The precedence component to intervention effects: Evidence from English passives
This talk substantiates and provides an account of the following novel generalization:
(1) When just one constituent can undergo passive A-movement in English, it is invariably the leftmost accessible passivizable constituent within vP.
I argue that (1) indicates that English passive A-movement exhibits precedence-based intervention. Precedence-based intervention is wholly unexpected under traditional analyses of intervention based solely on structural prominence determined via asymmetric c-command (e.g. the Minimal Link Condition (Chomsky 1995) or Attract Closest (Pesetsky 2000)). Instead, I argue that precedence-based intervention is derivable from procedural aspects of the search algorithm initiated by internal Merge--triggering probes. Specifically, sisters within the search domain of the probe are linearly ordered for evaluation and probing delves deeply into internally complex left sisters prior to considering right sisters (see Branan and Erlewine 2022; Chow 2022). A major conclusion is that linear precedence relations must be visible during the syntactic computation (see also Fukui 2001; Kayne 2011, 2022; Zwart 2011; Bruening 2014, 2018). Alternative conceivable analyses of the attested left-to-right intervention effects which rely on obligatorily right-branching (i.e. descending) verb phrases are shown to be empirically inadequate based on evidence from constituency tests (namely, do so--anaphora and VP-ellipsis) and the anti-c-command condition on parasitic gap licensing. Indeed, parasitic gap licensing reveals the need for a non-c-command--based account of binding and licensing asymmetries within vP, building on Bruening (2014), a.o. I conclude by considering the role that precedence might play elsewhere in syntax, for instance in intervention with Ā-movement.
Moving away from antilocality: A defense of very local movement
Most syntactic research standardly assumes that syntactic dependencies are subject to locality constraints: agreement and movement cannot cross certain elements (bounding nodes, phase boundaries, elements which bear the same features, etc.). A growing body of work also argues that there are antilocality constraints which impose a lower bound on syntactic dependencies – movement, and perhaps Agree, must cross a certain type of boundary. Focusing primarily on Spec-to-Spec Antilocality (Erlewine 2016 et seq.), which states that movement from Spec,XP “must cross a maximal projection other than XP”, we argue that such constraints are unlikely to be part of Universal Grammar. The argument is three-fold. First, we briefly discuss the trajectory of antilocality theories and argue that they were originally proposed as responses to theoretical idiosyncrasies which don’t extend beyond the frameworks they are couched in. Second, we focus on one of the most discussed empirical motivations for antilocality theories – constraints on subject extraction – and suggest that these data can be analyzed in other ways, while antilocality approaches fall short in explaining the patterns. And finally, we present a case study of possessor relativization in West Circassian which demonstrates that Spec-to-Spec Antilocality makes the wrong empirical prediction: very local movement exists.
Towards a theory of covert pied-piping and chain pronunciation
In this talk, I will discuss some preliminary results of an ongoing research project on discontinuous DPs, using data from English and Russian. I will show that while some instances of discontinuous DPs seem to freely violate the Subject Condition under any of its formulations, while others never do. I will try to argue that those types of discontinuous DPs that freely violate the Subject Condition do not involve genuine subextraction, but rather involve what has been called in the literature Distributed Deletion or Covert Pied-Piping (see Siewerska 1984, Sekerina 1997, Fanselow and Ćavar 2002, Pereltsvaig 2008, Goncharov 2015, Murphy 2021 among others). This will be done in three steps: (1) I will present arguments in favor of Covert Pied-Piping being an available parse for certain discontinuous DPs in English and Russian in principle; (2) I will argue that this parse is possible if and only if the first part of a discontinuous DP is an element that can otherwise occupy a left-peripheral position within the DP; furthermore, material that is base-generated in the left periphery of the DP (Spec,DP) cannot be genuinely extracted at all and only allows for a parse with Covert Pied-Piping; (3) I will argue that the Subject Condition can be freely violated in all and only those constructions with discontinuous DPs where the first part of the DP is an element that can otherwise occupy a left-peripheral position within the DP. In other words, (a) overt splitting of an element that has been base-generated in the Spec,DP from the rest of the DP can only involve covert pied-piping (no subextraction); (b) overt splitting of an element that optionally occupies a left-peripheral position within the DP from the rest of the DP is always ambiguous between a covert pied-piping parse and genuine subextraction parse; (c) overt splitting of an element that can never occupy a left-peripheral position within the DP from the rest of the DP always involves genuine subextraction.
With these tentative generalizations in mind, I will propose a hopefully restrictive theory of Covert Pied-Piping that builds on Cable's (2010) general theory of pied-piping, Fox and Pesetsky's (2005) Cyclic Linearization theory of phases and Richards' (2017) Contiguity Theory as the basic mechanism that determines pronunciation of chains.
