Here is a list of various research projects that I have worked on, or am currently working on:

  • The Anaphor Agreement Effect: further evidence against binding-as-agreement

    	Author = {Preminger, Omer},
    	Note = {{M}s.},
    	Title = {The {A}naphor {A}greement {E}ffect: further evidence against binding-as-agreement},
    	Url = {},
    	Year = {2019}}
    The Anaphor Agreement Effect (AAE; Rizzi 1990, Woolford 1999, a.o.) refers to a cross-linguistic ban on phi-feature agreement with anaphors. At first glance, the AAE seems to suggest a rather tight interaction between syntactic phi-feature agreement on the one hand, and binding on the other. This has led some (e.g. Reuland 2011) to take the AAE as evidence in support of a reduction of binding to phi-agreement.

    I show that, upon closer inspection, the AAE provides fairly strong evidence against the reduction of binding to syntactic phi-feature agreement. That is because a reductionist view of the AAE requires, somewhat paradoxically, assumptions about phi-agreement, and about the structure of anaphoric expressions, which break their compatibility with the very mechanism of binding-as-phi-agreement.

    I instead propose that the AAE arises due to what I call encapsulation: the binding index associated with anaphoric binding resides on a separate, higher projection than where valued phi-features reside. In addition to accounting for the AAE, the encapsulation hypothesis enjoys broad cross-linguistic morphological support (Middleton 2018).

    Finally, I offer a proposal on the nature of phi-feature matching between anaphors and their binders, given that syntactic phi-feature agreement cannot be what is responsible. I point out that even in the absence of any syntactic relation between binder and bindee (e.g. in cases of Donkey Anaphora), phi-feature matching is manifested between the two. This is not restricted to semantically interpreted features, either; the same applies to, e.g., grammatical gender on inanimates. Crucially, whatever non-syntactic mechanism underlies these cases is then sufficient to ensure phi-matching between an anaphor and its binder, as well, with no involvement of syntactic phi-feature agreement.
  • What the PCC tells us about “abstract” agreement, head movement, and locality

    	Author = {Preminger, Omer},
    	Doi = {10.5334/gjgl.315},
    	Journal = {Glossa},
    	Pages = {13},
    	Title = {What the {PCC} tells us about ``abstract'' agreement, head movement, and locality},
    	Volume = {4(1)},
    	Year = {2019}}
    It has become commonplace in syntactic theory to posit feature-valuation relations, such as agreement between a verbal head and a nominal argument, even in cases where there is no associated morpho-phonological covariance. Let us refer to such hypothesized feature-valuation relations, where the assumed exponents are all null, as "abstract" agreement. In the first part of this paper, I use the cross- and intra-linguistic distribution of Person Case Constraint (PCC) effects to argue that natural language does not allow abstract agreement in phi‑features (person, number, and gender/noun-class).

    Next, I turn my attention to clitic doubling. As far as PCC effects are concerned, clitic doubling behaves as though it were equivalent to overt agreement. In fact, the distribution of the PCC is hard to state unless we collapse the two. This is surprising because, quite simply, clitic doubling is not agreement; it behaves like movement, and unlike agreement, in crucial respects (most notably, in creating new antecedents for binding). Nor can this be because clitic doubling, qua movement, is contingent on prior agreement – since the claim that all DP movement depends on prior agreement is demonstrably false.

    I propose that clitic doubling necessarily involves a preliminary agreement step because it is an instance of non-local head movement – and movement of X0 to Y0 always requires a prior syntactic relationship between Y0 and XP. In cases of maximally local head movement (à la Travis 1984), this requirement is satisfied by c‑selection. But in non-local cases, it is phi‑agreement that fills this role. Thus, wherever clitic doubling is found, agreement has to have occurred, explaining why the two are interchangeable when it comes to conditioning the PCC.

    I conclude by discussing the nature of the ban on abstract phi‑agreement. Viewed as a grammatical principle, this ban would require simultaneous reference to syntax and morpho-phonology, mixing information from different grammatical modules into one constraint. Instead, I suggest that this ban is not a grammatical principle at all: it arises as the result of the acquisition strategy learners engage in when it comes to the placement of unvalued phi‑features on functional heads.
  • The Agreement Theta Generalization (with Maria Polinsky)

    	Author = {Polinsky, Maria and Preminger, Omer},
    	Journal = {Glossa},
    	Title = {The {{\em{{A}greement {T}heta {G}eneralization}}}},
    	Url = {},
    	Year = {to~appear}}
    In this squib, we propose a new generalization concerning the structural relationship between theta assigners and heads showing morpho-phonologically overt agreement, when the two interact with the same argument DP.

