2.3.1 The relationship between the verbal functional head and ergative P Let us now consider the role of the transitive verbal head in the structure in (15) above, repeated here:
(23) [vP [PP P [DP]] …]
This vP serves two main functions. First, it establishes a predication relation between its complement and specifier (Moro 2000; den Dikken 2006). This relational function serves as a connective element, creating a syntactic and semantic link between the specifier, on the one hand, and the lexical verb within its complement, on the other.
The second function of the transitive v head is to introduce an external theta role. This function of vP is standardly assumed in the derivation of such clauses, regardless of the content of the specifier position— that is, whether it is filled by a DP or another type of XP. If the specifier is a DP, theta- licensing proceeds directly.
However, even if the specifier is not a DP, theta- licensing is still established between the functional head and the DP within the larger specifier, bypassing any higher func- tional material in the specifier XP. This is not unexpected, given that theta roles are established on nominal expressions, not on clauses or PPs. For PP- ergatives, the DP complement of the base- generated PP is theta- licensed by the verbal functional head from the outside. The interpretation of the DP complement changes depending on the predicate with which it co- occurs (Williams 1985; Goodall 1997; Neeleman 1997).
A similar idea— that the semantic contribution of silent or light P heads depends on the constituents they bring together— is advanced by Caponigro and Pearl (2008) for spatial prepositions such as to, at, in. These authors also emphasize that semantic
“lightness” is a necessary but not sufficient condition for the licensing of light Ps in general, and silent Ps in particular.
The preposition heading the PP does not play any semantic role in the interpreta- tion of the ergative; it licenses its complement DP structurally, by assigning case to it, but not thematically. If we assume the standard distinction between functional
prepositions (such as English at, of, or by) and strong, verb- like prepositions that categorize prepositional roots (such as English beside),4 then we can conclude that the ergative preposition is distinctly functional in nature. It is semantically light and serves only to establish enough functional structure for the derivation to proceed.
On the semantic plane, the P head makes no contribution; it essentially serves as a structural building block. From the brief diachronic overview above, one could hypothesize that the original P that developed into the ergative was either a source (“from”) or a cause (“by,” “because of”).5
To recapitulate, in the configuration I propose for syntactically ergative languages, the verbal functional head gives the external argument its thematic interpretation, and the P head licenses its case. Kayne’s (1984) analysis of English dative construc- tions suggests a similar strategy for dividing labor between verbal and prepositional heads; Kayne proposes that the English double object construction includes a silent P, which originates outside the VP but inherits case from V and assigns it to its com- plement (see also Larson 1990), thus leading to “mediated” case assignment:
(24) give [PP e [Pat]] flowers
∣____Case_____↑∣_____↑
“Objective Case”
This implementation is conceptually similar to the one presented above; however, its applicability to the present context is complicated by the fact that mediated case assignment, on Kayne’s analysis, can only occur if the V and the P heads assign the same case (“objective” case in his terms), as in (24) above. It is not clear that it is either possible or desirable to impose this limitation on the derivation of the erga- tive. Abandoning the mediated case assignment mechanism essentially leads to the proposal that I sketched out above.
As a consequence of the proposal presented here, theta- licensing and case- licensing are dissociated: the DP inside the PP relies on the predicate for its thematic role, but on the preposition for its syntactic licensing. Crucially, theta- licensing and case- checking come from different heads, but these heads are both structurally local to the argument in question.
2.3.2 Ergative P and P- stranding
Assuming that vP in some ergative languages has a PP rather than DP in its specifier, further analytical steps are required to establish a connection between that specifier
4. See Corver and van Riemsdijk (2001) for details of the strong/ weak distinction in prepositions.
5. One could explore the semantics of this preposition further, but its meaning might differ depending on the animacy restrictions imposed by a particular language on its erga- tive. Languages differ in this regard, just as they differ in the extent to which inanimates are allowed as nominative subjects. I will not pursue this line of inquiry here.
and syntactic ergativity. In principle there is nothing that should prevent the PP in subject position from extracting with a gap, yet we have witnessed a robust ban on such extraction in some ergative languages. How can we relate the prepositional sta- tus of the ergative to the presence of syntactic ergativity?
