Ian Roberts's monograph is an ambitious attempt to provide a comprehensive account of a variety of phenomena that have been hypothesized to involve movement and subsequent adjunction of a head to a higher head (i.e. head movement), ranging from such traditional instances of head movement as V(erb)-to-T(ense) and T-to-C(omplementizer), to clitics in Romance and Slavic languages, Long Verb Movement in Breton, noun incorporation, and predicate clefts. Though data are mostly drawn from Indo-European languages, the variation exhibited by verbs and clitics in Romance provides a suitable challenge for the author; also, the resulting compilation of data from various sources makes this book a valuable reference tool. Regrettably, the main proposal runs into multiple problems, precluding its success.
After describing the goals and the limits of the book in the ‘Introduction’, Roberts turns to semantic effects of head movement in Chapter 1, ‘LF-effects of head movement’. The purpose of the chapter is to argue for the syntactic status of head movement: if head movement can license negative polarity items and affect quantifier scope, it cannot be a purely phonological operation.
If Agree is a head–head relation, the question arises why larger constituents that share all their features with heads (i.e. maximal projections) can move at all. Roberts addresses this question in Chapter 2, ‘Head movement and pied-piping’. Although the reasoning is somewhat obscured by the fact that the author unexpectedly switches from discussing Agree and Merge to treating movement as a single operation, it is clear that he relies on the definition in (1) to make possible both XP movement and head movement.
(1) A goal α is the closest one to a given probe if there is no distinct goal β such that for some X … , X c-commands α but does not c-command β. (Rackowski & Richards Reference Rackowski and Richards2005: 579)
As Roberts points out (35), in order to determine how this definition applies to head movement, it is necessary to decide if the minimal and the maximal projections of the same category are distinct. If they are not, as Roberts assumes, a head and its maximal projection are equidistant from the nearest probe, and either Xmin or Xmax can move (because either can be the closest goal for Agree).
The definition in (1) entails that if there is an XP in the specifier of the phrase headed by β (SpecβP) that bears appropriate features, that XP is a closer goal, and head movement of β will be blocked (37). Conversely, if the specifier is absent or not a suitable goal, either the head or the entire maximal projection can move. Roberts suggests that head movement becomes obligatory if the head βmin has some properties that the maximal projection βmax lacks, which is where the notion of a defective goal, i.e. a goal whose formal features are a subset of the features of the probe, comes in.
Chapter 3, ‘Cliticization’, presents the theoretical core of Roberts's approach to head movement, stated as the non-distinctness generalization given in (2), and then discusses its application to clitics. Crucially, under the assumption that (some) words are phases (Marantz Reference Marantz2001, Reference Marantz2006), the landing site of a clitic is at the edge of the minimal phase formed by its host, which permits Roberts to capture the elusive intuition that clitics are both part and not part of their host (42–43), and also permits their subsequent excorporation (Section 3.4).
(2) Incorporation can take place only where the label of the incorporee is nondistinct from that of the incorporation host. (57)
This generalization results from the reasonable assumption that a head Xmin (simple or complex) should be defined as a category dominating nothing but itself, see (3).
(3) The label L of category α is minimal iff α dominates no category β whose label is distinct from α's. (54)
In other words, a complex head can be formed if and only if the goal is defective with respect to the probe: the formal features of the former are a subset of those of the latter. Assuming that a clitic bears no Case or categorial features, and thus can function as a defective goal for v, Roberts argues that adopting the hypothesis in (2) yields an elegant account of cliticization as Agreement.
(4) Cliticization
(a) v [person:__, number:__][φperson:a, number:b]
before Agree
(b) v [person:a, number:b][φperson:a, number:b]
after Agree
Since Agree amounts to the copying of the values of the goal's features into the feature matrix of the probe, if the featural contents of the goal are a subset of those of the probe, we end up with two occurrences of the same set of features after Agree – precisely the situation arising after movement. As a result, Roberts contends, with a defective goal ‘Agree and Move/Internal Merge are formally indistinguishable’ (60). The configuration in (4b) leads to the formation of a chain (60–61), yielding, after chain reduction, the phonological effects of movement.
As is easy to see, Roberts's hypothesis is both Minimalist in spirit (stemming from an altogether natural definition of a minimal category) and elegant in execution (reducing the problematic phenomenon of head adjunction to the unproblematic Agree). In the remainder of this review I will try to evaluate to what extent it succeeds on both the theoretical side (by verifying whether agreement alone can yield the head adjunction structures that are viewed as the hallmark of head movement) and the empirical side (by evaluating how successfully the proposed mechanisms derive uncontroversial instances of head movement). In what follows I will argue that Roberts's assumptions are in part self-contradictory and that some of the computations used to derive complex heads seem to be irreducibly flawed. Since the main contribution of the book is a theoretical proposal and its highly detailed implementation, the ensuing discussion will of necessity be technical.