    This structural generalization bears directly on the proper modeling of syntactic agreement, as well as the prospects for reducing other syntactic (and syntacto-semantic) dependencies to the same underlying mechanism.

    (This work began as Section 7 of the unpublished manuscript "Agreement and semantic concord: a spurious unification" — see separate entry, below — but has now been expanded into a standalone squib.)
  • m-merger as relabeling: a new approach to head movement and noun-incorporation

    	Author = {Levin, Theodore and Preminger, Omer},
    	Note = {Poster presented at the {{\em{40th {G}enerative {L}inguistics in the {O}ld {W}orld ({GLOW}) Colloquium}}}},
    	Title = {\textsc{m\mbox{-merger}} as relabeling: A new approach to head movement and noun-incorporation},
    	Url = {},
    	Year = {2017},
    We propose a modification of Matushansky's (2006) syntactic approach to head movement. The modification has two main advantages, one conceptual and one empirical.

    The original Matushansky proposal treats X-to-Y head movement as movement of X to [Spec,Y], followed by m‑merger of X and Y. The latter operation is problematic on two fronts. First, it constitutes an interleaving of morphology within syntax. More importantly, it takes what was not previously a constituent – X and Y, to the exclusion of the material in [Compl,Y] – and turns it into something that behaves as a constituent, as far as subsequent syntactic operations are concerned. Outside of head movement, such manipulation of syntactic constituency is unheard of. We propose, instead, that the readjustment embodied by Matushansky's m‑merger happens not at the level of constituency, but in the labels that X and Y bear.

    Empirically, this allows us to explain why it is that – in any given language – the licensing conditions on reduced nominals (bare nouns, D‑less NPs, etc.) are always at least as stringent as those that apply to full-fledged DPs. These more stringent licensing conditions often include adjacency (as in cases of incorporation and pseudo-incorporation), which our proposal derives as a specific case of a general linearization schema for heads with the kind of complex labels formed by m‑merger.
  • Split ergativity is not about ergativity (with Jessica Coon)

    • download draft of paper; published in 2017, in The Handbook of Ergativity, ed. Jessica Coon, Diane Massam & Lisa Travis, 226-252. Oxford: Oxford University Press.
    	Address = {Oxford},
    	Author = {Coon, Jessica and Preminger, Omer},
    	Booktitle = {The Handbook of Ergativity},
    	Editor = {Coon, Jessica and Massam, Diane and Travis, Lisa},
    	Publisher = {Oxford University Press},
    	Title = {Split ergativity is not about ergativity},
    	Pages = {226--252},
    	Year = {2017}}
    It has been frequently noted in the literature on ergativity that few (if any) ergative systems are purely ergative. Rather, many ergative languages exhibit a phenomenon known as "split ergativity" in which the ergative pattern is lost in certain parts of the grammar.

    The central argument put forth in this paper is that split ergativity is epiphenomenal, and that the factors which trigger the appearance of such splits are not limited to ergative systems in the first place. In both aspectual and person splits, we argue, the split is the result of a bifurcation of the clause into two distinct case/agreement domains; this bifurcation results in the subject being, in structural terms, an intransitive subject. Since intransitive subjects do not appear with ergative marking, this straightforwardly accounts for the absence of ergative morphology in those cases. But crucially, such bifurcation is not specific to ergative-patterning languages; rather, it is obfuscated in nominative-accusative environments because, by definition, transitive and intransitive subjects pattern alike in those environments, and the terminology in question ('ergative' vs. 'non‑ergative') specifically tracks the behavior of subjects. Thus, the tendency noted above does not reflect any deep instability of ergative systems, nor a real asymmetry between ergativity and accusativity.

