« The Semantics and Polysemy of Goal Marking Postpositions in Japanese John Beavers Department of Linguistics Stanford University Stanford, CA, ...»
The Semantics and Polysemy of Goal Marking Postpositions in Japanese
Department of Linguistics
Stanford, CA, 94305-2150, USA
Japanese has three purported goal-marking postpositions, -ni and -e on the one hand and -made
on the other, where -ni/e only occur with “motion+path” predicates while -made occurs with all
motion predicates, including “motion+manner” predicates. Previous analyses assumed that all
of these postpositions are contributing goal semantics and that the distributional differences are due to additional semantics associated with -made, such as predicativity or path semantics. In this paper I show that the distributional differences are more complicated than just the occurrence with motion+manner predicates and that these distributional differences should instead be pinned down to something more fundamental, namely that -made is a generic event delimiter and -ni/e are argument markers. Finally, I generalize this analysis in light of non-motion uses of -ni/e and
-made in a way that captures all of their uses while avoiding any motion-based polysemy and also points towards possible areas of exploration in the interaction of events and predicates.
Keywords: lexical semantics, motion, Japanese, goals, typology 1 Introduction In this paper I’ll investigate the different lexical semantic properties of so-called “goal-marking” postpositions in Japanese, and the interrelationship between properties of events and predicates that these postpositions illustrate. The most widely cited data on distributional differences of goal-marking
postpositions are data as in (1):1
(1) (a) John-wa eki-made/ni itta/modotta/orita.
John-TOP station-until/to went/went-up/went.down ‘John went/went up/went down to the station.’ (b) John-wa kishi-made/*ni oyoida/tadayotta/hatta.
John-TOP shore-until/to swam/drifted/crawled ‘John swam/drifted/crawled to the shore.’ ¡ I would like to thank ﬁrst and foremost David Oshima for his native speaker judgments, thoughtful discussion, and patience in helping me get a handle on the data presented here. I’d also like to thank Tim Baldwin, Hana Filip, Beth Levin, Peter Sells, and Maarika Traat for their help and comments. Any errors and omissions are entirely my own.
In motion constructions -ni and -e appear to have the same distribution and thus I’ll not distinguish between them and simply refer to both as -ni. The primary difference in distribution is that -ni may occur in other verbal contexts, as I show below, but -e may not, while -e may appear marking arguments in nominal constructions but -ni may not, although neither point is relevant for this analysis. Suggestive of my ﬁnal analysis, I’ll uniformly gloss -made as ‘until’ and -ni as ‘to’.
In (1a) a motion event occurs in which the ultimate goal is expressible by either -ni or -made, but in (1b) only -made but not -ni is an acceptable goal marker. The difference seems to hinge on the lexical class of the motion verb. Following work on the typology of motion expressions (Talmy, 1975, 1985, inter alia), the verbs in (1a) are often loosely referred to as “motion+path” verbs, which conﬂate the expression of both motion and path semantics, and the verbs in (1b) are loosely termed “motion+manner” verbs which do not conﬂate path semantics but instead express the manner of motion in their semantics. Many analyses of goal-marking postpositions in Japanese have focused on this lexical distinction. Tsujimura (1994), for instance, pins the distinctions down to differences in the unaccusativity of these classes, wherein -made is adding predicative goal and path semantics to unergative motion+manner verbs, something that -ni cannot do. Following Talmy’s typology more directly, Inagaki (2001) claims that while -ni and -made are both goal-markers, i.e. contributing goal/locational semantics, the difference between them is mainly that -made also incorporates path semantics and thus may occur with motion+manner verbs as in (1b), whereas -ni has only location semantics and is only licensed when the verb contributes path semantics as in (1a). Details aside, most accounts have focused in some way or another on data like (1) and the lexical classes of the verbs.
In this paper, however, I present new evidence that such analyses are inadequate on two counts:
(a) even with motion+path verbs -ni and -made show differing distributional properties indicating further semantic differences and (b) both -ni and -made have other uses outside of motion constructions, so that attributing goal-marking or path semantics to either of them increases their polysemy unnecessarily. I argue that by looking at the subtler differences between -ni and -made both in and out of motion constructions we can instead posit the much more general semantics in (2) for each of them, providing a more uniﬁed analysis that covers all of their uses without polysemy.
