Semantic Coherence Research Articles

Semantic Web-Driven Targeted Adversarial Attack on Black Box Automatic Speech Recognition Systems

The susceptibility of Deep Neural Networks (DNNs) to adversarial attacks in Automatic Speech Recognition (ASR) systems has drawn significant attention. Most work focuses on white-box methods, but the assumption of full transparency of model architecture and parameters is unrealistic in real-world scenarios. Although several targeted black-box attack methods have been proposed in recent years, due to the complexity of ASR systems, they primarily rely on query-based approaches with limited search capabilities, leading to low success rates and noticeable noise. To address this, we propose DE-gradient, a new black-box approach using differential evolution (DE), a population-based search algorithm. Inspired by Semantic Web ideas, we introduce modulation noise to preserve semantic coherence while enhancing imperceptibility. In experiments on two public datasets, DE-gradient improved attack success rates by 19% and increased the signal-to-noise ratio (SNR) of silent parts from 27 dB to 54 dB, establishing a strong baseline for evaluating black-box adversarial attacks in ASR systems.

Memory-enhanced hierarchical transformer for video paragraph captioning

Video paragraph captioning aims to describe a video that contains multiple events with a paragraph of generated coherent sentences. Such a captioning task is full of challenges since the high requirements for visual-textual relevance and semantic coherence across the captioning paragraph of a video. In this work, we introduce a memory-enhanced hierarchical transformer for video paragraph captioning. Our model adopts a hierarchical structure, where the outer layer transformer extracts visual information from a global perspective and captures the relevancy between event segments throughout the entire video, while the inner layer transformer further mines local details within each event segment. By thoroughly exploring both global and local visual information at the video and event levels, our model can provide comprehensive visual feature cues for promising paragraph caption generation. Additionally, we design a memory module to capture similar patterns among event segments within a video, which preserves contextual information across event segments and updates its memory state accordingly. Experimental results on two popular datasets, ActivityNet Captions and YouCook2, demonstrate that our proposed model can achieve superior performance, generating higher quality caption while maintaining consistency in the content of video.

Intelligent Abstractive Summarization of Scholarly Publications with Transfer Learning

Intelligent abstractive text summarization of scholarly publications refers to machine-generated summaries that capture the essential ideas of an article while maintaining semantic coherence and grammatical accuracy. As information continues to grow at an overwhelming rate, text summarization has emerged as a critical area of research. In the past, summarization of scientific publications predominantly relied on extractive methods. These approaches involve selecting key sentences or phrases directly from the original document to create a summary or generate a suitable title. Although extractive methods preserve the original wording, they often lack the ability to produce a coherent, concise, and fluent summary, especially when dealing with complex or lengthy texts. In contrast, abstractive summarization represents a more sophisticated approach. Rather than extracting content from the source, abstractive models generate summaries using new language, often incorporating words and phrases not found in the original text. This allows for more natural, human-like summaries that better capture the key ideas in a fluid and cohesive manner. This study introduces two advanced models for generating titles from the abstracts of scientific articles. The first model employs a Gated Recurrent Unit (GRU) encoder coupled with a greedy-search decoder, while the second utilizes a Transformer model, known for its capacity to handle long-range dependencies in text. The findings demonstrate that both models outperform the baseline Long Short-Term Memory (LSTM) model in terms of efficiency and fluency. Specifically, the GRU model achieved a ROUGE-1 score of 0.2336, and the Transformer model scored 0.2881, significantly higher than the baseline LSTM model, which reported a ROUGE-1 score of 0.1033. These results underscore the potential of abstractive models to enhance the quality and accuracy of summarization in academic and scholarly contexts, offering more intuitive and meaningful summaries.

