Event extraction (EE) is a crucial information task that aims to extract event in texts. Previous methods for EE typically model it as classification task, which are usually prone the data scarcity problem. In this paper, we propose new learning paradigm of EE, by explicitly casting machine reading comprehension problem (MRC). Our approach includes an unsupervised question generation process, can transfer schema into set natural questions, followed BERT-based question-answering process...

This paper tackles the task of event detection (ED), which involves identifying and categorizing events. We argue that arguments provide significant clues to this task, but they are either completely ignored or exploited in an indirect manner existing approaches. In work, we propose exploit argument information explicitly for ED via supervised attention mechanisms. specific, systematically investigate proposed model under supervision different strategies. Experimental results show our...

Modern models of event extraction for tasks like ACE are based on supervised learning events from small hand-labeled data. However, training data is expensive to produce, in low coverage types, and limited size, which makes methods hard extract large scale knowledge base population. To solve the labeling problem, we propose automatically label via world linguistic knowledge, can detect key arguments trigger words each type employ them texts automatically. The experimental results show that...

The joint entity and relation extraction task aims to extract all relational triples from a sentence. In essence, the contained in sentence are unordered. However, previous seq2seq based models require convert set of into sequence training phase. To break this bottleneck, we treat as direct prediction problem, so that model can get rid burden predicting order multiple triples. solve propose networks featured by transformers with non-autoregressive parallel decoding. Unlike autoregressive...

We present an event extraction framework to detect mentions and extract events from the document-level financial news. Up now, methods based on supervised learning paradigm gain highest performance in public datasets (such as ACE2005, KBP2015). These heavily depend manually labeled training data. However, particular areas, such financial, medical judicial domains, there is no enough data due high cost of labeling process. Moreover, most current focus extracting one sentence, but usually...

Dianbo Sui, Yubo Chen, Kang Liu, Jun Zhao, Shengping Liu. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint (EMNLP-IJCNLP). 2019.

Frames defined in FrameNet (FN) share highly similar structures with events ACE event extraction program.An is composed of an trigger and a set arguments.Analogously, frame FN lexical unit elements, which play roles as triggers arguments respectively.Besides having structures, many frames actually express certain types events.The above observations motivate us to explore whether there exists good mapping from event-types if it possible improve detection by using FN.In this paper, we propose...

Identifying event instance in text plays a critical role building NLP applications such as Information Extraction (IE) system. However, most existing methods for this task focus only on monolingual clues of specific language and ignore the massive information provided by other languages. Data scarcity ambiguity hinder performance these approaches. In paper, we propose novel multilingual approach---dubbed Gated Multilingual Attention (GMLATT) framework---to address two issues simultaneously....

Traditional approaches to the task of ACE event detection primarily regard multiple events in one sentence as independent ones and recognize them separately by using sentence-level information. However, are usually interdependent information is often insufficient resolve ambiguities for some types events. This paper proposes a novel framework dubbed Hierarchical Bias Tagging Networks with Gated Multi-level Attention Mechanisms (HBTNGMA) solve two problems simultaneously. Firstly, we propose...

The International Classification of Diseases (ICD) provides a standardized way for classifying diseases, which endows each disease with unique code. ICD coding aims to assign proper codes medical record. Since manual is very laborious and prone errors, many methods have been proposed the automatic task. However, most existing independently predict code, ignoring two important characteristics: Code Hierarchy Co-occurrence. In this paper, we propose Hyperbolic Co-graph Representation method...

Unlike other domains, medical texts are inevitably accompanied by private information, so sharing or copying these is strictly restricted. However, training a relation extraction model requires collecting privacy-sensitive and storing them on one machine, which comes in conflict with privacy protection. In this paper, we propose privacy-preserving based federated learning, enables central no single piece of local data being shared exchanged. Though learning has distinct advantages...

Identifying causal relations of events is a crucial language understanding task. Despite many efforts for this task, existing methods lack the ability to adopt background knowledge, and they typically generalize poorly new, previously unseen data. In paper, we present new method event causality identification, aiming address limitations previous methods. On one hand, our model can leverage external knowledge reasoning, which greatly enrich representation events; other mine event-agnostic,...

The ambiguity in language expressions poses a great challenge for event detection. To disambiguate types, current approaches rely on external NLP toolkits to build knowledge representations. Unfortunately, these work pipeline paradigm and suffer from error propagation problem. In this paper, we propose an adversarial imitation based distillation approach, the first time, tackle of acquiring rawsentences our teacher module is devised learn representations ground-truth annotations. Then, set...

In the current era of big data, a huge amount data has been generated and collected from wide variety rich sources. Embedded in these are useful information valuable knowledge. An example is healthcare epidemiological such as related to patients who suffered epidemic diseases like coronavirus disease 2019 (COVID-19). Knowledge discovered helps researchers, epidemiologists policy makers get better understanding disease, which may inspire them come up ways detect, control combat disease. As "a...

In the current technological era, huge amounts of big data are generated and collected from a wide variety rich sources. These can be different levels veracity in sense that some them precise while others imprecise uncertain. Embedded these useful information valuable knowledge to discovered. An example is healthcare epidemiological such as related patients who suffered epidemic diseases like coronavirus disease 2019 (COVID-19). Knowledge discovered data-via science techniques machine...

Hang Yang, Dianbo Sui, Yubo Chen, Kang Liu, Jun Zhao, Taifeng Wang. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Modern models of event causality detection (ECD) are mainly based on supervised learning from small hand-labeled corpora. However, training data is expensive to produce, low coverage causal expressions, and limited in size, which makes methods hard detect relations between events. To solve this lacking problem, we investigate a augmentation framework for ECD, dubbed as Knowledge Enhanced Distant Data Augmentation (KnowDis). Experimental results two benchmark datasets EventStoryLine corpus...

Pengfei Cao, Xinyu Zuo, Yubo Chen, Kang Liu, Jun Zhao, Yuguang Weihua Peng. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Event argument extraction is an essential task in event extraction, and become particularly challenging the case of low-resource scenarios. We solve issues existing studies under situations from two sides. From perspective model, methods always suffer concern insufficient parameter sharing do not consider semantics roles, which conducive to dealing with sparse data. And data, most focus on data generation augmentation. However, these rely heavily external resources, more laborious create...

Xinyu Zuo, Pengfei Cao, Yubo Chen, Kang Liu, Jun Zhao, Weihua Peng, Yuguang Chen. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Jian Liu, Yubo Chen, Kang Jun Zhao. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint (EMNLP-IJCNLP). 2019.

Current models for event causality identification (ECI) mainly adopt a supervised framework, which heavily rely on labeled data training.Unfortunately, the scale of current annotated datasets is relatively limited, cannot provide sufficient support to capture useful indicators from causal statements, especially handing those new, unseen cases.To alleviate this problem, we propose novel approach, shortly named CauSeRL, leverages external statements identification.First all, design...

Tong Zhou, Pengfei Cao, Yubo Chen, Kang Liu, Jun Zhao, Kun Niu, Weifeng Chong, Shengping Liu. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Conventional approaches to event detection usually require a fixed set of pre-defined types. Such requirement is often challenged in real-world applications, as new events continually occur. Due huge computation cost and storage budge, it infeasible store all previous data re-train the model with data, every time arrive. We formulate such challenging scenarios incremental detection, which requires learn classes incrementally without performance degradation on classes. However, existing...