Let me start on a terminological note. Throughout this section, I will be using the term “subextraction” only in reference to the displacement of a subconstituent out of a DP or a PP, never out of a clausal argument.
Investigating subextraction out of the ergative expression is a not an easy task, for several independent reasons. For one thing, if an ergative expression is opaque to subextraction, this may happen because it is a subject, not because it is ergative. In many languages, subjects are islands in their own right, regardless of their categorial status as DPs or PPs (and, more generally, regardless of ergativity). Thus, we must first investigate absolutive subjects in order to ascertain that a given language does not have a general subject island constraint independent of ergativity. Since sub- jects of unaccusatives can be targeted for subextraction in their base positon (3b), the only reliable absolutives for this investigation are the subjects of unergatives, as shown below:
1. By “criterial” properties, I mean properties that define PPs as a class of expressions.
(3) a. unergative: [TP [Subject … t… ]]
b. unaccusative: [TP [vP [VP [Internal argument … t… ]]]
Let us imagine, for the sake of argument, that we find a language where subjects of unergatives are permeable and can be subextracted from. Such a language would be a good candidate for examining subextraction out of the ergative expression.
The next variable has to do with the nature of subextraction out of a PP. In the absence of P- stranding and pied- piping, discussed in chapter 2, movement of the entire complement of the P head is expected to be impossible (see Abels 2003 for further discussion). However, subextraction out of this complement to P should be possible. This prediction is captured by the Stranding Generalization, repeated below (Abels 2003: ch. 4):
(4) subextraction: [P [... t ...]]
This prediction, in turn, leads to the prediction that PP- ergatives should share sub- extraction properties with other PPs in a given language. In other words, those erga- tive expressions that are PPs should pattern just like other PPs in that language with respect to subextraction out of the DP complement.
Bringing all these factors together, several ingredients are needed to make predic- tions concerning subextraction out of the ergative expression in a language L:
(5) a. Language L is morphologically and syntactically ergative.
b. Language L has a reliable contrast between unaccusatives and unergatives.
c. Language L shows subextraction out of absolutive subjects of unergatives.
d. Language L has no subextraction out of complements to P heads.
If, in such a language, one finds that the ergative expression is an island for subex- traction, this ergative can be assimilated to other PPs. However, as (5c) states, in order to determine whether or not ergatives are islands, one first needs independent evidence that the language in question does not have subject islands and/ or does not respect the left branch extraction restriction, according to which movement of the leftmost constituent out of a noun phrase is impossible (Ross 1967: 114, 1986: 127;
Corver 1990). Since there are quite a few moving parts to the subextraction account, it is not easy to find languages that allow us to test the predictions spelled out above.
Fortunately, Chukchi turns out to be one such language. Chukchi is morphologi- cally and syntactically ergative, thus meeting condition (5a). In order to relativize the ergative, one needs to use an antipassive (already described in chapter 1). Consider the following examples:
(6) a. baseline sentence Chukchi
ətləg- e yaraq teykə- nen.
father- erg house.abs build- aor.3sg.3sg.
‘The father built the house.’
b. ABS object extracts with a gap at the extraction site [ətləg- e _ _ _ i teykə- lʔ- ən] yaraqi father- erg build- ptcp- abs house.abs
‘the house that the father built’
c. ERG subject cannot extract with a gap at the extraction site
*[ _ _ i yaraq teykə- lʔ- ən] ətləg- əni
house.abs build- ptcp- abs father- abs (‘the father who built the house’)
Next, we find that Chukchi has an established contrast between unaccusatives and unergatives, thus meeting condition (5b). Only subjects of unaccusatives can incor- porate; compare (7b), which is grammatical, and (8b), which is not:2
(7) unaccusative predicate Chukchi
a. Qora- ŋə wʔi- gʔi.
reindeer- abs die- aor.3sg
‘The/ A reindeer died.’
b. Qaa- wʔi- gʔi.
reindeer- die- aor.3sg
‘Reindeer died.’ (lit.: there was reindeer- dying) (8) unergative predicate
a. Qora- ŋə ekwet- gʔi.
reindeer- abs dash- aor.3sg
‘The/ A reindeer ran loose.’
b. *Qaa- kwatə- gʔe.
reindeer- dash- aor.3sg
(“There was the running loose of reindeer.”)
Chukchi speakers allow subextraction out of noun phrases, although there is some speaker variation (see also Fanselow and Féry 2006 for the same observation). In particular, subextraction out of unergative subjects is possible (9b), thus meeting condition (5c):
(9) a. [DP [AP Mətləŋqaw] ʔaaček- ət] aywe migčiret- gʔe. Chukchi fifth.adj youth- abs yesterday work- aor.3sg
‘The fifth young man worked yesterday.’
b. %[AP Mətləŋqaw]i aywe [DP ti [ʔaaček- ət]] migčiret- gʔe.
fifth.adj yesterday youth- abs work- aor.3sg
‘The fifth young man worked yesterday.’
