Bridging modalities: a unified framework for textual and multimodal dialogue discourse parsing

The increasing complexity of inter-hospital patient transfers, combined with the widespread use of heterogeneous and unstructured medical documentation, has intensified challenges related to data fragmentation, delayed clinical decision-making, and inefficient care coordination. Existing healthcare information systems and CRM platforms often operate in silos, limiting their ability to provide timely, standardized, and actionable patient insights. To address these limitations, this paper proposes a Salesforce Health Cloud–centric AI framework, where Salesforce Health Cloud acts as the primary clinical CRM platform orchestrating patient transfer workflows, while AI modules provide embedded document intelligence and predictive decision support. The proposed architecture combines an Adaptive Document Parsing and Structuring (ADPS) Framework for layout-aware document understanding, a Context-Aware Clinical Summarization (CACS) Engine for ontology-guided extraction of medically significant information, a FHIR-Compliant Interoperability Integration (FCI) Layer for standardized data exchange, and a Predictive Admission Intelligence Module (PAIM) for proactive triage and risk assessment. By unifying document intelligence, interoperability, CRM workflow automation, and machine learning–based prediction into a single end-to-end pipeline, the framework converts patient transfer documentation from a manual, error-prone process into an automated, decision-support-driven workflow. Experimental evaluation on real-world inter-hospital transfer datasets demonstrates high document structuring accuracy (94.2%), clinically relevant summarization performance (96.8%), substantial reductions in admission processing time (from 38 to 11 minutes), and strong predictive accuracy (97.1%) with reliable risk discrimination. The results confirm that the proposed approach enhances clinical efficiency, reduces administrative burden, and enables CRM platforms to function as active participants in clinical decision-making, establishing a scalable and intelligent foundation for digital transformation in healthcare transfer management.

Video Question-Answering (VideoQA) enables machines to interpret and respond to complex video content, advanc ing human-computer interaction. However, existing multimodal large language models (MLLMs) often provide incomplete or opaque explanations and existing benchmarks mainly focus on the correction of final answers, limiting insight into their reasoning processes and hindering both transparency and verifiability. To address this gap, we propose the Question Parsing, Video Alignment and Answer Aggregation framework (QPVA3), which leverages a compositional graph to drive visual and logical reasoning in VideoQA. Specifically, QPVA3 consists of three core components, the planner, executor, and reasoner to generate the compositional graph and conduct graph-driven reasoning. For the original question, the planner parses it into the compositional graph, capturing the underlying reasoning logic and structuring it into a series of interconnected questions. For each question in compositional graph, the executor aligns the video by selecting relevant video clips and generates answers, ensuring accurate, context-specific responses. For each question with its first-order descents, the reasoner aggregates answers by integrating rea soning logic with visual evidence, resolving conflicts to produce a coherent and accurate response. Moreover, to assess the performance of existing MLLMs in the reasoning processes of VideoQA, we introduce novel compositional consistency metrics and construct a VideoQA benchmark (QPVA3Bench) with 3,492 question-video tuples, each annotated with detailed composi tional graphs and fine-grained answers. We evaluate the QPVA3 framework on QPVA3Bench and 5 other VideoQA benchmarks. Experimental results demonstrate that our framework improves both consistency and accuracy compared to baselines, leading to a more transparent and verifiable VideoQA system. This approach has the potential to advance the field, as supported by our comprehensive evaluation and benchmarking efforts. Code and dataset are available at https://github.com/QPVA3/QPVA3-PAMI.

Substation drawing intelligent parsing framework with dense augmentation and semantic alignment

