Publishing Partner: Cambridge University Press CUP Extra Publisher Login
amazon logo
More Info

New from Oxford University Press!


Style, Mediation, and Change

Edited by Janus Mortensen, Nikolas Coupland, and Jacob Thogersen

Style, Mediation, and Change "Offers a coherent view of style as a unifying concept for the sociolinguistics of talking media."

New from Cambridge University Press!


Intonation and Prosodic Structure

By Caroline Féry

Intonation and Prosodic Structure "provides a state-of-the-art survey of intonation and prosodic structure."

Review of  Edge-Based Clausal Syntax

Reviewer: Matthew Reeve
Book Title: Edge-Based Clausal Syntax
Book Author: Paul M. Postal
Publisher: MIT Press
Linguistic Field(s): Syntax
Issue Number: 23.3084

Discuss this Review
Help on Posting
AUTHOR: Postal, Paul M.
TITLE: Edge-Based Clausal Syntax
SUBTITLE: A Study of (Mostly) English Object Structure
YEAR: 2011

Matthew Reeve, Research Department of Linguistics, University College London


Most current generative research assumes that single sister DPs of verbs can be
characterised as direct objects. In this monograph, Postal sets out to challenge
this conception, arguing instead that at least three distinct types of object
occur in such a configuration. According to the line of argument familiar from
past studies in Relational Grammar, what distinguishes these objects must
therefore be a primitive, rather than a structural configuration. That primitive
is the ‘edge’ of the title, a label which designates the grammatical function of
one constituent relative to another (containing) constituent. Because of the
highly technical nature of the analyses, which would require a great deal of
space to do justice to, I will primarily focus on the empirical aspects of the
book in the summary, while making some remarks on the specific analyses in the

Chapter 1, the introduction, begins by arguing that the large volume of concepts
accreted within “the various generative transformational grammar frameworks
emanating from the work of Noam Chomsky” (p. 1), which he refers to as ‘Barrel
A’, has consistently failed to provide insight and correct description regarding
English object structure. As such, Barrel A itself must be called into question.
Postal then enumerates various differences between Barrel A approaches and his
approach, which he refers to as ‘Metagraph Grammar’ (MG): the object of study (a
psychological/biological and proof-theoretic object versus a set-theoretic and
model-theoretic object); the testability of theories within these frameworks;
the fact that Barrel A is generally (partially) derivational while MG is
representational and has no notion of ‘operation’. The key difference between
the two approaches, however, is in the vocabulary of graph-theoretic concepts
used. Postal argues that Barrel A ignores the range of options available within
graph theory, essentially restricting itself to strings (of which ‘trees’ are
but a diagrammatic representation). He argues that trees with parallel lines
(i.e. with the same mother and daughter) should be part of syntactic theory,
with the lines being distinguished by ‘edges’ (basically, labelled branches), a
graph-theoretic concept not used within mainstream generative grammar. The rest
of the chapter provides a crash course in MG, defining its terms rigorously, and
illustrates how the approach can be applied to reflexive anaphora.