If A- bar movement targets a PP, then in principle the DP inside that PP could be extracted by stranding the P head or by pied- piping the P head to the landing site of movement. Both options are familiar from English:
(25) a. The house was destroyed by the storm
b. Whati was the house destroyed by _ _ _ i ? P- stranding b’. the stormi [that/ which the house was destroyed by _ _ i ]
c. By whati was the house destroyed _ _ _ i? pied- piping c’. the stormi [by which the house was destroyed _ _ _ i]
These examples demonstrates that even if a given ergative expression is a PP, there is no a priori reason why it cannot be A- bar moved with concomitant stranding or pied- piping of the P head. Let us now consider these two options in more detail.
Adposition stranding is a rather uncommon option in natural language design (Kayne 1984; van Riemsdijk 1975, 1978), and its best- known instances are found in North Germanic languages. A long- standing intuition about this pattern is that strandable adpositions comprise a special, well- defined subset among adpositions in general. In languages that permit this operation in principle, the adpositions that can be stranded have verb- like properties (van Riemsdijk 1975, 1978, 1990; Kayne 1981, 1984; more recently, Baker 2003: 303– 325) that categorize them as “strong.”
These strong adpositions are composed of lexical elements, in contrast to “weak”
adpositions, which are composed of functional elements (van Riemsdijk 1978, 1990; Kayne 1984; Corver and van Riemsdijk 2001). Stranding of weak prepositions is unacceptable and induces an ECP- type effect (Rizzi 1990: 109). Delving deeper into what makes strong prepositions verb- like, we find that one crucial difference between strong and weak adpositions is that the former are theta- role assigners, while the latter are not (Rooryck 1996).
As established above, the theta- licensing of the ergative is carried out externally by the transitive v or Voice head, rather than by the adpositional head itself. The P head in the ergative structure is therefore functional, not lexical: it lacks “heavy”
semantic content; it cannot assign a theta role; it cannot be separated from its com- plement. Thus, the stranding of ergative P heads in a language with PP ergativity should be impossible even if the language allows adposition stranding in principle.
2.3.3 Ergative P and pied- piping
A second possibility for A- bar movement of PPs, including the PP- ergative, involves pied- piping, which in this case entails the displacement of the DP contained in the ergative expression together with the P. The English examples below complement
the original example in (25c) by showing that pied- piping can target constituents of varying sizes.6
(26) a. [A picture of which football player]i does he own _ _ i ? b. [Of which football player]i does he own a picture _ _ i ?
Unlike P- stranding, which is unquestionably rare, pied- piping of a P by its com- plement is quite common. This pied- piping is often viewed as a last- resort option; if stranding is not available, then pied- piping becomes an escape hatch for movement.
The question that arises, then, is why PP- ergative languages do not make use of pied- piping. The answer to this question is nuanced and involves several moving parts, discussed in turn below. When all the parts are coordinated, what we arrive at is not an inviolable grammatical constraint of the “ergativity and pied- piping are incom- patible” variety but rather an interface condition concerning morphophonological visibility and the recoverability of P.
Most investigations of pied- piping have examined it specifically with respect to wh- questions. Since A- bar extraction is understood more broadly here, it is impor- tant to make a distinction between pied- piping in wh- questions, in focus construc- tions, in relativization, and in topicalization. The differing availability of pied- piping in these various A- bar operations may have to do with the semantic and morphopho- nological constraints underpinning pied- piping.
First, pied- piping is sensitive to the phonological content of the operator. Specifically, pied- piping is more acceptable in the presence of an overt operator (like the one found in interrogatives, for instance) than in the presence of a null operator (as found in some relative clauses and in topicalization). In other words, a silent A- bar movement operator blocks extraction of a PP, while an overt operator permits this extraction, if only marginally.
To see this distinction in action, consider the well- known constraints on A- bar movement of the indirect object in the English double- object construction. In rela- tive clauses or tough- movement double- object constructions, the operator cor- responding to the indirect object is silent, and its movement is ungrammatical. In wh- questions, the moved element is overt, and pied- piping is allowed, however mar- ginally (Hornstein and Weinberg 1981; den Dikken 1995; Basilico 1998). Consider the differences below between English relative clauses/ tough- movement construc- tions, on the one hand, and wh- questions formed from double- object constructions, on the other. Tough- movement clauses and relative clauses with a silent that are less acceptable, on balance, than wh- questions of dative objects— although even with wh- questions, one finds significant dialectal variation:
(27) a. *Shei is not easy [Opi to give gifts].
b. *the personi [Opi they sent ti this email]
6. While (26b) is acceptable to most speakers, (26a) is not; only some English speakers allow it.
(28) %Whoi did they send ti this email?
