As I already mentioned, the interpretation of syntactic ergativity adopted for the purposes is this book is restricted, and refers specifically to the inability of the erga- tive argument to be extracted with a gap under A- bar movement. Other researchers view syntactic ergativity as a broad phenomenon, so an explanation for my adoption of a notably narrower understanding of this phenomenon is in order.
Several researchers, most notably Konstantin Kazenin (1994) and Christopher Manning (1996), include a wide range of contrasts under the rubric of syntactic erga- tivity: the contrast between absolutive arguments (S and O) and the ergative argu- ment with respect to A- bar movement, coreference across clauses, scope, binding, quantifier float, raising, control, and possibly other dependencies. Based on these criteria, a number of languages could be characterized either as fully syntactically
ergative or as “mixed-pivot” languages (Manning 1996). It seems that this more inclu- sive approach to syntactic ergativity relies on two particular assumptions. First, sub- jects are uniformly defined by a cluster of properties that are responsible for the more
“active” status of subject in syntactic processes (as compared to more “inert” non- subjects). The corresponding properties do not necessarily have to be uniform, as long as the tally of what subjects can do is greater than the tally of what non- subjects can do. Second, the absence of a constituent from a particular position (say, a gap in a rela- tive clause or the missing subject of a clause conjoined with another clause) follows from a coherent set of underlying principles, with the result that all such absences are similar enough to warrant comparison. In other words, the following structures are viewed as qualitatively similar with respect to their silent elements:
(20) head noun X [RELATIVE CLAUSE … X … .] relativization
(21) [CLAUSE 1 X … .] and [CLAUSE 2 … X … .] conjunction reduction/
deletion under coreference (22) controller X [CONTROL CLAUSE … X … .] control structure
(purpose clause)
Manning (1996: 34) suggests that the three structures in (20) through (22) fol- low either the ergative pattern or the neutral pattern (where the absolutive and the ergative arguments are treated alike). Furthermore, they seem to be arranged hier- archically; for instance, the presence of syntactic ergativity under relativization is more likely than the presence of syntactic ergativity under conjunction reduction and purpose- clause formation. Thus, a language that exhibits syntactic ergativity under purpose- clause formation should also exhibit it under conjunction reduction and relativization, etc.
(23) Cline of syntactic ergativity (Kazenin 1994; Manning 1996)
relativization > conjunction reduction > purpose- clause formation
The problem with such a hierarchy is that it does not consider the possibility that the reasons for deletion of a constituent may vary across the three structures.
One can reject the machinery of a particular theory and its concomitant specific theoretical constructs— such as traces or null pronominals— while accepting that the silent elements in control clauses, relative clauses, and conjoined clauses have systematically different properties. Examples abound, and I limit the discussion here to a single illustration pertaining to silent subjects in Spanish embedded clauses.
Consider the paradigm in (24), below (see also Freidin 2007: 204). In all three sentences in (24), Juan is the subject of the matrix clause. However, the rules of core- ference in each clause are quite different, and clearly depend on the matrix verb as well as on the distance between the possible antecedent and the gap. Furthermore, only in (24b) and in the lower clause of (24c) can the empty category alternate with an overt pronoun without changing the interpretation.
(24) a. Juani quiere [que _ _ *i/ k se vaya]. Spanish Juan wants comp refl go
‘Juan wants for you/ for someone else to go.’
NOT: ‘Juan wants to go.’
b. Juani no cree [que _ _ i/ k se vaya].
Juan neg believes comp refl go
‘Juan does not believe that he is going/ you are going/ someone is going.’
c. Juani no quiere [que _ _ *i/ k crea [que __ i/ k se vaya]].
Juan neg wants comp believe comp refl go
‘Juan does not want you/ someone else to believe that he/ you/ someone else will go.’
Given the significant observed differences between different types of silent ele- ments, it is unclear how much can be gained by collapsing all of these elements into one group. Certainly, we stand to lose important structural distinctions when we dis- regard the smaller details of individual structures and of different types of silence. In admitting that not all types of silence are the same, we recognize that some types of silence are more constrained in terms of their distribution and interpretation— for example, the silent element in the Spanish example in (24a) is more constrained than the silent element in (24b). If not all types of silence are created equal, compar- ing them may not always be straightforward.
