« The Semantics and Polysemy of Goal Marking Postpositions in Japanese John Beavers Department of Linguistics Stanford University Stanford, CA, ...»
When the goal marked is the intended or contextual one, as in (11a), then either -ni or -made is acceptable. When the actual goal is not the intended one, as in (11b), then -made is the only natural way to mark the actual goal. In order for (11b) to be acceptable with -ni the context of the event would have to be reconstrued (for instance, John suddenly became violently ill and decided it would be best to simply stop at the tenth ﬂoor, just changing the intended goal) but even here -made is perhaps the more natural postposition. These data show that even with motion+path verbs -ni and -made are not entirely interchangeable. In the next section I will summarize the differences between -ni and -made we’ve seen so far and sketch a rough analysis that captures these additional observations.
4 Summary and Interim Analysis
In summary, we see the following distributional restrictions relative to motion predicates:
The analyses in 2 focused entirely on the ﬁrst restriction in (12), pinning the difference to the predicative/path semantics of -made vs. the non-predicative/location semantics of -ni. However, these analyses will not capture the remainder of the restrictions. In this section I’ll incorporate all of the new evidence in (12) and argue instead that the difference is really in the interactions with predicate properties vs. event properties.
Looking at the ﬁrst two properties in (12), restrictions on the predicate type and goal, only -ni but not -made has any such restrictions. The ﬁrst of these properties is entirely a property based on predicates, not events. The property of being directed motion with a goal is certainly a property of events, but -ni may only realize that goal with respect to speciﬁc predicate types. For example, an event of running to a store could be described by any appropriate predicate, such as hashiru ‘run’ or iku ‘go’, but -ni is only compatible with one of those, the motion+path class, regardless of the underlying nature of the event. Therefore, the predicate type restriction is purely a restriction based on properties of the predicate. The goal constraint, on the other hand, is more of an event property, since it refers to speciﬁc events in speciﬁc contexts. However, even here these constraints are somewhat mediated by the predicate, which must entail the goal and puts restrictions on acceptable goals (e.g.
interior spaces for certain uses of hairu ‘enter’, etc.). Regardless, from these two restrictions we see that -ni is constrained by both properties related purely to the predicates and properties of the underlying event however they’re mediated.
The second set of constraints in (12) are aspect and path, and here only -made but not -ni imposes any constraints. These properties, however, are properties of events and not of predicates. Looking ﬁrst at durativity, deﬁning durativity as subdividablity into smaller subevents, for instance, this would appear not to be true, since speciﬁc predicates can impose certain granularities on events that will affect their subdividability. For instance an event of entering a station may be construed duratively as in John walked through the door (in ﬁve seconds) or punctually as John entered the station (*in ﬁve seconds), so that the durativity appears to be a predicate property. However, this is not necessarily true, since speciﬁc contexts can trump the constraints predicates impose on durativity. This was seen above in (10), (*)John-ga too-made haitta ‘John entered until the tower’, infelicitous in the context of John standing just outside a tower but quite felicitous if the event of entering is contextually construed as complex. Therefore while predicates have some inﬂuence on aspect, ultimately these construals are subordinated by properties of the event.6 The same applies to the path structure as well, which is intimately tied to the aspectual structure as noted above. Therefore, it appears that -made, unlike -ni, is only sensitive to restrictions on properties of events. Collapsing the interrelated properties of (12)
into two categories, event vs. predicate properties, we get the simpler picture in (13):
(13) -ni -made Predicate Properties restricted no restriction Event Properties restricted restricted While -ni is sensitive to both kinds of properties (although mediated by the predicate in all cases),
-made only has restrictions based on events. This is a rough characterization of the data seen before, but it points to a difference between these two postpositions that can’t be boiled down to calling them just two different, largely synonymous goal markers. If we make the intuitive assumption that events have many autonomous properties (e.g. agents, goals, results, etc.), but predicates only pick out a subset of these to encode (e.g. agents and goals in the case of motion+path predicates), then we could
instead assume the following generalization:
(14) (a) -made marks goals of durative events (regardless of the predicate).