PFB-Diff: Progressive Feature Blending diffusion for text-driven image editing

Diffusion models have demonstrated their ability to generate diverse and high-quality images, sparking considerable interest in their potential for real image editing applications. However, existing diffusion-based approaches for local image editing often suffer from undesired artifacts due to the latent-level blending of the noised target images and diffusion latent variables, which lack the necessary semantics for maintaining image consistency. To address these issues, we propose PFB-Diff, a Progressive Feature Blending method for Diffusion-based image editing. Unlike previous methods, PFB-Diff seamlessly integrates text-guided generated content into the target image through multi-level feature blending. The rich semantics encoded in deep features and the progressive blending scheme from high to low levels ensure semantic coherence and high quality in edited images. Additionally, we introduce an attention masking mechanism in the cross-attention layers to confine the impact of specific words to desired regions, further improving the performance of background editing and multi-object replacement. PFB-Diff can effectively address various editing tasks, including object/background replacement and object attribute editing. Our method demonstrates its superior performance in terms of editing accuracy and image quality without the need for fine-tuning or training. Our implementation is available at https://github.com/CMACH508/PFB-Diff.

A linguistics-based approach to refining automatic intent detection in conversational agent design

In this paper, we propose Automatic Intent Detector (AID), a framework for automatic intent detection to facilitate the creation of a conversational agent. AID follows an eight-step process incorporating best practices from the current literature and introducing innovative approaches in certain steps. The most notable innovation within AID is the automatic labeling of clusters, which is based on detailed and sophisticated rules derived from linguistics. These rules focus on morphosyntactic analysis, while also taking into account an aspect of semantic role theory. Furthermore, as for the overall validation of the results obtained, it provides an approach based on the concepts of semantic coherence, variability, and label appropriateness. After describing AID at the technical level, we illustrate the experiments we conducted both on a dataset widely used as benchmark in the literature and on a real corporate dataset. Finally, we present a critical discussion on the results obtained.

A methodological exploration of the P-frame approach: Implications for developing phrase lists

The widespread availability of corpora and corpus tools has led to considerable advances in our understanding of phraseology and how it can be investigated. An increasingly influential approach to investigating phraseology is the retrieval and analysis of phrase frames (p-frames), recurrent word sequences with a variable slot. Although the p-frame approach is quite well-established, pedagogically oriented research in this area is still quite new, in particular the application of phrase frames to generate useful lists of phrases. This paper aims to contribute to methodological efforts in this direction by presenting a case study to illustrate decision-making throughout the process from corpus compilation to the generation of a final pedagogical list of phrases. Using a corpus of research article (RA) introductions in Health Sciences, this study shows how key decisions including p-frame length, frequency and range thresholds were arrived at. It also discusses exclusion criteria focusing on variability and predictability of p-frame fillers, and the fine-tuning of the resulting list, focusing on semantic coherence and pedagogical usefulness. Finally, we present the results of an initial evaluation of the list by stakeholders. The main contributions of the study to p-frame research methodology lie in the clarification of threshold settings, the potential contribution of variability and predictability to decision-making and the proposal of more in-depth concordance analysis of co-text to aid in the identification of phrases.

Efficient Headline Generation with Hybrid Attention for Long Texts

Headline generation aims to condense key information from an article or a document into a concise one-sentence summary. The Transformer structure is in general effective for such tasks, yet it suffers from a dramatic increase in training time and GPU consumption as the input text length grows. To address this problem, a hybrid attention mechanism is proposed. Both local and global semantic information among words are modeled in a way that significantly improves training efficiency, especially for long text. Effectiveness is not sacrificed; in fact, fluency and semantic coherence of the generated headlines are enhanced. Experimental results on an open benchmark dataset show that, compared to the baseline model’s best performance, the proposed model obtains a 14.7%, 16.7%, 14.4% and 9.1% increase in the F1 values of the ROUGE-1, the ROUGE-2, the ROUGE-L and the ROUGE-WE metrics, respectively. The semantic coherence of the generated text is also improved, as shown by a 2.8% improvement in the BERTScore’s F1 value. These results show that the effectiveness of the proposed headline generation model with the hybrid attention mechanism is also improved. The hybrid attention mechanism could provide references for relevant text generation tasks.