This study presents a max-margin–based approach for sentence boundary segmentation in Indonesian paragraphs, addressing a persistent challenge in Natural Language Processing applications. Conventional rule-based or sequential methods often struggle to distinguish ambiguous punctuation marks, particularly in contexts involving abbreviations, numerical expressions, hierarchical sentence structures, and direct quotations. To overcome these limitations, this research formulates sentence segmentation as a paragraph parsing task, enabling the model to capture both local boundary cues and global structural patterns within a paragraph. A manually annotated corpus of 12,000 paragraphs from news articles, public documents, and academic texts was developed to provide diverse linguistic structures and punctuation variations. The proposed model integrates local punctuation features, structural constraints from the Indonesian EYD standard, and global paragraph coherence through a max-margin discriminative parsing framework. Experimental results show that the model achieves strong performance on the test set, with a precision of 0.93, recall of 0.91, and F1-score of 0.92, significantly outperforming a rule-based baseline. Error analysis further highlights improvements in handling ambiguous cases such as abbreviations, numerical formatting, and direct quotations with nested punctuation. The findings demonstrate that a structured max-margin approach delivers more reliable sentence boundary segmentation and can enhance downstream NLP tasks requiring accurate sentence-level text processing.

This paper presents an extended statistical parsing framework for Tree-Adjoining Grammar (TAG) that incorporates part-of-speech (POS) information to enhance syntactic disambiguation, improve accuracy, and increase cross-lingual adaptability. While TAG provides a linguistically expressive mechanism for representing complex syntactic phenomena such as recursion and long-distance dependencies, however, conventional statistical TAG parsers remain largely constrained by their reliance on lexical anchors, which limits generalization across languages and leads to inefficiencies in ambiguous contexts. To address this, we improvise the statistical TAG formalism by conditioning derivation decisions on both lexical items and their associated POS tags, thereby enriching the feature space with syntactic category information. Beyond the baseline framework, this extended version introduces three major contributions. First, it integrates POS-based features into both generative and discriminative models, enabling robust handling of unseen or low-frequency lexical items. Second, it presents a cross-lingual evaluation using multilingual treebanks covering English to Indian language pairs, demonstrating consistent improvements in parsing accuracy and a 40–45% reduction in parsing time compared to conventional lexicalized TAG parser. Third, it provides an expanded analysis of computational efficiency, error patterns, and scalability across varying sentence lengths and linguistic families. Experimental results on a dataset of 15,000 annotated sentences reveal that the latest parser achieves significant gains in both accuracy and efficiency, with stable performance even in low-resource scenarios. The framework’s design further allows integration with neural embeddings, opening pathways toward hybrid symbolic–neural parsing models. Overall, the proposed POS enriched cross-lingual TAG framework offers a scalable, linguistically grounded, and computationally efficient solution for modern Natural Language Processing (NLP) tasks, including machine translation, information extraction, and question answering.

A Transformer-Enhanced Deep Learning Framework for Contextual Morphological Parsing in NLP Applications

Log parsing transforms log messages into structured formats, serving as the prerequisite step for various log analysis tasks. Although a variety of log parsing approaches have been proposed, their performance on complicated log data remains compromised due to the use of human-crafted rules or learning-based models with limited training data. The recent emergence of powerful large language models (LLMs) demonstrates their vast pre-trained knowledge related to code and logging, making it promising to apply LLMs for log parsing. However, their lack of specialized log parsing capabilities currently hinders their parsing accuracy. Moreover, the inherent inconsistent answers, as well as the substantial overhead, prevent the practical adoption of LLM-based log parsing. To address these challenges, we propose LILAC, the first practical Log parsIng framework using LLMs with Adaptive parsing Cache. To facilitate accurate and robust log parsing, LILAC leverages the in-context learning (ICL) capability of the LLM by performing a hierarchical candidate sampling algorithm and selecting high-quality demonstrations. Furthermore, LILAC incorporates a novel component, an adaptive parsing cache, to store and refine the templates generated by the LLM. It helps mitigate LLM's inefficiency issue by enabling rapid retrieval of previously processed log templates. In this process, LILAC adaptively updates the templates within the parsing cache to ensure the consistency of parsed results. The extensive evaluation on public large-scale datasets shows that LILAC outperforms state-of-the-art methods by 69.5% in terms of the average F1 score of template accuracy. In addition, LILAC reduces the query times to LLMs by several orders of magnitude, achieving a comparable efficiency to the fastest baseline.