A more famous version of this observation is punctual draw a circle, a predicate normally and almost necessarily used for durative events but that may be reconstrued as punctual, say by predicating of a drawing event by a special printer with a large circular jet that creates all points of a circle on a piece of paper simultaneously (Verkuyl, 1993).
(b) -ni marks goals of motion+path predicates.
This generalization assumes that the distributional differences of these postpositions stem from a radical difference in their basic functions that also explains some of the more subtle differences in their distributions. The predicate and event restrictions of -ni are explained since it is licensed only by motion+path predicates over motion events, allowing restrictions based on both predicates and events.
(For now I’ll assume that the event property of -ni, that it only realizes the contextual/intended goal, is mediated by the predicate and its relationship with the event and does not require speciﬁc semantics attributed to -ni, although I won’t offer an analysis of how this works.) On the other hand, the lack of predicate constraints on -made is understood as part of its independence from predicates and focus just on (durative) events. In the next section I’ll present additional data which not only further validates this distinction but allows us to generalize (14) even more by stripping away the explicit goal-marking semantics.
5 Additional Uses of -ni and -made Both -ni and -made have uses outside of motion constructions as well, and here the generalization in (13) is further supported without reference to goal-marking at all. 7 In general, -made serves a range
of functions in delimiting events:
In such uses Kuno (1973) deﬁnes -made as ‘continuously until/to X’, claiming it requires a durative verb.8 These uses of -made are similar to those in motion constructions, including the durativity requirements. By contrast, -ni is often analyzed as an argument marker (the glosses are from the
(16) (a) Mary-ga boku-ni kono hon-o kureta Mary-NOM I-DAT this book-ACC gave ‘Mary gave me this book’ (Dative/Recipient) Here the distribution of -ni vs. -e differs, wherein -e may only be used in motion constructions and not in any of the uses presented here. Therefore my ﬁnal analysis only applies to -ni but not -e.
Variant forms of -made, namely -made de ‘(do something) continuously until/up to X, (and stop it at X)’, where the action could potentially be continued sometime after the deadline, and -made ni ‘in the domain delimited by X at the farthest end’, likewise serve delimiting functions regardless of the predicate type.
(b) John-wa Mary-ni hon-o yom-ase-ru John-TOP Mary-DAT book-ACC read-CAUSE-NON.PAST ‘John will make Mary read a book’ (Causee) (c) Teeburu-no-ue-ni koppu-ga aru table-GEN-top-at cup-NOM exists ‘There are cups on the table.’ (Location in Existential) (d) E-ga doroboo-ni nusum-are-ta.
painting-NOM thief-by steal-PASS-PAST ‘The painting was stolen by the thief’ ((Logical) Subject of Passive) (Kuno (1973, (3a),(7),pp.127-139,(5a), p.352), Tsujimura (1996, (169), p.233)) In all of these cases -ni is marking a non-nominative, non-accusative participant in the predicate’s argument structure and the use of -ni is governed by particular verbs or construction, consistent with the more typical analysis of (non-motion) -ni as simply a dative case marker, e.g. the non-highest, non-lowest structural case marker as in Gamerschlag (2000) (cf. Kiparsky (1989), Wunderlich (1997), although I’ve generally avoided argument/case-marking terminology in an attempt to remain agnostic about the syntax of these constructions).
Incorporating these uses into the motion analysis in the previous section, it seems that there are some commonalities with (14):
-ni is always marking some kind of participant and -made always delimiting an event by some bound, although the nature of the participant/bound is different in different contexts. But if we assume that these postpositions carry goal-marking semantics in motion constructions and other semantics in other constructions, then this increases their polysemy signiﬁcantly. So is there some more general semantics we can posit that doesn’t require any polysemy?
