• No results found

Raising to object in PPs

In document The nanosyntax of case (sider 186-194)

4.7 Case and word order

4.7.4 Raising to object in PPs

In this section, I discuss a novel argument for the claim that case and position in the linear string are related. The example comes from the domain of PPs,

and the discussion draws on Caha (to appear). I will be using mostly German examples, but the same facts can be replicated in a number of other languages which I mention as we go.

What I want to show is that there are PPs where a measure phrase (base-generated above the preposition) is marked by the same case which normally occurs on the Ground argument (the semantic complement of P). Crucially, when this happens, the measure phrase fails to occupy its regular position, and necessarily switches to the word-order customary for Grounds. This can be captured if bearing a particular case entails movement to a particular position, but remains unaccounted for otherwise. Crucially, this behavior is exception-less, as far as I know, basing the conclusion on the sample of 53 languages discussed in Haspelmath (1997) (and a couple of others).

As a general background leading to the argument, consider a connection between the spatial and temporal domain in German, namely the homophony between spatial ‘in front of’ (68a) and temporal ‘before’ (68b), bothvor.

(68) a. vor

‘The dinosaurs died out before the ice age.’

(Haspelmath 1997:ex.10a) A typological study by Haspelmath (1997) reveals that such a connection between spatialin front of and temporalbefore is quite common in languages.

A possible interpretation of the pattern in (68) is that the adposition vor

‘in front of’ is syntactically and semantically identical to vor ‘before.’ The possibility of vor to be used in both contexts would then be due to the fact that space and time are (cognitively) structured in similar ways.34

Under such an interpretation, vor in (68a) locates the Figure along an axis projected from the center of the house through its “front.”35 In (68b), the same preposition does the same job; it locates a Figure (the event of dinosaurs’

dying out) on the (time-)axis projected through (what is conceptualized as) the

“front” of the period denoted by ice age. This will lead to the right semantics just in case the front part of an interval is its beginning. The explanation for this usually relies on a cognitive model of time; I do not go into this here in detail; see e.g. Jackendoff (1983:ch.10) and Haspelmath (1997:§4.2).

A use of vor which is interestingly different from (68) is in (69). Here again we see vor in a temporal use, but at first blush, the meaning is not

34See, e.g., Jackendoff (1983:ch.10) for such a proposal, and Haspelmath (1997:ch.1) for literature overview.

35See, e.g., Levinson (2003) for how the front axis is anchored in the object, and Zwarts and Winter (2000) for a formal implementation of “located along an axis.”

compositional;vor einem Monat does not mean: an event X is located along an axis projected through the beginning of a(n arbitrary) month(-long interval).

What (69) means is ‘a month before the utterance time.’

(69) vor before

ein-em a-dat

Monat month

‘a month ago’

There are at least two possible analyses of (69), which maintain the idea that the meaning of spatial and temporal vor are identical, or at least related.

The first account is based on the idea of a semantic shift, and I review it immediately below. I will, however, point out certain shortcomings of this solution and propose an alternative where the problems are avoided. The alternative will say that ‘a month’ in the example above is base-generated as a measure phrase, rather than the Ground argument.

I start with the approach based on a semantic shift due to Haspelmath (1997). What Haspelmath proposes is that in order to obtain the correct meaning of (69), we have to do two things. First, we have to make sure that einem Monat ‘a month’ denotes ‘the last month,’ i.e. the month long interval preceding the utterance time. Thenvor einem Monat will locate the Figure on the time axis as preceding the last month. But as Haspelmath observes, this is not enough, because while vor der Eiszeit in (68b) can mean ‘anytime, as long as that time precedes the ice age,’ vor einem Monat rather means

‘exactly before the last month.’ Hence, we have to add that vor in (69) is semantically enriched by a component of pragmatic strengthening. I state the two ingredients as (70):

(70) a. einem Monat ‘a month’ denotes ‘the last month’

b. vor in (69) is semantically enriched by a component of pragmatic strengthening

Such a solution has at least two problems. The first is that while vor now means roughly what we would expect (modulo the strengthening), the phrase einem Monat ‘a month’ does not. Rather than solving the puzzle, we shift it from the preposition on the noun phrase.

