WolofEntityLinking
收藏资源简介:
Wolof Entity Linking 是一个针对低资源非洲语言——沃洛夫语(Wolof)的实体链接基准数据集。该数据集旨在解决低资源语言实体链接任务中的一个核心挑战:正字法变异(即非标准化的拼写形式)。数据来源于 MasakhaNER 2.0 数据集的验证集和测试集,经过筛选保留了包含有效实体提及(排除仅含日期或无实体提及的句子)的 1,045 个沃洛夫语句子。数据集共包含 2,049 个实体提及,涵盖三种命名实体类型:地点(LOC)、组织(ORG)和人物(PER)。其中,1,704 个提及(In-KB)成功链接至 565 个唯一的 Wikidata 知识库条目(QID),而 345 个提及(Out-of-KB/NIL)则代表了 Wikidata 中不存在的实体。数据集以 JSON 格式提供,每个样本包含句子 ID、原始文本、分词列表以及详细的实体标注信息。实体信息包括字符/词索引边界、实体类型、链接状态(链接或 NIL)、对应的 Wikidata QID、标签、描述(法语或英语)以及元数据(标注来源、方法、NIL 原因等)。该数据集适用于命名实体识别(NER)、实体链接、低资源 NLP 研究以及跨语言知识库构建等任务。
Wolof Entity Linking is a benchmark dataset for entity linking in Wolof, a low-resource African language. It addresses a core challenge in entity linking for such languages: orthographic variation (i.e., non-standardized spellings). The data is derived from the validation and test sets of the MasakhaNER 2.0 dataset, filtered to retain 1,045 Wolof sentences containing valid entity mentions (excluding sentences with only dates or no mentions). The dataset includes 2,049 entity mentions covering three named entity types: location (LOC), organization (ORG), and person (PER). Among these, 1,704 mentions (In-KB) are successfully linked to 565 unique Wikidata knowledge base entries (QIDs), while 345 mentions (Out-of-KB/NIL) represent entities not present in Wikidata. The dataset is provided in JSON format, with each sample containing a sentence ID, original text, tokenized list, and detailed entity annotations. Entity information includes character/token index boundaries, entity type, linking status (linked or NIL), corresponding Wikidata QID, label, description (in French or English), and metadata (annotation source, method, NIL reason, etc.). It is suitable for tasks such as named entity recognition (NER), entity linking, low-resource NLP research, and cross-lingual knowledge base construction.




