So far in this chapter, I have considered explanations of syntactic ergativity that appeal to the general architecture of natural language. This makes perfect sense, given that the approach I am advocating— predicated on the notion that the ergative is the complement of an adposition— is itself syntactic. But what if the explanation for this phenomenon lies outside syntax? What if we are ignoring other possibilities just because of our overconfidence in syntactic structure?
In this section, I will explore, and ultimately reject, a non- syntactic explanation for syntactic ergativity. The explanation at hand appeals to language processing, and capitalizes on the observation that, in general, grammatical structures that impose a heavy processing load and tax working memory tend to be avoided (de Vincenzi 1991;
Townsend and Bever 2001; among others). Some researchers have extended this idea to an even stronger claim that language retains (“grammaticizes”) those structures that are most efficient for processing— consider, for instance, the model developed by Hawkins (especially Hawkins 2004), according to which grammatical complexity shows a roughly one- to- one correspondence with processing efficiency. To put this differently, structural properties of a language can be explained by appealing to pars- ing strategies and preferences; within the realm of parsing, structures that are more efficient are the ones that tend to survive.
If structures involving A- bar movement of the ergative expression are more dif- ficult to process (“less efficient”) than structures involving A- bar movement of the
absolutive, we can expect that the former will be used less and avoided more. If this pattern holds true, we can then expect that some languages will take this avoidance to its logical extreme and stop using structures with the ergative gap altogether. In other words, what may be a soft, gradient, processing- based constraint in one lan- guage (“Ergative gaps are hard to process”) may become a hard, categorical constraint (“Avoid ergative gaps”) in another. This leads to the expectation that in morphologi- cally ergative languages without syntactic ergativity (DP- ergative languages, in my terminology), structures with ergative gaps should be more difficult to process than structures with absolutive gaps. If we find that this expectation is confirmed experi- mentally, then we can conclude that the explanation for syntactic ergativity is rooted in processing principles rather than in the syntactic design of the relevant languages.
In order to test this hypothesis, it is necessary to explore the processing of erga- tive versus absolutive gaps in those languages that do not show syntactic ergativity.
Long- distance dependencies have been tested extensively in nominative languages;
subject and object gaps in relative clauses have been investigated in a number of languages, from English to Korean, using different methodologies (see Kwon et al.
2010, 2013 for an overview). With the exception of Chinese, for which different stud- ies have produced conflicting results, the overall outcome provides strong support for the accessibility hierarchy (Keenan and Comrie 1977): subject gaps are easier to process than object gaps. Thus, (26a) is overwhelmingly easier to process than (26b):
(26) a. The senator [who _ _ _ attacked the reporter] admitted the error.
b. The senator [who the reporter attacked _ _ ] admitted the error.
The existence of a large body of research on relative clauses in nominative- accusative languages provides a basis on which processing studies of ergative languages can build. Yet the data collected so far on relative- clause processing in ergative languages are scarce. Furthermore, for ergative languages, it is not sufficient to compare sub- ject and object gaps in transitive relative clauses as in (26a, b); the ergative and the absolutive object also need to be compared to the absolutive intransitive subject. The ergative and the absolutive intransitive subject have the same grammatical function (clausal subject), albeit with different case marking, while the absolutive intransitive subject and absolutive object share morphological marking but are different in their grammatical function. Accordingly, processing data need to be collected along both these axes of contrast.
As of the writing of this book, relative- clause processing has been experimen- tally tested in just a few morphologically ergative languages: Basque (Carreiras et al.
2010),6 Avar (Polinsky et al. 2012), Chol (Clemens et al. 2015), Q’anjob’al (Clemens et al. 2015), and Niuean (Longenbaugh and Polinsky, forthcoming- a, forthcoming- b).
These languages differ along two dimensions: headedness and the morphologi- cal expression of alignment. Avar and Basque are head- final, with prenominal relative clauses, while Chol, Q’anjob’al, and Niuean are strongly verb- initial, with 6. The Basque results are more equivocal than the others; for their discussion, see Clemens et al. (2015) and Longenbaugh and Polinsky (forthcoming- a).
postnominal relative clauses. Basque and Avar also pattern together in their use of ergative case- marking in addition to agreement; Chol and Q’anjob’al, on the other hand, express ergativity in agreement only, and their DPs all appear with null mark- ing. Niuean ergativity is limited to case- marking.
