Table-to-text generation aims to generate a description for factual table which can be viewed as set of field-value records. To encode both the content and structure table, we propose novel structure-aware seq2seq architecture consists field-gating encoder generator with dual attention. In encoding phase, update cell memory LSTM unit by field gate its corresponding value in order incorporate information into representation. decoding attention mechanism contains word level is proposed model...

Event extraction plays an important role in natural language processing (NLP) applications including question answering and information retrieval. Traditional event relies heavily on lexical syntactic features, which require intensive human engineering may not generalize to different datasets. Deep neural networks, the other hand, are able automatically learn underlying but existing networks do make full use of relations. In this paper, we propose a novel dependency bridge recurrent network...

Most existing knowledge base (KB) embedding methods solely learn from time-unknown fact triples but neglect the temporal information in base.In this paper, we propose a novel time-aware KB approach taking advantage of happening time facts.Specifically, use order constraints to model transformation between time-sensitive relations and enforce embeddings be temporally consistent more accurate.We empirically evaluate our two tasks link prediction triple classification.Experimental results show...

Generating texts from structured data (e.g., a table) is important for various natural language processing tasks such as question answering and dialog systems. In recent studies, researchers use neural models encoder-decoder frameworks table-to-text generation. However, these network-based approaches typically do not model the order of content during text When human writes summary based on given table, he or she would probably consider before wording. this paper, we propose an order-planning...

Answer selection plays a key role in community question answering (CQA). Previous research on answer usually ignores the problems of redundancy and noise prevalent CQA. In this paper, we propose to treat different text segments differently design novel attentive interactive neural network (AI-NN) focus those useful selection. The representations are first learned by convolutional networks (CNNs) or other architectures. Then AI-NN learns interactions each paired two texts. Row-wise...

Event extraction is a particularly challenging information task, which intends to identify and classify event triggers arguments from raw text.In recent works, when determining types (trigger classification), most of the works are either pattern-only or feature-only.However, although patterns cannot cover all representations an event, it still very important feature.In addition, identifying classifying arguments, previous consider each candidate argument separately while ignoring...

The reasoning abilities of Large Language Models (LLMs) have demonstrated remarkable advancement and exceptional performance across diverse domains. However, leveraging these capabilities to enhance LLM safety against adversarial attacks jailbreak queries remains largely unexplored. To bridge this gap, we propose Reasoning-to-Defend (R2D), a novel training paradigm that integrates reflections responses into LLMs' generation process, unlocking safety-aware mechanism. This approach enables...

The lack of labeled data is one the main challenges when building a task-oriented dialogue system. Existing datasets usually rely on human labeling, which expensive, limited in size, and low coverage. In this paper, we instead propose our framework auto-dialabel to automatically cluster intents slots. framework, collect set context features, leverage an autoencoder for feature assembly, adapt dynamic hierarchical clustering method intent slot labeling. Experimental results show that can...

Lexically constrained generation requires the target sentence to satisfy some lexical constraints, such as containing specific words or being paraphrase a given sentence, which is very important in many real-world natural language applications. Previous works usually apply beam-search-based methods stochastic searching lexically-constrained generation. However, when search space too large, always fail find optimal solution. At same time, cost steps correct optimization direction. In this...

Community question answering aims at choosing the most appropriate answer for a given question, which is important in many NLP applications. Previous neural network-based methods consider several different aspects of information through calculating attentions. These kinds attentions are always simply summed up and can be seen as ``single view", causing severe loss. To overcome this problem, we propose Multi-View Fusion Neural Network, where each attention component generates ``view'' QA pair...

Automatic event schema induction (AESI) means to extract meta-event from raw text, in other words, find out what types (templates) of may exist the text and roles (slots) each type.In this paper, we propose a joint entity-driven model learn templates slots simultaneously based on constraints same sentence.In addition, entities' semantic information is also considered for inner connectivity entities.We borrow normalized cut criteria image segmentation divide entities into more accurate...

Associative memories in the brain receive and store patterns of activity registered by sensory neurons, are able to retrieve them when necessary. Due their importance human intelligence, computational models associative have been developed for several decades now. In this paper, we present a novel neural model realizing memories, which is based on hierarchical generative network that receives external stimuli via neurons. It trained using predictive coding, an error-based learning algorithm...

In Semantic Role Labeling (SRL) task, the tree structured dependency relation is rich in syntax information, but it not well handled by existing models. this paper, we propose Syntax Aware Long Short Time Memory (SA-LSTM). The structure of SA-LSTM changes according to each sentence, so that can model whole an architecture engineering way. Experiments demonstrate on Chinese Proposition Bank (CPB) 1.0, improves F1 2.06% than ordinary bi-LSTM with feature engineered and gives state-of-the-art...

Multi-label text categorization is a type of categorization, where each document assigned to one or more categories.Recently, series methods have been developed, which train classifier for label, organize the classifiers in partially ordered structure and take predictions produced by former as latter classifiers' features.These predictions-asfeatures style model high order label dependencies obtain performance.Nevertheless, predictionsas-features suffer drawback.When training...

Table-to-text generation aims to generate a description for factual table which can be viewed as set of field-value records. To encode both the content and structure table, we propose novel structure-aware seq2seq architecture consists field-gating encoder generator with dual attention. In encoding phase, update cell memory LSTM unit by field gate its corresponding value in order incorporate information into representation. decoding attention mechanism contains word level is proposed model...

Explaining the predictions of AI models is paramount in safety-critical applications, such as legal or medical domains. One form explanation for a prediction an extractive rationale, i.e., subset features instance that lead model to give its on instance. Previous works generating rationales usually employ two-phase model: selector selects most important (i.e., rationale) followed by predictor makes based exclusively selected features. disadvantage these main signal learning select comes from...

Generating texts from structured data (e.g., a table) is important for various natural language processing tasks such as question answering and dialog systems. In recent studies, researchers use neural models encoder-decoder frameworks table-to-text generation. However, these network-based approaches do not model the order of contents during text When human writes summary based on given table, he or she would probably consider content before wording. biography, example, nationality person...

Controlling the style of natural language by disentangling latent space is an important step towards interpretable machine learning. After disentangled, a sentence can be transformed tuning representation without affecting other features sentence. Previous works usually use adversarial training to guarantee that disentangled vectors do not affect each other. However, methods are difficult train. Especially when there multiple (e.g., sentiment, or tense, which we call types in this paper),...

Abstract Changing an attribute of a text without changing the content usually requires first disentangling into irrelevant attributes and representations. After that, in inference phase, representation one is tuned to different value, expecting that corresponding can also be changed accordingly. The usual way disentanglement add some constraints on latent space encoder-decoder architecture, including adversarial-based mutual-information-based constraints. However, previous semi-supervised...

Abstract Plasma cell-free RNA (cfRNA) has recently emerged as a promising biomarker for non-invasive early cancer detection and treatment monitoring [8, 28, 31, 32, 38, 40]. Here, we introduce GeneLLM, novel large language model de-signed to interpret cfRNA sequences directly, bypassing the need genome annotations. GeneLLM significantly advances accuracy of various types. Our study demonstrates that this method achieves higher than traditional biomarkers effectively handles datasets from...

In this paper, we capture the argument relationships for Chinese semantic role labeling task, and improve task's performance with help of relationships.We split relationship between two candidate arguments into categories: (1) Compatible arguments: if one belongs to a given predicate, then other is more likely belong same predicate; (2) Incompatible less predicate.However, previous works did not explicitly model use simple maximum entropy classifier categories test its on Proposition Bank...