By contrast, when the operator is overt, extraction from the relative clause is roughly as acceptable as extraction from the wh- question (I set aside distinctions arising as a result of different relative- clause formation strategies: operator movement vs. head raising):
(29) %the person [who they sent ti this email]
These data indicate that the presence of an overt operator, which is common in cer- tain A- bar movement operations (wh- questions; some relative clauses) and less com- mon in others (tough movement; other relative clauses), improves the availability of pied- piping. Applying this logic to possible cases of syntactic ergativity, we can predict that syntactic ergativity should be more likely in contexts where the A- bar movement operator is null— namely, under tough movement, topicalization with a null operator, relativization with a null operator, or focusing with a null operator.
As I will show in chapter 4 (section 4.5 in particular), at least three languages with relativized syntactic ergativity are attested: Chukchi, West Greenlandic, and Asiatic Eskimo. Each of these languages permits the ergative expression to undergo wh- movement, but not relativization with a gap. Based on evidence provided in this section, we can assume that this pattern follows from the presence of an overt inter- rogative operator that facilitates pied- piping in wh- questions.
In addition to sensitivity to the phonological content of the operator, pied- piping is also sensitive to the phonological content of the P head itself. In particular, when the P head is null or extremely reduced phonologically, pied- piping cannot take place.
In other words, invisible P heads simply cannot be reliably evaluated as “real” parts of the construction. In the absence of an identifiable P head, a putative pied- piped PP would be indistinguishable from an extracted DP, making this structure unavailable to the learner (Klein 1993).
Languages with well- attested null adpositions are rather rare in the literature—
although, if my proposal concerning PP- ergatives is correct, there will soon be many additions to this list. Klein (1993) offers the following list: spoken Brazilian Portuguese,7 colloquial Puerto Rican Spanish, colloquial Venezuelan Spanish, some northern Greek dialects, seventeenth- century French, Quebecois French, Catalan, Haitian Creole, and Roviana. (Incidentally, Roviana is morphologically ergative and also shows syntactic ergativity: see Corston 1996; Corston- Oliver 2002; Todd 1978, 2000.) According to Dekydtspotter et al. (1998), null- preposition constructions are also attested in collo- quial varieties of modern French other than Quebecois, as well as in Yoruba. In none of these languages can null adpositions be pied- piped. Some languages exhibit a contrast between relative clauses with an invariant complementizer, which show no evidence of movement, and relative clauses with a relative pronoun, which permit A- bar move- ment. The following examples from Catalan illustrate the contrast. In (30a), we see the
7. See also Almeida and Yoshida (2007); Rodrigues et al. (forthcoming).
relative pronoun qui with the pied- piped overt preposition de. The null preposition is permitted only in (30b), in the colloquial register; in this case, the invariant comple- mentizer que, rather than the relative pronoun, must be used.
(30) a. el pare [de quii hem visitat el fils_ _ i] Catalan
det father of who we visited det son
b. el parei [que hem visitat el fils proi ]
det father comp we visited det son
‘the father whose son we visited’ (from Klein 1995: 92) c. el parei [que hem visitat soni fils]
det father comp we visited his son
‘the father whose son we visited’ (lit.: :the father that we visited his son) Klein (1993, 1995) shows that at least two generalizations can be made about null adpositional phrases: first, they cannot co- occur with agreeing complementizers or relative pronouns; second, they are typically replaced by overt resumptives, as in (30c), or silent resumptives, as in (30b).8 Both of these properties are characteristic of non- movement, further supporting the conclusion that null adpositions cannot be pied- piped.
The general conclusion we arrive at is that the pied- piping of adpositions is con- tingent on their phonetic content. If pied- piping were a purely syntactic operation, it would be surprising to find that it was sensitive to PF content. The sensitivity of pied- piping to the phonetic content of a moved element therefore suggests that the operation is subject to an interface condition:
(31) PF condition on pied- piping: Only overt adpositions can be pied- piped.
