Upload README.md with huggingface_hub

README.md

@@ -1,58 +1,10 @@
----
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: validation
-    path: data/validation-*
-dataset_info:
-  features:
-  - name: id
-    dtype: int32
-  - name: domain
-    dtype: string
-  - name: subdomain
-    dtype: string
-  - name: prompt
-    dtype: string
-  - name: response
-    dtype: string
-  - name: context
-    dtype: string
-  - name: final_answer
-    dtype: string
-  - name: text
-    dtype: string
-  - name: source
-    dtype: string
-  - name: license
-    dtype: string
-  - name: verified
-    dtype: bool
-  - name: src_lang
-    dtype: string
-  - name: tgt_lang
-    dtype: string
-  - name: task_type
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 7389464
-    num_examples: 9283
-  - name: validation
-    num_bytes: 807640
-    num_examples: 1032
-  download_size: 4104557
-  dataset_size: 8197104
----
 # AskVox Language Learning Dataset
 
-This dataset combines multiple public Kaggle datasets focused on multilingual text and language learning tasks.
+This dataset combines multiple **public Kaggle datasets** focused on multilingual text and language learning tasks.
 
-**Included Sources:**
-1. English Language Learning (ELL) – writing & proficiency
-2. Language Detection (17 languages)
+**Included Sources (all from Kaggle v1):**
+1. English Language Learning (ELL) – writing & proficiency (ELLIPSE corpus)
+2. Language Detection (17 languages, including Hindi, Swedish, Tamil, etc.)
 3. Global Language Learning Popularity (2020–2023)
 4. Adult Language Learning Profile
 
@@ -62,22 +14,27 @@ This dataset combines multiple public Kaggle datasets focused on multilingual te
 |---------------|------------------------------------------------------------------|
 | id            | Integer identifier                                               |
 | domain        | High-level domain, here `"language-learning"`                    |
-| subdomain     | More specific scope, here `"multilingual/text"`                 |
-| prompt        | Input sentence or question                                      |
+| subdomain     | More specific scope, here `"multilingual/text"`                  |
+| prompt        | Input sentence or question                                       |
 | response      | Translation, label, or target text (if available)               |
-| context       | RAG context: prompt (+ response if present)                     |
-| final_answer  | Ground-truth answer/target text for QA/RAG evaluation           |
-| text          | Combined view for retrieval: `[PROMPT] ...` + `[RESPONSE] ...`  |
-| source        | Original Kaggle source description                              |
-| license       | License note (derived from Kaggle open datasets)                |
-| verified      | Boolean quality flag                                            |
-| src_lang      | Source language (if available)                                  |
-| tgt_lang      | Target language or label language (if applicable)               |
-| task_type     | Task type (e.g. `language_detection`, `translation`, `generic`) |
+| context       | RAG context: prompt (+ response or language label if present)    |
+| final_answer  | Ground-truth answer/label for QA/RAG evaluation                  |
+| text          | Combined view for retrieval: `[PROMPT] ...` + `[RESPONSE] ...`   |
+| source        | Original Kaggle source description                               |
+| license       | License note (derived from Kaggle open datasets)                 |
+| verified      | Boolean quality flag                                             |
+| src_lang      | Source language (if available)                                   |
+| tgt_lang      | Target language or label language (normalized, e.g. 'swedish')   |
+| task_type     | Task type (`language_detection`, `translation`, `generic`, ...)  |
+
+**Notes:**
+- All v1 data comes from Kaggle public datasets.
+- Language labels are lightly normalized (e.g. `Sweedish` → `Swedish`).
+- Hindi and other languages appear both in text and as labels where available.
 
 **RAG Usage:**
 - Use `text` or `context` for embedding & retrieval.
 - Use `prompt` → `final_answer` / `response` pairs for QA or instruction tuning.
-- Filter by `domain`, `subdomain`, `task_type`, or languages when building indices.
+- Filter by `domain`, `subdomain`, `task_type`, or languages (`tgt_lang`) when building indices.
 
 **License:** Derived from Kaggle open datasets (CC0 / public domain or similar).