Search Results (161)

Aspect Sentiment Triplet Extraction (ASTE) aims to jointly identify aspect terms, opinion terms, and their associated sentiment polarities. Existing approaches, such as tagging or span-based modeling, often struggle with complex aspect–opinion interactions and long-distance dependencies. We propose a Knowledge-Augmented Graph Convolutional Network (KMG-GCN) that represents a sentence as a multi-channel graph integrating syntactic dependencies, part-of-speech tags, and positional relations. An adjacency tensor is constructed via a biaffine attention mechanism, while a multi-anchor triplet learning strategy with orthogonal projection enhances representation disentanglement. Furthermore, a pairwise refinement module explicitly models aspect–opinion associations, improving robustness against overlapping triplets. Experiments on multiple benchmarks demonstrate that KMG-GCN achieves state-of-the-art performance with improved efficiency and generalization. Full article

Background: By exploring the principles of Gestalt psychology, the neural mechanisms of perception, and computational models, scientists aim to unravel the complex processes that enable us to perceive a coherent and organized world. This multidisciplinary approach continues to advance our understanding of how the brain constructs a perceptual world from sensory inputs. Objectives and Methods: This study investigates the nature of visual perception through an experimental paradigm and method based on a comparative analysis of human and artificial intelligence (AI) responses to a series of modified square images. We introduce the concept of a “phenomenal gradient” in human visual perception, where different attributes of an object are organized syntactically and hierarchically in terms of their perceptual salience. Results: Our findings reveal that human visual processing involves complex mechanisms including shape prioritization, causal inference, amodal completion, and the perception of visible invisibles. In contrast, AI responses, while geometrically precise, lack these sophisticated interpretative capabilities. These differences highlight the richness of human visual cognition and the current limitations of model-generated descriptions in capturing causal, completion-based, and context-dependent inferences. The present work introduces the notion of a ‘phenomenal gradient’ as a descriptive framework and provides an initial comparative analysis that motivates testable hypotheses for future behavioral and computational studies, rather than direct claims about improving AI systems. Conclusions: By bridging phenomenology, information theory, and cognitive science, this research challenges existing paradigms and suggests a more integrated approach to studying visual consciousness. Full article

This entry introduces an integrated model of reading that situates the Lexical Quality Hypothesis (LQH) within the Reading Systems Framework (RSF). The LQH posits that skilled reading depends on high-quality lexical representations—precise and flexible mappings of orthographic, phonological, morpho-syntactic, and semantic features—stored in the mental lexicon. These representations facilitate automatic word identification, accurate meaning retrieval, and efficient word-to-text integration (WTI), forming the foundation of text comprehension. Extending this micro-level perspective, the RSF positions lexical quality (LQ) within a macro-level cognitive architecture where the lexicon bridges word identification and reading comprehension systems. The RSF integrates multiple knowledge systems (linguistic, orthographic, and general world knowledge) with higher-order processes (sentence parsing, inference generation, comprehension monitoring, and situation model construction), emphasizing the bidirectional interactions between lower-level lexical knowledge and higher-order text comprehension. Central to this model is WTI, a dynamic mechanism through which lexical representations are incrementally incorporated into a coherent mental model of the text. This integrated model carries important implications for theory refinement, empirical investigation, and evidence-based instructional practices. Full article

This paper presents a hybrid architecture for automatic sentiment analysis of Kazakh-language political discourse. The Kazakh language is characterized by an agglutinative structure, a complex word-formation system, and the limited availability of digital resources, which significantly complicates the application of standard neural network approaches. To account for these characteristics, a multi-level system was developed that combines morphological and syntactic analysis rules, ontological relationships between political concepts, and multilingual representations of the XLM-R model, used in zero-shot mode. A corpus of 12,000 sentences was annotated for sentiment polarity and used for training and evaluation, while Universal Dependencies annotation was applied for morpho-syntactic analysis. Rule-based components compensate for errors related to affixation variability, modality, and directive constructions. An ontology comprising over 300 domain concepts ensures the correct interpretation of set expressions, terms, and political actors. Experimental results show that the proposed hybrid architecture outperforms both neural network baseline models and purely rule-based solutions, achieving Macro-F1 = 0.81. Ablation revealed that the contribution of modules is unevenly distributed: the ontology provides +0.04 to Macro-F1, the UD syntax +0.08, and the rule-based module +0.11. The developed system forms an interpretable and robust assessment of tonality, emotions, and discursive strategies in political discourse, and also creates a basis for further expansion of the corpus, additional training of models, and the application of hybrid methods to other tasks of analyzing low-resource languages. Full article

