I have proposed that the theta- licensing of the ergative PP is done externally by the v head: the base- generated PP in spec,vP has a complement that is directly theta- licensed by v. As a result, the interpretation of the DP complement changes depend- ing on the predicate with which it occurs. Thus:
(17) vP
PP v’
P DP v VP
The important thing to note about this structure is that theta- licensing and case- licensing are done by separate heads and are therefore relatively independent of each other; in particular, case- licensing is severed from the v head. If this connection is severed completely in a given language, that language should lose the capacity to impose non- canonical (quirky) case- licensing on its subjects, since this operation is ordinarily done by the v head (Andrews 1976; Sigurðsson 1991; Massulo 1993;
Fernández- Soriano 1999; Fanselow 2002; among others). Such a language may still
possess a few verbs, specified in the lexicon, which require a particular lexical case;
however, the possibility of a large, semantically uniform class of verbs that license a particular case in subject position is not predicted in such a language; positing such a class would simply be redundant.
“Quirky subjects” are often semantically motivated by the predicates of their clauses. Dative subjects, for example, quite often appear with predicates denoting perception, cognition, attitude, potentiality, or happenstance (Bhaskararao and Subbarao 2004; Fanselow 2002, among others). Languages with well- known non- canonical subjects are mostly accusative (Icelandic, Romance, Slavic); however, such non- canonical subjects are also found in ergative languages— see Comrie (2001), Comrie and van den Berg (2006), Bhaskararao and Subbarao (2004), Joppen and Wunderlich (1995), and Režać (2008b), among others, for a helpful discussion. Indo- Aryan languages, Nakh- Dagestanian languages, and Basque exhibit experiencer sub- jects that are non- ergative, but nevertheless show all the subject properties (apart from agreement) that are observed for absolutive and ergative subjects (e.g., con- trol and binding). In Nakh- Dagestanian languages, subjects of potential verbs are encoded in one of the (many) locative cases, as illustrated below for Hinuq:
(18) a. De buƛe bul- an. Hinuq
1sg.erg house.abs build- fut
‘I will build a house.’
b. Di- qo buƛe bu- ɫ- o.
1sg- apudess house.abs build- pot- prs
‘I can build a house.’
(see also Cysow and Forker 2009; Forker 2010 and further references therein) The following Basque example illustrates a typical perception verb with a dative sub- ject; the dative can bind the absolutive (19a) but not vice versa (19b), as is typical of dative subjects (see Režać 2008b: 76 and discussion there):
(19) a. Kepa- rii bere buru- ai gustatzen zako. Basque Kepa- dat his head- det.abs like aux
‘Kepa likes himself.’
b. *Kepai bere buru- a- rii gustatzen zako.
Kepa.abs his head- det- dat like aux (Režać 2008b: 76)
At first glance, these examples may seem to contradict the prediction made earlier in this section, since quirky case and ergativity clearly do co- occur. However, such co- occurrence is found only in the absence of syntactic ergativity. The ergative languages with non- canonical (quirky) subjects are not syntactically ergative; in all of these languages, the ergative leaves a gap at the extraction site, just as the absolutive does.
An alternative strategy is found in Hindi, where relativization is expressed via cor- relatives (see Lipták 2009 and further references therein), and there is no evidence of syntactic ergativity.
If we turn to languages with syntactic ergativity, we find a striking absence of non- canonical subjects. There is no evidence of such subjects in Chukchi (Skorik 1962, 1977, and my personal field notes), in Polynesian languages, in Salish languages, in Northwest Caucasian languages, or in Mayan languages. Of course the lack of non- canonical subjects does not mean that anything goes in the subject position of PP- ergative languages; for instance, Polynesian languages, accusative and ergative alike, do not allow inanimate subjects, and as a result, sentences such as the English The storm destroyed the boat are expressed as statives; the PP denoting cause (the storm) does not have any subject properties. The following Tuvaluan sentence illustrates this pattern:
(20) Kaati kaa fiaffia koulua i taku gaaluega. Tuvaluan
perhaps fut be.happy 2sg.dual at my work
‘Perhaps my work will make you happy.’
(lit.: you will probably be happy with my work)
(Besnier 2000: 275)
While inanimate subjects in Polynesian must appear as stative agents, not all stative agents have to be inanimate. The stative construction in the example below is the same as in (20), but the stative agent is now animate:
(21) Ttou fenua ne puapuagaatia i tino Ppaalagi. Tuvaluan our island npst be.distressed at person Westerner
‘Westerners brought distress to our island.’
(lit.: our island is in distress because of Westerners) (Besnier 2000: 275)
Regardless, restrictions on animacy or other semantic properties of subjects do not seem to be tied to a particular alignment, so looking for canonical versus non- canonical encoding of allowable subjects is still a relevant practice.
There is a worry that the generalization I have drawn in this section may be too general; after all, I have mainly listed language groups, rather than individual lan- guages, and it is certainly the case that not all of the languages in these families display syntactic ergativity. It is important, therefore, to stress that the implica- tion considered here is unidirectional: specifically, syntactically ergative languages are not expected to have non- canonical (quirky) subjects. For languages without syntactic ergativity, there is simply no prediction. Some such languages will have quirky sub- jects, while others will not. As is typical of tetrachoric universals, where only one option out of four is ruled out, we are able to eliminate the presence of non- canonical subjects in PP languages, but the remaining three possibilities may have different explanations.