The authors base their arguments on Mayan languages, which they divide into two types: high- absolutive and low- absolutive. In high- absolutive languages, such as Q’anjob’al, the absolutive object must move up to TP in order to receive Case. In low- absolutive languages, there is no requirement for such movement; movement to spec,TP may occur for other reasons, but not for Case. The rest of the account follows straightforwardly, with just one additional assumption: the transitive vP in high- absolutive languages is a phase (a locality domain, according to the phase theory developed by Chomsky 1995, 2001, 2008). It is thus subject to the Phase Impenetrability Condition (Chomsky 2001). On the assumption that a vP has only one escape hatch (implemented as a requirement that only one edge feature be pres- ent, or as a general restriction on multiple specifiers), only one argument can escape that phase. The argument that escapes must be the object, since it requires Case. The result is that the ergative DP is “trapped” within the vP phrase. To represent this situation schematically,
4. Their proposal bears some similarity to the one developed in Merchant (2009); since their proposal is more articulated than the one in Merchant’s short article, I will concen- trate on representing Coon et al. (2014).
(11) InflP
ASP ABS b.
DP
DP VoiceP
Voice’
Voice ix Maria cl Maria
VP V seeil –a’
2pron
Influφ νPTV
νTV’
νTV[+EPP]
max
-ach
An intransitive vP is not a phase, so no constituents are trapped by it.
(12) InflP
DP
νPTTV
Influφ
ASP ABS b.
VP V sleepway –i
2pron
νTTV’
νTTV[+EPP]
max
-ach
On this approach, for the ergative to be prevented from extraction, it is crucial that it be generated very low and not raise out of vP.
Strong morphological support for this proposal comes from the surface linear- ization of agreement markers: in high- absolutive languages, the absolutive agree- ment marker is a prefix, while in low- absolutive languages, it is a suffix (see Tada 1993; Coon et al. 2014 for an extensive discussion). Compare the high- absolutive Q’anjob’al and the low- absolutive Chol. In Q’anjob’al, the absolutive marker ach is adjoined to the highest element of the clause, the aspectual marker max:
(13) Max- ach y- il- ’a ix Malin. Q’anjob’al
asp- abs.2sg erg.3sg- see- tr clf Maria
‘Maria saw you.’
In contrast, the absolutive exponent in Chol appears low:
(14) Tyi- k- wọy- is- ọ- yety. Chol
asp- 1sg.erg- sleep- caus- derived.tr- 2sg.abs
‘I made you sleep.’
Further support for this approach comes from the fact that certain objects in Mayan, such as regular reflexives and extended reflexives, allow the agent to be extracted from a transitive vP. These objects are taken to be caseless and therefore need not move for Case reasons. Thus in Q’anjob’al:
(15) Maktxel max y- il/ *il- on[- i] s- b’a? Q’anjob’al
who asp erg3- see/ *see- af- tr gen3- self
‘Who saw herself/ himself?’
This proposal allows the authors to account for a number of correlations, includ- ing the correlation between syntactic ergativity and the requirement that embed- ded clauses be intransitive (in other words, that the ergative cannot occur in the subject position of complement clauses). It also seems consistent with the divi- sion of absolutive case forms into true absolutives and “nominatives in absolu- tive clothing,” as proposed by Legate (2008a) and Aldridge (2004, 2007, 2008).
Specifically, the absolutive can be assigned either by the little v head inside the vP or by a higher projection in the inflectional domain (in Mayan, this is the aspec- tual head). High- absolutive languages have nominatives in the guise of abso- lutives, whereas low- absolutive languages have “true” absolutives, assigned by a lower verbal head. This is a pleasing result, and one that any account of ergativity should be able to capture.
