Update README.md

README.md

@@ -27,10 +27,42 @@ with corresponding options filled in were separated, shuffled and classified ind
 ## Training data
 
 [WinoGrande-XL](https://huggingface.co/datasets/winogrande) reformatted the following way:
-1. Each sentence was split on `_` placeholder symbol.
+1. Each sentence was split on "`_`" placeholder symbol.
 2. Each option was concatenated with the second part of the split, thus transforming each example into two text segment pairs.
 3. Text segment pairs corresponding to correct and incorrect options were marked with `True` and `False` labels accordingly.
-4. Text segment pairs were shuffled thereafter. 
+4. Text segment pairs were shuffled thereafter.
+
+For example,
+
+```json
+{
+  "answer": "2",
+  "option1": "plant",
+  "option2": "urn",
+  "sentence": "The plant took up too much room in the urn, because the _ was small."
+}
+```
+
+becomes
+
+```json
+{
+  "sentence1": "The plant took up too much room in the urn, because the ",
+  "sentence2": "plant was small.",
+  "label": false
+}
+```
+
+and
+
+```json
+{
+  "sentence1": "The plant took up too much room in the urn, because the ",
+  "sentence2": "urn was small.",
+  "label": true
+}
+```
+These sentence pairs are then treated as independent examples.
 
 ### BibTeX entry and citation info