Finally, I wish to briefly touch on a more general issue concerning the theoretical validity of pied- piping. A number of researchers have questioned whether pied- piping is a theoretically viable notion (cf. Heck 2008; Cable 2010a, b). The main conceptual reservation against pied- piping has to do with the desire to adopt a more general notion of Agree that is not subject to locality restrictions on feature checking. If Agree can apply in a remote fashion, crossing phrase boundaries (as long as the goal is c- commanded by the probe), the need for local feature- checking is obviated (Heck 2008). As an alternative, some scholars have proposed that wh- words may receive their interpretation through the help of an interrogative morpheme situated higher
8. Silent resumptives are also found with overt adpositions, a fact that casts doubts on some cases of apparent stranding. For instance, Quebecois French has examples of what looks like P- stranding, as in (i). However, the dependency fails to exhibit standard char- acteristics of movement, such as subjacency, suggesting instead a base- generated depen- dency between an antecedent and a null resumptive pronoun (Bouchard 1982; Abels 2003).
(i) la fillei [que je connais très bien [le gars qui sort avec proi/ *ti]]
the girl that 1sg know very well the guy comp goes with
‘the girl that I know very well the guy who goes out with’ (Vinet 1984)
in the structure. The interrogative complementizer (or any other operator) co- values its features with the features of this interrogative morpheme, which must move to C in time for interpretation. In some languages, Q can move by itself; in other lan- guages, Q carries along the material it is attached to. It is this latter case that results in the appearance of pied- piping (cf. Cable 2010b).
In sum, if pied- piping is available in a given ergative language and if the adposi- tion that licenses ergative case is overt, then syntactic ergativity is not expected to arise. Thus, syntactic ergativity is expected in some but not all PP- ergative languages.
If a PP- ergative language allows pied- piping of the ergative P, then syntactic ergativ- ity will not be observed or will be limited to a subset of constructions. The distribu- tion is as follows:
(32) Morphologically ergative languages DP-ergative PP-ergative:
Is the ergative-licensing P overt?
Yes No
Is pied-piping of PPs possible?
Yes No Syntactic ergativity
2.3.4 From a PP subject to syntactic ergativity
We are now ready to articulate the relationship between the PP structure of ergatives and syntactic ergativity. Several factors conspire to prevent the PP- ergative expres- sion from undergoing A- bar movement. Most obviously, the PP status of the ergative renders it an island, thus preventing extraction of its DP complement. To circumvent the ban on DP extraction, the entire PP could in principle be moved. Indeed, if the ergative PP is headed by an overt P, then this head can (in principle) be pied- piped, thus avoiding the phenomenon of syntactic ergativity. Alternatively, in some lan- guages, the P can be stranded, allowing the noun phrase complement of the ergative P to undergo A- bar movement.
If the ergative PP has a null P head, however, then the ergative argument it gov- erns should be inaccessible to A- bar movement. Again, movement with an overt operator (wh- movement or overt focus movement) may allow for the pied- piping of such a null head, so we predict the possibility of languages that exhibit syntactic ergativity in relativization but not in wh- questions.
Based on the preceding discussion, we can identify two major ingredients of syn- tactic ergativity.
(33) Prerequisites of syntactic ergativity
a. The ergative is expressed by a PP in subject position.
b. The ergative PP cannot be pied- piped or strand its P head under A- bar movement.
As I will show below, these very specific characteristics are associated, in a predict- able way, with a cluster of properties, some of which have been offered in previous accounts as independent realizations of syntactic ergativity.
Since the assumption that ergative PPs may be headed by a null P is fundamental to my account, it is incumbent upon me to provide more evidence showing how null Ps are detected. So far, the main pieces of evidence I have offered are the immobility of the complement of the P head and the parallel between my proposed null P head and those found with experiencer subjects and indirect objects. Additional evidence is quite nuanced, and is based on effects that may vary from language to language.
In particular, evidence comes from binding, from control and raising, and, in some instances, from adposition stacking. I will postpone discussion of the details of this evidence until chapters 4 and 5 (binding, raising, and control). In chapter 8, I will address the issue of potential triggers for a learner acquiring PP- ergatives.