Document annotator tokens (#178)

* fix typo
* clarify annotator tokens

---------

Co-authored-by: Roman Grundkiewicz <rgrundkiewicz@gmail.com>

EvalView/templates/EvalView/_instructions-esa.html

@@ -1,7 +1,7 @@
 <div class="row">
   <div class="col-md-12">
     <ul class="list-unstyled">
-      <li><strong>Higlighting errors:</strong>
+      <li><strong>Highlighting errors:</strong>
         <ul>
           <li>
             Highlight the text fragment where you have identified a translation error (drag or click start & end).

EvalView/templates/EvalView/_instructions-mqm.html

@@ -1,7 +1,7 @@
 <div class="row">
   <div class="col-md-12">
     <ul class="list-unstyled">
-      <li><strong>Higlighting errors:</strong>
+      <li><strong>Highlighting errors:</strong>
         <ul>
           <li>
             Highlight the text fragment where you have identified a translation error (drag or click start & end).

INSTALL.md

@@ -1,4 +1,4 @@
-## Setup
+# Setup
 
 1. Basic setup:
 
@@ -39,6 +39,9 @@ python3 manage.py StartNewCampaign Examples/MQM+ESA/manifest.json \
 python3 manage.py CreateInviteTokens test_group 20 --create-group test_group
 ```
 
+Add `--task-confirmation-tokens` if you want to generate annotator confirmation tokens.
+See [quality control](#Quality control) for more details.
+
 5. Optionally clean up everything
 
 ```
@@ -122,4 +125,13 @@ For task:
 - `batchNo`: task number
 - `randomSeed`: number used in batch generation
 - `requiredAnnotations`: how many annotations does a task need, in most cases use 1
-- `source/targetLanguage`: source and target language
+- `source/targetLanguage`: source and target language
+
+## Quality control
+
+With `--task-confirmation-tokens`, the annotators will be shown a random key/token if they fail the quality control and a correct one (matching the one in the CSV output with credentials) if they succeed.
+The quality control checks if the perturbed samples (`itemType=BAD`) have statistically lower scores than the original ones (`itemType=TGT`).
+Even without the switch, the campaign status page will show a p-value (last column for staff account) that corresponds to the outcome of this test.
+If it's close to 1, then the annotator is annotating randomly and is of poor quality.
+For values close to 0, the annotations are good.
+The threshold to generate the valid token for annotators is currently p<=10%.