For -made the answer seems obvious: if we suppose that all -made is doing is delimiting events, then it’s just a matter of pragmatic and lexical semantic inference, based on the type of bound the object of -made describes and the type of event it modiﬁes, that determines how that bound is interpreted. For instance, when the bound is a temporal expression as in (15a), the interpretation is that the event ends at that time (regardless of the nature of the event), and when the bound is a proposition as in (15d), the event ends at the time that proposition obtains (again regardless of the nature of the event). The interaction of spatial bounds and motion events is similar but a bit more complex. When
-made delimits an event of motion by a spatial bound then boundedness can’t simply be temporal, since spatial bounds don’t necessarily have temporal components. Instead, spatial bounds must have some other kind of interpretation, and in these cases the interpretation is always that the spatial bound is a goal. However, this is quite a natural inference. Motion+path predicates already entail delimitation by a goal, i.e. a spatial boundary at which the ﬁgure ultimately arrives, so any spatial bound
-made supplies must be reconciled with this entailment somehow, and the only possible interpretation is that -made’s complement is a realization of the entailed goal. For motion+manner predicates the inference is similar: these predicates entail motion of some ﬁgure through space and the most natural interpretation of a spatial bound would be that the ﬁgure arrives at this location, thus forcing a goal interpretation even if no goal is entailed by the predicate. Although this is only a sketch of an account, by relying on inferences about the type of bound -made takes as a complement and the type of event it modiﬁes we do not need to say anything more about -made except that it’s just a generic event delimiter.
A similar argument can be made for -ni. Goals, causees, passive agents, etc., are all participants entailed by certain predicates, and thus it’s possible to assume that the speciﬁc participant -ni marks is determined entirely by the predicate (via some kind of linking theory, e.g. as a dative case marker following Gamerschlag). Since each predicate entails the appropriate -role for the -ni-marked entity anyway, there’s no reason to assume that -ni itself must carry any semantics, and speciﬁcally it needn’t carry any speciﬁc goal marking semantics at all. Taking the generalizations over both of these
postpositions together, we can postulate the more general lexical semantics in (17):
(17) (a) -made marks arbitrary boundaries of durative events (where the interaction with predicate participants is determined pragmatically).
(b) -ni marks participants of predicates (where the speciﬁc participant is determined by the predicate).
discussed so far, unlike the analyses proposed in 2. Whether this is enough lexical semantics to attribute to these two markers so that, with a properly worked out theory of argument structure and event structure, all of the data is explained is a matter for future investigation.
6 A Word on Typology Before concluding it’s also worth comparing the analysis in (17) with those in 2 on typological grounds. Following recent typological work by Talmy (2000), Japanese can be considered a “verbframed” language, along with French and Spanish, wherein the characteristic typological pattern of expressing paths of motion is via motion+path verbs, as opposed to “satellite-framed” languages like English or German where the characteristic expression of path is always via some satellite (e.g.
PPs) and the verbs are typically motion+manner verbs (cf. Yoneyama (1986), Wienold (1995), Slobin (2000)). In each case, the path incorporating element deﬁnes what Talmy refers to as the “framing event”, roughly the part of the event that structures its overall nature and course, and the non-path element deﬁnes the “co-event”, which for motion events is typically a manner expression that is incidental to and parasitic on the framing event. Most important for the issue at hand is that in verbframed languages like Japanese satellites prototypically do not express path semantics. However, attributing path semantics to -made is exactly the analysis proposed by Inagaki (2001) (and to a certain degree by Tsujimura (1994)), thus making -made exceptional not only among Japanese postpositions but also in terms of this typology. Although these lexicalization patterns are by no means absolute (in particular, the satellite-framed language English has a wealth of Latinate motion+path verbs), the exceptionality in Inagaki’s analysis is striking.
On the account proposed in (17), however, this is no longer a problem. The proposed deﬁnition of