What does and doesn’t do so do?

To test whether a given string after the verb is a complement or an adjunct, one can apply the well-known do so test, first mentioned by Lakoff and Johnson (1966) and a staple in generative grammar textbooks:
(1) a. *I remained in England and John did so in France.
b. John solved the problem in England and I did so in France. (Lakoff and Ross 1966: II-7)
In this presentation, I discuss the do-so test: how does it work, what are its limits, and what does it prove, if anything, about the inner constituent structure of a VP? More specifically, the following questions will be addressed:
1. Can the do so test be tweaked to function as a diagnostic for the constituency of an entire VP? It would be a pity if there were no real pro-form for an entire VP, including adjuncts.
2. Does the do so test work with all kinds of predicates? What about stative ones, negative ones, or passive ones? And if it does not work with stative ones, as Lakoff and Ross (1966: II-5) claimed, why then did their demonstration of the complement/adjunct contrast include such a verb, as shown above, in (1a)? Furthermore, what do we do with so-called ‘argumental to-phrases’ (cf. Miller 1990; 2014; Huddleston and Pullum 2002: 1533; McInnerney 2024)?
3. Does the do so test demonstrate that the verb forms a tighter unit – an inner syntactic constituent – with its complement(s) than it does with any adjunct, or are there grounds for considering complements and adjuncts as ‘sisters’ of the verb, sitting shoulder by shoulder in a ‘flat’ VP? With Culicover and Jackendoff (2005: 124–126), I argue for the latter option, based on rare but possible cases in which do so picks up a discontinuous string of a verb and a complement.

Communication de Bert CAPPELLE (Université de Lille, Savoirs, Textes, Langage UMR 8163)