(6) a. It was the CAR (not the TRUCK) of which the driver _ _ arrived late/ was awarded a prize.
b. *It was the CAR (not the TRUCK) of which the driver _ _ caused a riot.
1. I would like to thank Rajesh Bhatt for these data.
Subextraction may be further limited by the left branch condition (Ross 1967, 1986), which blocks the movement of the leftmost constituent out of a noun phrase. Since Ross’s seminal work, it has been known that languages differ in terms of their toler- ance for left branch extraction (LBE), so before we compare the behavior of absolu- tives and ergatives with respect to the LBE, it is important to know if this type of extraction is independently allowed in a given language. As with a number of phe- nomena discussed in this work, the data on discontinuous constituents or extraposi- tion in various languages are scant, so it is often hard to tell what is going on. Our understanding of why some languages allow discontinuous constituents whereas others disfavor them is still far from complete.
Discontinuous constituents are typical of some Australian languages. Warlpiri, for example, clearly lacks syntactic ergativity and seems to be one of the better- established examples of a language in which the ergative shows properties of a structural case (Legate 2002, 2008a). In Warlpiri, discontinuity is equally possible for absolutive and ergative DPs, as shown in the examples below (discontinuous con- stituents are marked in brackets):
(7) a. [Maliki- rli- ji] yarlku- rnu [wiri- ngki]. Warlpiri dog- erg- 1sg.obj bite- pst big- erg
‘A big dog bit me.’ (Hale et al. 1995: 1434)
b. [Milpirri] ka- janapayi- ngki [muku]- rra ka- nyi.
cloud prs.IPFV- 3pl.obj all- thither carry- npst
‘The wind is blowing away all the rain- clouds.’ (Legate 2002: 87)
One might be tempted to dismiss data from Warlpiri (and some other ergative Australian languages with discontinuous constituents, listed in Legate 2002: 87) because of its putative pronominal- argument (non- configurational) structure. But ergative languages without any hint of non- configurationality also allow disconti- nuity of absolutive and ergative DPs alike. In chapter 11, I provide examples of dis- continuous constituents in Tsez which illustrate that, while there is no difference between absolutives and ergatives with respect to discontinuity in this language, both types of core arguments differ from other clausal constituents in allowing discontinuity at all. For the purposes of the discussion in this chapter, let me pres- ent some examples from Georgian which show that the absolutive and the ergative behave equivalently with respect to discontinuity and subextraction. In contrast to absolutive and ergative expressions, PPs in Georgian cannot be discontinuous. To examine discontinuity, let us start with the extraposition of relative clauses.
Georgian has two main types of relative clauses. The first type is formed with the wh- word romeli “which”; romeli appears at the left edge of the relative clause in the case corresponding to the case of the gap and hosts the focus enclitic - c.2 The second type of relative clause has the unchangeable complementizer rom; this type is
2. Romeli “which” is the most common wh- word used in relatives, but other wh- words such as “who,” “where,” etc., are possible as well (Foley 2013).
characterized as more colloquial in the literature (Hewitt 1987, 1995; Harris 1992).
The complementizer is typically used in the immediately preverbal position, but it can also appear in the second position of a relative clause. The examples below show non- extraposed (contiguous) relative clauses with both formations (since we do not know enough about the internal structure of Georgian relative clauses, I am again using the atheoretical notation “_ _ _ ” to represent the gap in such a clause):
(8) a. c’ignii [romel- ii=c Dato- m _ _ i dac’era] Georgian book which- abs- rel Dato- erg write.aor.3sg
‘the book which Dato wrote’
b. mosc’avlei [romel- mai=c _ _ i q’oveltvis icis p’asux- i]
student which- erg- rel always know.prs.3sg answer- abs
‘the student who always knows the answer’ (Foley 2013: (5a)) (9) a. c’ignii [Dato- m _ _ i rom dac’era]
book Dato- erg that write.aor.3sg
‘the book that Dato wrote’
b. mosc’avlei [_ _ i koveltvis rom icis p’asux- i]
student always that know.prs.3sg answer- abs
‘the student that always knows the answer’
Both types of relative clauses can extrapose from the absolutive or ergative head noun. Under extraposition of relative clauses, the head noun must appear with a demonstrative (although this demonstrative is optional when the relative clause is contiguous with the head noun). The following pairs of examples illustrate the extra- position of relative clauses from the absolutive subject, absolutive object, and erga- tive host, respectively. All speakers accept extraposition out of the object, but not all speakers accept extraposition out of subjects, regardless of case marking. Those speakers that accept extraposition out of subjects always prefer extraposed relative clauses with the agreeing relative pronoun (as in the (a) examples, below).
(10) a. %Es mosc’avle p’arḳ- ši tamašobs [romel- mai=c Georgian dem student.nom park- loc play.prs.3sg which- erg- rel
_ _ i koveltvis icis p’asux- i].
always know.prs.3sg answer- abs
‘The student is playing in the park who always knows the answer.’
b. %Es mosc’avle p’arḳ- ši tamašobs [koveltvis dem student.nom park- loc play.prs.3sg always _ _ i rom icis p’asux- i].
that know.prs.3sg answer- abs
‘The student is playing in the park who always knows the answer.’
(11) a. Nino *(im) c’ign- s k’itxulobs [romel- ii=c Nino dem book- dat read.prs.3sg which- abs- rel Dato- m _ _ i dac’era].