Testing plays a pivotal role in ensuring software quality, yet conventional Search Based Software Testing (SBST) methods often struggle with complex software units, achieving suboptimal test coverage. Recent work using large language models (LLMs) for test generation have focused on improving generation quality through optimizing the test generation context and correcting errors in model outputs, but use fixed prompting strategies that prompt the model to generate tests without additional guidance. As a result LLM-generated testsuites still suffer from low coverage. In this paper, we present SymPrompt, a code-aware prompting strategy for LLMs in test generation. SymPrompt’s approach is based on recent work that demonstrates LLMs can solve more complex logical problems when prompted to reason about the problem in a multi-step fashion. We apply this methodology to test generation by deconstructing the testsuite generation process into a multi-stage sequence, each of which is driven by a specific prompt aligned with the execution paths of the method under test, and exposing relevant type and dependency focal context to the model. Our approach enables pretrained LLMs to generate more complete test cases without any additional training. We implement SymPrompt using the TreeSitter parsing framework and evaluate on a benchmark challenging methods from open source Python projects. SymPrompt enhances correct test generations by a factor of 5 and bolsters relative coverage by 26% for CodeGen2. Notably, when applied to GPT-4, SymPrompt improves coverage by over 2× compared to baseline prompting strategies.

As a safety-critical application, autonomous driving requires high-quality semantic segmentation and real-time performance for deployment. Existing method commonly suffers from information loss and massive computational burden due to high-resolution input-output and multiscale learning scheme, which runs counter to the real-time requirements. In contrast to channelwise information modeling commonly adopted by modern networks, in this article, we propose a novel real-time driving scene parsing framework named NDNet from a novel perspective of spacewise neighbor decoupling (ND) and neighbor coupling (NC). We first define and implement the reversible operations called ND and NC, which realize lossless resolution conversion for complementary thumbnails sampling and collation to facilitate spatial modeling. Based on ND and NC, we further propose three modules, namely, local capturer and global dependence builder (LCGB), spacewise multiscale feature extractor (SMFE), and high-resolution semantic generator (HSG), which form the whole pipeline of NDNet. The LCGB serves as a stem block to preprocess the large-scale input for fast but lossless resolution reduction and extract initial features with global context. Then the SMFE is used for dense feature extraction and can obtain rich multiscale features in spatial dimension with less computational overhead. As for high-resolution semantic output, the HSG is designed for fast resolution reconstruction and adaptive semantic confusion amending. Experiments show the superiority of the proposed method. NDNet achieves the state-of-the-art performance on the Cityscapes dataset which reports 76.47% mIoU at 240+ frames/s and 78.8% mIoU at 150+ frames/s on the benchmark. Codes are available at https://github.com/LiShuTJ/NDNet.

There is scope to enhance agricultural measurement and control systems user interactivity, which typically necessitates training for users to perform specific operations successfully. With the continuous development of natural language semantic processing technology, it has become essential to augment the user-friendliness of multifaceted control and query operations in the agricultural measurement and control sector, ultimately leading to reduced operation costs for users. The study aims to focus on command parsing. The proposed AMR-OPO semantic parsing framework is based on the natural language understanding method of Abstract Meaning Representation of Rooted Markup Graphs (AMR). It transforms the user’s natural language inputs into structured ternary (OPO) statements (operation-place-object) and converts the corresponding parameters of the user’s input commands. The framework subsequently sends the transformed commands to the relevant devices via the IoT gateway. To tackle the intricate task of parsing instructions, we developed a BERT-BiLSTM-ATT-CRF-OPO entity recognition model. This model can detect and extract entities from agricultural instructions, and precisely populate them into OPO statements. Our model shows exceptional accuracy in instruction parsing, with precision, recall, and F-value all measuring at 92.13%, 93.12%, and 92.76%, correspondingly. The findings from our experiment reveal outstanding and precise performance of our approach. It is anticipated that our algorithm will enhance the user experience offered by agricultural measurement and control systems, while also making them more user-friendly.

