Miller and others built mathematical models and carried out experiments to test Chomsky's theory of transformational grammars. Studies involving memory for sentences and response times suggested that processing times were in some cases proportional to the number of transformations involved - though with some kinds of transformations such relationships were not found. Eg.
Active sentences were remembered better than passive sentences and showed less reaction time; similarly affirmative were easier than negative.
More strikingly, the difference between the response latencies (Affirmative active - Negative passive) was = (Affirmative active - Affirmative passive) + (Affirmative active - Negative active)
Was this proof of transformations operations taking a fixed amount of time? Later results were not so clear-cut (eg. of problem - John was reading a book vs A car hit John. - for the latter sentence the passive voice would be more ``natural'')