese emotional expression: (1) contextual variability (where yuàn may indicate moral outrage or intimate sorrow depending on usage), (2) culturally-conditioned manifestations (distinguishing between Confucian emotional moderation and Western affective directness), and (3) underlying worldview differences (contrasting Daoist perspectives on impermanence with Western philosophical individualism). These distinctions have substantial consequences for language instruction methodology. The study implements a comprehensive analytical framework combining: (1) systematic examination of emotion terminology across multiple literary forms; (2) assessment of translation variations in multiple English renditions of classical works; (3) practical classroom applications where learners construct detailed comparative ‘affect profiles.’ This tripartite methodology synthesizes anthropological linguistics with contemporary translation theory. Key research outcomes include: (1) identification of persistent translation challenges arising from the culturally-grounded nature of Chinese emotional lexicon (such as chóu conveying artistic sensibility rather than psychological distress); (2) demonstration of how context-sensitive teaching methods help students comprehend emotions as culturally-mediated concepts rather than direct equivalents; (3) empirical evidence showing enhanced intercultural awareness through this pedagogical approach. The results establish that conceptualizing emotional vocabulary as cultural signifiers rather than simple dictionary items: (1) improves cross-cultural communication skills; (2) produces quantifiable advances in cultural sensitivity; (3) provides transferable instructional techniques for linguistically complex concepts. This work significantly advances the fields of translation instruction and intercultural education by establishing practical, research-supported methods for navigating the intersection of language and culture. The findings confirm that meticulous examination of emotional language in literary translation constitutes a powerful mechanism for cultivating profound intercultural perspective in language learners.

Keywords: affective language, cross-cultural understanding, emotion translation, intercultural competence

Procedia PDF Downloads 14
2448 Nine Axes Program for Training Autistic Individuals Based on Sensory Threshold Diversity, Systematic Desensitization, and Modeling

Authors: Mahmoud Abdelrahman Abourehab

Abstract:

Autism spectrum disorder is often seen through a medical lens as a condition that needs to be treated. There is another way to view autism as a human diversity. This perspective, known as the neurodiversity paradigm, encourages us to recognize the unique ways autistic people experience the world rather than focusing only on their challenges. This paper outlines a tailored training program that harmonizes the dual perspectives of autism, leveraging inherent strengths while addressing challenges. This program blends both medical and neurodiversity viewpoints, aiming to help autistic individuals by supporting their needs and highlighting their strengths. The program revisits notions directly linked to neurodiversity, such as sensory thresholds, reaction time, sensory processing, and sensory integration. Dysfunction in the sensory thresholds of one or multiple senses leads to difficulties stemming from a mismatch with the responses of same-age typical development peers. Autistic individuals face differences in how they sense and process the world; some may be hypersensitive, while others might be hyposensitive. These differences can make everyday life challenging compared to typical development peers, or they may make them unique individuals. The proposed program in this paper consists of nine axes, the first of which assesses the eight senses, not just the usual five, but also proprioception, the vestibular sense, and interoception. The second axis measures the reaction time. Both axes underscore the necessity of building an "Autism Laboratory" for comprehensive assessment, creating a sensory perception profile, and identifying the autistic individual's strengths and weaknesses. The third axis is for sensory threshold regulation, placing the autistic individual in a fully controlled environment with aid devices designed to modify the intensity of sensory stimuli to align with their absolute and terminal sensory threshold levels. The fourth axis focuses on sensory deprivation with systematic desensitization, which involves depriving the autistic child of all sensory stimuli except for a particular stimulus. The fifth axis addresses the sensory processing of multiple stimuli within a single sense to guide the brain to focus on a specific stimulus among several. The sixth axis is for sensory integration across multiple senses; this entails presenting stimuli that engage more than one sense, such as sound and vision, to improve the child's sensory processing and integration capabilities. The seventh axis involves gradually increasing tolerance for sensations and pain perception using systematic desensitization techniques. The eighth axis is for teaching social expressions through modeling and engaging in one-to-one free play. The ninth axis is for utilizing the extraordinary abilities. The nine axes program outlined in this paper is a carefully organized, personalized, and individual-focused intervention that respects their unique sensory profiles and harnesses their extraordinary abilities. However, empirical validation is critical to determining its efficacy. Establishing an autism lab to assess and address sensory threshold diversity in autistic individuals is a compelling and worthwhile initiative. The paper also suggests building an industry focusing on manufacturing aid devices technologies designed to help autistic individuals regulate sensory thresholds, support them in navigating sensory difficulties, and empower them to leverage their distinctive abilities.

Keywords: modeling, reaction time, sensory integration, sensory processing, sensory threshold diversity, systematic desensitization.