Instance-level human parsing is aimed at separately partitioning the human body into different semantic parts for each individual, which remains a challenging task due to human appearance/pose variation, occlusion and complex backgrounds. Most state-of-the-art methods follow the “parsing-by-detection” paradigm, which relies on a trained detector to localize persons and then sequentially performs single-person parsing for each person. However, this paradigm is closely related to the detector, and the runtime is proportional to the number of persons in an image. In this paper, we present a novel detection-free framework for instance-level human parsing in an end-to-end manner. We decompose instance-level human parsing into two subtasks via a unified network: 1) semantic segmentation for pixel-level classification as a human part and 2) instance segmentation for mask-level classification as a person. The framework can directly predict the human-part semantic mask for all persons and binary masks for instance-level persons in parallel. The parsing result of each person can be acquired via a Hadamard product between the human-part semantic mask and the corresponding person's binary mask. Extensive experiments demonstrate that our proposed method performs favorably against state-of-the-art methods on the CIHP and MHP v2 datasets.

Previous human parsing models are limited to parsing humans into pre-defined classes, which is inflexible for practical fashion applications that often have new fashion item classes. In this paper, we define a novel one-shot human parsing (OSHP) task that requires parsing humans into an open set of classes defined by any test example. During training, only base classes are exposed, which only overlap with part of the test-time classes. To address three main challenges in OSHP, i.e., small sizes, testing bias, and similar parts, we devise an End-to-end One-shot human Parsing Network (EOP-Net). Firstly, an end-to-end human parsing framework is proposed to parse the query image into both coarse-grained and fine-grained human classes, which builds a strong embedding network with rich semantic information shared across different granularities, facilitating identifying small-sized human classes. Then, we propose learning momentum-updated prototypes by gradually smoothing the training time static prototypes, which helps stabilize the training and learn robust features. Moreover, we devise a dual metric learning scheme which encourages the network to enhance features' representational capability in the early training phase and improve features' transferability in the late training phase. Therefore, our EOP-Net can learn representative features that can quickly adapt to the novel classes and mitigate the testing bias issue. In addition, we further employ a contrastive loss at the prototype level, thereby enforcing the distances among the classes in the fine-grained metric space and discriminating the similar parts. To comprehensively evaluate the OSHP models, we tailor three existing popular human parsing benchmarks to the OSHP task. Experiments on the new benchmarks demonstrate that EOP-Net outperforms representative one-shot segmentation models by large margins, which serves as a strong baseline for further research on this new task. The source code is available at https://github.com/Charleshhy/One-shot-Human-Parsing.

MFFSP: Multi-scale feature fusion scene parsing network for landslides detection based on high-resolution satellite images

Logs are pervasive in modern computing systems, and are valuable to service and system management. Nevertheless, with the rapidly growing size and complexity of computing systems, the log volume is exploding, which makes automatic log analysis imperative. Generally, in automatic log analysis, the first and fundamental step is log parsing, to which a lot of effort has been devoted. However, in most existing log parsing methods, log messages are merely treated as plain text. In natural language processing (NLP) area, it is a common practice to represent words and sentences with vectors, then the similarity between two words or sentences can be measured by the distance between their vectors. Inspired by these, we put forward a novel log parsing framework, named LPV (Log Parser based on Vectorization), which performs log parsing by converting log messages and log templates into vectors, with the help of a vectorization method in NLP. LPV incorporates offline and online log parsing. In the offline log parsing, the central idea is to first represent log messages with vectors, so that the similarity between two log messages can be measured by the distance between their vectors, then we cluster log messages via clustering the vectors, and finally we extract log templates from the resultant clusters. By the end of the offline log parsing, each log template is assigned with an average vector, so that in the online log parsing, the similarity between an incoming log message and each log template can also be measured by the distance between their vectors. Extensive experiments have been conducted based on several public log datasets to evaluate LPV with three different vectorization methods. The results demonstrate that, with a proper vectorization method, LPV performs competitive with state-of-the-art log parsing methods, in both effectiveness and efficiency.

