11037 research outputs found
Sort by
Transition-based semantic role labeling with pointer networks
Get PDFSemantic role labeling (SRL) focuses on recognizing the predicate–argument structure of a sentence and plays a critical role in many natural language processing tasks such as machine translation and question answering. Practically all available methods do not perform full SRL, since they rely on pre-identified predicates, and most of them follow a pipeline strategy, using specific models for undertaking one or several SRL subtasks. In addition, previous approaches have a strong dependence on syntactic information to achieve state-of-the-art performance, despite being syntactic trees equally hard to produce. These simplifications and requirements make the majority of SRL systems impractical for real-world applications. In this article, we propose the first transition-based SRL approach that is capable of completely processing an input sentence in a single left-to-right pass, with neither leveraging syntactic information nor resorting to additional modules. Thanks to our implementation based on Pointer Networks, full SRL can be accurately and efficiently done in
, achieving the best performance to date on the majority of languages from the CoNLL-2009 shared task.Agencia Estatal de Investigación | Ref. PID2020-113230RB-C21Xunta de Galicia | Ref. ED431C 2020/11Xunta de Galicia | Ref. ED431G 2019/0
Sobre la norma de la prorrata en el IVA declarada contraria al Derecho comunitario: aún hay plazo para recuperar los ingresos firmes
Get PDF[Inicio] NTRODUCCIÓN
A ningún abogado tributarista ni asesor fiscal se le habrá pasado por alto la Sentencia del Tribunal de Justicia de las Comunidades Europeas de 6 de octubre de 2005, en el asunto C-204/03, que resuelve un recurso de incumplimiento planteado por la Comisión contra la Ley española reguladora de la prorrata en el Impuesto sobre el Valor Añadido, por su tra-tamiento de las subvenciones1. [...
The evolution of public relations trends over the last five years according to the most relevant professional reports in the sector
Get PDFEffectiveness of commercial anti-graffiti treatments in two granites of different texture and mineralogy
No full textThis paper presents a study of the efficiency of two chemically different anti-graffiti coatings (sacrificial and permanent anti-graffiti products) on two different compositional and textural granitic stones, Rosa Porriño and Albero. First, both uncoated and coated surfaces of the granites were characterized using stereomicroscopy and scanning electron microscopy, static contact angle measurements, colour spectrophotometry and gloss measurements. Results showed that both anti-graffiti products increased the static contact angle of the surfaces. The permanent anti-graffiti made them water-repellent without causing notable colour changes. Second, effectiveness of the anti-graffiti products was evaluated by means of the removal of two different spray graffiti paints (blue and silver colours) on both granites protected with the above-mentioned anti-graffiti products. The cleaning procedures were those recommended by the manufacturers. Fourier Transform Infrared spectroscopy and the previously mentioned techniques were used to assess the cleaning efficiency of the coated surfaces by detecting or not the presence of graffiti remains. As a result, textural differences in the granites, chemical composition of the graffiti paints and removal time were found to be the key parameters controlling the effectiveness of graffiti removal. On Albero granite, more residues of paint were found in its fissure system. Blue graffiti based on alkyd and polyester resins was more readily removed than silver paint. In general terms, graffiti extraction was more effective 30 days after painting than 3 days after.Universidade de Vig
Context is king: Large language models’ interpretability in divergent knowledge scenarios
Get PDFLarge language models (LLMs) have revolutionized the field of artificial intelligence in both academia and industry, transforming how we communicate, search for information, and create content. However, these models face knowledge cutoffs and costly updates, driving a new ecosystem for LLM-based applications that leverage interaction techniques to extend capabilities and facilitate knowledge updates. As these models grow more complex, understanding their internal workings becomes increasingly challenging, posing significant issues for transparency, interpretability, and explainability. This paper proposes a novel approach to interpretability by shifting the focus to understanding the model’s functionality within specific contexts through interaction techniques. Rather than dissecting the LLM itself, we explore how contextual information and interaction techniques can elucidate the model’s thought processes. To this end, we introduce the Context-Driven Divergent Knowledge Evaluation (CDK-E) methodology, along with the Divergent Knowledge Dataset (DKD), for evaluating the interpretability of LLMs in context-specific scenarios that diverge from the model’s inherent knowledge. The empirical results demonstrate that advanced LLMs achieve high alignment with divergent contexts, validating our hypothesis that contextual information significantly enhances interpretability. Moreover, the strong correlation between LLM-based metrics and semantic metrics confirms the reliability of our evaluation framework.Axencia Galega de Innovación (GAIN)Xunta de Galicia | Ref. ED431B 2024/36Fondo Europeo de Desarrollo Regiona
Effects of compost and technosol amendments on metal concentrations in a mine soil planted with Brassica juncea L.