Observe finer to select better: Learning key frame extraction via semantic coherence for dynamic facial expression recognition in the wild

In dynamic facial expression sequences captured in natural environments, the number and distribution of expression peak frames (key frames) show significant variation within video sequences. Efficient extraction and utilization of these key expression frames are crucial for improving the accuracy of expression recognition. To address the above challenges, this paper proposed the Key Frame Extraction via Semantic Coherence (KFE-SC) method. The method explores the spatial-temporal emotional semantic information within sequences, employing a coarse-to-fine approach to extract key expression frames, thus facilitating classification tasks. By introduce a Spatial Semantic Coherence (SSC) module and designing a Global-Local Feature Aggregation (GLFA) unit, the method explores the spatial semantic correlations at global-local levels in images. Through calculating weight proportions, the method achieves a coarse-grained extraction of key frames in the spatial dimension. Considering that video sequences contain both spatial and temporal semantic relevance, KFE-SC further introduces a Temporal Semantic Coherence (TSC) module, which include the Temporal Alignment Aggregation (TA2) unit to align short and long-term temporal information and explore temporal semantic relevance within the sequences, enabling fine-grained extraction of key frames in the temporal dimension through weight calculations.

ECCT: Efficient Contrastive Clustering via Pseudo-Siamese Vision Transformer and Multi-view Augmentation

Image clustering aims to divide a set of unlabeled images into multiple clusters. Recently, clustering methods based on contrastive learning have attracted much attention due to their ability to learn discriminative feature representations. Nevertheless, existing clustering algorithms face challenges in capturing global information and preserving semantic continuity. Additionally, these methods often exhibit relatively singular feature distributions, limiting the full potential of contrastive learning in clustering. These problems can have a negative impact on the performance of image clustering. To address the above problems, we propose a deep clustering framework termed Efficient Contrastive Clustering via Pseudo-Siamese Vision Transformer and Multi-view Augmentation (ECCT). The core idea is to introduce Vision Transformer (ViT) to provide the global view, and improve it with Hilbert Patch Embedding (HPE) module to construct a new ViT branch. Finally, we fuse the features extracted from the two ViT branches to obtain both global view and semantic coherence. In addition, we employ multi-view random aggressive augmentation to broaden the feature distribution, enabling the model to learn more comprehensive and richer contrastive features. Our results on five datasets demonstrate that ECCT outperforms previous clustering methods. In particular, the ARI metric of ECCT on the STL-10 (ImageNet-Dogs) dataset is 0.852 (0.424), which is 10.3% (4.8%) higher than the best baseline.

Boundary-Aware Dual Biaffine Model for Sequential Sentence Classification in Biomedical Documents.

Assigning appropriate rhetorical roles, such as "background," "intervention," and "outcome," to sentences in biomedical documents can streamline the process for physicians to locate evidence and resources for medical treatment and decision-making. While sequence labeling and span-based methods are frequently employed for this task, the former disregards a document's semantic structure, resulting in a lack of semantic coherence across continuous sentences. Span-based approaches, on the other hand, either necessitate the enumeration of all potential spans, which can be time-consuming, or may lead to the misclassification of sentences over extended spans. Consequently, an approach is required that models the semantic structure of documents explicitly and captures boundary information to achieve precise and effective sentence labeling in biomedical documents. To address these challenges, we propose a new approach, the boundary-aware dual biaffine model, which explicitly models the semantic structure of documents and incorporates boundary information via a dual biaffine layer. We introduce a dynamic programming algorithm to minimize missing labels and overlapping predictions, and achieve globally optimal decoding results. We evaluate our approach on three benchmark datasets, namely PubMed 20 k RCT, PubMed-PICO and NICTA-PIBOSO. The experimental results demonstrate that our approach outperforms strong baselines and achieves state-of-the-art performance on PubMed 20 k RCT and PubMed-PICO. Additionally, our method also achieves competitive results on NICTA-PIBOSO.

Enhancing Task-Oriented Dialogue Modeling through Coreference-Enhanced Contrastive Pre-Training

Pre-trained language models (PLMs) are proficient at understanding context in plain text but often struggle with the nuanced linguistics of task-oriented dialogues. The information exchanges in dialogues and the dynamic role-shifting of speakers contribute to complex coreference and interlinking phenomena across multi-turn interactions. To address these challenges, we propose Coreference-Enhanced Contrastive Pre-training (CECPT), an innovative pre-training framework specifically designed to enhance dialogue modeling. CECPT utilizes unsupervised dialogue datasets to capture both semantic richness and structural coherence. Our experimental results demonstrate that the CECPT model significantly outperforms established baselines in three critical applications: intent recognition, dialogue act prediction, and dialogue state tracking. These findings suggest that CECPT is more adept at following the information flow within dialogues and accurately linking statuses to their respective references.