Chapter 2, ‘Objects and Arrays’, provides the foundational argumentation for
dividing objects into three classes, which Postal calls ‘arrays’. It starts by
observing that while Chomsky (1965) defined ‘direct object’ as any NP occurring
in the context [VP V NP Y], he did not provide a ‘theoretical reconstruction’ of
any other object relation. Even such a familiar notion as ‘indirect object’ “has
not played an official role in Barrel A treatments of English” (p. 48). Even
more strikingly, Chomsky’s formulation means that single objects are always
direct objects. Postal argues, by contrast, that 2, 3 and 4 objects can all be
single objects. He starts, however, by discussing double object constructions,
of which the Barrel A literature “lacks any consensus analysis” (p. 50). The
basic features that remain unexplained are certain gaps in passivisation
possibilities (e.g. the ‘sell’/‘buy’ contrast), as well as gaps in
presentational ‘there’, locative inversion and middle paradigms. Postal provides
counterexamples to claims in the literature that these are due to factors such
as animacy or possession. A further claim that has been made is that the first
object is actually a direct object, which Postal also challenges, additionally
claiming that some single objects cannot be direct objects. In sum, then, this
supports the MG view that grammatical relations for these cases cannot be
defined in terms of phrase structure, word order or case, but must be
primitives. Postal then goes on to provide empirical evidence in favour of
distinguishing three arrays. Array 0 consists of prototypical direct objects,
with all the properties that these are typically held to have (e.g. they can
passivise and undergo ‘tough’-movement). Array 1, on the other hand, cannot be
passivised, nor can it participate in middles, ‘of’-nominalisations, nominalised
incorporation, ‘tough’-movement, ‘object deletion’, nominal ‘tough’-movement or
adjectival passives (which are grouped under the rubric of ‘Q-constructions’).
Array 2, in addition to disallowing passives and Q-constructions, disallows left
extraction, Heavy Noun Phrase Shift, Right Node Raising (RNR), left
subextraction, stranding under gapping and subdeletion, and such objects cannot
be in situ human relative pronouns. Summarising the implications of this
distinction, Postal argues that the three arrays do not differ in ways that
“would at one time have raised suspicions of constituency differences” (p. 69);
thus, the containing VP behaves identically with respect to VP-fronting, RNR,
VP-ellipsis, VP-relativisation and ‘strange’ coordination. The Array 0/1/2
distinction would thus be puzzling under Chomsky’s conception of grammatical
relations. Postal instead gives the three arrays distinct ‘edge labels’ (Array 0
objects are ‘2 objects’, Array 1 objects are ‘4 objects’ and Array 2 objects are
‘3 objects’). In addition, there are distinct edge labels for subjects (1),
semiobjects (5), quasiobjects (6), semiclauses (7), chômeurs (8), extraposees
(9) and genitives ({Oblique}).

Chapter 3, ‘Double Object Structures’ (DOCs), starts by recalling Fillmore’s
(1965) distinction between Class A and B double object constructions
(exemplified with ‘send’ vs. ‘buy’ respectively) on the basis of passivisation
possibilities. Postal notes that analyses of DOCs can be roughly divided into
two classes: ‘dative shift’ analyses, in which an underlying oblique becomes a
surface 2 object; and ‘invisible P’ analyses, in which the first object of a DOC
has an invisible preposition. Postal argues against both of these by arguing for
the existence of 3 objects and for the claim that the first object of some DOCs
is a 3 object. Thus, he provides evidence for a link between DOC first objects
and single Array 2 objects (here classed as 3 objects, remember): namely, the
evidence discussed in Chapter 2 for Array 2 (see above), which also holds for
DOC first objects in some varieties of English. In addition, there seem to be
some cases which genuinely do involve an analogue of ‘dative shift’ (in MG
terms, ‘advancement to 2’ and ‘demotion of earlier 2’), and which behave
differently from the Array 2 cases in allowing the relevant constructions.
Postal then turns his attention to DOC second objects, which, he argues, share
the properties of Array 1 single objects (here classed as 4 objects). Thus, DOC
second objects must be 4 objects. A potential problem is that while the 2/3
object contrast is ‘banal’ in the context of French and German (which
distinguish them in terms of case-marking), it is more difficult to find an
overt 2/4 contrast (though Postal suggests Hausa as a candidate language).
Returning to the Class A/B distinction, Postal argues that while both have the
same ‘surface’ grammatical relations (namely, 3 and 4 objects), they have
distinct ‘initial’ relations (Class A has 3-2; Class B has oblique-2). The
versions of DOCs with a DP-PP structure (Class A typically has a ‘to’-PP, Class
B a ‘for’-PP) are also distinct: Class A has initial 2-3 and final 2-5; Class B
has initial and final 2-oblique. Postal finally returns to the ‘invisible P’
analysis, arguing against it on the basis of contrasts between first objects and
PPs (e.g. word order and extraction possibilities).

Much of the rest of the work is dedicated to issues arising in the analysis of
passives. Postal’s aim from here on is “to document further ways in which this
division [between 2, 3 and 4 objects -- MJR] simplifies and regularizes the
statement of English grammatical facts and to explore various distributional
data that are related to the three-way object division but do not just follow
from it” (p. 143). Chapter 4, ‘Periphrastic and Nonperiphrastic Passives’, is
mainly devoted to elaborating an analysis of periphrastic passives, including
uncontroversial passives with auxiliaries, as well as ‘clause union passives’ in
French, which lack passive morphology, and other types of passive-like
construction such as clauses with ‘born’, middles and antipassives. An appendix
deals with the structure of clauses with adjectival predicates.

Chapter 5, ‘Passivization Targets: I’, deals with the question of what object
DPs can feed passivisations (in MG terms, what active clause DPs can correspond
to the final 1s of grammatical periphrastic passive and middle clauses). The
general aim of this chapter is to argue against the commonly assumed notion that
only ‘direct objects’ (2 objects in Postal’s terms) can be passivised (despite
the observations of Chapter 2 that single 3 and 4 objects cannot passivise).
Postal first discusses pseudopassives, arguing against the ‘reanalysis’ approach
which is so common in mainstream generative work (i.e. the passivised DP
‘becomes’ a direct object), arguing instead that the ‘prepassive arc’ (i.e. the
MG analogue of the pre-movement constituent) is a 3 arc. He then moves on to
passivisation of objects in DOCs, arguing that since some varieties allow
passivisation of DOC first objects, this is another instance of 3 object
passivisation. An appendix deals with the existence of ‘adjectival
pseudopassives’, apparently problematic in the face of the idea that adjectives
must modify nouns which are initial 2s (Postal here accepts a modified
reanalysis solution).

Chapter 6, ‘Passivization Targets: II’, primarily deals with two types of
passives which seem to be restricted to 2 objects even for speakers who accept
some 4 object passives (thus behaving similarly to middles), namely expletive
‘there’ passives and locative inversion passives. The restricted nature of these
constructions is unexpected in the ‘dative shift’ analysis that Postal is
arguing against. The data reveal a parallel between pseudopassives and DOC
passives: neither can be reduced to 2 object passivisation. Finally, Postal
argues that single objects of particle verbs are 3 objects, in contrast to the
Kayne (1985) analysis in which they are 1s (subjects) of small clauses. He
admits that the analysis of particle verbs is relatively incomplete, however.

Chapter 7, ‘Passivization Targets: III’, attempts to refine the analysis of
passivisation targets in English, explaining the restrictions on passivisation
discovered up to this point (middles only target ‘indigenous 2s’; pseudopassives
and DOC passives only target non-2s; expletive ‘there’ passives, locative
inversion passives, periphrastic ‘get’-passives and participial absolutes only
target 2s; and various consequences flowing from these).

Chapter 8, ‘Visser’s Generalization’, attempts an alternative analysis of the
eponymous generalization, which essentially says that subject control is
incompatible with object passivisation, whereas non-control and object control
cases with the same verbs do not preclude such passivisation. Postal’s claim is
that VG is not a single principle barring interaction of passive and subject
control, nor is it specific to passives. As expected by now, he argues that
capturing VG requires an appeal to the 2/3/4 object distinction.

Chapter 9, ‘Clauses with ‘That’ Clause Complements’, attempts to show that the
ideas introduced thus far provide an explanation and description of some
puzzling restrictions involving ‘that’-clause complements. It also claims that
such complements provide evidence for Postal’s claim that passives involve a
covert resumptive pronoun. Postal distinguishes four classes of verbs taking
‘that’-clause complements, again distinguishing them on the basis of
passivisation facts.

Finally, Chapter 10, modestly entitled ‘Results, If Any’, reiterates Postal’s
view that, since the all aspects of the account depend entirely on the
theoretical framework chosen, namely MG, they support the more general views
that define it. The result of the investigation is a “multitude of descriptive
conditions, some of which may be universals” (p. 389).


Postal’s book can be seen both as an extended argument for RG-type approaches
over generative approaches -- since, as the argument goes, the latter are unable
to distinguish these types of objects in an insightful way -- and as a goldmine
of interesting problems, both old and new, concerning anaphora, passives,
‘raising to object’, control, extraposition and islandhood, among other issues.
The book therefore ought not to be read merely by aficionados of RG -- on the
contrary, generative syntacticians owe it to their field to answer the
challenges that this book makes to generative argumentation in general and to
specific generative analyses in particular.

As well as being rich in interesting data and analytical detail, this book
provides an impressively thorough introduction to Postal’s MG framework. Indeed,
it seems that one of Postal’s aims in this book is to provoke cross-framework
debate: as well as arguing against the foundational assumptions of generative
work, he often engages with specific generative analyses (in addition, the
foreword to the book is written by Chris Collins, a prominent Minimalist). This
means that generativists should find it hard to ignore Postal’s claims. Another
striking fact about this book is the extent to which data that are problematic
for Postal’s analyses are readily discussed, as well as cases where there is no
particular evidence for or against a device that he posits (e.g. p. 147, which
proposes representing a participial clause as an initial 2 of its containing
auxiliary clause).

Nevertheless, a book which lays its cards on the table as much as this is bound
to raise some questions and concerns. One of these concerns relates to the
structure of the grammar that is envisaged. A much earlier paper of Postal’s
(1972) argued that ‘the best theory’ is one which minimises the number of
distinct components of the grammar. In the same vein, Postal’s aim in this book
is to account for all distinctions in acceptability/grammaticality in the same
manner: with conditions on relations between different types of arcs (and
lexical restrictions). Essentially then, we have one ‘component’ of grammar.
This is in contrast to much recent work in generative grammar, for example, in
the area of information structure, in which it is quite often proposed that
certain types of operations (and therefore the acceptability distinctions
bearing on them) do not reside in the syntax proper, but belong to more
‘surfacey’ components. Indeed, for focus-driven operations in certain languages
(e.g. Spanish, Italian), it is by now widely accepted that they do not merely
make reference to syntactic properties (e.g. Zubizarreta 1998, Reinhart 2006).
Similarly, it would be hard to argue that the relative acceptability of certain
cases of rightward dislocation (Heavy Noun Phrase Shift, ‘particle movement’),
which is based on parsing-motivated factors such as ‘heaviness’, was entirely
syntactically determined (e.g. Hawkins 1994). Yet Postal proposes a condition to
prevent weak definite pronouns following particles which states that the
relevant 4 arc of the pronoun must have ‘3 quace’ (i.e. also have properties of
3 objects). One question, then, would be how willing Postal would be to
‘insulate’ the syntax, and thus the sole source of dislocation possibilities,
from the phonology or the parser, for example. The alternative would be to claim
that not all acceptability distinctions between syntactic minimal pairs need to
be accounted for in the syntax.

The second question I would raise concerns the judgments themselves. Of course,
since Postal is an American English speaker and I am a British English speaker,
a certain amount of disagreement is very likely. Yet the very large number of
times that I disagreed with Postal’s judgments on data which were crucial for
the analyses being proposed was slightly concerning. To give just one example,
‘buy’ in the sense of ‘buy an argument’, is claimed not to participate in ‘Q
constructions’ (see above), which is taken as evidence for including it in a
distinct array (Array 1). For me, it can easily participate in four of the eight
constructions. Of the other verbs given in this section (also purportedly
belonging to Array 1), ‘croak’ allows at least one (though one which is not
permitted for ‘buy’), ‘dig’ (meaning ‘like’) allows three, ‘quit’ allows at
least two, ‘want’ probably allows at least two, while only ‘give’ (as in ‘give
milk’) and ‘hear’ (probably) allow none (while Postal sometimes notes the
existence of ‘different dialects’, of which I was quite often a speaker, there
are more cases where no such distinction is noted). Furthermore, this
disagreement does not just represent an American/British divide -- Levine (2001)
and Postal’s wife often disagree with his judgments as well. As Levine notes,
the existence of such disparities gives rise to methodological concerns. If,
when using grammaticality judgments, we really are only studying idiolects or
I-languages, then asking other speakers to corroborate a set of judgements is
fairly useless, since we have no antecedently given idea of how (un)stable a
community of I-languages should be. In this case, my disagreements with the
judgements are irrelevant. However, this makes inter-theory comparison extremely
difficult -- a case of not being able to see the wood (i.e. what is more
‘universal’) for the trees (the more variable facts). The suspicion might even
arise that almost as many versions of this book could be written as there are
native speakers of English; until we know what the theoretical significance of a
particular dialect difference is, it is hard to say.

I have said little about the MG machinery put to work in this book. This is
partly because I am not a specialist in MG/RG and partly because it would not be
possible to do so in any depth in the space allotted. Nevertheless, I do have
some questions about certain theoretical aspects of the proposal. First, Postal
argues in Chapter 1 that “Barrel A ignores graph theory” (p. 8) because it does
not make use of parallel branches and edges. Yet in order to capture the
analogue of ‘surface structure’ or ‘Spell-Out’ within MG, Postal posits
‘S(urface)-graphs, which are connected, rooted and have no overlapping distinct
arcs. In other words, they are like trees in mainstream generative grammar. This
might suggest that there is something fundamental about such trees that is being
missed in the enriched formalism (of course, it could just be that the fact that
such trees essentially correspond to strings, they are more suitable for a
pronounced structure -- but I see no reason inherent to Postal’s system as to
why S-trees should not in principle have overlapping distinct arcs). A further
concern I have is about the use of the notion of ‘quace’. Roughly speaking, this
means that an arc with edge label 2 and 3 quace has properties of both 2 and 3
arcs. There is nothing wrong with this in principle, but it sometimes seems to
me that the use of quace is not very constrained; it is never stated what the
restrictions on bearing particular quace values might be.

Nevertheless, the above points do not seriously detract from what is a very
impressive work by one of the giants of the field. Postal presents a good amount
of convincing evidence for the 2/3/4 object distinction, and I agree with
Collins when he says in the foreword that although he “disagree[s] with Postal’s
claim that the book argues in some fundamental way for a
graph-theoretical/relational approach to syntax, over a Merge-based approach,
[…] one can ask whether relational approaches to syntax are in some sense more
conducive than other approaches to the discovery of generalizations like those
found in this book” (pp. xiii-xiv). This book thus sets Minimalists an
interesting challenge: to capture the distinctive properties of the various
object types in a satisfying and natural way.


Chomsky, Noam. 1965. Aspects of the theory of syntax. Cambridge, MA: MIT Press.

Fillmore, Charles. 1965. Indirect object constructions in English and the
ordering of transformations. The Hague: Mouton.

Kayne, R. S. 1985. Principles of particle constructions. In Grammatical
representation, eds. J. Guéron, H.-G. Obenauer & J.-Y. Pollock, 101-140.
Dordrecht: Foris.

Levine, Robert D. 2001. The extraction riddle: just what are we missing? Journal
of Linguistics 37, 145-174. [Review of Paul M. P. (1998). Three investigations
of extraction. Cambridge, MA: MIT Press.]

Postal, Paul M. 1972. ‘The best theory’. In: S. Peters (ed.), Goals of
linguistic theory. Englewood Cliffs, NJ: Prentice-Hall.

Zubizarreta, Maria L. 1998. Prosody, Focus, and Word Order. Cambridge, MA: MIT

Matthew Reeve currently teaches syntax at University College London, where he obtained his PhD in 2010. His primary research interests are in the interfaces between syntax, semantics and information structure, and the syntax of English and Russian in general.

Format: Hardback
ISBN: 0262014815
ISBN-13: 9780262014816
Pages: 488
Prices: U.S. $ 70

Format: Paperback
ISBN: 0262512750
ISBN-13: 9780262512756
Pages: 488
Prices: U.S. $ 35