An important consequence of the division of Mayan languages into the high- absolutive and low- absolutive types is the prediction this division makes concerning the appearance of the absolutive in non- finite embedded clauses. The logic is as fol- lows: in those languages where the absolutive (either subject or object) is assigned by the high clausal head (T or Infl), absolutive DPs cannot be licensed in embedded phrases that lack that licensing head— that is, in non- finite embedded clauses. In contrast, in low- absolutive languages, the absolutive can appear in non- finite embed- ded clauses, because it does not rely on a higher head for licensing. This prediction is confirmed. In high- absolutive Q’anjob’al, there is no way to generate the absolu- tive in embedded clauses (16); instead, the language uses a special nominalization, the so- called “crazy antipassive” (T. Kaufman 1990), illustrated in (17). According to Coon et al. (2014), the agent focus marker - on- in Q’anjob’al serves as a licensor for the object case in the absence of the high aspectual head that would otherwise license the absolutive.
(16) *Chi uj [hin y- il ix Malin]. Q’anjob’al
asp be.able.to abs1 gen3- see clf Maria (‘Maria can see me.’)
(17) Chi uj [hin y- il- on[- i] ix Malin]. Q’anjob’al
asp be.able.to abs1 gen3- see- af- iNtr clf Maria
‘Maria can see me.’
Example (18) is another instance of the crazy antipassive, this time used in a nominalization:
(18) [J- aw- on ixim awal yet Q’anjob’al
gen1pl- plant- af clf cornfield when
s- tiempo- hal awnub’] kawal yowalil.
gen3- time- abstract planting.season very important
‘To plant corn in the planting season is very important for us.’
(lit.: our planting of corn in the planting season is very important)
The proposal thus makes correct predictions about the syntax of non- finite comple- ments in high- absolutive Mayan languages.
The properties of the Coon et al. (2014) proposal that I have outlined in this sec- tion are intriguing, but there are also reasons to remain skeptical. Syntactic phases provide one of the major building blocks of Coon et al.’s proposal; this is related to their desideratum that the transitive subject be generated below the vP phase. As I mentioned above, the authors assume that intransitive vP is not a phase, while transitive vP is. Transitive vP therefore provides an escape hatch that can be utilized by an argument that needs to move out of vP: in high- absolutive languages, this will crucially be the object (which needs to move to receive Case), rather than the sub- ject. In Chomsky’s original proposal (1995), which was built mainly upon Spanish data, unaccusative vPs were not considered phases (in the terminology employed by Chomsky 2001, they are “weak phases”), but all vPs that assigned an external theta role were (strong) phases. Thus, the distinction between phase and non- phase (or equivalently, strong and weak phase) relied on the presence/ absence of an external argument, not an internal argument. Chomsky’s proposal and Coon et al.’s analysis are compatible as long as all intransitive verbs are unaccusative (as they seem to be in Chol); however, in a language with a different distribution of intransitive verbs, this analysis would not stand up.5 In my view, the crucial reliance on phases is a vulner- ability of Coon et al.’s (2014) proposal.
This caveat notwithstanding, a phase- based approach to syntactic ergativity leads to several follow- up questions and predictions.
The first prediction follows from the single- occupancy restriction on the phase escape hatch: since the absolutive has to go through the phase edge and raise to TP, all other constituents inside the vP, not just the ergative, should be “trapped”. I am not aware of any languages in which overt or covert object shift to the edge of the vP blocks wh- movement of VP- internal elements. Recall that agent focus (AF) mark- ing on Mayan verbs signals that the absolutive does not get case outside the vP; the other constituents should therefore not be “trapped” inside that vP. By this logic, we might predict that when vP- internal constituents need to move out of a vP, AF will be
5. The analysis would become even more problematic if we were to assume that all vPs, even passives/ unaccusatives, are phases (Legate 2003).
employed. However, the wh- movement of low adjuncts in Q’anjob’al does not require AF marking on the verb, contrary to expectation:
(19) Baq’in max tz’ib’- e- j/ *tz’ib’- e- n Q’anjob’al
when asp write- der- tr/ *write- der- af naq Pedro tx’am carta?
clf Pedro clf letter
‘When did Pedro write a/ the letter?’