Human parsing is a fine-grained semantic segmentation task, which needs to understand human semantic parts. Most existing methods model human parsing as a general semantic segmentation, which ignores the inherent relationship among hierarchical human parts. In this work, we propose a pose-guided hierarchical semantic decomposition and composition framework for human parsing. Specifically, our method includes a semantic maintained decomposition and composition (SMDC) module and a pose distillation (PC) module. SMDC progressively disassembles the human body to focus on the more concise regions of interest in the decomposition stage and then gradually assembles human parts under the guidance of pose information in the composition stage. Notably, SMDC maintains the atomic semantic labels during both stages to avoid the error propagation issue of the hierarchical structure. To further take advantage of the relationship of human parts, we introduce pose information as explicit guidance for the composition. However, the discrete structure prediction in pose estimation is against the requirement of the continuous region in human parsing. To this end, we design a PC module to broadcast the maximum responses of pose estimation to form the continuous structure in the way of knowledge distillation. The experimental results on the look-into-person (LIP) and PASCAL-Person-Part datasets demonstrate the superiority of our method compared with the state-of-the-art methods, that is, 55.21% mean Intersection of Union (mIoU) on LIP and 69.88% mIoU on PASCAL-Person-Part.

It is fundamental to cut multi-domain proteins into individual domains, for precise domain-based structural and functional studies. In the past, sequence-based and structure-based domain parsing was carried out independently with different methodologies. The recent progress in deep learning-based protein structure prediction provides the opportunity to unify sequence-based and structure-based domain parsing. Based on the inter-residue distance matrix, which can be either derived from the input structure or predicted by trRosettaX, we can decode the domain boundaries under a unified framework. We name the proposed method UniDoc. The principle of UniDoc is based on the well-accepted physical concept of maximizing intra-domain interaction while minimizing inter-domain interaction. Comprehensive tests on five benchmark datasets indicate that UniDoc outperforms other state-of-the-art methods in terms of both accuracy and speed, for both sequence-based and structure-based domain parsing. The major contribution of UniDoc is providing a unified framework for structure-based and sequence-based domain parsing. We hope that UniDoc would be a convenient tool for protein domain analysis. https://yanglab.nankai.edu.cn/UniDoc/. Supplementary data are available at Bioinformatics online.

This article presents a novel log abstraction framework based on neural open information extraction (OpenIE) and dynamic word embedding principles. Though various log parsing frameworks are proposed in the literature, the existing frameworks are modeled on predefined heuristics or auto-regressive methodologies that work well in offline scenarios. However, these frameworks are less suitable for dynamic self-adaptive systems, such as the Internet of Things (IoT), where the log outputs have diverse contextual variations and disparate time irregularities. Therefore, it is essential to move away from these traditional approaches and develop a systematic model that can effectively analyze log outputs in real-time and increase the system up-time of IoT networks so that they are almost always available. To address these needs, the proposed framework used OpenIE along with term frequency/inverse document frequency (TF/IDF) vectorization for constructing a set of relational triples (aka triple-sets). Additionally, a dynamic pretrained encoder–decoder architecture is utilized to imbibe the positional and contextualized information in its resultant outputs. The adopted methodology has enabled the proposed framework to extract richer word representations with dynamic contextualization of time-sensitive event logs to enhance further downstream activities, such as failure prediction and prognostic analysis of IoT networks. The proposed framework is evaluated on the system event log traces accumulated from a long range wide-area network (LoRaWAN) IoT gateway to proactively determine the probable causes of its various failure scenarios. Additionally, the study also provided a comparative analysis of its mathematical representations with that of the current state-of-the-art (SOTA) approaches to project the advantages and benefits of the proposed model, particularly from its data analytics standpoint.

Characterizing ingredients in commercially packaged baked products sold in the U.S.: An application of IngID

Human parsing is an important technology in human–robot interaction systems. At present, the distribution of multi-category human parsing datasets is unbalanced, and the samples present a long-tailed distribution, which directly affects the performance of human parsing. Meanwhile, the similarity between different categories leads the model to predict false parsing results. To solve the above problems, a general decoupled training framework called Decoupled Training framework based on Pixel Resampling (DTPR) was proposed to solve the long-tailed distribution, and a new sampling method named Pixel Resampling based on Accuracy distribution (PRA) for semantic segmentation was also proposed and applied to this decoupled training framework. The framework divides the training process into two phases, the first phase is to improve the model feature extraction ability, and the second phase is to improve the performance of the model on tail categories. The training framework was evaluated in MHPv2.0 and LIP datasets, and tested in both high-precision and real-time SOTA models. The MPA metric of model trained by DTPR in above two datasets increased by more than 6%, and the mIoU metric increased by more than 1% without changing the model structure.