Upload ARETA gold reference dataset

README.md

@@ -0,0 +1,70 @@
+---
+language:
+- ar
+configs:
+- config_name: annotations
+  default: true
+  data_files:
+  - split: test
+    path: gold_ref_hf_annotations.csv
+- config_name: contexts
+  data_files:
+  - split: test
+    path: gold_ref_hf_contexts.csv
+---
+
+# ArabiGEE Data
+
+This dataset accompanies the paper **ArabiGEE: A Hierarchical Taxonomy for
+Arabic Grammatical Error Explanation**. It contains Arabic grammatical error
+data and gold reference explanation annotations. The data is split into two CSV
+tables to avoid repeating full sentence contexts across annotation rows.
+
+## Citation
+
+If you use this dataset, please cite:
+
+**ArabiGEE: A Hierarchical Taxonomy for Arabic Grammatical Error Explanation**
+
+## Tables
+
+### annotations
+
+`gold_ref_hf_annotations.csv` contains one row per annotated error. A word pair
+can have multiple annotation rows, linked by the same `pair_id` and ordered with
+`annotation_index_in_pair`.
+
+Columns:
+
+- `context_id`: joins to `gold_ref_hf_contexts.csv`
+- `pair_id`: unique word-pair identifier in this export
+- `source_pair_id`: original pair id from the source dataset
+- `annotation_index_in_pair`: 1-based index for multiple annotations on the same pair
+- `erroneous_word`: erroneous surface form
+- `correct_word`: corrected surface form
+- `areta_label`: original ARETA edit label
+- `lex_code`, `orth_code`, `morph_code`, `synt_code`: explanation code by tier
+- `lex_explanation_text`, `orth_explanation_text`, `morph_explanation_text`, `synt_explanation_text`: explanation text by tier
+
+### contexts
+
+`gold_ref_hf_contexts.csv` contains one row per source sentence/context.
+
+Columns:
+
+- `context_id`: joins to `gold_ref_hf_annotations.csv`
+- `source_dataset`: source corpus label
+- `source_sentence_id`: original sentence id from the source dataset
+- `src_context`: erroneous sentence/context
+- `target_context`: corrected sentence/context
+
+## Usage
+
+```python
+from datasets import load_dataset
+
+annotations = load_dataset("REPO_ID", "annotations", split="test")
+contexts = load_dataset("REPO_ID", "contexts", split="test")
+```
+
+Join `annotations` to `contexts` with `context_id`.