A simple, brute force strategy would be to write a parser to read each NOTES and split the words into a single vertical table.  Then simply run a count for each of the distinct words in the vertical table.  An alternative implementation is to have two columns in the vertical table (WORD and COUNT) and just update count if the word already exists.

The same results can be achieved by using one of the "split" functions available on this site inside of a derived subquery.

Have a good day

Wayne