A second prediction that follows from the phase- based approach to ergativity con- cerns variable binding. The inspiration for this prediction comes from analyses of object shift, which presumably involves movement of the object to a phase edge.
A shifted object is able to bind a lower constituent, but not vice versa (Larson 1988, 1990 and references therein):
(20) a. I showed every athletei hisi coach.
b. *I showed himi every athlete’si coach (21) a. We will sell no juicei after itsi expiration date.
b. *We will not sell iti after any juice’si expiration date
These asymmetries in binding are accounted for under an analysis in which the object shifts and is therefore structurally superior to the constituent it binds (Holmberg and Platzack 1995; Zwart 2001; Bošković 2004b; Woolford 2007; among others). The underlying idea is that once the object moves up, it should be able to bind an element lower in the structure.
Assuming such an account of binding, we can predict that in the Mayan languages where the absolutive raises out of its base position (thus undergoing an object shift of sorts), it should be able to bind a variable in a lower constituent. Double object constructions would provide a desirable test case, but, unfortunately, Mayan lan- guages do not have such constructions (Coon 2013b, Coon et al. 2014). Examples similar to (20) and (21) are therefore simply untestable. Variable binding in adjunct clauses could likewise provide a potential testing ground for this contrast. Consider a high adjunct clause with a bound expression in it, such as the one in (22) below. We would expect that a quantified absolutive object should bind such a variable only if it raises outside the vP. This should be possible in a high- absolutive language, but not in a low- absolutive language:
(22) Pedro congratulated every studenti [after hei spoke].
Unfortunately, there are two further complications. First, it is not entirely clear what quantified expressions in Mayan are, or whether these languages even have genuine quantifiers that can directly bind the reference of a nominal. In particular, existen- tial quantification is expressed strictly predicatively (Judith Aissen, pers. comm., and Jürgen Bohnemeyer, pers. comm.). Second, even if quantified expressions
were readily available, most Mayan languages do not seem to allow binding under (apparent) quantification: these languages exclusively use demonstratives to refer to anaphoric antecedents. This sort of e- type expression strategy is not sensitive to c- command. Thus, we are prevented from testing variable binding predictions in the Mayan cluster of high- absolutive versus low- absolutive languages. Should more lan- guages become available, variable binding could provide an effective test of the high- versus low- absolutive account.
Another problem for the phase- based approach has to do with syntactic contrasts between ergative and accusative arguments. Recall that for the ergative to be pre- vented from extraction, it must be generated very low and stay within vP. However, this restriction is no different from the restriction on the checking of the accusa- tive: an object must raise to an outer spec,vP to check accusative case against v in a spec- head relationship. Such movement clearly does not prevent the nominative or other internal arguments from extracting. This takes us back to the puzzle we started out with at the very beginning of this work: What motivates syntactic ergativity in such a way that the ergative is different from the accusative?
There may be a way to incorporate the advantages of the phase- based analysis into the PP- ergative hypothesis. Recall the reasoning presented in chapter 6: The PP- ergative is not “nominal” or DP- like enough to satisfy the subject criterion on the higher inflectional head. It therefore stays in spec,vP, and some other means are needed to satisfy the EPP on the T head above. The raising of the absolutive may be one such means (in addition to, or in lieu of, verb raising). On such an account, the absolutive in high- absolutive languages does indeed raise, but for EPP rather than Case reasons. The raising of the absolutive is thus motivated by the defective EPP features on the ergative; the ergative is unable to satisfy the EPP, so the next available DP raises to meet that need. To represent this move- ment schematically,
(23) TP
DPabs [EPP]
T'
T vP
PP v'
v VP
V DPabs
If this is correct, then the existence of high absolutives can in fact be motivated by PP- ergativity, rather than by restrictions on transitive phases. In other words, PP- ergativity would motivate the high- absolutive structure, rather than being a side effect of the high- absolutive parameter.
The approach presented in this section crucially relies on the idea that the expla- nation for syntactic ergativity should reside with the status of the absolutive, and that the status of the ergative is just a side effect of absolutive case assignment.