No full textMining activities often cause important impacts on soil and water quality. The main objective of this study was to evaluate the effect of amendments (compost and technosol made from waste) on metal concentrations in a mine soil planted with Brassica juncea. A greenhouse experiment with cylinder pots was carried out during 11 months. The mine soil was collected from the settling pond of the depleted copper mine of Touro (Galicia, Northwest Spain). A series of characteristics were analysed including soil pseudototal metal concentrations, soil CaCl2-extractable (phytoavailable) metal concentrations and metal concentrations in soil pore water. The results showed that at depth 0–15 cm SCP (mine soil + compost, grown with B. juncea) had a significantly lower CaCl2-extractable Cu, Pb, Ni and Zn concentration than STP (mine soil + technosol, grown with B. juncea) over the time (P < 0.05). At depths 15, 30 and 45 cm, STP and SCP had lower Cu pore water concentration than S over the time. The highest translocation factor (TF) values for all metals (Cu, Pb, Ni and Zn) were observed at time 1 (3 months) in the settling pond soils treated with technosol and B. juncea L. The conclusions of this experiment revealed that SCP compared to STP caused a higher reduction on Cu, Pb, Ni and Zn phytoavailable concentrations in the first depths
Microwave hydrodiffusion and gravity versus conventional distillation for Acacia dealbata flowers. Recovery of bioactive extracts for cosmetic purposes
No full textNatural extracts recovered from underutilized wild Acacia dealbata flowers using microwave hydrodiffusion and gravity (MHG) were compared with those obtained from conventional steam distillation. Several irradiation powers (50–125 W) were studied. MHG solid phases were treated by solid-liquid extraction (SLE) with ethanol solvent. Their total phenolic content, antioxidant capacity, total carotenoid content, color characteristics, pH value and solar protection factor (SPF) were tested. Selected extracts were incorporated to an oil-in-water emulsion cosmetic model made with three thermal spring waters and preliminary sensory analysis was made. These creams were chemically and rheologically analyzed jointly with their bioactive capacity by an acceleration oxidation test. The optimum MHG extract was collected at 75 W for 180 min, since this flower liquor exhibited the highest total phenolic content (around 0.15 mg GAE/g flower dry weight) and antioxidant capacities (about 0.39 mg Trolox eq/g flower dry weight and 0.36 μg β-carotene/g flower dry weight) of microwave extracts. Distillation extracts presented higher concentrations (about 0.20 mg GAE/g flower dry weight and around 0.58 mg Trolox eq/g flower dry weight), but with larger (two times) estimated specific energy requirements. Selected extracts provided sunscreen creams with similar chemical (i.e., pH and SPF values) and bioactive properties (i.e., thiobarbituric acid reactive substance outcomes) that cosmetics prepared with BHT or (±)-α-tocopherol commercial antioxidants. The viscosity features of the different cosmetic samples were similar, except for those sun creams elaborated with one thermal spring water. This sample presented lower apparent viscosity profiles, which could be an important advantage from the skin application point of view.Ministerio de Economía y Competitividad | Ref. CTM2015-68503-RMinisterio de Economía y Competitividad | Ref. IJCI-2016-27535Ministerio de Economía y Competitividad | Ref. RYC2018-024454-
Avaliación emocional e cognitiva desde a fala: unha investigación de solucións de aprendizaxe automática cara a aplicacións clínicas fiables
No full textThe escalating global prevalence of mental health disorders, including Major Depressive Disorder (MDD) and Alzheimer’s Disease (AD), underscores the urgent need for accessible, non-invasive diagnostic tools. Speech, as a biomarker integrating cognitive, emotional, and physiological information, offers a promising avenue for early detection. This thesis investigates machine learning (ML) solutions to automate the assessment of emotional and cognitive states through speech analysis, focusing on developing trustworthy clinical applications.
Using a diverse experimental framework comprising five datasets (ADReSS, AcceXible-MCI, DAIC-WOZ, RADAR-MDD, and Androids), this work evaluates paralinguistic (e.g., acoustic prosody, Low-Level-Descriptors, self-supervised features) and linguistic (e.g., GloVe, lexical diversity) features. Key findings reveal that Transformer-based models, leveraging semantic content, achieve state-of-the-art performance in detecting MDD and cognitive decline, while self-supervised acoustic features demonstrate cross-lingual robustness, particularly in low-resource settings. Personalized, speaker-dependent frameworks significantly outperform generalized models, highlighting the importance of tailoring systems to individual variability in symptom expression.
The study identifies noun/adverb ratios and emotional valence as statistically significant indicators of cognitive impairment and MDD, respectively. Furthermore, ensemble strategies combining top-performing models enhance detection accuracy (e.g., 90.04 % AUC for MDD detection in DAIC-WOZ). A critical contribution is the integration of explainable AI (XAI) techniques, such as SHAP values, to elucidate model decisions, fostering clinician trust by aligning feature importance with clinical markers such as lexical impoverishment.
