Postal, Paul M. (1998) Three Investigations of Extraction. Cambridge,
Mass.: MIT Press. Hardback edition: February 5, 1999 [sic], x+215pp. incl.
index. ISBN 0-262-16179-6. $32.50.
Reviewed by Kleanthes K. Grohmann
Synopsis:
If you're a (non-)native speaker of English, work on syntactic phenomena
and want to experience a sensory overload of complex data, rush out and buy
this book! It's fantastic: If you thought that extraction simply involves
displacement of an element and a well understood relationship to its
original position (regardless of theoretical framework) with more or less
sorted out conditions and restrictions, think again. Postal painstakingly
(yet often elegantly) aims at showing that none of the cuts linguists have
made ever since and including Ross' (1967) seminal dissertation are even
remotely adequate. Most importantly--from a current generative perspective
which I will focus on here (the P&P approach from GB to Minimalism
developed ever since Chomsky 1981)--our often mentioned and upheld belief
that extraction phenomena can roughly be split into a subject vs. object
asymmetry (Kayne 1984), an argument vs. adjunct distinction (cf. Cinque
1990) or in terms of referentiality (Rizzi 1990, 1992) is simply false.
Instead, Postal proposes two basic classes of extraction types with one of
them to be split into two more groups; presence of a resumptive pronoun and
island sensitivity are crucial ingredients of his cuts. (**Author's note:
Any difficulties on part of the reader following sentences in this review
are intended and should reflect the work discussed.)
Structure:
Postal lays out his main claims in a compact and detailed introduction
which basically consists of remarks on each of the following three chapters
(pp. 1-24). To illustrate the main themes of each chapter, he employs
examples from these chapters as well as data that did not find its way into
the main text (presumably for space reasons, but see below for exceptions).
Chapter 2 is a rather rich exposition of the different types of extraction
types in which Postal argues (quite convincingly) for an elaborate set of
distinctions (25-49). The last two chapters deal with specific types of
constructions and island phenomena: chapter 3 (51-95) discusses the
Coordinate Structure Constraint (CSC) in more detail than has ever been
done in the P&P-approach (see also p. 184, fn. 3) and chapter 4 (97-163)
Right Node Raising (RNR). The last twenty pages of the book are devoted to
appendices A-C on islands vs. non-islands (165-172), RNR as extraction
(173-174) and comments by a reviewer (175-180). The useful index follows 20
pages of notes and the extensive yet slightly dated bibliography (see
below). If you want a measurement of "sensory overload of complex data,"
the book features a total of 539 numbered examples (many of which with more
than three sentences) which, confirmed by a quick look through other syntax
books, weighs in at twice as much as most (with the possible exception of
Postal 1974, if memory serves me right).
Critical Evaluation:
The obvious goal of this book is to present a deep study of all
instances of "L(eft)-extraction," classify them into clear categories and
relate them to each other, despite many differences, on the basis of a
basic set of shared properties. Rather than dividing them purely along the
lines of (certain) island-sensitivities, adjunct/argument asymmetries and
the like, Postal proposes two basic types of extractions: A- and
B-extractions (where 'A' and 'B' are obviously arbitrary labels). We will
also see that the former need (?) to be further distinguished into A1- and
A2-extractions and the presence of (often null) resumptive pronouns is
crucial for the overall approach.
Just to give you a feeling of what kind of constructions we are talking
about here, consider (1), a subset of Postal's (1) on p. 1 (refraining from
Postal's intricate coindexation system of extractee and gap position which
should be clear):
(1) a. Who did they nominate t to be director?
b. No such gorilla did I ever see t.
c. Frank, I would never hire t.
d. It was Frank who they hired t.
All examples involve a (sometimes null) constituent moved to the left and
its corresponding gap. (For ease of exposition, indicated as t but, as
Postal points out (p. 181, fn. 1), "[t]he gap/coindexing notation in (1)
and throughout this book is a descriptive device representing no commitment
to the linguistic reality of either traces or coindexing" which reflects
again Postal's non-commitment to a particular theoretic framework within
the generative tradition; see also his remark "in the work of Chomsky and
those he influenced" on p. 36.) Examples such as (1a,b) above (question
extraction and negative NP-fronting, respectively) illustrate
A-extractions, while (1c,d) (topicalization and clefting) are instances of
B-extractions.
The properties of NP-extraction, hence applying to all cases of (Postal's)
(1), in English are a visible gap, unbounded "distance" between gap and
binder, island-sensitivity, "licensing" of P(arasitic)-gaps, inducing
strong crossover violations, same determination of crossing dependencies,
across-the-board instances, stranding of prepositions in the same contexts
and their being subject to the same specific "pure extraction" constraints
(p. 2; all quotation marks from Postal). While these properties seem to
paint a unitary picture with regard to extractions, chapter 2 argues for a
distinction into A- and B-extractions on the basis of systematic adherence
to these properties by one set of data (A-extractions, namely all but
B-extractions such as "most types of question extraction and restrictive
relative extraction as well as negative extraction, pseudoclefting and
others" (p. 4)) and a more constrained behavior of another (B-extractions
such as topicalization, NP clefting and nonrestrictive relative clause
extraction).
An A-extraction site, unlike that of a B-extraction, forms an
"anti-pronominal context" (AC), that is to say a position incompatible with
weak definite pronouns. This, Postal argues, can be accounted for by
assuming an invisible resumptive pronoun (RP) in the B-extraction site. In
order not to lose Ross' (1967) generalization that presence of RPs allows
island violations (which neither A- nor B-extractions do), Postal proposes
that the RP itself extracts into a sisterhood position to the extractee in
which it can be controlled and consequently deleted, hence rendered
invisible. Let me illustrate this with an example from inalienable
possession contexts, corresponding to Postal's (8) and (9) on p. 28:
(2) a. They touched his ear.
b. They touched him on the/his ear.