As noted above, Roberts assumes that a moved head may be interpreted in its derived position. Likewise, the comparison of Scandinavian object shift and clitic movement (44–50) makes it clear that from the author's standpoint both phenomena have a semantic effect, viz. the moved item becomes specific as a result of movement to the edge of a phase – the vP phase for object shift, and the v phase for cliticization. It is therefore essential that the head of the clitic chain be interpretable; however, under standard assumptions, valuation, as in (4b), does not render uninterpretable features interpretable.
From the technical point of view, although two occurrences of the same feature bundle are produced in (4b), it is not the case that the outcome of a movement operation (Agree+Internal Merge) in this configuration would give rise to the structure in (4b): irrespective of whether the moved φP adjoins to v* or occupies Specv*P, it would place yet a third bundle of the relevant features at the landing site, which, unlike the valued uninterpretable features on v*, must be interpretable.
A closer inspection of some tree representations provided by the author suggests that complex heads are, after all, created by movement rather than Agreement. Thus the structure in (5) contains the verbal root (Vmin), which clearly cannot have appeared there as a result of an Agree operation: there exist no features whose value could be set to yield a lexical root (and even if the phonology of a given terminal can be treated as a formal feature, under most assumptions it is valued and interpretable).
(5)
Likewise, pied-piping of a clitic together with the head that it has adjoined to (needed by Roberts for compound tenses (78, 89), causative constructions (96), clitic clusters (141–142), among other constructions), cannot be due to agreement, since feature valuation does not involve pied-piping. Similarly, if a complex head is a pre-syntactically constructed feature bundle, neither its internal hierarchy nor its spellout may depend on its external syntax, and syntactic operations such as Agree cannot take place inside it. Nonetheless, surface agreement between the past participle head Partmin and the object clitic forming part of vAspmin with which Partmin combines is claimed to result from a syntactic operation similar to spec-head agreement (89), and various clitic orderings are argued to be derived in syntax (Section 3.6.2).
If we assume that Roberts's incorporation is actually a movement operation, these problems disappear, but other, well-known problems resurface. Thus, the clitic in (5) does not c-command its trace, which requires the development of an alternative approach to the interpretation of movement (227, fn.19). Like in more standard approaches, it becomes unclear why a defective goal cannot land in a specifier rather than adjoin to a head; ascribing this impossibility to the EPP (Extended Projection Principle) (166) amounts to restating the problem. Conversely, the possibility of excorporation out of a complex head and the lack of the Head Movement Constraint are predicted – a conclusion argued for by the author.
Setting these issues aside, I will now address the complications caused by Roberts's hypothesis that the goal of head movement is necessarily defective. It is clear that there exist instances of head movement where the features of the probe do not stand in an obvious superset relation with those of the goal. Two strategies can be envisaged for dealing with such situations: (i) ‘enrichment’ of the probe and (ii) ‘impoverishment’ of the goal.
Assuming that the featural content of the probe is richer than what is apparent from its surface representation is the strategy adopted for second-position clitics in Slavic. Roberts assumes that Slavic pronominal clitics are D(eterminer)Ps (67) and, crucially, that C has D-features. As a result, Slavic clitics are defective with respect to C (but not v) and therefore cliticize to it. The limitations of this type of approach are clear: in the reductio ad absurdum example in (6), the raising of the French finite verb, its accompanying clitics and the negative marker to C should require C to bear the features of nearly every overt terminal in the clause and perhaps those of some covert ones.
(6) Ne m' en parlent- ils pas?neg isg -datloc speak-pres-3pl 3mpl-nom not ‘Don't they speak to me of it?’
The ‘impoverishment’ strategy is used with Romance clitics. Assuming that they are φPs rather than DPs and that v bears φ-features (56) yields an ideal environment for ‘incorporation’, as long as these φPs have no Case features or categorial features (57, 139–149). This incorrectly predicts that Romance clitics should be allowed in positions where accusative Case licensing is absent (for example, inside noun phrases) and also that multiple different clitics on a single head are impossible. To circumvent the latter issue, Roberts derives clitic clusters via movement of a complex head, as detailed below.
Instead of the usual assumption that the surface difference in form between direct and indirect object clitics reflects their Case, Roberts suggests (57) that it is due to the functional head they incorporate into (v or v appl), which would seem to entail that the formal makeup of the two v heads is different. Regrettably, this conclusion somewhat undermines the assumption that v is a defective goal with respect to v appl, essential for Roberts's account of clitic clusters.
- (7)
(a) Isabelle donne une lettre à Roger. Isabelle gives a letter to Roger ‘Isabelle gives Roger a letter.’
(b) Isabelle mele donne. Isabelle isg-dat 3sg-acc gives ‘Isabelle gives it to me.’
Roberts hypothesizes (137) that in the underlying representation the indirect object c-commands the direct object, as in (8a). To obtain the correct word order for the full noun phrases (NPs) in (7a), VP-movement to Specv applP is postulated (139). The initial steps necessary to derive (7b) are indicated in (8).