Event-Oriented State Alignment Network for Weakly Supervised Temporal Language Grounding.

Weakly supervised temporal language grounding (TLG) aims to locate events in untrimmed videos based on natural language queries without temporal annotations, necessitating a deep understanding of semantic context across both video and text modalities. Existing methods often focus on simple correlations between query phrases and isolated video segments, neglecting the event-oriented semantic coherence and consistency required for accurate temporal grounding. This can lead to misleading results due to partial frame correlations. To address these limitations, we propose the Event-oriented State Alignment Network (ESAN), which constructs "start-event-end" semantic state sets for both textual and video data. ESAN employs relative entropy for cross-modal alignment through knowledge distillation from pre-trained large models, thereby enhancing semantic coherence within each modality and ensuring consistency across modalities. Our approach leverages vision-language models to extract static frame semantics and large language models to capture dynamic semantic changes, facilitating a more comprehensive understanding of events. Experiments conducted on two benchmark datasets demonstrate that ESAN significantly outperforms existing methods. By reducing false high correlations and improving the overall performance, our method effectively addresses the challenges posed by previous approaches. These advancements highlight the potential of ESAN to improve the precision and reliability of temporal language grounding tasks.

Pre-Processing of Categorical Features Within Medical Analysis Systems.

The complexity of the cancer problem domain presents challenges not only to the medical analysis systems tasked with its analysis, but also to the users of such systems. While it is desirable to assist users in operating these medical analysis systems, prior groundwork is required before this can be achieved, such as recognising patterns in the way users create certain analyses within these systems. In this paper, we use machine learning algorithms to analyse user behaviour patterns and attempt to predict the next user interaction within the CARESS medical analysis system. Since an appropriate pre-processing scheme is essential for the performance of these algorithms, we propose the usage of a Natural Language Processing (NLP)- inspired approach to preserve some semantic cohesion of the mostly categorical features of these user interactions. Furthermore, we propose to use a sliding window that contains information about the latest user interactions in combination with Latent Dirichlet Allocation (LDA) to extract a latent topic from these last interactions and use it as additional input to the machine learning models. We compare this pre-processing scheme with other approaches that utilise one-hot encoding and feature hashing. The results of our experiments show that the sliding window LDA scheme is a promising solution, that performs better for our use case than the other evaluated pre-processing schemes. Overall, our results provide an important piece for further research and development in the area of assisting users in operating analysis systems in complex problem domains.

Differences in the structure of scientific discourse with individual and collaborative authorship of publications

Within the framework of scientific discourse, the article occupies a central place in the construction and dissemination of knowledge in the scientific community. At the same time, collaborative authorship is becoming increasingly widespread in the scientific community. The syntactic and semantic differences between the texts of collaborations and individual authors present an urgent problem and remain poorly understood. This determines the purpose of the study — to identify differences in the structure of scientific discourse with individual and collaborative authorship of publications. At the preliminary stage of the study, an unstructured research interview was conducted on the specifics of writing scientific articles individually and as part of author teams. The main research is devoted to the analysis of the semantic and syntactic structure of texts using the method of relational situational analysis (RSA), which allows you to automatically identify the correspondence of the syntaxemic structure of a sentence to the logical structure of the actions described in this sentence. The study compared the texts of 201 scientific articles in Russian from the leading psychological journals included in WoS and Scopus, divided into corpora by type of authorship (individual-collaborative) and genres (theoretical articles and reviews vs empirical methodological articles). During the study, it was revealed that the articles written in collaboration have higher lexical coherence, they are more syntactically correct. The volume of the text of a scientific article on psychology is on average 9% more with individual authorship compared with collaborative writing. The semantic and syntactic differences between the texts of articles by individual authors and collaborations have a genre differentiation. In theoretical articles and reviews of collaborations, the syntactic complexity and semantic coherence of the text are higher, the traditional structure of the statement prevails: the thesis is briefly postulated and then fully justified. With individual authorship, the main idea is more often formulated after the context, while reformulation is more often used in the texts of individual authors. The meta-discourse in theoretical articles also differs significantly: the author’s position is more represented by individual authors, and “reader involvement” is in the case of co-authorship. In empirical and methodological articles, the main differences are revealed in the specifics of meta-discourse: the author’s position, unlike the genre of theoretical articles, is more inherent in the texts of collaborations.