In generative grammar, and in minimalism in particular, extraction is accounted for under Move (or, strictly speaking, under copying or internal re- Merge of an already merged syntactic object); together with Agree and external Merge, Move is part of the narrow syntax (for arguments, see Safir 2008, among others). The output of narrow syntax is then enriched by further vocabulary to fully represent logical structure. It’s within this logical structure that operations such as scope- reading and operator- variable binding occur (Reuland 2011). Coreference across clauses is part of an even broader system— that of discourse. The discourse component of the gram- mar situates the logical syntax in a larger context that includes world knowledge, speaker intent, and the full linguistic context. Discourse is where reference relations are established; it includes coreference relations across clauses and sentences, and possibly reference of the imperative addressee. On this approach, generalizations related to A- bar movement pertain to narrow syntax, as do generalizations related to obligatory control and raising (although these two phenomena do not involve A- bar movement). What happens in conjunction reduction pertains in large measure to discourse; together, these facts argue for treating conjunction reduction effects separately from movement effects.17
The hierarchical relationship between structure- building mechanisms (includ- ing extraction), on the one hand, and logical- syntactic and discourse relation- ships, on the other, gives us a natural way of dividing the dependencies discussed
17. In functional approaches to syntax, all types of silence (if they are allowed at all) are represented along a cline; this amounts to treating all the variants of silence as stemming from the same source, with their differences being determined by specific lexical verbs
here; the boundary between extraction (A- bar movement) and A- movement, binding, and various anaphoric dependencies related to binding is not ad hoc.
Some of these dependencies— most notably coreference across clauses— belong to a part of language that lies outside the purview of syntax and concerns the dis- course integration of language material. These facts alone call for a more restric- tive view of syntactic ergativity, which excludes these phenomena as diagnostics for syntactic ergativity. In addition, as I will show below, the division of erga- tive languages into two types based on their A- bar movement properties actually allows us to account for a larger cluster of syntactic dependencies in a principled way. In other words, what has been presented as a series of implicational proper- ties by other researchers, as in (23), can be shown to follow from a single contrast between two types of ergative languages, differing in the nature of their ergative expression.
The result is a more predictable, uniform syntax, which is always desirable. If the cost of that predictability involves giving up some of what is traditionally assumed to be part of syntax, that does not seem to be an unwelcome trade- off.
Accordingly, narrowing down the domain in which syntactic ergativity is evalu- ated allows us to treat (with certainty) cross- linguistic instances of syntactic erga- tivity as realizations of the same phenomenon, not different phenomena whose connection to each other may not be entirely clear or apparent. By starting small, we can maintain the explanatory power of our observations.
Limiting the range of phenomena under consideration allows us to arrive at generalizations drawn from the same grammatical domain. But in addition to that, there is also a more practical consideration for starting small, which has to do with the accuracy of cross- linguistic comparisons. In a broad cross- linguistic compari- son, it is not uncommon to compare phenomena from several languages even if we are not certain how similar these phenomena are. In these cases, we run the risk of
or constructions. That, in turn, entails the possibility that a particular instantiation of silence (“deletion” of an element) may display properties associated with different con- texts. A cline of properties, such as the one shown in (23), is therefore useful in capturing the facts as they are observed, but, crucially, it also allows for a mixture of properties— for instance, if a silent element is observed in a purpose clause, its properties might show some of the signatures of a silent element in a relative clause, plus additional purpose- clause- specific properties. The result is an absence of discrete categories. While this approach has benefits, failure to recognize discrete categories (including discrete types of silent elements) prevents us from establishing a priori a set of features that determine how a given category enters into a relationship with other categories. This, in turn, could lead to missed generalizations.
In unification- based frameworks, it is common to distinguish between grammatical rela- tions and thematic (argument) structure, with relativization, wh- question formation, and topicalization being representative of the former and control (purpose) clauses and bind- ing representative of the latter (it is not entire clear where conjunction reduction belongs).
A cline such as in (23) is useful in that it treats phenomena that are linked to grammatical relations as more constrained than phenomena associated with argument structure. Put differently, there is an expectation that cross- linguistic variation is more likely to occur at the level of grammatical relations than at the level of argument structure.
comparing structures whose similarities are only superficial. Purpose clauses are a particularly telling case; in some languages, they are always non- finite, in oth- ers, they are always finite, and for quite a few languages, they can appear in both structures. Without taking finiteness into consideration, it is difficult to compare the types. And if finiteness is taken into consideration, the cline in (23) needs to be revised.
To offer another illustration, consider coordination: how clear can it be, from a cursory survey of cross- linguistic data, that a particular instance of coordination represents coordination of clauses as opposed to coordination of verb phrases?
