Ah, that's interesting. Because that describes very well what I felt when I read the early Chomskian works. Very cool theories and models with very little data to support them. At some point, I was wonderering whether the authors of those theories had any practical experience with languages apart from their own (English) mother tongue. And then the models became more and more complex, seemingly in an attempt to make them somehow work with real data, and at the end they became so complex that any data could be matched and no scientifically verifiable predictions could be made.

I hear lots of broad brush criticism of generative linguistics, but in almost all cases it turns out that the proposed alternative approaches are just working on different problems, and aren’t really alternatives at all.

The point to make is that interpretation makes use of multiple sources of information, of which syntax is just one. You don't need an elegant, tight fitting, constrained syntactic structure to use language. Which incidentally is why we can understand things even they are not grammatical. I firmly believe that the issue of "grammaticality" simply boils down to convention. Not to mention that many constructed examples used to motivate formal grammar research (be it syntax or semantics) are borderline acceptable themselves to many native speakers.

https://en.wikipedia.org/wiki/Dependency_grammar

I didn't work in the field at all after finishing the studies, so I cannot really tell you any actual implications of the different grammar approach anymore.