    In an ergative system that exhibits this type of split, ergative-absolutive alignment is always associated with a fixed set of substantive values (e.g. perfective for aspectual splits, 3rd person for person splits). The account we present derives this universal directionality of splits by connecting the addition of extra structure to independently attested facts: the use of locative constructions in progressive and non-perfective aspects (Bybee et al. 1994, Laka 2006, Coon 2013), and the requirement that 1st and 2nd person arguments be structurally licensed (Bejar & Rezac 2003, Preminger 2014).
  • How can feature-sharing be asymmetric? Valuation as UNION over geometric feature structures

    • download draft of paper; published in 2017, in A Pesky Set: Papers for David Pesetsky, ed. Claire Halpert, Hadas Kotek & Coppe van Urk, 493-502. Cambridge, MA: MITWPL.
    	Address = {Cambridge, MA},
    	Author = {Preminger, Omer},
    	Booktitle = {A {P}esky {S}et: Papers for {D}avid {P}esetsky},
    	Editor = {Halpert, Claire and Kotek, Hadas and van Urk, Coppe},
    	Pages = {493--502},
    	Publisher = {MITWPL},
    	Title = {How can feature-sharing be asymmetric? {V}aluation as \textsc{union} over geometric feature structures},
    	Year = {2017}}
    In this paper, I review two recent developments in the theory of syntactic agreement – feature sharing, and feature-geometric agreement – along with brief synopses of the kinds of facts that have motivated each of the two. I then present an apparent puzzle that arises when these two results are juxtaposed with one another: given that a complete lack of values is itself a valid phi-featural representation (viz. 3rd person singular), why would feature-sharing consistently choose the more marked of its two operands as the output of the sharing operation? I propose a solution in terms of a UNION operation, defined not over sets, but over feature-geometric representations of the kind put forth by Harley & Ritter (2002) and others.
  • Agreement and semantic concord: a spurious unification (with Maria Polinsky)

    	Address = {Storrs, CT},
    	Author = {Preminger, Omer},
    	Note = {Paper presented at the {U}niversity of {C}onnecticut Linguistics Colloquium},
    	Title = {Upwards and onwards},
    	Url = {},
    	Year = {2015}}
    	Author = {Preminger, Omer and Polinsky, Maria},
    	Note = {Ms.},
    	Title = {Agreement and semantic concord: a~spurious unification},
    	Url = {},
    	Year = {2015}}
    This work is a much expanded version of Preminger 2013 (TLR), arguing against the spate of recent proposals aiming to reverse the direction of the formal operation underlying agreement in phi-features. We contend that these proposals are an outgrowth of a reasonable -- but ultimately, ill-fated -- urge to reduce any and all correspondence between two elements (be it agreement, or semantic concord) to the same underlying operation. In other words, they constitute a spurious unification.

    We focus here on Bjorkman & Zeijlstra's (2014) recent "hybrid" proposal -- which is, in title, an attempt to argue for the reversal of agreement, but which sanctions both directions of agreement under different sets of circumstances, for the sake of unifying phi-feature agreement with semantic concord. First, we address instances where Bjorkman & Zeijlstra attempt to argue from data that, upon closer inspection, fail to distinguish regular from reverse agreement. These include fully local agreement (in Bantu), whose irrelevance to the debate was already noted in Preminger 2013; and agreement asymmetries between SV and VS word orders.

    Next, we address long-distance agreement (LDA) in Tsez and in Basque, the two empirical domains put forth in Preminger 2013 as challenges to reverse agreement. Bjorkman & Zeijlstra attempt to reanalyze these two domains within their hybrid system. We present empirical and conceptual arguments against their analysis. We then review further crosslinguistic evidence demonstrating the same basic point: that a reversal in the direction of agreement is empirically unsupported.

    Finally, we argue that even if it were successful, Bjorkman & Zeijlstra's proposal would not have achieved what such a unification sets out to achieve: a reduction in the amount of machinery required in the overall theory. Instead, they have to propose several principles which overgenerate and make incorrect predictions.
  • Nominative as no case at all: an argument from raising-to-accusative in Sakha (with Jaklin Kornfilt)

    • download draft of paper; published in 2015, in the Proceedings of the 9th Workshop on Altaic Formal Linguistics (WAFL 9), MIT Working Papers in Linguistics 76, ed. Andrew Joseph & Esra Predolac, 109-120. Cambridge, MA: MITWPL.
    	Address = {Cambridge, MA},
    	Author = {Kornfilt, Jaklin and Preminger, Omer},
    	Booktitle = {Proceedings of the 9th {W}orkshop on {A}ltaic {F}ormal {L}inguistics ({WAFL}~9)},
    	Editor = {Joseph, Andrew and Predolac, Esra},
    	Number = {76},
    	Pages = {109--120},
    	Publisher = {MITWPL},
    	Series = {MIT Working Papers in Linguistics},
    	Title = {Nominative as \emph{no case at all}: An argument from raising-to-\textsc{acc} in {S}akha},
    	Year = {2015}}
    In this paper, we present a novel argument that the proper grammatical representation of cases like nominative and absolutive – and, potentially, genitive – is as the *absence* of otherwise assigned case. This contrasts with a positively-specified view of, e.g., nominative, where it is actively "assigned" by the grammar (e.g. Chomsky 2000, 2001). This result provides further evidence that the assignment of case is not a consequence of agreement (see Bobaljik 2008, and Preminger 2011, 2014).

    The argument is based on raising-to-accusative constructions in Sakha (Turkic).
  • Case in Sakha: are two modalities really necessary? (with Ted Levin)

    	Author = {Levin, Theodore and Preminger, Omer},
    	Doi = {10.1007/s11049-014-9250-z},
    	Journal = {Natural Language \& Linguistic Theory},
    	Number = {1},
    	Pages = {231--250},
    	Title = {Case in {S}akha: Are Two Modalities Really Necessary?},
    	Volume = {33},
    	Year = {2015}}
    Baker & Vinokurova (2010) argue that the distribution of morphologically observable case in Sakha (Turkic) requires a hybrid account, which involves recourse both to configurational rules of case assignment (Bittner & Hale 1996, Marantz 1991, Yip, Maling & Jackendoff 1987), and to case assignment by functional heads (Chomsky 2000, 2001). In this paper, we argue that this conclusion is under-motivated, and present an alternative account of case in Sakha that is entirely configurational. The central innovation lies in abandoning Chomsky's (2000, 2001) assumptions regarding the interaction of case and agreement, and replacing them with Bobaljik's (2008) and Preminger's (2011, 2014) independently motivated alternative, nullifying the need to appeal to case assignment by functional heads in accounting for the Sakha facts.
  • Agreement and its failures

    	Address = {Cambridge, MA},
    	Author = {Preminger, Omer},
    	Number = {68},
    	Publisher = {MIT Press},
    	Series = {Linguistic Inquiry Monographs},
    	Title = {Agreement and its failures},
    	Year = {2014}}
    In this monograph, I show that the typically obligatory nature of predicate-argument agreement in phi-features (person, number, and gender/noun-class) cannot be captured using a model based on "derivational time-bombs." In such a model, there are elements of the initial syntactic representation that cannot be allowed to persist in the final, end-of-the-derivation structure, and it is the application of agreement that eliminates these elements from the representation. (This includes, but is not limited to, the 'uninterpretable features' of Chomsky 2000, 2001.) Instead, an adequate model of phi-agreement requires recourse to operations as primitives – operations whose invocation is obligatory, but whose successful culmination is not enforced by the grammar.

    I also discuss the implications of this conclusion for the analysis of dative intervention. This leads to a novel view of how case assignment interacts with phi-agreement, and furnishes an argument that both phi-agreement and so-called "morphological case" must be computed within the syntactic component proper.

    Finally, I survey other domains where the empirical state of affairs proves well-suited for the same operations-based logic: Object Shift, the Definiteness Effect, and long-distance wh-movement.

    This work is based on data from the K'ichean branch of Mayan (primarily from Kaqchikel), as well as from Basque, Icelandic, French, and Zulu.
  • The role of case in A-bar extraction asymmetries: evidence from Mayan (with Jessica Coon & Pedro Mateo Pedro)

    	Author = {Coon, Jessica and Mateo Pedro, Pedro and Preminger, Omer},
    	Doi = {10.1075/lv.14.2.01coo},
    	Journal = {Linguistic Variation},
    	Number = {2},
    	Pages = {179--242},
    	Title = {The Role of Case in \mbox{A-bar} Extraction Asymmetries: Evidence from {M}ayan},
    	Volume = {14},
    	Year = {2014}}
    Many morphologically ergative languages display asymmetries in the extraction of core arguments: while ABS arguments (transitive objects and intransitive subjects) extract freely, ERG arguments (transitive subjects) cannot. This has been labeled "syntactic ergativity" (see, e.g., Dixon 1972, 1994; Manning 1996). These extraction asymmetries are found in many languages of the Mayan family, where in order to extract transitive subjects (for relativization, focus, or interrogative-formation), a particular construction known as Agent Focus (AF) must be used. Crucially, other Mayan languages – though still morphologically ergative – exhibit no such extraction restrictions.

    In this paper, we offer a proposal for (i) why some morphologically ergative languages exhibit extraction asymmetries, while others do not; and (ii) how the AF construction in Q'anjob'al (Mayan) circumvents this problem. We adopt recent accounts which argue that ergative languages vary in the locus of ABS case assignment (Aldridge 2004, 2008; Legate 2002, 2008), and propose that the same variation is attested within the Mayan family itself.

    Based primarily on comparative data from Q'anjob'al and Chol, we argue that the inability to extract ERG arguments does not reflect a problem with properties of the ERG subject itself, but rather reflects locality properties of ABS case assignment in the clause. We show how the AF morpheme -on circumvents this problem in Q'anjob'al by assigning case to the internal argument.
  • The absence of an implicit object in unergatives: new and old evidence from Basque

    	Author = {Preminger, Omer},
    	Doi = {10.1016/j.lingua.2011.04.007},
    	Journal = {Lingua},
    	Number = {3},
    	Pages = {278--288},
    	Title = {The absence of an implicit object in unergatives: New and old evidence from {B}asque},
    	Volume = {122},
    	Year = {2012}}
    Basque unergatives have long been held as evidence that unergative verbs have implicit objects (Bobaljik 1993, Hale & Keyser 1993, Laka 1993, Levin 1983, Ortiz de Urbina 1989, Uribe-Etxebarria 1989). Recently, it has been shown that the presence of absolutive agreement-morphology in Basque is not a reliable indicator of a successful agreement relation with a nominal target (Preminger 2009, LI). Building on this, I present two new arguments (and one old one) that Basque unergatives *lack* an implicit object.

    Since the subject of these verbs is nonetheless ergative-marked, these facts furnish an argument against a case-competition account of ergative case in Basque (i.e., against ergative being a dependent case; Marantz 1991). At first glance, this seems to favor an account of ergative as inherent case. However, previous work on Basque provides evidence against such an account: (i) raising-to-ergative constructions (Artiagoitia 2001), and (ii) the existence of ergative-marked arguments that are unambiguously Themes (Etxepare 2003, Holguín 2007, among others).

    These facts thus point to the need for a new theory of ergative case that is compatible (at the very least) with:
    1. the existence of ergative noun-phrases without a case-competitor
    2. the assignment of ergative case in non-thematic positions
    3. a lexically-determined distinction between unergatives and unaccusatives

    I conclude by discussing what such a theory of ergative case might look like.
  • Asymmetries between person and number in syntax: a commentary on Baker’s SCOPA

    	Author = {Preminger, Omer},
    	Doi = {10.1007/s11049-011-9155-z},
    	Journal = {Natural Language \& Linguistic Theory},
    	Number = {4},
    	Pages = {917--937},
    	Title = {Asymmetries between person and number in syntax: A commentary on {B}aker's {SCOPA}},
    	Volume = {29},
    	Year = {2011}}
    This paper is a commentary on Baker's "When Agreement is for Number and Gender but not Person." In many contexts, the behavior of person agreement departs from that of number and/or gender agreement; the central hypothesis advanced by Baker (2011) – the Structural Condition on Person Agreement (or SCOPA) – is an attempt to derive these departures from a single, structural condition on the application of person agreement.

    In this commentary, I explore Basque data which demonstrates that SCOPA is overly restrictive, as well as a handful of other empirical patterns that SCOPA fails to address but which I believe should be treated as part of the same empirical landscape.

    I propose an alternative to SCOPA, one which is able to handle these additional patterns, and proceed to show how it can be derived from the assumption that person and number phi-probes are situated in consecutive but separate heads along the clausal spine (Preminger 2009, 2011, building on Anagnostopoulou 2003, Bejar & Rezac 2003, Shlonsky 1989, Sigurdsson & Holmberg 2008, Taraldsen 1995, and others).
  • Transitivity in Chol: a new argument for the Split VP Hypothesis (with Jessica Coon)

    	Address = {Amherst, MA},
    	Author = {Coon, Jessica and Preminger, Omer},
    	Booktitle = {Proceedings of the 41st meeting of the {N}orth {E}ast {L}inguistic {S}ociety ({NELS}~41)},
    	Editor = {Fainleib, Lena and LaCara, Nick and Park, Yangsook},
    	Publisher = {GLSA},
    	Title = {Transitivity in {C}hol: A New Argument for the \mbox{Little-\textit{v}} {H}ypothesis},
    	Year = {2011}}
    In this paper, we provide a new argument in favor of the Split-VP Hypothesis (Bowers 1993, Chomsky 1995, Collins 1997, Kratzer 1996, inter alia), also known as the Little-v Hypothesis -- the idea that external arguments are base-generated outside the syntactic projection of the stem (i.e., outside of VP proper). More specifically, the hypothesis is that external arguments are base-generated in the specifier of a projection which:
    1. endows the stem with its categorial status as verb
    2. assigns structural Case to the complement of V
    3. assigns the external theta-role to the subject

    While our argument shares some similarities with the one put forth by Kratzer (1996), the data we examine here establishes more directly that these three properties are intrinsically interrelated.
  • Nested interrogatives and the locus of [wh]

    • download draft of paper; published in 2010, in The Complementizer Phase: subjects and operators, ed. E. Phoevos Panagiotidis, 200-235. Oxford: Oxford University Press.
    	Address = {Oxford},
    	Author = {Preminger, Omer},
    	Booktitle = {The {C}omplementizer {P}hase: Subjects and Operators},
    	Editor = {Panagiotidis, E. Phoevos},
    	Pages = {200--235},
    	Publisher = {Oxford University Press},
    	Title = {Nested Interrogatives and the locus of \emph{wh}},
    	Year = {2010}}
    Most of the literature on utterances with multiple-wh constructions is concerned with examples like (1), where more than one wh-phrase is involved in a single interrogative structure:

    (1) English:

    There is, however, a second way in which multiple wh-elements can interact – namely, when one question is embedded within another question:

    (2) Hebrew:
    ∼’What is the thing x such that Dan forgot who ate x?'(lit.: ‘What did Dan forget who ate?’)

    Unlike (1), which is a questions about pairs (namely, pairs of <eater, eatee>), (2) is a question about singletons (only an eatee). The question about an eater is embedded within the larger question, and interrogation regarding the eater is not directly part of the matrix speech-act in (2). I refer to the kind of constructions exemplified by (2) as nested interrogatives.

    I use the behavior of nested interrogatives in Hebrew, with respect to phenomena such as superiority and syntactic islandhood, to argue that overt displaced wh-elements in Hebrew occupy a position *lower* than the overt complementizer (the C head), whereas intermediate wh-movement crucially involves a position *higher* than the complementizer. As a result, Hebrew appears at first glance to massively violate the "wh-Island Condition." More subtle investigation, however, reveals that this condition – conceived of as a ban against two (or more) elements passing through the position above the complementizer in the same clause – is still operative, even in Hebrew, and a lower final landing site for wh-elements simply makes its effects harder (but crucially, not impossible) to detect.
  • Breaking agreements: distinguishing agreement and clitic doubling by their failures

    	Author = {Preminger, Omer},
    	Doi = {10.1162/ling.2009.40.4.619},
    	Journal = {Linguistic Inquiry},
    	Number = {4},
    	Pages = {619--666},
    	Title = {Breaking agreements: distinguishing agreement and clitic doubling by their failures},
    	Volume = {40},
    	Year = {2009}}
    In this paper, I propose a novel way to distinguish between "pure" agreement and clitic doubling. The innovation lies in examining what happens when the relation in question fails to obtain:

    Given a scenario where the relation R between an agreement-morpheme M and the corresponding full noun phrase F is broken – but the result is still a grammatical utterance – the proposed diagnostic supplies a conclusion about R as follows:
    - if M shows up with default phi-features (rather than the features of F) → R is agreement
    - if M disappears entirely → R is clitic doubling

    The workings of the proposed diagnostic are demonstrated using a family of constructions in "substandard" Basque (Etxepare 2006). Besides supporting the proposed diagnostic, the analysis of Basque provides a new perspective on the typological status of the Basque agreement system, as well as evidence against the traditional analysis of unergatives in Basque as being underlyingly transitive.