Clausal coordination and VP- coordination are often difficult to tell apart without an extensive language- particular analysis. The difference is not always clear even for such well- studied languages as English; consider examples such as (25a, b), for which one may need to posit IPs at a lower level of representation or posit various types of null categories in the second conjunct (Van Valin 1986; Goodall 1987; Burton and Grimshaw 1992; McNally 1992; among others):
(25) a. The criminal will be arrested and will confess to the crime.
(Burton and Grimshaw 1992: 310)
b. The employees complained and were given more vacation time.
(McNally 1992: 336)
Along the same lines, consider the ambiguity in the following sentence; the mate- rial deleted in each of the readings is indicated by strikethrough.
(26) Must and will she make John go to the dentist?
(i) Must she make John go to the dentist and will she make John go to the dentist?
(ii) Must John go to the dentist and will she make John go to the dentist?
Under reading (i), John is construed as the object in both instances; under read- ing (ii), John is the subject of the event of going and the object of her causing him to go (Wilder 1997: 92). In (i), the two constituents joined by and are verb phrases, which share the same subject. Under reading (ii), the two constituents are clauses, not verb phrases.
If we cannot definitively determine what happens in English, how can we be sure whether a paratactic combination of two clauses in a lesser- known language, avail- able only from a grammatical description, stands for conjunction reduction?
Next, conjunction reduction and reduction in purposive clauses are not always easy to tell apart without an in- depth linguistic analysis. Imagine a language— let’s call it English- 1— which has ergativity, pro- drop, and the functional element lest.
Furthermore, imagine that we only have a couple of examples of sentences with that functional element, such as those shown below:
(27) Johni was afraid to speak out [lest _ _ _ i be punished for his words]. English- 1
Is the bracketed clause coordinate or subordinate? Is lest a conjunction like and or a complementizer like for? Is the entire sentence an instance of conjunction reduc- tion or reduction in a purpose clause? Without clear answers to such questions, the comparison of English- 1 with Tongan or Basque would be meaningless.18 The limita- tions in identifying a particular phenomenon as an instance of conjunction reduc- tion across clauses— or as an instance of a finite purpose clause— can compromise our subsequent comparisons, offering yet another argument for limiting syntactic ergativity to a smaller class of phenomena rather than adopting the more inclusive approach.
In summary, the proposal here is to limit syntactic ergativity to extractions across clauses— those representing long- distance syntactic dependencies— and to exclude referential dependencies such as coreference across clauses. In some languages, we may find that the two sets of phenomena are subject to similar constraints, but there is enough evidence not to expect that, and combining syntactic and referential dependencies may result in missed generalizations.
1.3.2 Not all A- bar movement phenomena are created equal
We are left with syntactic dependencies grouped under A- bar movement. Within the A- bar movement family, there could be difficulty in detecting whether or not a set of phenomena share a structural similarity. There are at least two reasons for this difficulty. First, in some languages, wh- questions, focus constructions and topic con- structions are actually biclausal and are based on relativization. For such languages, relativization is in fact the main A- bar phenomenon. This is the case for Tongan, which I will discuss in chapter 10; Tongan wh- questions, focus, and left- dislocated topic constructions are biclausal and include a headless relative clause. The only other type of A- bar movement in Tongan is short topicalization, observed in some embedded clauses (see chapter 10, section 10.5).
Among the subtypes of A- bar movement, relativization is probably the most read- ily detectable without an in- depth syntactic analysis; relative clauses exhibit similar characteristics across different languages, and it seems a reasonable assumption that relative clauses are universal, or at least very common. This does not mean that all relative clauses are built alike, but we have accumulated a set of clear diagnostics that allow us to test the analyses available (cf. Bianchi 2002; Bhatt 2002; Hulsey and Sauerland 2006, and further references therein). In other words, in a cross- linguistic comparison, however coarse- grained, relativization is easily visible, unlike some other operations. That already makes it a good starting point for assessing the range of syntactic ergativity.
The second reason for the heterogeneity of A- bar movement processes within a given language has to do with differences in pied- piping, which I will discuss more
18. For instance, Manning (1996: 34, 60– 74) writes that, for certain languages, the con- junction reduction analysis or purposive clause analysis would have to be abandoned if the relevant constructions were analyzed “properly.”
in chapter 2. Some A- bar operations, including wh- questions in particular, tend to be more amenable to pied- piping than other A- bar operations, such as topicaliza- tion. We can therefore expect to find morphologically ergative languages where syn- tactic ergativity is observed under topicalization and relativization but not under wh- question formation. As I will show in the subsequent chapters, such languages do exist.
