peneira-recursos
收藏资源简介:
Peneira资源包是一个用于加利西亚语规范性校对的开放数据集,专注于校正西班牙语干扰、代词位置、短语结构和拼写问题。它包含三个核心组件:EVAL-204评估集(含204个系列,包括193个严格评估项和11个有文档的排除项,以及te=CD子集)、干扰清单(含153个规范性条目,分为125个明确条目、9个模糊条目、4个偏好条目和15个模式)和架构说明文档(描述“有疑问时保持沉默”原则)。资源基于RAG/ILG(NOMIG 2003, DRAG)规范性标准,使用Salamandra-7B作为基础模型。请注意,这些资源未经独立外部语言学审查,是“尽力而为”的资源而非规范性权威,训练数据集和模型权重不包含在内。数据集采用CC BY 4.0许可证发布,所有内容均为原创作品,不包含维基百科文本。
The Peneira Resource Package is an open resource developed by ARGALLA for the Galician language normative proofreading assistant project. This dataset focuses on normative correction of Galician, specifically targeting issues caused by Spanish interference, pronoun placement, phrase structure and spelling. The resource package includes three core components: 1) The EVAL-204 evaluation set, which contains 204 entries (193 strict evaluation items and 11 documented exclusion items) along with the te=CD subset; 2) The interference inventory, which encompasses 153 normative entries (125 clear entries, 9 ambiguous entries, 4 preference entries and 15 patterns); 3) The architecture specification document, which describes the "When in Doubt, Keep Silent" principle. These resources are based on the RAG/ILG normative standards (NOMIG 2003, DRAG) and use Salamandra-7B as the base model. It should be noted that these resources have not undergone independent external linguistic review, and are "best-effort" resources rather than normative authorities. The training datasets and model weights are not included in this resource package. This dataset is released under the CC BY 4.0 license, and all content is original work that does not include Wikipedia texts.
Peneira · 资源包 (v1.0)
许可证: CC BY 4.0
语言: 加利西亚语
标签: 加利西亚语、语法、规范性校正、评估
名称: Peneira · recursos (eval-204 + inventario + arquitectura)
该项目由 ARGALLA 开发,是一个用于加利西亚语的规范性辅助校正工具(针对卡斯蒂利亚语干扰、代词放置、动词短语、正字法)。外部审查尚在进行中,资源为“尽力而为”性质,不作规范性权威。
资源内容
| 资源 | 文件 | 说明 |
|---|---|---|
| Held-out EVAL-204 | eval_set.jsonl、eval_extension_v3.jsonl、eval_sintaxe_v2.jsonl、eval_sintaxe_extension_v3.jsonl、eval_teche_v32.jsonl、EVAL-204-manifesto.md |
密封评估集(204系列/193严格/11个记录在案的排除项)+ te=CD 子集 |
| 干扰清单 | inventario.jsonl、INVENTARIO-auditado.md |
153个规范性条目(125个明确项、9个歧义项、4个偏好项、15个模式) |
| 架构 | ARQUITECTURA.md |
速查表+核心+三个守卫·“存疑时保持沉默”原则 |
未包含在此的内容
- 训练数据集及模型的权重/GGUFs
- 基准测试数字(架构描述仅说明评估方法,而非结果)
许可证与引用
- 采用 CC BY 4.0 许可证,三个资源为原创作品,不含维基百科文本,不继承相同方式共享条款。
- 规范性来源:RAG/ILG(NOMIG 2003,DRAG)
- 基础模型:Salamandra-7B(BSC-LT)
- 训练所用维基语料(Wikipedia,CC BY-SA)未包含在此。
报告错误
欢迎在本仓库提交 issue,内容包括:片段、建议的更正及规范性参考(如有)。所有基于规范的更正都将推动“待外部审查”的进展。




