Measure More, Question More: Experimental Studies on Transformer-based Language Models and Complement Coercion