Procedia PDF Downloads 14
2447 Navigating Workplace Dynamics with Wisdom: The Role of Wise Emotion Regulation Ability

Authors: Krishna Singh Bhandari

Abstract:

This study pursued a twofold objective: to establish and validate the construct of wise emotion regulation ability (WERA) by operationalizing its theoretical framework and testing its empirical distinctiveness from general emotion regulation, and to examine WERA's moderating effects on the relationship between workplace factors and employee withdrawal from work. Data were collected from 152 employees across manufacturing and IT sectors using standardized survey instruments. For the first objective, exploratory factor analysis identified latent dimensions of WERA, while confirmatory factor analysis validated a hypothesized three-factor structure comprising awareness, self-management, and adaptability. The construct demonstrated satisfactory internal consistency through reliability testing (Cronbach's α and composite reliability). The second objective employed multiple moderation analyses through hierarchical multiple regression in RStudio, utilizing factor scores computed via the regression method to model latent relationships while accounting for measurement error. Results revealed that supervisory support positively predicted work engagement, while interpersonal conflict negatively predicted the same outcome. Notably, these associations were stronger among employees with lower levels of emotion self-management (a WERA factor). Additionally, employees with lower emotion identification abilities were more vulnerable to psychological disengagement at work when supervisory support was lacking. These findings suggest that WERA constitutes a distinct construct with three empirically validated dimensions that significantly moderate workplace relationships. The study contributes to organisational behaviour literature by demonstrating how emotion regulation wisdom differentially affects employee responses to workplace interpersonal dynamics, highlighting WERA as a potential target for organizational development interventions aimed at improving employee adjustment and performance in complex interpersonal work environments.

Keywords: wise emotion regulation ability, supervisory support, work engagement, psychological withdrawal at work

Procedia PDF Downloads 17
2446 Research on the Detection and Judgment Strategy of Single-Frequency Interference Within the Signal Band of the Peak-Shaped Spectrum

Authors: Wenxiu Zheng, Xiangxin Qi, Fukang Yang

Abstract:

Aiming at the challenge of distinguishing between the frequencies of peak-shaped spectrum signals and single-frequency interference, this study proposes a detection and identification strategy for single-frequency interference. By employing extremely narrow FIR bandpass filters to process the peak-shaped spectrum signal contaminated by single-frequency interference through frequency-specific filtering at each spectral point, signals are extracted corresponding to individual frequency components. The instantaneous envelope of these signals is analyzed, and a parameter is introduced defining the energy fluctuation level of the envelope. Leveraging the significant disparity in this parameter between interference signals and peak-shaped spectrum signals, the method effectively differentiates signal frequency components from interference frequency components. Simulation results demonstrate that the proposed approach successfully separates peak-shaped spectrum frequencies from single-frequency interference frequencies.

Keywords: peak-shaped spectrum signal, single-frequency interference detection, Hilbert transform, envelope energy fluctuation level

Procedia PDF Downloads 17
2445 The Experiencing Wisdom of Tamil Proverbs: A Reflection of Folk Experience and Lifelong Relevance

Authors: Sasikumar Ponnalagu

Abstract:

This paper explores the intrinsic connection between Tamil proverbs and the lived experiences of rural folk, focusing on how these oral expressions have evolved into repositories of cultural, ethical, and practical knowledge. Proverbs, or "pazhamozhigal," encapsulate the worldview, values, and social structures of Tamil society. They originate from community experiences and have been transmitted orally through generations as part of the Tamil folk tradition. The study categorizes proverbs based on themes such as marriage customs, health practices, daily life, agriculture and commerce, divine beliefs, morality, education, unity, and self-restraint.The paper delves into the origins of proverbs, illustrating how they were formed not by individuals but by collective folk wisdom responding to everyday situations. It traces their presence in classical Tamil literature, including Sangam poetry and works like "PazhamozhiNaanooru," and analyzes how these proverbs reflect historical and socio-cultural contexts. The article also examines their pedagogical role in conveying complex life lessons through succinct language and vivid imagery.In the modern context, despite the rise of urbanization and digital communication, traditional proverbs still find relevance. The emergence of neo-proverbs adapted to digital lifestyles and online culture illustrates the adaptability of proverbial wisdom. Through qualitative analysis and literary references, this paper argues that Tamil proverbs—though ancient—remain timeless tools for communication, reflection, and education, reinforcing their role as a bridge between tradition and modernity.

Keywords: tamil proverbs, folk wisdom, oral tradition, cultural heritage, life philosophy

Procedia PDF Downloads 18
2444 Students’ Perspectives and Practices of Using ChatGPT for Self-Regulation in an Interpreting Course in Vietnamese Higher Education

Authors: Le Mai Van, Nguyen Thi Minh Thao

Abstract:

Recent advancements in artificial intelligence have increasingly been harnessed to support students’ self-regulated learning (SRL) within digital educational contexts. This research explores Vietnamese students’ perspectives and practices concerning the use of ChatGPT for SRL in interpreting. Drawing on Zimmerman’s Cyclical Model of SRL, the study examines how students interact with ChatGPT across the phases of forethought, performance, and self-reflection. A mixed-methods design was employed, combining survey questionnaires (n=178), focus-group interviews (n=24), and guided reflections (n=50) to capture a comprehensive view of student experiences. Findings indicate that students perceive ChatGPT as a beneficial tool for online interpreting practices, primarily due to its flexibility, accessibility, and ability to deliver immediate language assessment and scores. However, its effectiveness varies due to students’ proficiency levels and readiness to utilize the tool. Additionally, cultural nuances in interpreting performances are recorded as ChatGPT’s hindrance in the evaluation process. These results highlight the significance of digital literacy and metacognitive awareness in leveraging ChatGPT and suggest the need for pedagogical support in integrating this AI tool into SRL environments.

Keywords: perspectives, practices, self-regulated learning (SRL), interpreting

Procedia PDF Downloads 17