Over 25 years ago Collins and Poras (1994) studied successful companies and noted their ability to often embrace what appear to be competing constructs.
This may be akin to machine learning (statistical) and knowledge engineering (linguistics-taxonomies). On many occasions I’ve seen one ‘camp’ explain all the pitfalls of the other, without acknowledging the limitations of their own approach.
Back in 2015 we investigated how both approaches could be combined in a case study and series of experiments on unstructured text.
The ‘Best of Both Worlds’ paper was published in the Knowledge Organization Journal Here