Note on bias

README.md

@@ -2,11 +2,11 @@
 
 Pairs of words and phrases in exclusive language and alternative words and phrases in inclusive language.
 
-Inclusivity aims to comprehend all [dimensions of diversity](https://www.charta-der-vielfalt.de/en/understanding-diversity/diversity-dimensions/) (age, ethnic background and nationality, gender and gender identity, physical and mental abilities, religion and worldview, sexual orientation, social background, and more); but currently focuses almost exclusively on gender inclusion, since gender exclusion is very dominant in German language.
+Inclusivity aims to comprehend all [dimensions of diversity](https://www.charta-der-vielfalt.de/en/understanding-diversity/diversity-dimensions/) (age, ethnic background and nationality, gender and gender identity, physical and mental abilities, religion and worldview, sexual orientation, social background, and more); but currently focuses almost exclusively on **gender inclusion**, since gender exclusion is very dominant in German language.
 
 ## Dataset structure
 
-**Train/test split:** This is not primarily a dataset that to be used directly for machine learning, so there is no train/test split.
+**Train/test split:** There is no train/test split, just a "train" dataset.
 
 - **`exclusive`**: Exclusive words and phrases in the singular. For the dimension of gender, these are certain words and phrases in the grammatical masculine. Note that the grammatical masculine is only exclusive if it is used in a _generic_ sense: "Die Doktoren" may be accurately used to describe three male doctors, but the same phrase is exclusive when it intends to refer to a group that also (potentially) includes women and nonbinary people. The relation between exclusive and inclusive phrases is n-to-n: An exclusive phrase may occur in multiple rows with various inclusive phrases associated, and vice versa.
 
@@ -26,6 +26,10 @@ Inclusivity aims to comprehend all [dimensions of diversity](https://www.charta-
 
   - _diversifix_: Entries added by ourselves or our community, also under the CC-0 license.
 
+## Bias
+
+The entries from the `dereko` source have been extracted according to their frequency in the corpus. This means, for example, that there are words referring to people from larger countries but not from some smaller countries; or, more accurately, countries that are considered important from the perspective of German-speaking journalism are more prevalent in the dataset.
+
 ## License
 
 Mixed license. All data is open, but a part of it only noncommercially. See the description for the `source` column above for details.