Basic NLP starts with simple techniques like stemming, synonyms, typo correction etc. But many users expect more, e.g. contextual disambiguation or broader synonym/similarity resolution. These users would see a search engine as dysfunctional if it could not even find pages in which a plural form or a spelling variation of a query term occurs.
However, there is an inherent contradiction to those users who want to find a specific string. Google, Bing and others apply complex algorithms ("weird Black Box AI") in order to interpret the user intent and the meaning of web pages in order to optimize towards the majority of internet users. To some degree, they do try to also serve users of (large enough) minority use cases in which the user feels like they knew exactly what words (strings) they were looking for. But again, the different use cases pose contradictory interests that are probably not resolvable in a "general purpose" approach.
Quotes used to be an exact search, but it now overrides that basically everywhere and continues to search for what it thinks you mean (and has for like a decade now). Even though the "Missing: ___ | Must include: ___" popup just wraps that word in quotes in the search. Which is infuriating, since they know damn well adding quotes won't actually search for that word.
In reality, you need to click "tools" -> "all results" and change it to "verbatim" which does at least still appear to work... for now.
Any morphological variation would already be covered with stemming alone, so this particular point doesn’t speak in favor of more advanced interpretations, does it?
Having an path which uses advanced AI powered tools doesn’t necessarily imply one should drop any ability to search with more technical query interface for advanced users. And it’s doubtful big companies wouldn’t have the budget to maintain even two completely distinct tools, would it be required.
How are they not resolvable? If Google respected quotation marks (as it used to do, in the old times) that would be enough to accommodate both kinds of users (or rather both kind of search needs - I'm the same person and do both).