I'm getting the impression that recreating functionality from Lucene on top of RDBMS abstractions is certainly possible, but requires a lot of effort from devs and you still need to understand stuff like tokenization, stemming, inverted indexes etc. if you want to use it. At this point one may as well just learn Solr or Elasticsearch. And these are still better at searching at the low level, although of course you sacrifice consistency guarantees. This is a legitimately different problem arena, not NoSQL for the sake of it.