The experimental investigation of Avar in Polinsky et al. (2012) and in subsequent work followed two methodologies: self- paced reading and sentence- picture match- ing. In both cases, the participants were tested on three conditions: absolutive sub- ject gap, ergative gap, and absolutive object gap, as shown in the following sample sentences from the self- paced reading study.
(27) a. extraction of the intransitive (absolutive) subject Avar [ _ _ _ i artistka- yal- da ask’o- y č’:u- n y- ik’- ara- y] yasi
actress- obl- loc near- ii standing- ger ii- be- ptcp- ii girl.abs
GAP HEAD NOUN
‘the girl that stood next to the actress’
b. extraction of the transitive (ergative) subject
[ _ _ _ _ i yas repetici- yal- de y- ač:- un y- ač’- ara- y] artistkai girl.abs rehearsal- obl- loc ii- bring- ger ii- come- ptcp- ii actress.abs
GAP HEAD NOUN
‘the actress that brought the girl to a rehearsal’
c. extraction of the absolutive object
[artistka- yał _ _ _ _ i repetici- yal- de y- ač:- un y- ač’- ara- y] yasi actress- erg rehearsal- obl- loc ii- bring- ger ii- come- ptcp- ii girl.abs
GAP HEAD NOUN
‘the girl that the actress brought to a rehearsal’
In both studies, the intransitive subject gap showed a significant advantage over the other two argument gaps. In the reading study, the processing patterns for the erga- tive subject gap and the absolutive object gap were roughly equal, but followed very different time courses; this is shown in Figure 1.
At word 2, a significant slowdown occurred when the reader encountered an erga- tive (in absolutive- gap relative clauses); the slowdown observed in response to an absolutive (in ergative- gap relative clauses) was much smaller. An explanation for this discrepancy comes from the fact that the ergative is morphologically dependent on the absolutive; thus, the presence of the ergative allows the parser to project a transitive clause containing an absolutive object, and the processing of that projec- tion causes a slowdown. The appearance of the absolutive, on the other hand, does not lead to any such commitment, and hence causes no slowdown. Similarly, pro- cessing studies of Japanese and Korean have reported a slowdown at or immediately after an accusative; again, since the accusative is the dependent case, its presence allows the parser to project the remaining transitive clause with a nominative in it (Miyamoto and Nakamura 2003; Ueno and Garnsey 2008: 665; Kwon et al. 2006).
Outside the relative clause, there is a strong effect of grammatical function; in particular, ergative subjects are read faster than absolutive objects. Cumulative reading times for the absolutive object gap and the ergative gap are not different.
This is essentially a null result, and the most plausible explanation for this result is that the two types of gaps enjoy different advantages, which cancel each other out. The absolutive- object gap is cued by the presence of the ergative in the rela- tive clause, which gives that gap an advantage. However, object gaps seem generally more difficult to parse than subject gaps, which puts the absolutive- object gap at a disadvantage. The ergative gap derives no cueing effect from morphology, but since it is a subject gap, it still enjoys a processing advantage as the head noun of the rela- tive clause.
The reading- time results described above were replicated in a sentence- picture- matching task. On this latter task, Avar participants had to select the appropriate picture to match the content of a given sentence.7 The pictures depicted either (a) a transitive action where one kind of participant acts on another kind of participant, while the latter acts on another participant of the first kind (e.g., a brown goat rams a sheep, while the sheep rams a grey goat, as shown in Figure 2) or (b) an intransitive action, where two different kinds of participants do one thing, and another one of these kinds does something else (e.g., a boy and a girl sleep while another boy laughs, as shown in Figure 3).
Each test participant saw 8 examples of each relative clause type and 12 fillers (consisting of questions concerning the unambiguous participant). Order of presen- tation, left- to- right direction of action, and target participants were counterbalanced.
