共查询到18条相似文献,搜索用时 46 毫秒
1.
2.
3.
4.
共指消解作为自然语言处理中的一个重要问题一直受到学术界的重视。二十多年来,基于规则的和基于统计的不同方法被提出,在一定程度上推进了该问题研究的发展,并取得了大量研究成果。该文首先介绍了共指消解问题的基本概念,并采用形式化的方法对该问题做了描述;然后,针对国内外近年来在共指消解研究中的方法进行了总结;之后,对共指消解中重要的特征问题进行了分析与讨论;最后,历数了共指消解的各种国际评测,并对未来可能的研究方向进行了展望。 相似文献
5.
6.
对象共指消解是语义Web研究中的一个关键问题。虽然目前已有许多不同的对象共指消解方法,但是它们的效率还不能满足实际使用的要求。MapReduce框架具有简单性和较强的计算能力,已被广泛用于各种数据并行处理任务。基于MapReduce的两个不同阶段,分别提出了两种并行算法来消解对象共指。具体地,给定一个初始训练集合和一个阈值,算法能够高效地发现一组具有可判别度的属性,并且满足它们的确信度高于预先给定的阈值。这些具有高可判别度的属性将被用于识别拥有相似取值的对象共指。基于真实数据集,通过人为增大数据集规模,验证了基于MapReduce算法的有效性。 相似文献
7.
指代消解处理是自然语言处理的关键环节,也是众多语言工程项目的核心任务。本文针对指代消解的一些基本问题进行阐述,按照时间线索,对国内外各类指代消解技术方法的研究情况进行分析,阐明了指代消解技术目前的主流方法和技术线路,最后对未来汉语指代消解技术的研究前景加以展望。 相似文献
8.
9.
一种基于图划分的无监督汉语指代消解算法 总被引:6,自引:2,他引:4
指代消解是自然语言处理领域中的一个重要问题。针对当前中文指代标注训练语料非常缺乏的现状,本文提出一种无监督聚类算法实现对名词短语的指代消解。引入图对名词短语的指代消解问题进行建模,将指代消解问题转化为图划分问题,并引入一个有效的模块函数实现对图的自动划分,使得指代消解过程并不是孤立地对每一对名词短语分别进行共指决策,而是充分考虑了多个待消解项之间的相关性,并且避免了阈值选择问题。通过在ACE中文语料上的人称代词消解和名词短语消解实验结果表明,该算法是一种有效可行的无监督指代消解算法。 相似文献
10.
11.
Peter Schüller 《人工智能实验与理论杂志》2018,30(4):525-546
AbstractWe describe the first automatic approach for merging coreference annotations obtained from multiple annotators into a single gold standard. This merging is subject to certain linguistic hard constraints and optimisation criteria that prefer solutions with minimal divergence from annotators. The representation involves an equivalence relation over a large number of elements. We use Answer Set Programming to describe two representations of the problem and four objective functions suitable for different data-sets. We provide two structurally different real-world benchmark data-sets based on the METU-Sabanci Turkish Treebank and we report our experiences in using the Gringo, Clasp and Wasp tools for computing optimal adjudication results on these data-sets. 相似文献
12.
13.
14.
15.
随着深度学习的兴起与发展,越来越多的学者开始将深度学习技术应用于指代消解任务中.但现有的神经指代消解模型普遍只关注文本的线性特征,忽略了传统方法中已证明非常有效的结构信息的融入.以目前表现最佳的Lee等提出的神经网络模型为基础,借助成分句法树对上述问题进行了改进:1)提出了一种枚举句法树中以结点为短语的抽取策略,避免了... 相似文献
16.
Order-sorted logic programming with predicate hierarchy 总被引:1,自引:0,他引:1
Ken Kaneiwa 《Artificial Intelligence》2004,158(2):155-188
Order-sorted logic has been formalized as first-order logic with sorted terms where sorts are ordered to build a hierarchy (called a sort-hierarchy). These sorted logics lead to useful expressions and inference methods for structural knowledge that ordinary first-order logic lacks. Nitta et al. pointed out that for legal reasoning a sort-hierarchy (or a sorted term) is not sufficient to describe structural knowledge for event assertions, which express facts caused at some particular time and place. The event assertions are represented by predicates with n arguments (i.e., n-ary predicates), and then a particular kind of hierarchy (called a predicate hierarchy) is built by a relationship among the predicates. To deal with such a predicate hierarchy, which is more intricate than a sort-hierarchy, Nitta et al. implemented a typed (sorted) logic programming language extended to include a hierarchy of verbal concepts (corresponding to predicates). However, the inference system lacks a theoretical foundation because its hierarchical expressions exceed the formalization of order-sorted logic. In this paper, we formalize a logic programming language with not only a sort-hierarchy but also a predicate hierarchy. This language can derive general and concrete expressions in the two kinds of hierarchies. For the hierarchical reasoning of predicates, we propose a manipulation of arguments in which surplus and missing arguments in derived predicates are eliminated and supplemented. As discussed by Allen, McDermott and Shoham in research on temporal logic and as applied by Nitta et al. to legal reasoning, if each predicate is interpreted as an event or action (not as a static property), then missing arguments should be supplemented by existential terms in the argument manipulation. Based on this, we develop a Horn clause resolution system extended to add inference rules of predicate hierarchies. With a semantic model restricted by interpreting a predicate hierarchy, the soundness and completeness of the Horn-clause resolution is proven. 相似文献
17.
This survey presents the concept of Big Data. Firstly, a definition and the features of Big Data are given. Secondly, the different steps for Big Data data processing and the main problems encountered in big data management are described. Next, a general overview of an architecture for handling it is depicted. Then, the problem of merging Big Data architecture in an already existing information system is discussed. Finally this survey tackles semantics (reasoning, coreference resolution, entity linking, information extraction, consolidation, paraphrase resolution, ontology alignment) in the Big Data context. 相似文献