Non-veridical uses of Japanese expressions of temporal precedence

This paper discusses two Japanese expressions for asserting that an eventuality a precedes another eventuality b: B mae ni A and B-nai uchi ni A (where A and B are clauses referring to a and b); cf. (1). Both expressions are non-veridical in that they do not entail B. This property is shared with English A before B, where it has been studied extensively (Anscombe 1964; Heinämäki 1972, 1974; Valencia et al. 1992; Ogihara 1995; Beaver and Condoravdi 2003). We compare the two Japanese expressions with each other and with before, with special emphasis on the counterfactual and modal implicatures carried by their non-veridical uses.

(1)	a.	[Boku-wa] I-top ziko-ga accident-sub okoru occur mae-ni before-loc basu-o bus-acc orita exit-past
	b.	[Boku-wa] I-top ziko-ga accident-subj okoranai occur-neg uti-ni within-loc basu-o bus-acc orita exit-past
		I got off the bus before there was an accident.

It is also often observed that English A before B on its non-veridical use supports (i) the implicature that B is or was likely at the time of A; or (ii) the counterfactual implicature if not-A, would (have been) B. In fact, (i) and (ii) are often conflated and treated as different ways of stating the same condition (e.g., Beaver and Condorvadi). They are not equivalent, however: The counterfactual does not require for its truth that B was likely at the time of A. Consider (2), and suppose an accident occurred but was highly unlikely and not foreseeable at the time the speaker got off. Then (2a) is false, but the posterior counterfactual (2b) is true. In fact, (2a) is intuitively more similar to the prior predictive conditional (2c). Thus a counterfactual analysis would make different predictions from a likelihood analysis, and the latter would seem to be more suitable for before.

(2)	a.	I got off the bus before I got injured.
	b.	If I hadn't gotten off the bus, I would have been injured.
	c.	If I don't get off the bus, I will get injured.

In Japanese, on the other hand, in the situation described, both (3a) and the counterfactual (4a) are true, whereas (3b), like (2a) and the prior predictive conditional (4b), implies that the danger was imminent when the speaker got off the bus.

(3)	a.	[Ziko-de] accident-caus kega suru get injured mae-ni before-loc basu-o bus-acc orita exit-past
	b.	[Ziko-de] accident-caus kega sinai get injured-neg uti-ni within-loc basu-o bus-acc orita exit-past
		I got off the bus before I got injured.

(4)	a.	Basu-o bus-acc ori-tei-nakat-tara get off-perf-neg-cond kega si-tei-ta get injured-perf-past
		If I hadn't gotten off the bus, I would have been injured.
	b.	Basu-o bus-acc ori-nakere-ba get off-neg-cond kega suru get injured
		If I don't get off the bus, I will get injured.

Our analysis proceeds from the standard possible-worlds account of English before, which can be adapted to mae-ni fairly straightforwardly. (In doing so, we discuss certain unresolved problems facing Beaver and Condoravdi's implementation). As for B-nai uti ni A, B-nai denotes a state, and uti ni locates A within this state and presupposes that the state is bounded. (The latter is a lexical property commonly found in temporal connectives across languages, which in Japanese distinguishes uti from mama and aida 'during/while'.) This presupposition in turn implies that B is expected to occur, for that is the only way for a state of B's not-occurring to end. This inference, while defeasible, strengthens the meaning from while not-B to before B. It is also the reason why B-nai uti ni A implies that B is/was imminent at the time of A, unlike B mae ni A (also unlike B-nai mama).

We conclude with general remarks about related phenomena at the interface between modality and temporal reference, and point out some residual problems which seem to call for a more fundamental rethinking of the technical apparatus than we are proposing in this paper.