Cross Clausal Dependencies in Turkish: Argument - Adjunct Asymmetries
Cross clausal dependencies have been noted to be limited to a subset of configurations where an embedded clause functions as a complement to an embedding predicate. In such configurations, a long distance syntactic dependency between a syntactic object inside the embedded clause and another syntactic object in the matrix clause leads to a grammatical result. In contrast, configurations where a clause is merged as a specifier or an adjunct famously block cross clausal syntactic dependencies. This asymmetry between complements and non-complements have been formulated as the Condition on Extraction Domains (CED) (Huang 1982; Chomsky 1986; Cinque 1990; Manzini 1992). CED is generally considered as an indication for locality of syntactic operations and several major theories of locality have been proposed in the literature: 1) absolute locality domains (Barriers, Phases) (Chomsky 1986, 2000, 2001), 2) relative locality domains (Rizzi 1990), and 3) path based locality domains (Pesetsky, 1982; McFadden & Sundaresan 2018; Newman & Branan).
In this talk, I present data showcasing the behavior of -DIK clauses which embed clauses through nominalization as a 1) complement, 2) adjunct, and 3) relative clause. Testing -DIK clauses across 1) preposing, 2) postposing, 3) scope-shifting overt QR, and 4) relativization, we observe the unique picture in (1).
(1) Cross Clausal Dependencies in Turkish DIK Clauses
(a) Complement clauses with Genitive subjects are completely transparent.
(i) Subject extraction shows A properties.
(ii) Object extraction shows A-bar properties.
(b) Adjunct clauses with Nominative subjects are completely opaque. (c) Adjunct clauses with Genitive subjects are partially transparent.
(i) Subjects are available for cross clausal dependencies.
(ii) Objects are unavailable for cross clausal dependencies.
A detailed investigation of the morphological properties of these three clause types reveals that in general case marked arguments (genitive and accusative) are available for cross clausal dependencies whereas unmarked (nominative) arguments are not. Adopting Öztürk and Taylan’s (2016) account of genitives, we show that only arguments that raise to Spec, DP (above the nominalized TP/CP) become available for cross clausal dependencies by virtue of being at the Phase edge. We implement our analysis through a Phase based account (Chomsky 2000, 2001, et seq.) while acknowledging that path-based accounts could also derive the same facts.
When wh-phrases are their own interveners
Much work on syntactic locality has shown that processes like wh-movement are subject to several kinds of locality restrictions. In addition to being sensitive to intervening wh-phrases, wh-movement must proceed successive cyclically through various points in the clause, and in some cases/languages, may not cross intervening arguments (see e.g. Branan and Erlewine (2022) for a recent overview). Sensitivity to intervening arguments is known to be quite fine-grained: according to insights from Keenan & Comrie (1977) and others, languages might differ with respect to what kinds of arguments count as interveners for a wh-element, and might also treat arguments vs. adjuncts differently.
In this talk, I propose that all of these locality restrictions and their various levels of granularity are interconnected. More specifically, I suggest that they reduce to a particular view of how selection influences the projection of category information from daughter nodes to their mothers (following Zeijlstra 2020). I show that by examining the nature of selection and projection, we can leverage the architecture of grammar to predict the requirement for wh-movement to be successive-cyclic: the projection rule makes it so that wh-phrases create their own barriers for extraction if their wh-features get too high, meaning they have to move outside the scope of their own features in order to extract. The theory entails that movement must be successive cyclic, but does not say through which positions. By varying the different allowed parameters in this theory, I show that it also captures variable sensitivity to the Keenan & Comrie hierarchy. Thus, the various locality requirements governing wh-movement can be reduced to basic principles governing selection and projection.
When to revisit? Investigating (un)ambiguity in temporal clauses
Ever since the seminal work of Geis in the 1970s, it has been known that temporal adverbial clauses in English (and a number of other languages) show the same kind of ambiguities as also observed in wh-questions and (even more relevantly) relative clauses, so that a sentence like “She arrived just when I predicted she would arrive” can mean either that her arrival coincided with my act of predicting, or with the time given in the prediction. This follows fairly directly from an analysis of such temporal adverbial clauses as a type of free relative clause denoting a definite description of a time, which can be formed syntactically by A’-movement of a temporal operator, or whatever analysis is given to such long-distance dependencies. Here we’d like to report on joint work in progress (Caroline Heycock, Elise Newman, Rob Truswell) where we investigate cases of temporal clauses where long-distance movement seems to be either excluded or at the least heavily disfavoured, and explore the possibility that the data motivate a distinction between temporal clauses that are descriptions of time intervals and those that are descriptions of events.