While differing in details from Coon et al.’s approach, the approach advocated in
Aldridge (2004, 2008) and Legate (2008a) also places the explanatory burden on the absolutive.
The licensing of the ergative is contingent on the presence of a transitive light verb, which means that the absolutive has to be licensed in the structure. To reca- pitulate what was discussed above, this predicts that there cannot be any PP- ergative without an absolutive DP in the same clause. The licensing of that absolutive happens at the level below the v head. The approach to licensing I have developed in this book is different from the proposals offered by Legate (2008) and Aldridge (2004, 2008), who suggest that the absolutive corresponds to nominative case, and is assigned to both the intransitive subject and the direct object by T. The existence of syntactic ergativity is the primary motivation for the proposal that T assigns the absolutive case in such languages. For Aldridge and Legate, syntactic ergativity follows from the assignment of the absolutive, not from the status of the ergative. For example, Aldridge (2008: 983– 984) writes:
The crux of Aldridge’s proposal is that transitive, but not intransitive, v can carry an EPP feature in syntactically ergative languages. The effect of this constraint in recent minimalist theory of Chomsky (2001) is to force absolutive objects to move to the outer edge of the vP, from where they will be able to undergo further move- ment, specifically to the specifier of CP, as in wh- movement or relativization. They will also be interpreted in a position external to VP, thereby receiving wide- scope, presuppositional interpretations.
. . . Aldridge . . . proposes that absolutive case is assigned directly by T or v. T assigns case in intransitive clauses; v does so in transitive clauses. In a transitive clause, v carries an absolutive case feature that it assigns to the direct object. . . . Transitive v also carries an EPP feature, which draws the absolutive NP to its outer specifier, where it is visible to a probe in the next phase, for example, a [wh] fea- ture on C, as in the case of wh- movement. This will allow the absolutive NP to be extracted in cases of A’- movement.
. . . Since the source of absolutive case is v in transitive clauses, absolutive case is still available in non- finite clauses in this type of language. Indeed, in Tagalog and Inuit languages, controlled PRO can appear in the ergative subject position, while absolutive case appears on the object. Absolutive case is still available in a non- finite clause, because it is assigned by v and therefore is not affected by the finiteness of T.
Aldridge proposes yet another corollary to syntactically ergative languages, one that follows from her account of the high- assigned absolutive: the absolutive expression is expected to take wide scope. The critical evidence comes from Inuit, where the absolutive object receives a wide- scope interpretation (Bittner 1994, 1995; Bittner and Hale 1996a, b), whereas the instrumental object in the antipassive receives an obligatory narrow- scope reading (Aldridge 2008: 975). As with some other subtle phenomena discussed throughout this work, we are in need of more data from dif- ferent languages to test this hypothesis. At least some languages that I have been able to test do not conform to the scope generalization proposed by Aldridge.
For example, in Chukchi, only surface scope is allowed, so the readings of clauses depend on the word order:
(24) a. Gamga- ŋawəskət- e ŋərnə- ŋinqey winren- nen. Chukchi every- woman- erg three- boy.abs help- aor.3sg.3sg
b. Gamga- ŋawəskət- e winren- nen ŋərnə- ŋinqey.
every- woman- erg help- aor.3sg.3sg three- boy.abs
‘Every woman was helping three boys.’ (every > three; *three > every) (25) a. ŋərnə- ŋinqey gamga- ŋawəskət- e winren- nen.
three- boy.abs every- woman- erg help- aor.3sg.3sg b. ŋərnə- ŋinqey winren- nen gamga- ŋawəskət- e.
three- boy.abs help- aor.3sg.3sg every- woman- erg
‘Every woman was helping three boys.’ (three > every; * every > three)
Thus, the prediction of a wide- scope absolutive in syntactically ergative languages is not supported by the facts, however limited. A better understanding of scope rela- tions in individual ergative languages would have to build upon an understanding of quantifiers in such languages as well as differences between determiners, demon- stratives, and other types of modifiers.