1100 1050 1000 950 900 850 800
W1 W2 W3 W4 W5 (RC W8
Predicate)
W6 (Head Noun)
W7 (Spill Over) Absolutive Subject Gap Absolutive Object Gap Ergative Gap Figure 1 Reading time results for Avar, in ms (from Polinsky et al. 2012: 272)
7. The pictures were developed by Annie Gagliardi and have been used in a series of experiments on adult and child language comprehension.
There were three sets of pictures with human participants (as in Figure 3), three sets with animate participants (as in Figure 2), and two sets with inanimate participants.
The pictures were presented on a computer screen using the program ExperigenRT (Becker and Levine 2013). In an experiment conducted in 2013, 45 native speakers of Avar (avg. age 40.7; 32 females) saw a picture, heard a recorded prompt (“Where is X?” or “Show X”), and had to push a radio button corresponding to the correct figure out of the three in a given picture. Their response times were recorded from the onset of each prompt.
In picture- matching tasks, there are always a certain number of errors, with respondents picking the wrong participant in the picture. The error rate on rela- tive clauses with the absolutive subject, absolutive object, and ergative gap did not differ significantly: 22.2% errors on absolutive subject gaps, 21.8% errors on absolutive object gaps, and 24.1% errors on the ergative subject gap.8 The response times, normalized by the length of individual recordings, did not differ significantly for the correctly identified relative clause heads or the misidentified ones, indicating that the respondents did not have any particular difficulty with the misidentified heads. Because there was no significant difference between the overall response times recorded in the experiment and the response times for only those trials in which participants responded accurately (p = 0.896), I will
Figure 2 Sample transitive action
Figure 3 Sample intransitive action
8. These percentages may seem quite high, especially when compared to results reported for picture- matching studies conducted in English, where the error rate rarely reaches 10%.
However, English testing often involves younger subjects and, more importantly, subjects who are familiar and comfortable with test- taking, the use of computers, and the notion of multiple choice. The results reported here are a sobering reminder that our expecta- tions, formed on the basis of work with particular populations, need to be tempered once we leave the comfortable confines of university testing environments. The Niuean results reported below are quite comparable to the Avar results with respect to error rate and response time. See Clemens et al. (2015) and Dąbrowska (2010, 2012) for more discussion of the challenges inherent in native-language testing.
only report mean response times for trials in which participants selected the cor- rect response.
The response times (for accurate responses) are summarized in the table below:
The absolutive subject gap was processed significantly faster than the other two types of gap (p = 0.08); meanwhile, the difference between the response times to the ergative gap and the absolutive object gap was not significant (p = 0.35). These results are consistent with the reading- time results. Crucially, they reinforce the finding that relative clauses with an ergative gap do not impose a heavier processing load than relative clauses with an absolutive object gap.
Avar relative clauses are prenominal and participial (although they allow for IP- level adverbs and include negation, aspect, and modals, suggesting that they are structurally quite extensive). Could it be possible that some of the results observed in Avar arose due to the particular head- final properties of the language and the participial nature of its relative clauses? To test this hypothesis, we also conducted a picture- matching study of Niuean, which for all intents and purposes is the mirror image of Avar: VSO, with postnominal relative clauses, and no agreement whatso- ever. Forty- one native speakers of Niuean (avg. age 42;8; 21 females) were tested in Auckland in 2014 using the same picture- matching methodology as in the Avar experiment, with the same distribution of prompts (8 sentences per condition, 3 human, 3 animate, and 2 inanimate participants). To illustrate, the questions cor- responding to Figure 2 were as follows:
(28) a. extraction of the absolutive object Niuean
Ko fe e kotii [ne poka he mamoe_ _ _ i]?
pred where abs goat nfut ram erg sheep
HEAD NOUN GAP
‘Where is the goat that the sheep is ramming?’
b. extraction of the ergative subject
Ko fe e kotii [ne poka _ _ _ i e mamoe]?
pred where abs goat nfut ram abs sheep
HEAD NOUN GAP
‘Where is the goat that is ramming the sheep?’
Experimental questions corresponding to Figure 3 paralleled the following:
Table 1. RESPONSE TIMES (IN MS) TO SENTENCE- PICTURE MATCHING TASK, AVAR SUBJECTS (N=45)
Gap Type Mean RT (ms)
Intransitive Absolutive Subject 3580
Ergative Subject 4620
Absolutive Object 4809