A Formal Foundation for A and A-bar Movement

It seems a fact that movement dependencies come in two flavours: ”A” and ”A-bar”. Over the years, a number of apparently independent properties have been shown to cluster together around this distinction. However, the basic structural property relating these two kinds of movement, the ban on improper movement (‘once you go bar, you never go back’), has never been given a satisfactory explanation. Here, I propose a timing-based account of the A/A-bar distinction, which derives the ban on improper movement, and allows for a simple and elegant account of some of their differences. In this account, ”A” dependencies are those which are entered into before an expression is first merged into a structure, and ”A-bar” dependencies are those an expression enters into after having been merged. The resulting system is mildly context-sensitive, providing therefore a restrictive account of possible human grammars, while remaining expressive enough to be able to describe the kinds of dependencies which are thought to be manifest. It’s common to describe the syntax of natural language in terms of expressions being related to multiple others, or moved from one position to another. Since Ross (1967), much effort has been put into determining the limitations on possible movements. A descriptively important step was taken by classifying movement dependencies into two basic kinds: those formed by the rule move NP, and those formed by move wh-phrase (Chomsky, 1977). This bipartition of movement dependencies is a formal rendering of the observation that wh-movement, topicalization, and comparative constructions seem to have something in common, that they do not share with passive and raising constructions, which in turn have their own particular similarities. Whereas syntactic theories such as Head-Driven Phrase Structure Grammar (Pollard and Sag, 1994) and Lexical-Functional Grammar (Bresnan, 2001) have continued UCLA Working Papers in Linguistics Mathematics of Language 10 to cash out this intuitive distinction between dependency types formally, in terms of a distinction between lexical operations and properly syntactic operations, this distinction has no formal counterpart in theories under the minimalist rubric (Chomsky, 1995). This theoretical lacuna has led some (Sportiche, 2005) to explore the hypothesis that this perceived distinction between movement types is not an actual one; i.e. that the differences between Whconstruction types on the one hand and passive construction types on the other are not due to differences in the kinds of dependencies involved. A problem besetting those minimalists eager to maintain the traditional perspective on the difference between whand NPmovement dependencies, is that there is no principled distinction between long-distance dependency types available in the theory; a theory with one kind of long-distance dependency does not lie well on the procrustean bed of one with two. The contribution of this paper is to provide a non-ad hoc minimalist theory with two kinds of movement dependencies, which have just the kind of properties which have become standardly associated with move NPand move wh-phrase-related phenomenon, respectively. It is important to note that it is not a particular analysis of a particular language which I will claim has these properties, but rather the theoretical framework itself. Once we are in posession of a theoretical framework in which we have two movement dependency forming operations that interact in the appropriate way, we are in a position to determine whether the old intuitions about movement dependencies coming in two types were right; we can compare the relative elegance of analyses written in one framework to those written in the other. In §1 I describe the kinds of properties which are constitutive of the empirical basis for the bifurcation of movement into two types. Recent minimalist accounts of some of these properties (Lasnik, 1999; Boeckx, 2000) will form the conceptual background of my own proposal, developed in §2. The formal architecture of minimalist grammars (Stabler, 1997) is presented in §2.1, and it is extended in §2.2 in accord with my proposal. In §2.3, I present an analysis of passivization in English (drawing on the smuggling account proposed in Collins (2005)) written within the framework of §2.2. A and A-bar Movement 3 1 On A and A-bar movements Many differences between NP and wh-phrase movement have been suggested in the literature, such as whether they license parasitic gaps, whether they can move out of tensed clauses, whether they bar the application of certain morpho-phonological processes, and whether they incur crossover violations (see e.g. Mahajan (1990)). (These questions uniformly receive a negative answer with respect to NP movements, and a positive one with respect to wh-phrase movements.) A perusal of these properties makes clear that they are highly