Typy danych a koherencja semantyczna definicji kognitywnej (przykład polskiej ojczyzny)

The central concept of cultural linguistics (ethnolinguistics) – regardless of its dialectological, etymological, onomastic or cognitive orientation – is the linguistic worldview (LWV). In synthesizing ethnolinguistic publications (dictionaries or lexicons), while reconstructing “portions” of LWV (stereotypes, terms, concepts, culturemes, keywords), linguistic data (systemic, text-based, surveys) and co-linguistic data are taken into account. The former are especially used in analyses based on “national” and “comparative” ethnolinguistics (cf. e.g. Axiological Lexicon of Slavs and their Neighbors, LASiS 2015–2019); both the former and the latter are used within the framework of “folk” ethnolinguistics (cf. Dictionary of Folk Stereotypes and Symbols, SSiSL 1996–2022). The type of ethnolinguistically relevant data usually goes hand in hand with the types of descriptions: separative (LASiS) or holistic (SSiSL). This article presents the concept of the Polish homeland, reconstructed on the basis of both linguistic and co-linguistic data (as in SSiSL). Questions are asked concerning the semantic coherence of the cognitive definition of homeland, i.e., which characteristics of homeland are present in which types of data and how coherent are the judgments about homeland contained in those different types.

Development of a Brief Assessment Tool to Evaluate Early Low Nutrition Risk in Community Elderly. Creation of the tool and examination of its reliability and criterion-related validity.

To date simple assessment tool to evaluate early low nutrition risk in general older population has not been available. This study aimed to create such tool and examined its reliability and criterion-related validity. 1,192 community elderly with a mean age of 74.7(5.8) years responded to a questionnaire consisting of 48 (Hatoyama) or 34 items (Kusatsu), which have been reported to be associated with nutritional state in older people. Item analysis was conducted on the 34 common items, and items were selected based on the following criteria: adequate pass rates and discriminative power, no gender and regional differences, and a certain level of commonality based on factor analysis. Next, the factor structure of the candidate items was examined through exploratory factor analysis, and confirmatory factor analysis was conducted as the final scale structure. Furthermore, Spearman's partial rank correlation coefficients (sex- and age-adjusted) between the created index and important health indicators were examined to determine the criterion-related validity. Finally, we obtained a semantic coherence of 4 factors (named health beliefs, dietary status, physical activity, and food-related quality of life) totaling 13 items; confirmatory factor analysis of the 4-factor solution yielded good model fit values, χ2 (59) =275.4 (p<0.001), CFI=0.930, and RMSEA=0.056. The factor loadings for each factor ranged from 0.43 to 0.82, indicating adequate loadings. The reliability of the index was shown to be high by Good-Poor analysis and Cronbach's α. The index showed statistically significant correlations with all health indicators. We have developed a simple assessment tool to evaluate early low nutrition risk in general older population.

Natural language processing in at-risk mental states: enhancing the assessment of thought disorders and psychotic traits with semantic dynamics and graph theory.

Verbal communication has key information for mental health evaluation. Researchers have linked psychopathology phenomena to some of their counterparts in natural-language-processing (NLP). We study the characterization of subtle impairments presented in early stages of psychosis, developing new analysis techniques and a comprehensive map associating NLP features with the full range of clinical presentation. We used NLP to assess elicited and free-speech of 60 individuals in at-risk-mental-states (ARMS) and 73 controls, screened from 4,500 quota-sampled Portuguese speaking citizens in Sao Paulo, Brazil. Psychotic symptoms were independently assessed with Structured-Interview-for-Psychosis-Risk-Syndromes (SIPS). Speech features (e.g.sentiments, semantic coherence), including novel ones, were correlated with psychotic traits (Spearman's-ρ) and ARMS status (general linear models and machine-learning ensembles). NLP features were informative inputs for classification, which presented 86% balanced accuracy. The NLP features brought forth (e.g. Semantic laminarity as 'perseveration', Semantic recurrence time as 'circumstantiality', average centrality in word repetition graphs) carried most information and also presented direct correlations with psychotic symptoms. Out of the standard measures, grammatical tagging (e.g. use of adjectives) was the most relevant. Subtle speech impairments can be grasped by sensitive methods and used for ARMS screening. We sketch a blueprint for speech-based evaluation, pairing features to standard thought disorder psychometric items.