The second point of criticism concerns the way the proposal distributes the work-load between the preposition and the noun phrase. Empirically, (70a) and (70b) are related. That is because we cannot allow ‘a month’ to denote ‘the last month’ across the board, but rather just in case it is the complement of vor. At the same time, vor obligatorily undergoes pragmatic strengthening just in case it is followed by a complement that undergoes the relevant semantic shift. However, the factual relatedness of these processes is theoretically obscured by the fact that (70a) targets the noun phrase, and (70b) targets the preposition. As things stand, we are left without a deeper

understanding of why these processes should work in tandem, except for saying that they do.

I do not think that one can improve much on the analysis with the assump-tions Haspelmath starts from. Specifically, he assumes that einem Monat is a semantic complement of the preposition, i.e., the Ground. This assumption in turn stems from the fact that the phrase ‘a month’ follows vor and bears dative. The step from “X is a syntactic complement of a preposition” to “X is a semantic complement of the preposition” is, however, not necessary, if the semantic role and the case selecting position are distinct (the standard assumption for VPs).

Hence, I propose instead that the phrase einem Monat is generated as a measure phrase, grounding the proposal in the paraphrase of vor einem Monat: ‘a month before utterance time.’36 The analysis is depicted below in (71): the semantic complement of vor is a silent deictic element UT, the utterance time.

(71) [ a month [ before = vor [ UT ] ] ]

Vor ‘before’ projects an axis through the “front” part of UT, i.e., in the direction towards past. The measure phrase ‘a month’ indicates the distance from UT to the Figure on this axis (see Zwarts and Winter 2000 for the precise semantics of measure phrases assumed here). This analysis gives us compositionally the meaning of (69) without the need to say anything special about the semantics of the expressions involved; ‘before’ means ‘before,’ ‘a month’ means ‘a month,’ and there is no pragmatic strengthening.37

36This analysis is inspired by van Riemsdijk’s (2007) analysis of Englisha month ago. In English,a month shares distribution with measure phrases, which in general appear to the left of the adposition.

37The analysis should be improved by clarifying the status of the unpronounced UT. I see two options. First, UT is pronounced as a part of the adposition, hence, the lexical entry ofvor ‘before, in front of’ is something like (i):

(i) /vor/ [ P [ UT ] ]

By the Superset Principle, when the complement is different than UT, the adposition will lexicalize just P. (There is a potential issue here with the Anchor Condition, but assuming that P is a lexical category, UT is likely to be an argument selected by some functional head in its projection. This makes the issue disappear, since the lexical P will be lower than the head selecting the argument UT, and count as an anchor.)

This analysis allows for a neat statement of the difference between the English before andago:

(ii) a. ago [ P [ UT ] ]BEFORE b. before [ P ] BEFORE

Alternatively, UT is absent in syntax altogether, and vor ‘before’ appears without

The proposal gains independent support from two facts. First, phrases similar toeinem Monat surface as measure phrases in a variety of languages (see also van Riemsdijk 2007). I illustrate this on Farsi:38

(72) Farsi, Haspelmath (1997), Marina Pantcheva (p.c.)

a. piˇs

‘two hours before the operation’

c. do

(72a) shows that the prepositionpiˇsmeans ‘before,’ and the Ground is marked by æz. In (72b), we add a measure phrase to (72a). In (72c), the measure phrase still precedes the adposition, and it is formally identical to the measure phrase in (72b). This in turn makes the measure phrase analysis of do sa’æt

‘two hours’ in (72c), the analogue ofeinem Monat, quite straightforward for Farsi.39

Second, even within German, one finds parallels between measure phrases and the object ofvor in the meaning ‘ago.’ For instance, van Riemsdijk (2007) notes that measure phrases productively combine with modifiers such ashalb

‘half,’ but ordinary Grounds do not. Now note thathalb ‘half’ is fine in (73a), but not in (73b):

The semantically neat and independently confirmed analysis in (71) leads to an obvious question, which is the main topic of this section: how come that any complement. This is parallel to adjectives: tall means in fact tall-er than standard, where the ‘standard’ of comparison is usually assumed to be missing in syntax, but filled in by default. The same reasoning applies to vor, which means

‘beforeut’ by default, i.e. when no complement is present.