2. In Polinsky (1994), I actually suggested that subjects of unergatives can also incorporate.
However, on closer inspection, the apparently unergative verbs that allow subject incorpora- tion are simply ambiguous between unergative and unaccusative— the same way as some Italian verbs, such as correre “run,” can co- occur with either avere or essere (cf. Centineo 1996).
In contrast, subextraction out of PPs is impossible, which meets our condition (5d):
(10) a. [PP [DP [AP Rəmagtə- ken] nəmnəm- ək] yaačə] ətri Chukchi far- adj village- loc behind 3pl.abs
ga- managra- npə- len- at.
PRF- tent- erect- PRF- 3pl
‘They put up a temporary tent behind the faraway village.’
b. *[AP Rəmagtə- ken] ətri [[PP [[DP ti nəmnəm- ək] yaačə] ga- managra- npə- len- at.
far- adj 3p.abs village- loc behind prf- tent- erect- prf- 3pl (‘They put up a temporary tent behind the village, the faraway one.’)
And finally, subextraction out of the ergative expression is impossible, on par with (other) PPs, but in contrast with the absolutive subject:
(11) a. [DP [AP Mətləŋqaw] ʔaaček- a] aywe keyŋ- ən penrə- nen. Chukchi fifth.adj youth- erg yesterday bear- abs attack- aor.3sg.3sg
‘The fifth young man attacked the bear yesterday.’
b. *[AP Mətləŋqaw]i aywe [DP ti ʔaaček- a] keyŋ- ən penrə- nen.
fifth.adj yesterday youth- erg bear- abs attack- aor.3sg.3sg These data provide independent evidence that ergative expressions in Chukchi are indeed PPs.
Halkomelem is another language that allows for the testing of subextraction, fol- lowing the conditions spelled out in (5a– d). Halkomelem is syntactically and mor- phologically ergative; it allows possessor extraction out of absolutives, but not out of ergatives (Gerdts 1988: 76). In (12a) and (13a), we observe the baseline sentences with the absolutive subjects, and (12b) and (14b) show subextraction out of those subjects. Such subextraction is possible for all intransitive subjects, regardless of the unergativity/ unaccusativity of the verb (Donna Gerdts, pers. comm.; Gerdts and Hukari 2001, 2006).
(12) a. Ni q’wəl [θə ripe det
stθu:m- s ɬe sɬéniʔ]. Halkomelem
aux berry- 3poss det woman
‘The woman’s berries got ripe.’
b. ɬe sɬéniʔi [ni q’wəl [θə stθu:m- s ti]]
det woman aux ripe det berry- 3poss
‘the woman whose berries got ripe’
(13) a. Ni xčénəm [kwθə sqéʔəq- s ɬe sɬéniʔ]. Halkomelem aux run det younger. brother- 3poss det woman
‘The woman’s younger brother ran.’
b. ɬe sɬéniʔi [ni xčénəm [kwθə sqéʔəq- s ti]]
det woman aux run det younger. brother- 3poss
‘the woman whose younger brother ran’ (Gerdts 1988: 74)
However, subextraction out of the ergative subject is impossible:
(14) a. Ni q’wəl- ət- əs [kwθə sqéʔəq- s ɬe sɬéniʔ]
aux bake- tr-3erg det younger. brother- 3poss det woman Halkomelem kwθə scé.ʔtən.
det salmon
‘The woman’s younger brother baked the salmon.’
b. *ɬe sɬéniʔi [ni q’wəl- ət- əs [kwθə sqéʔəq- s ti] ]
det woman aux bake- tr- 2sg.sbj det younger. brother- 3poss kwθə scé.ʔtən]
det salmon
(‘the woman whose younger brother baked the salmon’) (Gerdts 1988: 74) If Halkomelem did not allow subextraction out of subjects in general, then we would (incorrectly) predict both (12b) and (14b) to be bad.3 Other PPs in Halkomelem are also islands for subextraction (Gerdts 1988: 59– 84); thus, the characteristics of the ergative match the characteristics of (other) PPs.
In Upriver Halkomelem, quantifiers can extract out of absolutive DPs (subjects and objects) but not out of ergative DPs, as described by Wiltschko (2006b: 207, 226). Again, it seems that this asymmetry cannot be attributed to the unaccusative/
unergative distinction (Galloway 1993: 245– 255; Wiltschko 2006b), so it must have something to do with the difference between absolutives, on the one hand, and erga- tives, on the other.
Finally, in Tzotzil, as described in Aissen (1996), there is a subtler contrast with respect to subextraction. Absolutive subjects of unaccusatives are completely trans- parent to subextraction, as the following example demonstrates:4
(15) a. I- cham x- ch’amal li Xun- e. Tzotzil
compl- died gen- child the Xun- enc
‘Xun’s child died.’
b. Buch’ui i- cham [x- ch’amal ti]?
who compl- died gen- child
‘Whose child died?’ (Aissen 1996: 456)
Extraction from absolutive subjects of unergative predicates is possible but is degraded compared to extraction from unaccusative subjects. Aissen contrasts (15b) with the following example,
3. An alternative to the subextraction analysis would be to assume that the possessor in the relevant construction first undergoes raising to an argument position and then extracts. Possessor raising is independently attested in Halkomelem (Gerdts 1988, 1999);
however, it would be surprising to find possessor raising with unergatives, since such rais- ing is normally limited to internal arguments (Deal 2015).
