-
into a TO path). Now, imagine an enclosure around point A with an opening
at its south side and point B to its north. Because of the enclosure, one can
only go from A to B going southwards, through the opening. To go from A
toward B, however, one should go north. Crucially, the TOWARD path in this
situation cannot be extended in the same direction to become a bounded to B
path. Thus, instead of directionality, toward rather seems to expresses
orientation. When modifying a motion event with this expression, the moving object of
course ends up closer to the ground. And by continuing along this direction, one
will generally end up at this ground too. But this need not be, as this example
shows.
As a second objection, the path reference that is assumed by Jackendo
may follow from world knowledge instead of being part of the lexical semantics
of directionality (cf. the procedure for the development of a spatial ontology
in [
1
] and the principle of conceptual abstraction in [
16, 595
]). We know that
it takes a path from A to B to go from A to B as we cannot but traverse
all points in between when going there. Crucially, however, such paths are not
necessarily what is referred to by directionality expressions. In fact, directionality
expressions are probably better analyzed as predicates (cf. [
16
], [
17
]).
Finally, if directional PPs referred to paths, it should be possible to combine
an expression of duration with the continuation along such a path. But this is
not possible with Goal directionality as illustrated by the ungrammaticality of
He is walking into the building (*for hours). (cf. also [
18
]). Note that this is
not due to the semantics of the verb to walk : If explicit reference is made to a
path, by substituting through for into in the example, it is possible to use an
expression of duration. Thus, the contrast between these sentences shows that
(Goal) directionality is probably not about paths.
2
Directionality in terms of locations
A more careful procedure than the one used by [
7
] can be followed in the
collection of the data. It has been observed that more grammatical means of expression
tend to make less idiosyncratic meaning distinctions ([
14, 178
], [
6, 178
], [
1, 1035
]).
Also, it has been found that spatial cases primarily express directionality ([
8
],
[
13
], [
2
]). Finally, cross-linguistic agreement is said to suggest relative uniformity
in the way people conceptualize a domain [
4
]. So not only can we indeed expect
directionality distinctions to be expressed by spatial case systems, also, we can
expect whatever directionality distinctions that are made by spatial case to be
of a more fundamental, conceptual type, especially when they show up in
language after language. In a cross-linguistic study of spatial case inventories that is
thus motivated, Lestrade [
10
] identi es three basic distinctions of directionality:
Place, Goal, and Source.
This kernel of directionality could be described in terms of paths, but,
arguably, it is preferable to use locations only as we need these anyway for the
con guration function. Then, Goal and Source directionality denote a change of
location, and Place denotes an absence of such a change. To de ne Goal and
Source, we need some ordered dimension: Goal directionality denotes a change
into some location, Source does the opposite.
An ordered dimension can be provided for free by the extended event
structure of the verb. Pustejovsky ([
12
]) argues that Davidsonian event arguments
may have internal structure. For our present purposes, only the structure in
which there is a strict partial order between the two subevents is relevant:
(1)
a. [e3 e1 < e2 ] =def < (fe1; e2g; e3)
b. 8e1; e2; e3[< (fe1; e2g; e3) $ e1 e3 ^ e2
8e[e e3 ! e = e1 _ e = e2 ]]
e3 ^ e1 < e2 ^
In this de nition, event e3 is a complex event structure that consists of two
subevents, e1 and e2, where e1 and e2 are temporally ordered such that each is
a logical part of e3, the rst subevent precedes the second, and there is no other
event that is part of e3 [
12, 69
]. For example, the verb build is analyzed into a
development process and a resulting state.
Pustejovsky [
12, 74
] explicitly allows for adverbial phrases to take scope over
both the entire event and over individual subevents. Thus, we have three logical
possibilities for spatial modi cation of motion verbs, which nicely corresponds
to the empirically established kernel: the spatial modi cation of the entire event
is called Place directionality (note the di erent use of this term here from the
one by Jackendo in the above); the modi cation of the rst subevent is called
Source, and the modi cation of the second subevent is called Goal. For example,
depending on the type of directionality that is imposed by the spatial modi er
and assuming the structure in (1), a walking event e3 of subject x modi ed by
location y can be decomposed as follows: [walk(e3; x) ^ loc(e3; x; y)] for Place,
[walk(e1; x) ^ loc(e2; x; y)] for Goal, and [loc(e1; x; y) ^ walk(e2; x)] for Source.
In principle, the explicit spatial modi cation of one subevent by location y
does not exclude the additional implicit modi cation of the second subevent by
this same location. Following a suggestion of Hendriks et al. [5, chapter 8], we
can ensure a change of location in a system of pragmatic contrasts (cf. also [
15
]):
The speaker would have modi ed the whole event if the location had scope over
the whole event, so if she only modi es the rst subevent, we know that the
locative function does not apply to the second one by pragmatic implicature.
By only using existing structures that have been established independently
from present purposes, the proposal meets the criterion of Bateman et al. [
1
]
to exclude the contribution of world knowledge in the development of a spatial
ontology. Also, the account straightforwardly accounts for syncretism patterns in
directionality systems. It has been observed that such syncretisms occur between
Place and Source, between Place and Goal, or between all three distinctions, but
not between Source and Goal to the exclusion of Place (cf. [
2
], [
10
], [
11
]). This
naturally follows from the present proposal: If a language has a special marker
for the spatial modi cation of the rst subevent, the second subevent and the
entire event will be treated uniformly as its complement (and the other way
around), but taking together the two subevents would render the entire event
(cf. (1-b)) and therefore could not be distinguished from it. Finally, the temporal
speci cation of a spatial modi cation does not impose any speci c ontological
category to this modi cation and is thus compatible with both geometrical and
functional representations of space [
3
].
In conclusion, it was argued that directionality is best analyzed as the
locative modi cation of an extended event structure. This accounts for the
empirically established kernel of directionality, correctly predicts attested syncretism
patterns, and does not stipulate any additional machinery.