constructionand analysis-specific. In other words, a theoretical framework cannot derive these differences between NP and wh-phrase movement simpliciter, but may at most derive them relative to particular analyses of these constructions. The only analysis independent property of NP and wh-phrase movement types is the so-called ‘ban on improper movement’, which states that NP movement of an expression may not follow its movement as a wh-phrase. This relational property of NP (henceforth: ‘A’) and wh-phrase (henceforth: ‘A-bar’) movements is widely accepted, and was motivated by the desire to rule out sentences such as 1 below. (1) *[S John seems [S t [S t wanted to sleep ] ] ] In 1, the first movement (to SPEC-S) is an A-bar movement, and the second (to the matrix clause subject position) an A movement. The unacceptability of 1 contrasts with the well-formed 2 below, which one can interpret as suggesting that it is the second movement in 1, from SPEC-S to the subject position in the matrix clause, which leads to the deviance of 1. (2) [S Who does [S Mary believe [S t [S t wanted to sleep ] ] ] ] In the government and binding (GB) framework (as described in Chomsky (1981)) and the minimalist program (MP) (as in Chomsky (1995)), the ban on improper movement must simply be stated as such; movement from an A position may target an A-bar position, but movement from an A-bar position may only target other A-bar postions (see Müller and Sternefeld (1993) for a particularly articulated view). In LFG and HPSG, where A movements are taken to be resolved lexically (by means of redundancy rules and the like) and A-bar movements resolved grammatically, the ban on improper movement follows from the architecture (grammatically UCLA Working Papers in Linguistics Mathematics of Language 10 complex expressions are simply not the kinds of things that lexical processes apply to). In the grammatical architecture I will develop in §2, A movements are those which occur before, and A-bar movements those which occur after, an expression has been first merged into a structure. The ban on improper movement is then just a simple consequence of the structure of derivations. Strictly speaking, the ban on improper movement is the only property of movement types which a grammatical framework can be said to derive. However, the following more analysis-specific property of movement types listed above will be shown to follow naturally from the architecture of the system in §2. A and A-bar movements differ systematically as to whether they create new binding possibilities. Consider sentences 3 and 4 below. In 3, the reflexive pronoun himself cannot be bound by the quantified noun phrase every boy, whereas in 4, after movement, it can. (3) *It seems to himself that every boy is wonderful. (4) Every boy seems to himself to be wonderful. This situation contrasts with the one laid out in 5 below, where we see that a wh-moved expression is not able to bind the reflexive pronoun. Sentence 6 shows that it is indeed the failed attempt at binding that results in the ungrammaticality of 5, as this movement is otherwise fine. (5) *Which boy does it seem to himself that Mary loves? (6) Which boy does it seem that Mary loves? The difference between these movement types can be summed up in the following diagram, with A movement of XP being able to, while A-bar movement of XP being unable to, bind the pronoun pro: XP . . . [. . . pro . . . t . . .] This is called ‘crossover’ in the literature (Postal, 1971). Strong crossover is when the bound expression c-commands the source position of the movement, and weak crossover is when the bound expression is properly contained in such a c-commanding phrase. Weak crossover violations have been argued to be ameliorable under certain conditions (Lasnik and Stowell, 1991). A and A-bar Movement 5 Attempts to account for these phenomena have been numerous in the GB framework, and have continued into the MP (see Ruys (2000) for an accessible typology). One option is to rule out rebinding by A-bar movements by denying the ability to bind pronouns from A-bar positions, and another is to require that no closer potential binders may intervene between an A-bar trace and its antecedent. Given the framework developed below, we are in a position to stipulate that an expression may bind only those expressions that it c-commands when first merged (in other words, that binding is determined by c-command in the derivation tree). 