38Piˇsalso means spatial ‘front,’ similarly to the Germanvor. The use ofpiˇsin spa-tial contexts, however, is restricted to particle-like uses. I thank to Marina Pantcheva for her help with the Farsi data. She notes that there is variation among speakers concerning (72), which I ignore here.

39Thus, the syntax of the measure phrase in (72c) is different in Farsi and in Ger-man. I discuss this in more detail below.

if einem Monat is a semantic measure phrase, it does not share syntax with other measure phrases? Consider (74): here the measure phrase precedes the adposition, and it is marked accusative.

(74) einen

‘a month before the concert’

To answer this question, a syntactic analysis must find (i) a way to force the appearance of dative on the measure phrase, rather than the usual accusative, and (ii) a way to make the measure phrase follow the adposition. (i) and (ii) are obviously related: to bear dative and to follow the adposition are the two properties that syntactic objects of adpositions have in German. And it is this connection between case and word order what is the core of my argument here.

Thus, looking at cross-linguistic parallels of the expression ‘a month ago,’ a generalization emerges: whenever the semantic measure phrase bears the same case which normally occurs on Grounds (as in German, but unlike in Farsi), it occupies the same position in the linear string as the Ground. This means that it follows the adposition in prepositional languages (such as German), and precedes it in postpositional languages.

To illustrate the latter point, I give an example below from Tamil.

(75) Tamil (diacritics omitted)

‘three hours ago’ (Tamil, Haspelmath 1997) In (75a), we see the postposition munnaale ‘before,’ k-selecting a dative on the Ground. (Munaale is like the Germanvor, and it also means ‘in front of’

in the spatial domain.) In (75b), the same expression (munaale) is used to mean ‘ago,’ with the measure phrase bearing dative. Crucially, it occurs to the left of the adposition, as regular Grounds do. Tamil is thus like German, but with postpositions, rather than prepositions.40

40The split in PPs between languages where the measure phrase raises to object (German, Tamil), and languages where it does not (Persian, English), is found also in VPs.

As Svenonius (2002) points out, Finnish temporal adverbials are promoted to nom-inative in passive (i), but Icelandic temporal adverbials are not (ii).

(i) a. Olen

To point out what is crucial: there appear to be no languages where case marking of the measure phrase and its position in the linear string are not in harmony. Thus, Haspelmath’s sample does not include a language where the measure phrase acquires the case of the syntactic object, but not the position, as in (76) or (77).

(76) An unattested language 1

a. P > Ground-Kx

b. Measure-Kx > P (77) An unattested language 2

a. Ground-Kx > P

b. P > Measure-Kx

Thus, acquiring the case of the syntactic object entails occupying the same position as the syntactic object. To capture the generalization, k-selection must be the result of overt movement. The details of the derivations follow.

As highlighted above, we start from the base-generated structure (78).

(78) [ a month [ BEFORE [ UT ] ] ]

The facts discussed here require that adpositional phrases include a k-selecting position which attracts a KP*. I understand this position as the PP analogue to a structural object position in VPs, and I label it k-S-OP, for Object-of-P k-selector:41

‘We/they/one stayed there a whole week’ (Finnish, Svenonius (2002)) (ii) a. Mar´ıa

‘Reading went on all day.’

c. *pad

This can be captured if Finnish temporal adverbials raise to the object position, but Icelandic ones don’t. The distinction between the two types of languages is blurred in active sentences ((ia) vs. (iia)), because both types of adverbials end up in the accusative (but for different reasons). The difference only shows up in the passive ((ib) vs. (iic)), where only adverbials which have been raised to object prior to the application of passive can move to the nominative position.

