A normal word can be used in legislation. Language is intepreted so its application is consistent or its subject matter is identifiable. Someone reading the text can might form a concept of where it fits into the body of law. The courts and individuals are bound by its propositions.  

Humpty dumpty words mean whatever you say they mean. They are not part of any law in which it appears, it is the law. They are not the facially appealling narratives they present themselves as. Logical constructions in legislation are subordinate to an ideology which is not bound by logic, but by power.  Such is the nature of hate speech laws.

They cannot be defined like normal words, and their inconsistencies will show up in language models which integrate them. They will prove to be purpose built, since the semiotic key is the political project, not its ostensible role in language. Interrogating a language model on chat GPT shows clear evidence in the case of "discrimination", it being unable to thread the needle between discernment and bigotry.

To explain its usage, providing examples of sentences which contain it are insufficient. They are endlessly capable of generating homonyms. You need to understand its strategic application as a weapon.

....

At this moment I discovered 

 https://unconstrainedanalytics.org/wp-content/uploads/2020/05/WARNING-ON-RACISM-131119-2.pdf

"From the beginning, therefore, the Islamophobia LOE the Organization of Islamic CoOperation (OIC) operationalizes, that Islamic Movement groups like the Muslim Brotherhood execute domestically, reflects the full leveraging and co-option of a Neoclassic Marxist campaign, a classic “aufheben der Kultur” effort – and not the other way around."

Further examination of these methods from Unconstrained Analytics: https://unconstrainedanalytics.org/report-re-remembering-the-mis-remembered-left-the-lefts-strategy-and-tactics-to-transform-america/