With the increasing complexity of software systems, the number of security vulnerabilities contained within software has risen accordingly. The existing shift-left security concept aims to detect and fix vulnerabilities during the software development cycle. While CodeQL stands as the premier static code analysis tool currently available on the market, its high barrier to entry poses challenges for meeting the implementation requirements of shift-left security initiatives. While large language model (LLM) offers potential assistance in QL code development, the inherent complexity of code generation tasks often leads to persistent issues such as syntactic inaccuracies and references to non-existent modules, which consequently constrains their practical applicability in this domain. To address these challenges, this paper proposes CQLLM (CodeQL-enhanced Large Language Model), a novel framework for automating the generation of CodeQL security vulnerability detection code by leveraging LLM. This framework is designed to enhance both the efficiency and the accuracy of automated QL code generation, thereby advancing static code analysis for a more efficient and intelligent paradigm for vulnerability detection. First, retrieval-augmented generation (RAG) is employed to search the vector database for dependency libraries and code snippets that are highly similar to the user’s input, thereby constraining the model’s generation process and preventing the import of invalid modules. Then, the user input and the knowledge chunks retrieved by RAG are fed into a fine-tuned LLM to perform reasoning and generate QL code. By integrating external knowledge bases with the large model, the framework enhances the correctness and completeness of the generated code. Experimental results show that CQLLM significantly improves the executability of the generated QL code, with the execution success rate improving from 0.31% to 72.48%, outperforming the original model by a large margin. Meanwhile, CQLLM also enhances the effectiveness of the generated results, achieving a CWE (Common Weakness Enumeration) coverage rate of 57.4% in vulnerability detection tasks, demonstrating its practical applicability in real-world vulnerability detection. Full article

This study introduces a multi-mechanism entity–relation extraction model designed to address persistent challenges in natural language processing, including syntactic complexity, long-range dependency modeling, and suboptimal utilization of contextual information. The proposed architecture integrates several complementary components. First, a pre-trained Chinese-RoBERTa-wwm-ext encoder with a whole-word masking strategy is employed to preserve lexical semantics and enhance contextual representations for multi-character Chinese text. Second, BiLSTM-based sequential modeling is incorporated to capture bidirectional contextual dependencies, facilitating the identification of distant entity relations. Third, the combination of multi-head attention and gated attention mechanisms enables the model to selectively emphasize salient semantic cues while suppressing irrelevant information. To further improve global prediction consistency, a Conditional Random Field (CRF) layer is applied at the output stage. Building upon this multi-mechanism framework, an adaptive dynamic network is introduced to enable input-dependent activation of feature modeling modules based on sentence-level semantic complexity. Rather than enforcing a fixed computation pipeline, the proposed mechanism supports flexible and context-aware feature interaction, allowing the model to better accommodate heterogeneous sentence structures. Experimental results on benchmark datasets demonstrate that the proposed approach achieves strong extraction performance and improved robustness, making it a flexible solution for downstream applications such as knowledge graph construction and semantic information retrieval. Full article

This paper investigates the nature of adjunct-inclusive interpretations in Japanese, which has long been debated in the literature. Previous studies have disagreed on whether these interpretations arise from V-stranding VP-ellipsis or adjunct ellipsis. This study argues that adjunct-inclusive interpretations fall into two distinct types: one semantically encoded and structurally represented, and another pragmatically inferred, depending on context beyond sentence structure. Using anaphoric expressions and negation as diagnostics, this study shows that adjunct-inclusive interpretations involving (i) omission of both adjunct and object in transitive sentences and (ii) adjunct omission in intransitive sentences are syntactically represented, supporting the existence of V-stranding VP-ellipsis. By contrast, adjunct-inclusive interpretations where only the adjunct is omitted and the object is contrastively focused are derived from pragmatic inference via free pragmatic enrichment, rather than from syntactic structure. These findings provide empirical and theoretical support for the view that Japanese does not allow syntactic adjunct ellipsis but does allow V-stranding VP-ellipsis. More broadly, this study contributes to the understanding of the syntax–pragmatics interface in ellipsis, showing that not all implicit interpretations reflect syntactic structure and highlighting the importance of carefully distinguishing between semantic and pragmatic sources in analyzing ellipsis phenomena. Full article