The thesis underscores the viability of speech-based ML tools as scalable screening aids in healthcare, addressing ethical considerations through transparent methodologies. Future work will explore longitudinal monitoring and multimodal integration to refine diagnostic precision and adaptability across languages and demographics.La creciente prevalencia mundial de trastornos de salud mental, incluyendo el Trastorno Depresivo Mayor (TDM) y la Enfermedad de Alzheimer (EA), subraya la urgente necesidad de herramientas de diagnóstico accesibles y no invasivas. El habla, como biomarcador que integra información cognitiva, emocional y fisiológica, ofrece una prometedora vía para la detección temprana. Esta tesis investiga soluciones de aprendizaje automático (AA) para automatizar la evaluación de estados emocionales y cognitivos mediante el análisis del habla, centrándose en el desarrollo de aplicaciones clínicas fiables. Utilizando un marco experimental diverso que comprende cinco conjuntos de datos (ADReSS, AcceXible-MCI, DAIC-WOZ, RADAR-MDD y Androids), este trabajo evalúa características paralingüísticas (p. ej., prosodia acústica, descriptores de bajo nivel, características autosupervisadas) y lingüísticas (p. ej., GloVe, diversidad léxica). Los hallazgos clave revelan que los modelos basados en Transformers, que aprovechan el contenido semántico, alcanzan un rendimiento de vanguardia en la detección del TDM y el deterioro cognitivo, mientras que las características acústicas autosupervisadas demuestran robustez interlingüística, especialmente en entornos de bajos recursos. Los marcos personalizados, dependientes del hablante, superan significativamente a los modelos generalizados, lo que destaca la importancia de adaptar los sistemas a la variabilidad individual en la expresión de los síntomas. El estudio identifica las proporciones sustantivo/adverbio y la valencia emocional como indicadores estadísticamente significativos de deterioro cognitivo y TDM, respectivamente. Además, las estrategias de conjunto que combinan los modelos de alto rendimiento mejoran la precisión de la detección (p. ej., 90,04 % de AUC para la detección del TDM en DAIC-WOZ). Una contribución fundamental es la integración de técnicas de IA explicable (XAI), como los valores SHAP, para dilucidar las decisiones sobre los modelos, lo que fomenta la confianza del profesional clínico al alinear la importancia de las características con marcadores clínicos como el empobrecimiento léxico. La tesis subraya la viabilidad de las herramientas de aprendizaje automático (ML) basadas en el habla como herramientas escalables de detección en el ámbito sanitario, abordando consideraciones éticas mediante metodologías transparentes. El trabajo futuro explorará el monitoreo longitudinal y la integración multimodal para mejorar la precisión diagnóstica y la adaptabilidad en diferentes idiomas y grupos demográficos.A crecente prevalencia global de trastornos de saúde mental, incluíndo o trastorno depresivo maior (MDD) e a enfermidade de Alzheimer (EA), subliña a necesidade urxente de ferramentas de diagnóstico accesibles e non invasivas. A fala, como biomarcador que integra información cognitiva, emocional e fisiolóxica, ofrece unha vía prometedora para a detección precoz. Esta tese investiga solucións de aprendizaxe automática (ML) para automatizar a avaliación dos estados emocionais e cognitivos mediante a análise da fala, centrándose no desenvolvemento de aplicacións clínicas fiables. Usando un marco experimental diverso que comprende cinco conxuntos de datos (ADReSS, AcceXible-MCI, DAIC-WOZ, RADAR-MDD e Androids), este traballo avalía características paralingüísticas (por exemplo, prosodia acústica, descritores de baixo nivel, características autosupervisadas) e lingüísticas (por exemplo, GloVe, diversidade léxica). Os principais descubrimentos revelan que os modelos baseados en Transformer, que aproveitan o contido semántico, conseguen un rendemento de última xeración na detección de MDD e o deterioro cognitivo, mentres que as funcións acústicas autosupervisadas demostran robustez multilingüe, especialmente en entornos con poucos recursos. Os marcos personalizados e dependentes do falante superan significativamente os modelos xeneralizados, destacando a importancia de adaptar os sistemas á variabilidade individual na expresión dos síntomas. O estudo identifica as proporcións nome/adverbio e a valencia emocional como indicadores estatisticamente significativos de deterioro cognitivo e MDD, respectivamente. Ademais, as estratexias de conxunto que combinan modelos de alto rendemento melloran a precisión da detección (por exemplo, 90,04 % AUC para a detección de MDD en DAIC-WOZ). Unha contribución fundamental é a integración de técnicas explicables de IA (XAI), como os valores SHAP, para dilucidar as decisións do modelo, fomentando a confianza do clínico aliñando a importancia das características con marcadores clínicos como o empobrecemento léxico. A tese subliña a viabilidade das ferramentas de ML baseadas na fala como axudas de cribado escalables na atención sanitaria, abordando consideracións éticas a través de metodoloxías transparentes. Os traballos futuros explorarán o seguimento lonxitudinal e a integración multimodal para refinar a precisión do diagnóstico e a adaptabilidade entre idiomas e datos demográficos.Xunta de Galicia | ED481A-2020/27