Dato- erg write.aor.3sg
‘Nino is reading the book that Dato wrote.’
b. Nino *(im) c’ign- s k’itxulobs [Dato- m _ _ i rom dac’era ].
Nino dem book- dat read.prs.3s Dato- erg that write.aor.3sg
‘Nino is reading the book that Dato wrote.’
(12) a. %(Im) mosc’avle- mi icis p’asux- i [romel- ii=c _ _ i
dem student- erg know.prs.3sg answer- abs which- abs- rel p’arḳ- ši tamašobs].
park- loc play.prs.3sg
‘The student knows the answer that is playing in the park.’
b. %(Im) mosc’avle- mi icis p’asux- i [ _ _ i p’arḳ- ši dem student- erg know.prs.3sg answer- abs park- loc rom tamašobs].
that play.prs.3sg
‘The student knows the answer that is playing in the park.’
Although extraposition from subjects is not agreeable to all Georgian speakers, it crucially differs from the completely impossible extraposition out of PPs. Compare the examples above and the following example, which shows that extraposition out of a locative is unacceptable for all speakers:
(13) *Im xi- is kveš bavšvebi tamašoben [romel- ii=c dem tree- gen under children.nom play.prs.3pl which- abs- rel _ _ i čvens baɣ- ši izrdeba].
our garden- loc grow.prs.3sg
(‘The children are playing under the tree that grows in our garden.’)
The contrast between (12) and (13) indicates that Georgian ergatives pattern with other case- marked arguments and not with PPs.3
In addition to relative clause extraposition, which allows us to group ergatives with other core cases, Georgian also allows discontinuous DPs with adjectival and some other modifiers; in the discontinuous structure, either the modifier or the noun can appear first. In (14a), below, we find the baseline sentence, with the adjec- tival modifier and the head noun appearing contiguously; in (14b), the adjective is fronted, stranding the noun; in (14c), the noun is fronted. The fronting has observ- able information- structural consequences (Nash 2002).
3. Although these data serve to illustrate the contrast between core arguments (includ- ing ergatives) and PPs in Georgian, they deserve further scrutiny beyond the depth offered here, because the syntax of Georgian extraposition is still unclear. Extraposed relative clauses in English are generally more acceptable when they modify an indefinite noun phrase, whereas Georgian extraposed relatives must occur with definite expressions, and the demonstrative on the head noun is obligatory. This suggests that extraposed relative clauses may in fact be correlatives, with the demonstrative serving as the proform in the matrix clause. Foley (2013) offers a correlative analysis of preposed relative clauses but a different analysis for the clause types shown here. I leave the actual analysis of Georgian extraposition to further work.
(14) a. Keto- m ik’ida saint’ereso c’igni. Georgian Keto- erg bought interesting book.abs
‘Keto bought an interesting book.’
b. Saint’ereso Keto- m ik’ida c’igni.
interesting Keto- erg bought book.abs
‘As for interesting things, Keto bought a book.’
c. C’igni Keto- m ik’ida saint’ereso.
book.abs Keto- erg bought interesting
‘Of books, Keto bought an interesting one.’
Now, Georgian has case concord, with modifiers matching the case of the noun they modify (the facts in (15) are simplified):
(15) Georgian case concord
Modifier Head noun
nom/ abs - i - i
erg - m(a) - m(a)
dat - s (if discontinuous), null otherwise -s
vocative - o - o
When a noun phrase is discontinuous, the modifying expression must appear in the matching form, indicating the presence of connectivity effects. Thus, if the ergative DP is discontinuous, the adjective must bear the ergative case- marking:
(16) Axal- *(ma) gamaoca mankana- m. Georgian
new- erg 1sg.astonished car- erg
‘As for new things, the car astonished me.’
‘As for cars, the new one astonished me.’
Discontinuous constituents in Georgian include ergative, nominative/ absolutive, dative, and vocative DPs. Unlike these forms, PPs cannot be discontinuous, no mat- ter whether the modifier is overtly marked or not (17b):
(17) a. Nino camovida pat’ara kalaki- dan. Georgian Nino.nom arrived small town- from
‘Nino arrived from a small town.’
b. *Pat’ara(- dan) camovida Nino kalaki- dan.
small- from arrive Nino.nom town- from (‘Towns, Nino arrived from a small one.’)
There are two possible analyses of these discontinuous constituents in Georgian, which involve either subscrambling (including left branch extraction, which is clearly
possible in Georgian), or the duplication of identical DPs in which a portion is elided.
These two options are schematically shown below for example (16), with irrelevant details omitted:
(18) a. Subscrambling (cf. Abels 2003)
Axal- mai gamaoca [ti mankana- m].
new- erg 1sg.astonished car- erg b. Partial deletion under identity (Nash 2002)
[DPAxal- ma mankana- m] gamaoca [DP axal- ma mankana- m].
new- erg car- erg 1sg.astonished new- erg car- erg
Adjudicating between these two analyses would require further work; if the sub- scrambling analysis shown in (18a) is correct, this provides a clear indication that the ergative in Georgian patterns with all the core cases and against PPs. If the ellip- sis analysis (18b) is correct, the split construction cannot be taken as evidence for the DP status of the ergative; however, the deletion analysis would need to explain the inability of PPs to form discontinuous constituents. Nothing in the structure of (18b) immediately accounts for the DP– PP distinction. Whichever the winning analysis of case concord and discontinuous constituents in Georgian is, Georgian ergative forms clearly pattern with other DPs and not with PPs.