2 Trace deletion, derivationally Without a formal apparatus to hang an account of differences between movement on, researchers in the GB tradition have attempted to capture the difference between A and A-bar movements in terms of properties of source positions: traces. It was discovered that under a certain network of assumptions, A-bar traces behaved as R-expressions, and A traces as anaphors. In the MP, it has been suggested for diverse reasons that A-bar traces should be treated formally as copies of the moved expression, while A traces should be treated formally as unstructured objects (Fox, 1999; Lasnik, 1999). This is the idea upon which this paper builds. But currently there is nothing more than arbitrary stipulation (why are some traces copies, and others not? why does the ban on improper movement hold?). To excavate the idea, we should get clear on what, exactly, traces are (for). In mainstreamminimalism, movement chains are licensed derivationally: only well-formed chains are built in the first place. Therefore, traces are not needed for evaluating well-formedness of a syntactic representation (their role in government and binding theory). Instead, traces (qua copies) play a role primarily at the interfaces, in particular the syntax-semantics interface, where they determine the positions in which an expression may take scope, as per Chomsky (1995). The distinction between structured and unstructured traces (i.e. copies versus ‘traditional’ traces) is intended to indicate the possibility or not of reconstruction (with expressions being reconstructible into structured trace positions, but not into unstructured trace positions). A ‘copy’ indicates that an expression is present at a particular UCLA Working Papers in Linguistics Mathematics of Language 10 location in the structure for the purposes of reconstruction, while (unstructured) traces indicate that it is not. The intuition is simply that an expression may be interpreted in any position in which it is present; it is present in its A-bar positions, but not (necessarily) in its A positions. This is easier to understand if we think not about derived structures, but about the derivation itself: talk of ‘copies’ versus ‘traces’ is recast in terms of whether (copies) or not (traces) the object which is entering into these various dependencies is already present in the derivation at the time the dependency in question is entered into. The basic formal idea behind this intuition is to incorporate both transformations, as well as ‘slash-feature percolation’ (Gazdar, 1981) into a single formalism. Then we may have derivations involving slash-features, in which the object entering into the dependency in question is not present at the time the dependency is established: 1. [V write ] 2. [V P/DP was written ] In addition, we may have derivations using transformations, in which the object entering into the dependency in question is present at the time the dependency is established: 1. [S′ that [S book was written ] ] 2. [N book [S′ that [S t was written ] ] ] The present derivational reconstruction of the representational traces versus copies account of the A/A-bar distinction has the distinct advantage of giving a unified and intuitive account of various properties of A and A-bar movement. In particular, the ban on improper movement is forced upon us in this timing-based perspective on long-distance dependency satisfaction. In the next section I show how to incarnate this derivational perspective on A and A-bar movement in a formal system. In so doing we gain a better understanding not only of the mechanisms involved, but also of the various analytical options which the mechanisms put at our disposal. A and A-bar Movement 7 2.1 Minimalist Grammars Minimalist grammars (Stabler, 1997) provide a formal framework within which the ideas of researchers working within the minimalist program can be rigorously explored. A minimalist grammar is given by a four-tuple 〈V,Cat, Lex,F〉, where • V , the alphabet, is a finite set • Cat, the set of features, is the union of the following pair of disjoint sets: – sel × Bool, where for ∗ 〈x, 0〉 ∈ sel×Bool, we write =x, and call it a selector feature ∗ 〈x, 1〉 ∈ sel×Bool, we write x, and call it a selectee feature – lic× Bool, where for ∗ 〈y, 0〉 ∈ lic×Bool, we write +y, and call it a licensor feature ∗ 〈y, 1〉 ∈ lic×Bool, we write -y, and call it a licensee feature • Lex, the lexicon, is a finite set of pairs 〈v, δ〉, for v ∈ V ∪{ǫ}, and δ ∈ Cat • F = {merge,move} is the set of structure building operations Minimalist expressions are traditionally given in terms of leaflabelled, doubly ordered (projection and precedence) binary trees. The leaves are labelled with pairs of alphabet symbols (V ∪ {ǫ}) and feature sequences (Cat). A typical expression is given in figure 1, where the precedence relation is indicated with the leftright order, and the projection relation is indicated with less-than (<) and greater-than (>) signs. The projection relation allows for the definition of the important concepts ‘head-of’ and ‘maximal projection’. Intuitively, one arrives at the leaf which is the head of a complex expression by always descending into the daughter which is least according to the projection relation. In the tree in figure 1, its head is 〈will, +ks〉, which is also (trivially) the head of its root’s left daughter. The UCLA Working Papers in Linguistics Mathematics of Language 10