(2) (a) -made marks arbitrary boundaries of durative events, which for motion events is pragmatically construed as a goal.
(b) -ni marks participants of predicates, which for motion events is the goal. 2 In 2 I’ll detail two previous analyses that attribute speciﬁc goal semantics to each marker. In 3 I’ll discuss some new data that highlights aspectual differences between -ni and -made within motion constructions, and I’ll sketch a rough lexical semantic characterization of their behavior in motion
into the typology of motion constructions of Talmy (2000), and I’ll conclude in 7.
2 Background Most of the literature on goal marking in Japanese has focused on data as in (1), showing that -ni, unlike -made, can only occur with motion+path verbs. Typically these analyses assume -ni and -made are two different kinds of goal markers. Tsujimura (1994), for instance, claims that -ni realizes the inherent goal of directed motion verbs whereas -made is actually predicative, predicating a result state of the ﬁgure. She cites as evidence a class of unaccusative mismatches that occur with motion+manner verbs, wherein normally unergative motion+manner verbs act more like unaccusative motion+path verbs when they cooccur with -made, as can be seen from the use of numeral quantiﬁers as in (3) (coindexation added):
By “participant” I mean an entity whose involvement in the event is entailed by the verbal predicate, e.g. predicate arguments. My intention in using a generic term such as “participant” (rather than, say, “argument”) is simply to indicate that the entity marked by -ni is entailed by the predicate itself without making any claims about how that realization occurs syntactically.
(3) (a) ?*Kodomo-ga [¡ inu-to awatete san-nin hashitta/aruita].
£ ¢ child-NOM dog-with hurriedly three-cl ran/walked ‘Three children ran/walked hurriedly with a dog.’ (b) Kodomo-ga [¡ inu-to awatete san-nin kooen-made hashitta/aruita].
¤ ¢ child-NOM dog-with hurriedly three-cl park-up-to ran/walked ‘Three children ran/walked hurriedly to the park with a dog.’ (Tsujimura, 1994, cf. (16)) Assuming that numeral quantiﬁers must be in a mutual c-command relationship with their antecedents, (3a) is predicted to be ungrammatical since the unergative subject is base-generated in a speciﬁer position, but (3b), with the -made phrase, is acceptable, indicating that the subject in this case is base-generated in direct object position, necessary for the predication by -made. Since -ni isn’t predicative it may only occur with motion+path verbs already predicating a result state. So far this explains the occurrence of -made with motion+manner verbs; it does not explain the fact that
-made, like -ni, can also occur with predicative motion+path verbs. To account for this Tsujimura hypothesizes two variants of -made, one predicative and the other a non-predicative goal-marker on par with -ni.
Inagaki (2001) proposes a similar analysis by trying to derive the predicativity from the types of semantics that are incorporated into each lexical item (following Hale and Keyser (1993)). In particular, he argues that -ni only marks location, whereas -made incorporates both location and path se mantics (a point I’ll return to in 6). Motion+path verbs likewise incorporate path semantics, whereas motion+manner verbs do not, thus requiring the presence of some other lexical item to explicitly
express goals. The four possible combinations are outlined in (4):
(5) (John is some distance from the train station.) John-wa (eki-ni) 30-pun-de/*kan itta/modotta/orita.
John-TOP station-to 30-minutes-in/for went/went.up/went.down ‘John went/went up/went down (to the station) in 30 minutes’ These verbs are only acceptable with -de ‘in’ temporal modiﬁers and not -kan ‘for’ (see, for instance, Dowty (1979) on these tests in general and Yoneyama (1986) on Japanese). This is true regardless of whether the goal is overtly realized or not, and there is always a lexically encoded entailment of some sort of arrival at a new (perhaps unspeciﬁed) location. When the goal is not
overtly realized, the goal is inferred from context:
Therefore, when not explicitly stated the entailed goal itself is inferable from context. The most important fact is that these verbs always entail some goal regardless of whether it’s overtly realized.
