Upload folder using huggingface_hub

README.md

@@ -92,6 +92,32 @@ print(dataset)
 These splits may represent domain structure (for example part-of-speech groups) rather than train/validation/test partitions.
 If you need ML evaluation splits, create an explicit derived split plan in downstream preprocessing.
 
+## Summary
+WordNet is published as lexical-resource views rather than benchmark train/validation/test tasks.
+
+## Intended Use
+- Lexical semantic analysis and symbolic NLP workflows.
+- Knowledge graph and relation-structure exploration.
+- Input resource for downstream task-specific data construction.
+
+## Schema Notes
+- The same part-of-speech split names (noun, verb, adj, adv) can appear across multiple configs.
+- The aggregate config combines variant outputs while preserving split names.
+
+## Evaluation Protocol
+- For ML benchmarks, define explicit train/validation/test derivations from the lexical splits.
+- Record any filtering or balancing logic to keep experiments reproducible.
+
+## Limitations
+- Not a native benchmark dataset with pre-defined evaluation partitions.
+- Coverage and lexical conventions follow WordNet source files.
+
+## Config Selection Guidance
+- Use the data config for synset-level lexical structure.
+- Use the index config for lemma and offset lookup behavior.
+- Use the exceptions config for inflection-to-base mappings.
+- Use aggregate when you want one entry point and can filter by variant in downstream code.
+
 ## Standardization
 - split_aliases: {"dev": "validation", "val": "validation", "valid": "validation"}
 - rename_map: {}

dataset.json

@@ -133,6 +133,39 @@
     }
   },
   "dataset_id": "wordnet",
+  "documentation": {
+    "dataset_card": {
+      "custom_sections": [
+        {
+          "content": [
+            "Use the data config for synset-level lexical structure.",
+            "Use the index config for lemma and offset lookup behavior.",
+            "Use the exceptions config for inflection-to-base mappings.",
+            "Use aggregate when you want one entry point and can filter by variant in downstream code."
+          ],
+          "title": "Config Selection Guidance"
+        }
+      ],
+      "evaluation_protocol": [
+        "For ML benchmarks, define explicit train/validation/test derivations from the lexical splits.",
+        "Record any filtering or balancing logic to keep experiments reproducible."
+      ],
+      "intended_use": [
+        "Lexical semantic analysis and symbolic NLP workflows.",
+        "Knowledge graph and relation-structure exploration.",
+        "Input resource for downstream task-specific data construction."
+      ],
+      "limitations": [
+        "Not a native benchmark dataset with pre-defined evaluation partitions.",
+        "Coverage and lexical conventions follow WordNet source files."
+      ],
+      "schema_notes": [
+        "The same part-of-speech split names (noun, verb, adj, adv) can appear across multiple configs.",
+        "The aggregate config combines variant outputs while preserving split names."
+      ],
+      "summary": "WordNet is published as lexical-resource views rather than benchmark train/validation/test tasks."
+    }
+  },
   "pipeline": {
     "required_splits": [
       "noun",