add examples of parsing annotations

README.md

@@ -168,6 +168,49 @@ Volunteers and Expert annotators
 
 ## Considerations for Using the Data
 
+## Accessing the annotations 
+
+Each example text has multiple annotations. These annotations may not always agree. There are various approaches one could take to calculate agreement, including a majority vote, rating some annotators more highly, or calculating a score based on the 'votes' of annotators. Since there are many ways of doing this, we have not implemented this as part of the dataset loading script.  
+
+An example of how one could generate an "OCR quality rating" based on the number of times an annotator labelled an example with `Illegible OCR`:
+
+```python 
+from collections import Counter
+
+
+def calculate_ocr_score(example):
+    annotator_responses = [response['response'] for response in example['annotator_responses_english']]
+    counts = Counter(annotator_responses)
+    bad_ocr_ratings = counts.get("Illegible OCR")
+    if bad_ocr_ratings is None:
+        bad_ocr_ratings = 0
+    return round(1 - bad_ocr_ratings/len(annotator_responses),3)
+ 
+ 
+ dataset = dataset.map(lambda example: {"ocr_score":calculate_ocr_score(example)})
+ ```
+        
+To take the majority vote (or return a tie) based on whether a example is labelled contentious or not:
+
+```python
+
+def most_common_vote(example):
+    annotator_responses = [response['response'] for response in example['annotator_responses_english']]
+    counts = Counter(annotator_responses)
+    contentious_count = counts.get("Contentious according to current standards")
+    if not contentious_count:
+        contentious_count = 0
+    not_contentious_count = counts.get("Not contentious")
+    if not not_contentious_count:
+        not_contentious_count = 0
+    if contentious_count > not_contentious_count:
+        return "contentious"
+    if contentious_count < not_contentious_count:
+        return "not_contentious"
+    if contentious_count == not_contentious_count:
+        return "tied"
+```
+
 ### Social Impact of Dataset
 
 This dataset can be used to see how words change in meaning over time