All said, relativization seems to be both the clearest phenomenon to test (since we do not run much of a risk of confusing it with something superficially similar but structurally different) and the phenomenon where the contrast between the ergative and the absolutive is most clear- cut: under syntactic ergativity, the absolutive can extract with a gap, while the ergative cannot.
To summarize, the understanding of syntactic ergativity adopted here subsumes evidence from A- bar movement, including relativization, wh- question formation, focusing, and topicalization. In an ideal situation, a given language would offer all these phenomena for study, but some languages may have a more restricted inven- tory of A- bar operations. In fact, I will argue below that many properties associated with the broader understanding of syntactic ergativity discussed above (especially in the work by Kazenin 1994 and Manning 1996) simply follow from the division of ergative languages into subtypes which will be introduced below.
1.3.3 Some methodological odds and ends
In my desire to start small, I have tried to maintain the connection between syntac- tic and morphological ergativity. The methodological consequence of this approach is that I will focus on morphologically ergative languages that also have syntactic ergativity. This allows me to avoid at least some controversy with respect to deter- mining which languages are syntactically ergative. Western Austronesian languages are a prime example of the (possible) controversy at work here. Some researchers argue that these languages are syntactically ergative (Manning 1996; Aldridge 2004, 2007, 2008), while others consider them to be syntactically accusative (Kroeger 1993; Pearson 2005). In addition, a number of researchers focus on their subject- only characteristic, mentioned above, wherein only one prominent argument is vis- ible to syntax and can undergo extraction (Gọrtner et al. 2006; Chung and Polinsky 2009). These languages are not morphologically ergative, however, which allows me to exclude them for present purposes. Once we have a better understanding of syn- tactic ergativity in morphologically ergative languages, it should become possible to expand the ensuing account to languages without morphological ergativity.
In developing an account of syntactic ergativity, I have tried to balance the need to perform a thorough linguistic analysis of each language under discussion with the desire to establish broader cross- linguistic generalizations. A micro- typological investigation is an ideal, albeit not always available, way of balancing such needs (Baker and McCloskey 2007; Polinsky and Kluender 2007); on this approach, one starts with an in- depth investigation of (approximately) three to seven languages in order to build a fine- grained characterization of a given phenomenon. More sweeping
generalizations can then be drawn on the basis of the data from these languages and tested against a larger sample. The present work aims to conduct this type of micro- to- macro investigation; I will focus especially on a thorough accounting of a small number of languages, but I will also occasionally survey more languages and try to make more wide- ranging conclusions as well.
APPENDIX
Compensatory strategies under syntactic ergativity
Presumably all languages have the need to form questions such as “Who did that?” or to form relative clauses that somehow restrict the reference of the logical subject in a description of an event with two participants. If forming such questions or relative clauses via A- bar movement of the transitive subject is impossible, we can expect a syntactically ergative language to work around the ban on ergative extraction. The purpose of this appendix is to discuss the most common ways of coping with this restriction in syntactically ergative languages. Let me emphasize that it is the inabil- ity of the ergative to undergo A- bar movement that remains constant: across the board, we find that something prevents the ergative (but never the absolutive) from undergoing A- bar movement. The ways of circumventing this restriction may vary and are not specific or unique to ergative languages. This, in turn, means that the list of workaround strategies presented here is not exhaustive; I have done my best to identify different strategies, but there is no principled approach that could limit their number or content. It is possible that a syntactically ergative language may also use different makeshift measures to relativize the ergative and to topicalize it. My primary focus in this appendix will be on strategies for relative clause formation in the absence of ergative extraction. These strategies include: resumption, antipassiv- ization, agent focus, anti- agreement, and the use of nominalizations.
The resumption strategy was illustrated by the Tongan examples above; see espe- cially example (9) above. We will also see this option at work in some Mayan lan- guages (discussed in chapter 4).
Another strategy is to first turn the ergative subject into an absolutive (retaining its subjecthood) and then extract it. This switch from ergative (transitive subject) to absolutive (intransitive subject) reflects a pattern that we will see recurring through- out this book: the behavior of transitive subjects differs from the behavior of intran- sitive subjects with respect to a number of properties.
The conversion of a transitive subject to an intransitive subject can be achieved through antipassivization. To understand the antipassive, let’s start with the hierar- chy of grammatical relations shown below. Within the elements of this cline, most linguistic theories recognize, at minimum, the difference between core arguments (subject, object) and the rest:
(28) subject > object > non- core argument > non- argument