4. I have slightly modified the glosses from the original.
(16) a. Te ta=xjalav s- tzeb li Maruch- e. Tzotzil there incompl=weave gen- daughter the Maruch- enc
‘Maruch’s daughter is weaving there.’
b. ??Buch’ui te ta=xjalav [s- tzeb ti]?
who there incompl=weave gen- daughter
‘Whose daughter is weaving there?’ (Aissen 1996: 461)
But it is significant that extraction out of the ergative expression is completely unacceptable:5
(17) *Buch’ui y- elk’an chij [ti [x- ch’amal ti]]? Tzotzil who erg3- steal sheep gen- child
(‘Whose child stole sheep?’) (Aissen 1996:460)
The only way to rescue questions such as (17) is by pied- piping the entire possessive expression to the left periphery of the question (Aissen 1996: 460).
Tzotzil is also unusual among the languages considered in this section in that its syntactic ergativity is not categorical. As Aissen (1999) observes, the extraction of the ergative agent can occur either from clauses with a special dependent verb form (agent focus, abbreviated as AF below) or from regular transitive clauses. AF verbs are found only under agent extraction, but agent extraction is not limited to AF clauses.6
The subextraction facts reviewed here are significant in two respects. First, they provide further evidence for the principled distinction between absolutives and PP- ergatives that I have pursued here. While PP- ergatives differ from absolutives with respect to subextraction, DP- ergatives do not; I will discuss relevant instances of subextraction in chapters 7 and 11. Languages with subextraction from both absolu- tives and ergatives include at least Warlpiri (Hale et al. 1995; Simpson 1991), Tsez ( chapter 11), and Georgian. Significantly (in the absence of complications from other factors), all these languages that allow subextraction from an ergative argument belong to the DP- ergative type, while languages that allow subextraction only from absolutives belong to the PP- ergative type.
Second, these facts inform our understanding of mechanisms that allow or block subextraction from arguments on a more general plane. Setting ergativity aside, subextraction from all arguments is categorically impossible in some languages, for example, Basque or in Polynesian languages. Some researchers have attempted to explain the ban on subextraction by appealing to agreement. The logic goes as fol- lows. Outside the realm of ergative languages, agreement with subjects is common, and subjects are often opaque. The common lack of transparency for subextraction
5. Aissen does not provide the baseline sentence for this question.
6. It is possible that Tzotzil may be undergoing a change from a PP- ergative type to a DP- ergative type, a transition I will discuss in chapter 8; this may explain the variation in its agent extraction.
among subjects may be due to their need to move in order to receive case. Such movement is accompanied by agreement, and, for some researchers, agreement is viewed as necessary for case to be licensed on subjects (cf. Bhatt 2005). Thus, the opacity of some arguments to subextraction may be attributed to the combined factors of case- related movement and agreement. This line of reasoning has been advanced, for example, for Basque (Laka and Uriagereka 1987; Uriagereka 1999;
Boeckx 2008; Lohndal 2011), where the verb arguably agrees with absolutive, erga- tive, and dative arguments, and where subjects and objects are equally opaque for subextraction.7
As we expand our empirical coverage of subextraction, however, it becomes more difficult to maintain the view that agreement leads to opacity of DPs.
Counterarguments against this view come from several quarters. Starting with the licensing of ergatives, the ergative rarely needs to move for case, so even if it enters an agreement relation, this agreement may not be accompanied by movement. That means that agreement cannot serve as motivation for the opacity of ergatives. On the contrary, the absolutive often moves for case and is agreed with while remain- ing transparent. On a more general note, agreement may actually be a prerequisite for subextraction, not a barrier to it— at least in some instances. This is particularly apparent in those languages that manifest a clear interaction between agreement and subextraction, such as in Dinka (van Urk and Richards 2015). All these consider- ations suggest that there may be a number of reasons for the opacity of arguments to subextraction, and agreement is not among these reasons.
Although PPs are islands for subextraction, and some ergatives, as I claim, are PPs, it is important to keep in mind that subextraction restrictions are not limited to a particular alignment type. The ergative Polynesian languages are among ergative languages whose clausal constituents never undergo subextraction, but subextrac- tion is equally impossible in the nominative- accusative Polynesian languages. And in another sign that languages that generally disallow subextraction from their argu- ments do not discriminate, subextraction is completely impossible in the syntacti- cally ergative Tongan ( chapter 10) and morphologically ergative Niuean ( chapter 8).
To conclude, subextraction is a broad phenomenon, and one where we are still look- ing for generalizations and clearer answers. For the purposes of this book, the only aspect of this phenomenon that matters is the difference between ergative and abso- lutive subjects with respect to subextraction.