Add dataset preparation notes

README.md

@@ -490,6 +490,23 @@ Source datasets:
 - [masakhane/InjongoIntent](https://huggingface.co/datasets/masakhane/InjongoIntent)
 
 
+## Dataset Preparation in MTEB
+
+This repository is a staging copy of `masakhane/InjongoIntent` for MTEB. The intended long-term canonical benchmark copy is `mteb/InjongoIntent`.
+
+### Transformations
+- Harmonized text columns across configs: `text` / `utterance` / `sentence` -> `text`
+- Harmonized label columns across configs: `label` / `intent` / `labels` -> `label`
+- Preserved the MTEB-facing subset names, including `gaz` and `swh`, while sourcing from the original Hub configs
+- Applied dataset cleaning before upload to reduce duplicates and train-test leakage in the benchmark copy
+
+### Label Schema
+- Integer intent labels from the source dataset are preserved after harmonization
+
+### Splits and subsets
+- 17 language-specific configs with cleaned train/dev-or-validation/test style splits
+- Language coverage matches the benchmark task: 16 African languages plus English
+
 ## How to evaluate on this task
 
 You can evaluate an embedding model on this dataset using the following code: