It is well known that ellipsis is restricted to occurring in certain syntactic environments. This restriction is known as ellipsis licensing, and while most of the literature on ellipsis concentrates on the analysis of the relationship between antecedent and ellipsis, there has been little work on the peculiarities of licensing, with the notable exception of Lobeck's (Reference Lobeck1995) Government and Binding theoretical approach. In The syntactic licensing of ellipsis, Lobke Aelbrecht approaches the puzzle of licensing from the perspective of modern-day Minimalism, centering her theory around the mechanisms of Agree and phases, as proposed by Chomsky (Reference Chomsky, Martin, Michaels and Uriagereka2000) . The book consists of five chapters. Chapter 1, ‘What is ellipsis?’, establishes the basic assumptions that offer a starting point for an investigation of ellipsis. Chapter 2, ‘Dutch modal complement ellipsis’, provides a thorough description of the book's core empirical focus, viz. modal complement ellipsis (MCE), which is an ellipsis construction found in Germanic and Romance that resembles English V(erb)P(hrase)-ellipsis. Chapter 3, ‘Ellipsis licensing’, develops an Agree-based theory of ellipsis licensing and discusses how it accounts for the MCE data. Chapter 4 is concerned with ‘Extending the analysis to other ellipses’. The book concludes with Chapter 5, ‘Conclusion and issues for further research’.
In Chapter 1, Aelbrecht identifies her study as an extended defence of the ‘PF-deletion’ approach to ellipsis, according to which the ellipsis site contains a full syntactic structure that is simply deleted at Phonetic Form (PF). She compares PF-deletion to nonsyntactic approaches, in which the ellipsis site contains no structure (e.g. Culicover & Jackendoff Reference Culicover and Jackendoff2005), and to so-called LF-copying (or proform) approaches, in which the ellipsis site is empty at spellout (except perhaps for a null proform) and is filled in only by copying of structure from the antecedent in the Logical Form (LF) component (e.g. Chung, Ladusaw & McCloskey 1995). Aelbrecht reviews some of the arguments from Merchant (Reference Merchant2001) for the PF-deletion account, and pays particular attention to the argument from extraction; that is, the fact that some ellipsis sites allow extraction under certain conditions. This is demonstrated by English VP-ellipsis, as in (1).
(1) I know which puppy you should take home,
but I don't know which one she should [take home twhich puppy].
(9, ex. (17a))
The availability of extraction is taken to be an argument for PF-deletion and against the proform analysis, since the ability to overtly extract from an ellipsis site indicates the presence of structure in that site that can be targeted by movement operations. Importantly, Aelbrecht proposes that while extraction unambiguously diagnoses the presence of structure, the absence of extraction does not diagnose the absence of structure, since other factors may interfere in extraction. This point becomes crucial to Aelbrecht's theory since one of its main concerns is to derive constraints on extraction from ellipsis from the interaction of ‘derivational ellipsis’ and the timing of movement.
Chapter 2 introduces MCE, focusing on data from Dutch. Aelbrecht begins by discussing the syntax of Dutch modals. She analyses the different kinds of modals (epistemic, deontic, dynamic) and argues that they are all raising verbs of a specific class (to be distinguished from auxiliaries), which select a Tense Phrase (TP). With this background, she provides a thorough description of MCE. Modal complement ellipsis occurs only with root modals (deontic and dynamic), and the deletion in MCE targets full constituents, including VP-adverbs and auxiliaries. Aelbrecht further provides evidence from there-expletives, which she takes to indicate that deletion does not target the full TP-complement of the modal head. In Dutch MCE with expletive subjects, the associate of the expletive is not deleted, as is shown in (2).
(2) Ermochtenweleensvriendinnetjesblijvenslapen,maarermochtengeenvriendjes[blijven slapen].therewere.allowedprtoncegirlfriendsstaysleepbuttherewere.allowednoboyfriendsstay sleep‘Girlfriends were allowed to sleep over, but boyfriends weren't.’
(56, ex. (72c))
Aelbrecht proposes that this is explained by assuming that the associate geen vriendjes ‘no boyfriends’ is in SpecTP and that deletion targets the complement of T.
The most important data come from extraction. Since Aelbrecht analyses modals as raising verbs, it follows that MCE must allow A-extraction, and she demonstrates that MCE is also compatible with other A-movements like passivisation and unaccusative raising (60f.). However, it turns out that MCE does not allow for A′-extraction (see (3)), either by scrambling or wh-movement, even though the non-elliptical equivalents are grammatical.
(3) *IkweetnietwieKaatwouuitnodigen,maarikweetwelwiezemoest[twieuitnodigen]IknownotwhoKaatwantedinvitebutIknowaffwhoshemust.pstinvite‘I don't know who Kaat wanted to invite, but I do know who she had to.’
(63, ex. (81a))
In this respect Dutch MCE is unlike VP-ellipsis in English (and indeed its MCE counterpart in French).
Chapter 3 of the book introduces the mechanisms that will provide an account of this asymmetry. Aelbrecht adopts the basic assumptions of Merchant (Reference Merchant2001), which derives ellipsis from the checking of a morphosyntactic feature on the elided constituent (called the E-feature) against the ‘licensing head’. According to Merchant, the E-feature issues the instruction to delete the given constituent at PF and is optionally included in the numeration. The licensing head is a syntactic head which needs to be in a head–complement relation with the elided constituent for ellipsis to occur. Thus, in Merchant's analysis, the licensing head checks the E-feature, deleting the constituent. Aelbrecht brings the analysis up to date by reframing Merchant's analysis in terms of Chomsky's Agree operation. Accordingly, ellipsis occurs when the licensing head is merged into the structure and enters into an Agree relation with the E-feature on a given head. As soon as this happens, the complement of the head is rendered unavailable for further syntactic computations, and vocabulary insertion at PF is blocked. Aelbrecht calls this process ‘derivational ellipsis’.
A crucial difference between the Agree model and Merchant's feature-checking predecessor is that the licensing head and the E-feature–bearing constituent can be in a non-local configuration. This has two main consequences, which form the empirical basis of Aelbrecht's theory. First, the system can now explain cases like (2), where the licensing head and the deletion site are non-adjacent. Aelbrecht proposes that the E-feature in MCE is located on T, and that Mod is the licensing head, so when the licensor is merged and Agrees with the E-feature, it deletes the complement to the exclusion of the associate in SpecTP. This also allows her to explain the fact that TP-adjuncts can be excluded from MCE, while VP-adjuncts cannot. The second outcome of adopting an Agree-based model is that it provides an explanation for the extraction facts. Simple subject extraction is allowed, since this involves movement from vP to SpecTP, prior to merger of the licensing head Mod. However, when the trigger for movement is merged after the licensing head Mod, then extraction should be blocked. This is the case with object wh-movement, where the trigger in the Complementiser head (C) takes the ModP as its complement: when C is merged, Mod has already Agreed with the E-feature on T, deleting its vP and therefore rendering it unavailable for extraction. On the assumption that wh-movement must stop off at the phase edge, the object wh-phrase is located in SpecvP and thus included in the deletion site. The proposed analysis therefore derives the extraction possibilities from the timing of extraction with respect to ellipsis in the derivation.
Chapter 4 extends the analysis to other ellipsis constructions: sluicing, VP-ellipsis, pseudogapping and British do. Aelbrecht provides an overview of the properties of each construction and posits a unique E-feature and matching licensor for them. However, the analysis for sluicing is not very different from that given in Merchant (Reference Merchant2001), and it makes no new predictions. The analysis of VP-ellipsis, on the other hand, is quite different, since Aelbrecht's system makes predictions about the extraction facts: as demonstrated in (1), English VP-ellipsis allows for wh-extraction, so this must be accounted for by its licensing syntax. Aelbrecht notes the assumption in the previous literature that the licensing head for English VP-ellipsis is T; she adopts this assumption, and makes the additional assumption that the E-feature occurs on the head of VoiceP, a functional projection above vP, which constitutes the phase head rather than v (following Baltin Reference Baltin2007). Importantly, since VoiceP is a phase head, wh-extraction proceeds via the specifier of this phrase, so when T and the E-feature on Voice enter into an Agree relation, the extracted wh-phrase will be in SpecVoiceP, thus allowing extraction. Pseudogapping receives a similar analysis, where extraction proceeds via the VoiceP edge to a Focus projection below T.
Finally, Aelbrecht proposes that her framework also provides an account of British do, an ellipsis construction found in British English where a standard VP-ellipsis construction contains a ‘superfluous’ do at the edge of the ellipsis, illustrated in (4).
(4) Luis will run the race, and Nana will do [run the race], too.(194, ex. (81a))
Baltin (Reference Baltin2007) notes that British do contrasts with standard VP-ellipsis by blocking extraction of wh-phrases and inverse scope readings with quantified subjects and objects, although it does allow for A-extraction. Aelbrecht proposes to account for this by positing that the do in this construction is a spellout of little v, and that the ellipsis seen in these constructions is deletion of the VP-complement. Thus, v is both the E-feature–bearing element and the licensing head; extraction is blocked because there is no phase edge between licensor and E-feature.
Aelbrecht's book gives a thorough analysis and overview of the issues that need to be considered when we try to account for the seemingly elusive generalisations concerning ellipsis licensing. Its theoretical rigour is demonstrated by the fact that, in true Minimalist spirit, it does not introduce any new technology to the Agree system to derive its results; instead, it takes the existing system and puts it to good use, exploiting the fact that Agree establishes non-local dependencies to derive the apparently non-local nature of the relation between licensors and deletion sites. Its empirical rigour is demonstrated by the thoroughness of the analysis of MCE and the clear presentation of predictions and their consequences throughout the book.
However, the proposed theory also faces a number of problems. Although Aelbrecht adopts the Agree system wholesale from Chomsky (Reference Chomsky, Martin, Michaels and Uriagereka2000), she overlooks important technical issues in the implementation of this system in doing ellipsis by Agree. In her theory, the Agree relation between the E-feature and the licensor is a case of Reverse Agree: the interpretable category feature on the licensor values a matching uninterpretable one that is a subpart of the E-feature. Such cases are described as Reverse Agree since they are the opposite of the basic schema in Chomsky's (Reference Chomsky, Martin, Michaels and Uriagereka2000) system, where interpretable features are borne on the goal rather than the probe. Aelbrecht cites Zeijlstra's (Reference Zeijlstra2008) proposals regarding Negative Concord (NC) as precedent for Reverse Agree, but this overlooks an important characteristic of the Agree relations in these cases, namely that the relations are not confined to the vP phase (or VoiceP in Aelbrecht's system). Aelbrecht argues (146–154) that examples like (5) below are ruled out by her theory because the Agree relation cannot apply across the VoiceP phase boundary (here the E-feature would be on the lower VoiceP and the licensor would be must), but as we can see in the Italian example in (6), such a dependency is possible with Agree in NC.
(5) *Brian loves watching fat guys wrestleand Griffin must [VoiceP love [VoicePwatching fat guys wrestle]],as well.
(153, ex. (121a))
(6) Nonmi[VoicePpiace [VoicePvederenessunosoffrire]].(Italian)notto.melikesee.infno-oneget.hurt‘I don't like to see anyone getting hurt.’
It is thus necessary to explain why Agree can apply across phase boundaries with NC but not with ellipsis, but this issue is not discussed in Aelbrecht's book.
There are also problems with the two predictions that provide the empirical basis for Aelbrecht's approach. For one, the theory is unable to deal with the fact that in English, unlike in Dutch, Agree can establish a relation between expletive and associate in VP-ellipsis, where the expletive is contained in the ellipsis site, as in (7).
(7) Rab said there would be punch and pie, and there was [punch and pie].
It is important to bear in mind that, in the Agree system adopted by Aelbrecht, movement and Agree are essentially the same operation, with movement involving the addition of an EPP-feature to the probe. Thus, if movement from an ellipsis site is banned, Agree should also be banned. But the example in (7) shows that this is not the case. Aelbrecht's theory is also unable to explain the fact that English VP-ellipsis can optionally include some auxiliaries, something the author simply dismisses as a ‘peculiar fact’ in a footnote on page 171. This is inaccurate. Rather, in Thoms (to appear), it is shown that the inclusion or exclusion of auxiliaries within the ellipsis site is a syntactic property of the auxiliaries in question.
The analysis of British do is also problematic, since it overlooks the fact that this construction is acceptable with comparatives, pseudogapping (for many speakers) and some kinds of inverse scope readings (see Thoms Reference Thoms2010 for extensive discussion of this construction). Finally, the proposed analysis even struggles to deal with MCE, as pointed out by the author herself, since Quantifier Raising can escape the MCE site in cases of antecedent-contained deletion. Given these facts, it is hard to see how the extraction facts can be taken to be a successful prediction of the theory.
Beyond these issues, there is a larger question about the nature of the theory proposed. Aelbrecht assumes that ellipsis licensing is a fact of lexical variation, where E-features appear on certain constituents, certain lexical heads have the property of being licensing heads, and certain heads are phases. In effect, the theory describes the basic phenomenon of licensing by stipulating the presence of features where it needs them. While it has been common practice to assume that all morphosyntactic variation should be rooted in the lexicon (going back to Borer Reference Borer1984), it makes for a theory that is sketchy and lacking in explanatory power (see also Boeckx Reference Boeckx2010 for criticisms from the perspective of modern Minimalism). This is particularly disappointing when it is used in the context of ‘licensing heads’ borrowed from Lobeck's (Reference Lobeck1995) classic account. The success of Lobeck's theory was that it united the mechanisms for licensing the empty categories created by movement with the mechanisms for licensing ellipsis: both traces and ellipsis sites needed to be properly governed by a licensing head of some kind. This unity was lost in the move to Minimalism, in which government has no status and conditions on traces are derived from conditions on derivations. Therefore, the licensing head has no theoretical justification beyond its use in the theory of ellipsis, and the elegance of Lobeck's analysis is lost.
Despite these problems, Aelbrecht's book is a valuable and insightful contribution to the study of ellipsis. Its virtue is that it clearly delineates the issues and challenges for further research on the theory of ellipsis. Any future work would do well to meet Aelbrecht's standards of theoretical and empirical rigour.
