首页> 外文会议>Workshop on universal dependencies 2017 >Universal Dependency Evaluation
【24h】

Universal Dependency Evaluation

机译:普遍依赖评估

获取原文
获取原文并翻译 | 示例

摘要

Multilingual parser evaluation has for a long time been hampered by the lack of cross-linguistically consistent annotation. While initiatives like Universal Dependencies have greatly improved the situation, they have also raised questions about the adequacy of existing parser evaluation metrics when applied across typologically different languages. This paper argues that the usual attachment score metrics used to evaluate dependency parsers are biased in favor of analytic languages, where grammatical structure tends to be encoded in free morphemes (function words) rather than in bound morphemes (inflection). We therefore propose an alternative evaluation metric that excludes functional relations from the attachment score. We explore the effect of this change in experiments using a subset of treebanks from release v2.0 of Universal Dependencies.
机译:长期以来,由于缺乏跨语言一致的注释,多语言分析器评估受到了阻碍。尽管诸如“通用依赖项”之类的计划极大地改善了这种状况,但它们也引发了有关在类型上不同的语言中应用现有解析器评估指标是否足够的问题。本文认为,用于评估依赖性解析器的常用依附评分标准偏向于分析语言,在语法语言中,语法结构倾向于以自由语素(功能词)而不是绑定语素(词尾变化)编码。因此,我们提出了一种替代评估指标,该评估指标从附件得分中排除了功能关系。我们使用Universal Dependencies版本v2.0中的一部分树库在实验中探索这种变化的影响。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号