Análise de rede sociais de autores e instituições participantes nos anais do WIDAT 2017 a 2023 por coautoria e palavras-chave

The Information, Data and Technology Workshop (WIDaT), with its six editions from 2017 to 2023, published 210 articles in the proceedings with the participation of 395 authors linked to 80 institutions based in 46 cities and 7 countries. The objective of the research is to reveal affinity groups between authors and institutions participating in WIDaT. Qualitative research of an applied nature, used information visualization techniques with the support of social network analysis methods supported by software, such as: OpenRefine, Gephi and Looker Studio. The main result was the analysis of social networks of authors and institutions participating in the six events. The network of authors by co-authorship showed a giant component composed of the majority of authors, while the network of authors by keywords amplified the density of connections, making it a 'small world'. The network of co-authorship institutions revealed strong collaboration between institutions. The network of institutions using keywords identified seven clusters, complemented by keyword clouds, which revealed a semantic cohesion of the topics covered by the institutions belonging to each cluster. Observing the most common keywords for a given cluster can motivate approaches to research collaboration, both for those already consolidated and to strengthen the development of new investigations and studies.

SAC-GAN: Structure-Aware Image Composition.

We introduce an end-to-end learning framework for image-to-image composition, aiming to plausibly compose an object represented as a cropped patch from an object image into a background scene image. As our approach emphasizes more on semantic and structural coherence of the composed images, rather than their pixel-level RGB accuracies, we tailor the input and output of our network with structure-aware features and design our network losses accordingly, with ground truth established in a self-supervised setting through the object cropping. Specifically, our network takes the semantic layout features from the input scene image, features encoded from the edges and silhouette in the input object patch, as well as a latent code as inputs, and generates a 2D spatial affine transform defining the translation and scaling of the object patch. The learned parameters are further fed into a differentiable spatial transformer network to transform the object patch into the target image, where our model is trained adversarially using an affine transform discriminator and a layout discriminator. We evaluate our network, coined SAC-GAN, for various image composition scenarios in terms of quality, composability, and generalizability of the composite images. Comparisons are made to state-of-the-art alternatives, including Instance Insertion, ST-GAN, CompGAN and PlaceNet, confirming superiority of our method.

Coh-Metrix 시스템을 통한 고등학교 영어 모의시험 독해지문의 학년 내 연계성 분석

This corpus-driven research aims to examine the intra-grade continuity of reading passages within high school English mock exams, specifically the national academic achievement tests for seniors in 2023. These tests were conducted quarterly by four regional educational offices: Seoul in March, Incheon in April, Gyeonggi Province in July, and Seoul in October. A corpus of 100 reading passages, 25 from each test, was analyzed via a range of Coh-Metrix measures, including word features, lexical diversity, connectives, readability, referential cohesion, syntactic complexity, semantic cohesion, and so on. These Coh-metrix results demonstrated that there was no compelling evidence to support strong intra-grade continuity, highlighting the necessity for meticulous calibration of text difficulty, which should incrementally escalate with successive test administrations. The analysis of basic metrics, lexical frequency and diversity, syntactic complexity, and readability revealed no consistent trends across test administrations. Additionally, lexical attributes such as concreteness, familiarity, imagery, and age of acquisition did not show significant continuity. Moreover, the analysis of syntactic complexity also showed no consistent pattern. The current inconsistency in text difficulty suggests that students may not be receiving the progressive challenge needed to develop their reading comprehension skills effectively. Drawing upon these findings, theoretical and pedagogical implications are elucidated.