Maskrio/nya-ir-miracl-id-preprocessed

该数据集是MIRACL数据集（Zhang等人，2023）的一个派生工作，专注于印尼语（id）子集，通过五种不同的预处理策略来研究印尼语中-nya附着词处理如何影响检索质量。预处理策略包括：保持原样（keep）、无条件剥离后缀（naive_strip）、使用PySastrawi的RemoveInflectionalPossessivePronoun访问器（sastrawi_clitic）、用标记<NYA>替换剥离的后缀（sentinel）以及基于规则的前驱词替换（rule_resolved）。数据集包含开发集查询和训练集语料库的JSONL文件，每个文件有id和contents字段。遵循Apache-2.0许可证，与MIRACL一致。This dataset is a derivative work of MIRACL (Zhang et al., 2023) restricted to the Indonesian (id) subset, preprocessed five different ways to study how Indonesian `-nya` clitic handling affects retrieval quality. Preprocessing strategies include: keep (baseline pass-through), naive_strip (unconditional stripping of suffix), sastrawi_clitic (using PySastrawis RemoveInflectionalPossessivePronoun visitor), sentinel (replacing stripped suffix with <NYA> token), and rule_resolved (rule-based antecedent replacement). The dataset includes JSONL files for dev queries and train corpus, each with id and contents fields. Licensed under Apache-2.0, matching MIRACL.