Topic modeling of large news streams is widely used to reconstruct economic and political narratives, which requires coherent topics with low lexical overlap while remaining interpretable to domain experts. We propose TF-SYN-NER-Rel, a structural–semantic term weighting scheme that extends classical TF-IDF by integrating positional, syntactic, factual, and named-entity coefficients derived from morphosyntactic and dependency parses of Russian news texts. The method is embedded into a standard Latent Dirichlet Allocation (LDA) pipeline and evaluated on a large Russian-language news corpus from the online archive of Moskovsky Komsomolets (over 600,000 documents), with political, financial, and sports subsets obtained via dictionary-based expert labeling. For each subset, TF-SYN-NER-Rel is compared with standard TF-IDF under identical LDA settings, and topic quality is assessed using the C_v coherence metric. To assess robustness, we repeat model training across multiple random initializations and report aggregate coherence statistics. Quantitative results show that TF-SYN-NER-Rel improves coherence and yields smoother, more stable coherence curves across the number of topics. Qualitative analysis indicates reduced lexical overlap between topics and clearer separation of event-centered and institutional themes, especially in political and financial news. Overall, the proposed pipeline relies on CPU-based NLP tools and sparse linear algebra, providing a computationally lightweight and interpretable complement to embedding- and LLM-based topic modeling in large-scale news monitoring. Full article

This study, based on naturalistic discourse in Mano and on both morphosyntactic and prosodic characteristics, analyses the Mano constructions formed with the marker lɛ́, including the identifying construction, referent introduction, focus, relativization, and hanging topic. While the identifying construction can be treated as a separate predication, and lɛ́ within it as a predicator, in all the other constructions lɛ́ does not have a predicative function. For Mano lɛ́, we suggest an invariant function instead, that of attention shift. Depending on both the structural and the pragmatic grounds, attention shift can be interpreted as having a predicative or a non-predicative function. We finally suggest that mapping recurrent constructions on interactants’ actions requires no definition of the notion of “clausehood”: NP-based constructions can be deployed for performing a communicatively self-sufficient action of an attention shift. This would present them as “clausal” in a speech-act-based analysis, and non-clausal from the perspective that defines clauses as subject–predicate structures—but this question does not arise in our approach that links syntactic structures to communicative action. The analysis is nested in the approach to polysemy as a “family of constructions” and to information structure as diverse interpretive effects, rather than a closed set of discrete universal categories. Full article

This study investigates how large language models (LLMs) handle Chinese zero anaphora under symmetric minimal-pair conditions designed to neutralize shallow syntactic cues. We construct a Winograd-style benchmark of carefully controlled sentence pairs that require semantic interpretation, pragmatic inference, discourse tracking, and commonsense reasoning rather than structural heuristics. Using GPT-4, ChatGLM-4, and LLaMA-3 under zero-shot, one-shot, and few-shot prompting, we assess both accuracy and the reasoning traces generated through a standardized Chain-of-Thought diagnostic. Results show that all models perform consistently on items solvable through local cues but display systematic asymmetric errors on 19 universally misinterpreted sentences that demand deeper discourse reasoning. Analysis of these failures reveals weaknesses in semantic role differentiation, topic-chain maintenance, logical-relation interpretation, pragmatic inference, and long-distance dependency tracking. These findings suggest that while LLMs perform well on simpler tasks, they still face challenges in interpreting contextually omitted arguments in Chinese. The study provides a new controlled evaluation resource, an interpretable error analysis framework, and evidence of differences in symmetric versus asymmetric reasoning behaviors in LLMs. Future research could expand the current benchmark to longer discourse contexts, incorporate multi-modal or knowledge-grounded cues, and explore fine-tuning LLMs on discourse data, helping clarify whether asymmetric patterns stem from deeper reasoning challenges or from interactions between models and the evaluation format. Full article

Constructed-response items offer rich evidence of writing proficiency, but the linguistic signals they contain vary with grade level. This study presents a cross-sectional analysis of 5638 English Language Arts essays from Grades 6–12 to identify which linguistic features predict proficiency and to characterize how their importance shifts across grade levels. We extracted a suite of lexical, syntactic, and semantic-cohesion features, and evaluated their predictive power using an interpretive dual-model framework combining LASSO and XGBoost algorithms. Feature importance was assessed through LASSO coefficients, XGBoost Gain scores, and SHAP values, and interpreted by isolating both consensus and divergences of the three metrics. Results show moderate, generalizable predictive signals in Grades 6–8, but no generalizable predictive power was found in the Grades 9–12 cohort. Across the middle grades, three findings achieved strong consensus. Essay length, syntactic density, and global semantic organization served as strong predictors of writing proficiency. Lexical diversity emerged as a key divergent feature, it was a top predictor for XGBoost but ignored by LASSO, suggesting its contribution depends on interactions with other features. These findings inform actionable, grade-sensitive feedback, highlighting stable, diagnostic targets for middle school while cautioning that discourse-level features are necessary to model high-school writing. Full article