41The PP–VP parallelism idea is not new, and it is defended in detail by den

(79) [ k-S-OP [ a month [before[ UT ] ] ] ]

This position attracts a KP* of a particular size to its Spec, thus determin-ing its case. Prototypically, the Ground argument moves. However, since in examples like vor einem Monat ‘a month ago’ the Ground is null, in some languages, the measure phrase moves to this position instead of the Ground:42 (80) [ a month [ k-S-OP [ a month [before[ UT ] ] ] ] ]

As a result of this movement, the measure phrase (i) takes on the case k-selected by k-S-OP, and (ii) comes to occupy the same position in the linear string as any other argument attracted to this position (i.e., the Ground in the prototypical case). This overt movement into the k-selecting position is the crucial ingredient from which the split between attested and unattested languages follows. Without this movement, there is no guarantee that Grounds and measure phrases end up in the same position when bearing the case k-selected by the k-S-OP.

It is also worth pointing out that if this analysis is correct, we have an example of raising into an oblique case position.43

Dikken (to appear). Den Dikken labels the k-selecting position Aspect, stressing thus the parallel to VPs, where Asp is assumed to be the k-selector of the object case. Den Dikken’s Asp head has a predecessor (called Place) in the work of Koopman (2000).

See also Boˇskovi´c (2004).

42A clarification is needed. When I say that the Ground is null, this is intended to generalize over the two possible analyses of the element UT highlighted in ftn. 37.

Thus, recall that under one possible scenario, the element UT is absent in syntax altogether, and hence, it cannot move to the k-selecting position simply because it does not exist.

Under the alternative analysis, UT is pronounced as a part of the adposition. If in such an entry, the Ground occupies the complement position, then it can be pro-nounced (spelled out) as part of the adposition only if it does not move from that position.

43A prediction of the Peeling theory is that measure phrases are base-generated bearing a case which is bigger than the dative in the Case sequence. This is confirmed by languages like Hungarian, which has measure phrases in the instrumental case. The idea is that Hungarian PP measure phrases stay in situ, and show the original oblique marking.

(i) k´et perc-cel a koncert el˝ott 2 minutes-insthe concert before

‘2 minutes before the concert’ (Hungarian, Andr´ea Markus, p.c.) In German, the measure phrase appears in the accusative, a sign of the fact that it has been displaced from the base-generated position.

English apparently allows the measure phrase to surface in either position (in some cases), as pointed out to me by Peter Svenonius (p.c.). Thus, (iia) and (iib) seem to mean the same:

The structure (80) is where the derivation stops in postpositional lan-guages, yielding the order O > P. Since in prototypical PPs (with an overt Ground), Spec,k-S-OP is targeted by the movement of the Ground, the struc-ture predicts that when the Ground and the measure phrase co-occur, the Ground ends up to the left of the measure phrase (nothing else said, i.e., if the measure itself does not move, or a constituent which contains it). This is attested in some of the postpositional languages. I give an example from Japanese:

(81) sensoo-no war-gen

ni-zi-kan 2-hour-period

mae-ni before

‘2 hours before the war’ (Japanese, K. Takamine, N. Yamato, p.c.) In prepositional languages, the preposition precedes the object. This is the result of a movement which brings the constituent (spelled out by the prepo-sition) to the left of the KP*, as shown in (82). The derivation stops here.

(82) [before[ UT ] ] [ a month [ k-S-OP [ a month [before[ UT ] ] ] ] ] To sum up the argument: bearing a particular case entails occupying a par-ticular position in the linear string, regardless of the base-generated position.

This shows that movement to a k-selecting position is overt (across languages), as predicted by the Peeling theory.

In document The nanosyntax of case (sider 186-194)