(8)
The complex v and the indirect object clitic in SpecvP are then attracted to v applmin in the order indicated below, deriving the surface word order; finally, the entire complex moves to v*:
(9)
The derivation in (9) is problematic. Under Rackowski & Richards's definition in (1), the dative clitic is closer to v applmin than v and should move first, yielding an incorrect word order. In fact, this is precisely how subject agreement happens (159): it is the specifier of the complement of T (the external argument) rather than the head of its complement (v*) that controls agreement on the finite verb. Suppose, however, that this scenario follows from Roberts's surmise that ‘movement of a more deeply embedded category must precede movement of a less deeply embedded one’ (51), also used to derive French third-person clitic clusters (142).
(10) β is more deeply embedded than α iff there is at least one more category γ in the phase containing α and β that c-commands β and does not c-command α. (51)
Under the definition in (10), the derived head in v is more deeply embedded than the indirect object and should move first, yielding the right word order (indirect object-direct object-V) in the clitic cluster in (7b). The question now arises why the valued φ-features of v do not enter into an agreement relation with those of v applmin, incidentally removing the trigger for dative clitic movement (φ-agreement). That such a scenario is possible is shown by Roberts's analysis of French past participle agreement (76–78): Partmin merges as the sister of v*P, but agrees with v* rather than with the external argument in Specv*P. I conclude that neither definition of locality yields the correct results; the complexity of the computations involved has kept me from verifying whether similar issues arise in other cases of cliticization to an agreeing head: restructuring (Section 3.4.2), causative verbs (Section 3.4.3), subject clitics (Section 3.5.1) and locative clitics (Section 3.6.3).
Can this contradiction be disentangled? While merging Partmin directly with VP would reduce French past participle agreement to subject agreement, the clitic cluster dilemma is much more complex, since any attempt to resolve it necessarily has repercussions elsewhere in the book, given the complexities of verb–clitic ordering, the excorporation analysis of compound tenses, the defective goal hypothesis and the interplay with XP movement.
I have provided the detailed explanation of this case in order to demonstrate the extreme complexity of computation involved, but the problem illustrated is not an isolated case. Space limitations prevent me from addressing in full other instances where I believe the technical description to be incomplete or erroneous, but the reader can evaluate for herself the following three representative examples.
(i) To deal with Slavic second-position clitics (section 3.3), Roberts assumes that unlike Romance clitics, which are φPs without a DP layer, ‘C-oriented’ clitics are DPs, and C in languages with second-position clitics has D-features in addition to φ-features. No independent evidence is provided for either hypothesis, but it is mentioned that lexical noun phrases, including proper names, contain an additional NP phase. It is unclear to me whether this entails, among other things, that clitic pronouns in Czech or Slovak are not nominal, while Romance clitics are not definite.
(ii) Obligatory clitic doubling with definite subjects in Montesover Trentino is handled by stipulating that although ‘the subject DP as a whole cannot be a defective goal, since it bears a Case feature; … the structural position of the Case feature in DP is not D’ (108–109), and therefore, ‘features of the D of the subject DP may be realized in the probe T as a consequence of the Agree relation between the two elements’ (109). The question arises why subjects are not doubled everywhere and why in languages with ‘C-oriented’ clitics, where C is endowed with the right feature specification ([uD, uφ]), arguments are not obligatorily doubled. The apparent accessibility of D also gives rise to some concerns, as does the predicted lack of case-marking on it.
(iii) The elegant account of the Person-Case Constraint (143–145), based on the assumption that third person is the unmarked value of the person feature, and that therefore, ‘the relation between two occurrences of [the person feature underspecified for value, i.e. the third person clitic,] can be construed as absence of Agree’ (144), leaves open the question why it can be so construed in French but not in Italian, and, what is more, only in a cluster of two third-person clitics, with total disregard of the number feature.
Turning to an illustrative case in Chapter 4, ‘Verb movement and incorporation’, V-to-T movement in Romance and lack thereof in English are attributed to the presence of an uninterpretable [uV] feature on T or its absence. In the latter case, affix hopping is claimed to amount to the valuation of [uT] features on V by [iT] on T, in some unspecified way made possible by the fact that T and V (or more precisely, v*) both have uninterpretable φ-features (valued by the subject and the object, respectively). Auxiliaries, on the other hand, having no argument structure, lack V-features but bear interpretable T features, which, Roberts claims, makes them incorporate into T – presumably presupposing an agreement relation between two interpretable features (162). Similarly unsatisfactory proposals are made regarding other cases of verb movement.
Finally, Chapter 5, ‘Head movement and the theory of movement’, is also not free from incorrect predictions and unmotivated stipulations in its discussion of verb fronting. The author rejects the head movement constraint, while maintaining the possibility of excorporation; the overlap in the empirical consequences of the two choices is not discussed.
To summarize, the hypothesis that head movement involves defective goals necessitates questionable assumptions, ‘impoverishing’ or ‘enriching’ the goals and probes involved. In addition, technical decisions made at various points in the book sometimes contradict each other and have undesirable consequences that are often left untested. The combination of these two factors limits the value of the monograph.