With these observations in mind, I’ll now address how -ni and -made interact with the aspect and paths involved in these motion events. First, while many motion+path verbs allow goal-marking by
both postpositions as in (1a), not all necessarily do, as shown in (7):
(7) John-wa eki-ni/*made haitta.
John-TOP station-to/until went.in ‘John went into the station.’ The difference between the verbs in (1a) and the verb hairu ‘go in’ in (7) is simply that the verbs in (1a) are typically accomplishment verbs and hairu is an achievement verb, in the Vendler aspectual classiﬁcation system (Dowty (1979); see Kindaichi (1976), Tsujimura (1996) for a discussion of aspectual classes in Japanese). Accomplishments are verbs that are telic and are durative (have some sort of process associated with them), whereas achievements are telic but are typically perceived of as instantaneous. Therefore on a ﬁrst cut it would appear that -made phrases must combine with durative verbs. However, as is typical with aspectual classiﬁcations, the pattern is more complicated when contextual factors massage the interpretation of events described by particular predicates. For instance, -made can occur with a verb like hairu ‘go in’ as in (8) in the given context (cf.
(8) (John is outside a wall around the perimeter of the tower’s area, which is a short walk from the door of the tower, and his goal is the chamber inside the tower) ?John-wa too-made haitta.
John-TOP tower-until went.in ‘John went in as far as the tower’ Although the context is a bit forced in (8), it does have the effect of changing the instantaneousness of the event to something slightly more durative (by creating a more complex ground into which the ﬁgure must enter, a point I’ll return to in a moment) and the context considerably improves the reading of -made with the verb hairu. This suggests that -made must mark the goal of an event that is somehow durative (regardless of the verb), unlike -ni, which, as shown in (1a) and (7), can occur with both durative and non-durative events. The ﬂip side to this aspectual difference is the different implications both -ni and -made make about the path of motion. First, when -made is used, there is an implication that the path of motion must somehow be complex, so that there are other possible goals before or after the goal marked by -made. This is a difﬁcult implication to test, but certain contexts
highlight this. Consider the pairs in (9):
Although not entirely unacceptable, in the context given for (9a) where John is on the ﬁrst ﬂoor of a two story building, -made is a signiﬁcantly less natural goal marker than -ni (marked by ?#).
The same sentence with -made becomes much more acceptable/natural in the context that there are additional ﬂoors beyond the second ﬂoor.5 This is not simply a restriction against -made marking the
endpoint of speciﬁc paths as the data in (10) suggests:
(10) (John is on the ﬁrst ﬂoor of a ten story building) John-ga juk-kai-ni/made nobotta John-NOM 10th-CL.ﬂoor-to/until went.up ‘John went up to the tenth ﬂoor.’ This is likely due to the construal of an event of going from one ﬂoor to the next as somehow discrete, without intermediate goals.
Even when the explicit goal is the ultimate endpoint of the path, -made is acceptable as a marker of that goal provided that there were intermediate possible goals. This distinction can also be found in the English John went up until the second ﬂoor, which sounds far more natural in the context of there being multiple ﬂoors beyond the second ﬂoor and not as natural (as John went up to the second ﬂoor) when there are only two stories. That this is the case shouldn’t be surprising given what we’ve already seen about the aspectual nature of -made. The aspect of motion verbs is often analyzed in terms of a homomorphic mapping between the path of motion and the motion event (see Dowty (1991), Tenny (1994), Jackendoff (1996), Krifka (1998)). Under such an assumption, the durativity and path complexity requirements are in fact one and the same, since a complex path would entail a complex (i.e. durative) event and vice versa. The acceptability of -made with hairu in a more durative context in (8) also necessitated understanding the path as more complicated than a typically binary distinction of inside/outside. The particle -ni, on the other hand, does not require any such path complexity and marks a goal regardless of the nature of the path.
There is one further distributional difference between -made and -ni. For a motion event with a complex path, only -made is a natural marker of a goal that is not the contextually intended goal of