Validación de una versión informatizada de la Escala de Apoyo Social de Harter en preadolescentes y adolescentes
Get PDFThe aim was to adapt Harter’s Social Support Scale into a computerized format and validate it in a sample of preadolescents and adolescents. A
total of 654 students (9-17 years old) from schools in Galicia (Spain) participated. The adapted version features a Likert response scale and adjustments to reflect diverse family structures, thanks to its computerized format. Confirmatory factor analysis showed that, among the models tested, the
one composed of four interrelated factors (Perceived Social Support from Parents, Peers, Teachers, and Close Friends) provided the best fit to the
data. Nonetheless, the hierarchical model with a general perceived social support factor also showed a good fit. The scores obtained for each factor
demonstrated high internal consistency. The negative correlations found between the scores on the four social support dimensions and victimization
in face-to-face bullying support the criterion validity of the instrument. The tool is particularly relevant for assessing perceived social support and
identifying psychosocial needs during key transitions, such as from childhood to adolescence, when issues like bullying frequently arise.
Keywords: supportive relationships, instrument, evaluation, child and adolescent population, school bullyingValidación de una versión informatizada de la Escala de Apoyo Social de Harter en preadolescentes y adolescentes. El objetivo fue adaptar la
Escala de Apoyo Social de Harter a un formato informatizado y validarla en una muestra de preadolescentes y adolescentes. Participaron 654 estudiantes de 9 a 17 años de Galicia (España). La versión adaptada incluye una escala de respuesta tipo Likert y ajustes para reflejar la diversidad
familiar, gracias a su formato informatizado. El análisis factorial confirmatorio mostró que, de los modelos puestos a prueba, el que mejor ajustó a
los datos fue el compuesto por cuatro factores interrelacionados (Apoyo Social Percibido de Padres, Compañeros/as, Profesores/as y Amigos/as
Íntimos/as). No obstante, el modelo jerárquico con un factor general de apoyo social percibido también mostró un buen ajuste. Las puntuaciones
obtenidas en cada factor presentaron una alta consistencia interna. Las correlaciones negativas halladas entre las puntuaciones en las cuatro dimensiones de apoyo social y la victimización en acoso escolar presencial respaldan la validez de criterio de la prueba. Su aplicación puede resultar
especialmente útil para evaluar el apoyo social percibido y detectar necesidades psicosociales en etapas críticas como la transición de la infancia
a la adolescencia, donde emergen problemáticas como el acoso escolar
Expression of a novel D4 Dopamine receptor in the lamprey brain. Evolutionary considerations about dopamine receptors
Get PDFNumerous data reported in lampreys, which belong to the phylogenetically oldest branch of vertebrates, show that the dopaminergic system was already well developed at the dawn of vertebrate evolution. The expression of dopamine in the lamprey brain is well conserved when compared to other vertebrates, and this is also true for the D2 receptor. Additionally, the key role of dopamine in the striatum, modulating the excitability in the direct and indirect pathways through the D1 and D2 receptors, has also been recently reported in these animals. The moment of divergence regarding the two whole genome duplications occurred in vertebrates suggests that additional receptors, apart from the D1 and D2 previously reported, could be present in lampreys. We used in situ hybridization to characterize the expression of a novel dopamine receptor, which we have identified as a D4 receptor according to the phylogenetic analysis. The D4 receptor shows in the sea lamprey a more restricted expression pattern than the D2 subtype, as reported in mammals. Its main expression areas are the striatum, lateral and ventral pallial sectors, several hypothalamic regions, habenula, and mesencephalic and rhombencephalic motoneurons. Some expression areas are well conserved through vertebrate evolution, as is the case of the striatum or the habenula, but the controversies regarding the D4 receptor expression in other vertebrates hampers for a complete comparison, especially in rhombencephalic regions. Our results further support that the dopaminergic system in vertebrates is well conserved and suggest that at least some functions of the D4 receptor were already present before the divergence of lampreys.Ministerio de Educación y Ciencia | Ref. BFU2006-14127Ministerio de Ciencia e Innovación | Ref. BFU2009-13369Universidade de Vig