Add training data

README.md

@@ -1,3 +1,58 @@
 ---
 license: mit
 ---
+
+# ZNO dataset
+
+This dataset contains 4670 machine-readable questions and answers from
+Ukrainian External independent testing (called _ЗНО_/_ZNO_ in Ukrainian).
+
+Question subjects are:
+
+- History of Ukraine
+- Ukrainian language and literature
+
+Currently, only train subset (3063 question/answers) is released. We will
+release test set soon.
+
+
+## File format
+
+Every line in a .jsonl file contains a structure like this:
+
+```js
+{
+  "question": "На другий склад падає наголос у слові",
+  "answers": [
+    { "marker": "А", "text": "начинка" },
+    { "marker": "Б", "text": "випадок" },
+    { "marker": "В", "text": "дрова" },
+    { "marker": "Г", "text": "загадка" },
+    { "marker": "Д", "text": "русло" }
+  ],
+  "correct_answers": ["Д"]
+}
+
+```
+
+Currently, all questions have exactly one correct answer, stored in
+`correct_answers[0]`.
+
+
+## Pre/postprocessing
+
+* Question text is converted to Markdown.
+* Questions with images are skipped (~600 cases).
+* Open-ended questions ("Write your thoughts about...") are skipped.
+* Questions that require matching multiple statements to multiple choices are skipped.
+* Links to full texts of literature pieces are removed.
+
+
+## Sources:
+
+- https://zno.osvita.ua/comtests.html
+
+
+## Other
+
+This dataset is used in [UNLP 2024 Shared Task](https://github.com/unlp-workshop/unlp-2024-shared-task)