In Talmy's (1991; 2000) binary typology, verb(V)-framed languages systematically encode PATH (or `direction') in verbs, e.g. {`cross the river swimming' / `enter the house running'}, whilst satellite(S)-framed languages do so in adpositions, e.g. {swim across the river / run in(to) the house}. Several researchers have suggested that this typology might be formalized as a parameter at the whole-language level (e.g. Jackendoff, 1990; Snyder, 1995); in the wake of such proposals, Inagaki (2001) provides his own formal analysis of the phenomenon, and uses the results of a bidirectional study involving English learners of Japanese (V-framed) and Japanese learners of English (S-framed) to argue that interlanguage argument structures are the result of full transfer of L1 parameter settings in this domain (in support of the Full Transfer/Full Access Hypothesis of Schwartz and Sprouse, 1994, 1996). I draw on an original L1 study in which utterances with directional predicates were elicited from Japanese, French and English children and adults. The results of this study indicate that Talmy's descriptive generalization resists formalization as a parameter; therefore, there can be no transfer of a path parameter setting in L2 acquisition. An analysis of utterances containing MANNER verbs and PPs with directional interpretation revealed that the three languages fell into discrete response categories, which implies that Japanese and French do not share a single parameter setting. Moreover, both lexicalization types were found in each language, so that the `English setting' characterizes several Japanese utterances and vice-versa. In addition, semantic features and principles of syntactic computation appear to be uniform across the three languages, such that all allow certain classes of MANNER verb to combine with locational P with a directional interpretation, among other commonalities. Differences are argued to be between individual lexical items rather than particular languages, and the relevant (arguably universal) syntactic principles appear to be in place from the earliest tested stages of development. The elicited production data of the above-mentioned L1 experiment contained 68 Japanese utterances of the opposite conflation type, all confirmed as grammatical in colloquial speech by native informants. However, the relevant test sentences in Inagaki (2001) were presented as written stimuli, which created a bias for rejection of this colloquial pattern by native speakers. It follows that the broad acceptance of this alternative conflation pattern by English learners of Japanese is not necessarily a transfer effect; learners may have generalized across narrow conflation classes on the basis of L2 input. That said, verb classes differ between Japanese and English, and transfer effects certainly do obtain in many instances; I argue that such cases can be most fully explained on the assumption of Lefebvre's (1998) Relexification Hypothesis as a model of transfer in L2 acquisition (Sprouse, to appear). These related investigations of the acquisition of directional verbs and adpositions by children and adults call into question the idea of a `path parameter' at the level of the whole language, and point toward a lexicalist account of variation in the linguistic expression of motion events.