​I was helping to expose one of your test cases :)​

'product' is a lexeme - a headword - a basic unit of meaning that has a 'set of forms' and those have 'a set of definitions'

  Here are some of its forms:

     1. product

     2. products

     3. producing

     4. production

     5. ...

​Here are some definitions against PRODUCT:

    1. product - ​a commodity offered for sale.

    2. product - a quantity obtained by multiplication of two or more numbers

    3. product - anything that is produced.

​But a thought just occured to me...

A. In order to model this perhaps would be to have those headwords stored in Wikidata.  Those headwords ideally would not actually be a Q or a P ... but what about instead ... L​  ?  Wrapping the graph structure itself ?  Pros / Cons ?

B.  or do we go with Daniel's suggestion of linking out to headwords and not actually storing them in Wikidata ?  Pros / Cons ?