Traditional degree semantics approaches have aimed to pin down the inherent class of adjectives. This paper presents a novel dynamic perspective, where the classification of an adjective is dynamic and syntactically dependent. Using measurement theory and fuzzy set analysis, the proposed framework defines dynamic patterns of adjective classes with a set of axioms and integrates these patterns with syntactic structures to explain the flexibility and constraints observed in adjectival expressions. Employing Mandarin data, the paper illustrates how different syntactic constructions select specific adjective classes, thereby affecting their distribution and interpretation. This approach not only accommodates cross-linguistic variations but also provides a more comprehensive understanding of the semantics of adjectives. Full article

The automatic construction of learners’ dictionaries requires robust methods for identifying non-literal word combinations, or collocations, which represent a significant challenge for second-language (L2) learners. This paper addresses the critical initial step of accurately extracting collocation candidates from corpora to build a learner’s dictionary for Italian. The adopted method and the implemented application are significant for learning the Italian language. We present a comparative study of three methodologies for identifying these candidates within a 41.7-million-word Italian corpus: a Part-Of-Speech-based approach, a syntactic dependency-based approach, and a novel Hybrid method that integrates both. The analysis yielded 2,097,595 potential collocations. Results indicate that the Hybrid method achieves superior performance in terms of Recall and Benchmark Match, identifying the most significant portion of candidates, 42.35% of the total. We conducted an in-depth analysis to refine the extracted dataset, calculating multiple statistical metrics for each candidate, which are described in detail in the paper. Such analysis allows for the classification of collocations by relevance, difficulty, and frequency of use, forming the basis for the future development of a high-quality, web-based dictionary tailored to the proficiency levels of Italian learners. Full article

As software systems grow increasingly complex and interconnected, detecting vulnerabilities in source code has become a critical and challenging task. Traditional static analysis methods often fall short in capturing deep, context-dependent vulnerabilities and adapting to rapidly evolving threat landscapes. Recent efforts have explored knowledge graphs and transformer-based models to enhance semantic understanding; however, these solutions frequently rely on static knowledge bases, exhibit high computational overhead, and lack adaptability to emerging threats. To address these limitations, we propose DynaKG-NER++, a novel and lightweight framework for context-aware vulnerability detection in source code. Our approach integrates lexical, syntactic, and semantic features using a transformer-based token encoder, dynamic knowledge graph embeddings, and a Graph Attention Network (GAT). We further introduce contrastive learning on vulnerability–patch pairs to improve discriminative capacity and design an attention-based fusion module to combine token and entity representations adaptively. A key innovation of our method is the dynamic construction and continual update of the knowledge graph, allowing the model to incorporate newly published CVEs and evolving relationships without retraining. We evaluate DynaKG-NER++ on five benchmark datasets, demonstrating superior performance across span-level F1 (89.3%), token-level accuracy (93.2%), and AUC-ROC (0.936), while achieving the lowest false positive rate (5.1%) among state-of-the-art baselines. Sta tistical significance tests confirm that these improvements are robust and meaningful. Overall, DynaKG-NER++ establishes a new standard in vulnerability detection, balancing accuracy, adaptability, and efficiency, making it highly suitable for deployment in real-world static analysis pipelines and resource-constrained environments. Full article

This study investigated how Japanese speakers interpret structurally ambiguous wh-questions, testing whether filler–gap resolution is guided by syntactic resolution based on hierarchical structure or linear locality based on surface word order. We combined behavioral key-press responses with fine-grained eye-tracking data and applied cluster-based permutation analysis to capture the moment-by-moment time course of syntactic interpretation as sentences were processed in real time. Key-press responses revealed a preference for resolving the dependency at the main clause (MC) gap position. Eye-tracking data showed early predictive fixations to the MC picture, followed by shifts to the embedded clause (EC) picture as the embedded event was described. These shifts occurred prior to the appearance of syntactic cues that signal the presence of an EC structure, such as the complementizer -to, and were therefore most likely guided by referential alignment with the linguistic input rather than by syntactic reanalysis. A subsequent return of the gaze to the MC picture occurred when the clause-final question particle -ka became available, confirming the interrogative use of the wh-phrase. Both key-press and eye-tracking data showed that participants did not commit to the first grammatically available EC interpretation but instead waited until clause-final particle information confirmed the interrogative use of the wh-phrase, ultimately favoring the MC interpretation. This pattern supports the view that filler–gap resolution is guided by structural locality rather than linear locality. By using high-resolution temporal data and statistically robust analytic techniques, this study demonstrates that Japanese comprehenders engage in predictive yet structurally cautious parsing. These findings challenge earlier claims that filler–gap resolution in Japanese is primarily driven by linear locality and instead showed a preference for resolving dependencies at the structurally higher MC position, consistent with parsing biases previously observed in English, despite typological differences in word order between the two languages. This preference also reflects sensitivity to language-specific morpho-syntactic cues in Japanese, such as clause-final particles. Full article