Situational Opacity
This talk is centered around a curious set of facts involving extraction: certain domains are transparent for extraction, of both the A and Ā varieties, just in cases where some semantic variable --- such as an individual (Grano & Lasnik 2018), event (Truswell 2007 et seq.), time (Hiraiwa 2005, Huang 2019) or situation variable --- stands in some sort of semantic relationship with a comparable variable in the domain containing the landing site for extraction. I will review a number of cases from a variety of languages to try and convince you that this generalization does indeed hold. Having convinced you that it does, I'll then note that this presents a puzzle for the commonly adopted Y-model of the grammar: a semantic dependency here seems to condition the availability of a syntactic one, yet the syntactic calculation must, on this model, proceed blindly with respect to the semantics.
I will then discuss options that we might consider for resolving this puzzle. I will focus on particular on what I view, at the moment, as a particularly promising candidate: that the apparently problematic relationship between a syntactic and semantic dependencies is a perfectly respectable one between two semantic dependencies. The working idea is that there is a general semantic condition on the interpretation of movement chains that requires them to be evaluated with respect to the same context (see Gluckman 2018 for a similar idea). In the cases where movement is not allowed, the absence of a semantic relationship between variables in the two domains results in the insertion of an operator at LF at the edge of the relevant domain which: a) assigns a value to situation variables contained within the domain, and b) requires this value to be distinct from the value assigned to situation variables in other domains. One consequence of this is that the head and tail of a movement chain spanning such a domain will necessarily be evaluated with respect to different contexts, violating the aforementioned condition on the interpretation of movement chains. In contrast, the presence of a prior semantic relation spanning the relevant domain will block insertion of the relevant operator --- perhaps for reasons of economy --- and as a consequence extraction from the domain will be licit, as both tail and head of the movement chain will be evaluated with respect to the same context.
Successive cyclicity and DP intervention
The well-known requirement that movement must proceed successive-cyclically through intermediate landing sites is standardly attributed to the presence of locality domains (phases) along the extraction path. Correspondingly, the existence of clause-medial intermediate landing sites is commonly taken as evidence for the existence of a clause-medial phase. We argue that at least some instances of successive cyclicity through clause-medial positions are better understood as the result of intervention by the external-argument DP, not phasehood. Building on recent proposals about the principles that govern the behavior of complex probes, we propose that C in these cases can only attract the structurally closest DP. Elements separated from C by an intervening DP must first “leapfrog” around the intervening DP. In languages where such leapfrogging is impossible, a local-subject-only extraction restriction arises; in language where such leapfrogging is possible, extracting elements across the local subject is possible but must proceed through a clause-medial intermediate position, resulting in successive cyclicity.
Head movement and linear edges
We present some of our work on an algorithm for linearization which is considerably more relaxed than that commonly found in models of the relation linear order to phrase structure, most prominently Kayne (1994). We discuss how this approach to linearization allows us to understand a variety of effects commonly attributed to head movement. The core idea in deriving these effects is that "head movement" involves nothing more than the linearization of a head at the left edge of its corresponding phrase. We further present two approaches to disharmonic head orders we are considering. Time permitting, we may discuss how the system allows us to understand certain differences between scrambling in Japanese --- which may have semantic effects --- and scrambling in Tagalog, which does not.
Deriving syntactic assimilation vs. dissimilation in the local phase
The formal operation of Agree in Minimalism (Chomsky, 2001) is hardwired to always and only yield outputs that are assimilatory in nature. The existence of phenomena which are both syntactic and dissimilatory (e.g. case, cf. Preminger, 2014; Levin, 2015; Yuan, 2021) and, I argue, certain cases of local anaphora (Lidz, 2001; Reuland, 2011) thus poses a challenge. To accommodate these facts, I develop a revised model of Agree (renamed RELATE), the core intuition underlying which is the observation of a “Category Process Correlation (CPC)”, namely that syntactic dissimilation overwhelmingly obtains between two non-distinct categories while syntactic assimilation obtains between two distinct ones. I show that, under RELATE, we can not only derive cases of syntactic assimilation (e.g. φ-agreement) and dissimilation (e.g. case and local anaphora) under a single operation, but can also correctly predict why cases of syntactic dissimilation are exceptional and pertain primarily to nominals. The latter follows directly from the way in which the CPC interacts with the distribution of individual categories across phases. I conclude by showing that the current model also makes a number of empirical predictions which seem, by and large, to be fulfilled.
Phases, modularity and reuse of structures
In this talk I'll discuss a reimplementation of phase theory called Phase Stitching, which is based on taking the idea seriously that phase-based derivations look like modular design. Phases, with their opaque domains, are modules, and their escape-hatch edges are the transparent interfaces linking the modules together. Phase Stitching implements this by having phases built independent of each other, only to be Stitched together via unification of their matching edges. I'll explore in particular one set of consequences of this approach, which is that the construction of individual phasal modules and the Stitching into larger structures need not be thought of as parts of unified derivations. The phasal modules are constructed independent of each other, can be combined in different orders and reused, and could even in principle be stored, opening up the possibility of dealing with certain `lexical' effects in a system that is still syntax all the way down.