Searle is ignored by AI researchers because it has no relevance--it has been thoroughly debunked.
The question of whether the man in the chinese room understands chinese is the wrong question. Someone reading symbols from outside of the room is not interacting with the man, they are interacting with the symbols--the algorithm. The algorithm itself does understand chinese! This algorithm is made real, it is reified by the actions of the man in the room to create and sustain it. The algorithm exists on an entirely new layer of abstraction from the man. So while it is true that the man does not understand chinese, the algorithm itself demonstrates every possible requirement for understanding--semantics as you would put it.
A human is analogous to the algorithm and the man shuffling papers is analogous to our neurons. Of course our neurons do not "understand" what they are processing, but as a whole our neurons create an entity that itself has full understanding.
Semantics do in fact come from symbols. When I enter my password into my computer and it uses that password and allows me access to some resource as a consequence, that is semantics. The password string exists in a specific context within the computer system, from which it derives meaning. The difference between the computer system and a person is that the semantic meaning is not generic and integrated. The password exists within the semantic context of granting or revoking access to a resource. It can only exist here as there is no mechanism in standard programming languages for that semantic meaning to be integrated with other parts of the system (without explicitly programming every single case). The brain on the other hand has generic semantics that effortlessly allows integration of different kinds of semantics into a single whole. This is the only meaningful difference.