(3) a. What part of the/his body did they touch (him on) t?
b. His ear, they never touched (*him on) t.
(3a) is an A-extraction of body part NPs, possible with both varieties in
(2); the corresponding B-extraction (3b) is only well-formed with the first
type, as illustrated in (2a).
Illustrating the AC a little bit better, (4)--Postal's (36) on p. 34--shows
that the position of the inalienable NP that does not allow B-extraction is
antipronominal (while the one that does is not):
(4) a. They wanted to touch his arm and they did touch it.
b. * They wanted to touch him on the/his arm and they did touch him
on it.
On p. 32, Postal summarizes the results achieved by the detailed contrast
of A- and B-extractions (partly in chapter 1, partly in chapter 2). Given
that the data fall indeed the way he argues, the environments of
B-extractions can be seen as a proper subset of those of A-extractions as
no context allows B- but not A-extractions. (The fact that some
environments allow B- but not certain A-extractions can be captured by
further distinguishing A-extractions; see below.) The cases Postal looked
at in particular are existential 'there'-constructions, change-of-color
contexts, name positions, inalienable possession contexts, predicate
nominals, adverbial NPs, extraposed prepositional phrases, infinitival
extraposition, exceptive shifting, temporal NPs and lastly, idiomatic verb
+ NP structures; counting in "2.2.13 A Locative Case" (right after the
other cases), which shows another A/B-extraction contrast, I add up
altogether thirteen "English contexts that permit A- but not B-extractions"
(not twelve, as Postal states on p. 32).
This considerable database arguing for two types of extraction now begs for
a theoretical explanation. As mentioned above, the Resumptive Pronoun
Hypothesis shall play a crucial role. Building the analysis on the presence
of invisible RPs in English, Postal defends by independent motivation for
RP-extraction in a number of languages (such as German, Hebrew, Irish) and
the fact "that even certain visible English RPs are subject to
L-extraction" (p. 13). As a first stab, RPs are a crucial ingredient of
B-extractions in which the invisible (=null) RP is extracted and then
deleted (under some form of control, cf. pp. 15-17, 36-38 and 65), while
A-extractions do not contain such a pronominal element.
Rather than going through all the details of the rather complex analysis
that Postal develops in this book, I will restrict the remainder of the
discussion to the general line of reasoning, basic patterns and a
simplified summary. To exemplify what kind of details I will leave out,
consider (4) (Postal's (29) on p. 12) and Postal's immediately following
paragraph dealing with further partitioning of both islands and RPs.
(5) WHAT primary RP {...} <(secondary RP) who saw t> was a video
|_______| |______________| |________________|
where {...} represents 'the secret police arrested everyone'
"In (29) [(4)] the path between the extractee WHAT-1 and its extraction
site can span the relative clause island boundary because the former links
to a (primary) RP, in accord with (21a) [Ross' (1967) observation that
constituents can be extracted out of islands when linked to an RP]. That RP
must itself extract because it is a controlled RP and must extract to the
point of the extractee to which it is linked and hence out of the same
island because it is primary. Moreover, the primary RP can extract from
that island because the latter is unlocked. But to take advantage of that
property, the primary RP extractee must, in accord with Ross's principle,
itself link to a secondary RP. That controlled RP must also extract and
does, but nothing forces it to extract further than to the boundary of the
lowest island the primary RP extracts from, here the relative clause. There
is and can be no extraction from an island of the secondary RP because,
given my interpretation of Ross's principle as (21a), that would require a
tertiary RP, banned by (27d) [which says that "[t]ertiary RPs (i.e., RPs
linked to extracted secondary RPs) are excluded entirely"]." (p. 13)
In the light of (4) I hope I can be forgiven for not going through every
aspect of the approach in excruciating detail. But to return to the
distinction of A- vs. B-extractions, the RP-approach is not enough. While
all B-extractions contain an (invisible) RP, it is not the case that no
A-extractions do.
(6) a. How many bags do you wonder whether I think t are on the table?
b. * How many bags do you wonder whether I think there are t on the
table?
The a-example in (6)--Postal's (66) on p. 44--must obviously be an
A-extraction (because B-extractions are extremely sensitive to islands). On
the other hand, the position t in (6b) is what he calls a "wide
antipronomainal context," which Postal, exploring Obenauer's (1984) and
Cinque's (1990) studies, argues to be of such nature that extraction from a
selective island is not possible. ((6a) is an obvious example of extraction
out of selective islands.) Rather than going the Cinque-route and
distinguishing different strengths of islands, Postal maintains that a
selective island is just that, namely an island; extractability must come
from somewhere else, and this somewhere else, he argues, relates to RPs.
Given that both constructions in (6) are A-extractions and one is
grammatical, the other one not, we are forced to say that there are two
types of A-extractions. Moreover, if B-extractions differ from
A-extractions in requiring an RP and if an (invisible) RP allows extraction
out of (selective) islands, we could say that the A-extraction in (6a) also
contains an RP, unlike the one in (6b). This is exactly the path Postal
pursues:
(7) L-extraction types (slightly adopted from (74), p. 45)
a. B-extractions: those that require RPs in their extraction sites
b. A-extractions: those that do not require RPs in their extraction
sites
i. A1-extractions which allow RPs in their extraction sites
ii. A2-extractions which forbid RPs in their extraction sites
B-extractions are thus the more restrictive instances of L-extraction and,
among other things, very sensitive to islands. A-extractions fall into (as
Postal says, "at least") two major classes: those that are very
island-sensitive (A1-extractions) and those that may violate what he calls
"selective islands" (A2-extractions). Simply put, selective islands allow
"only a very limited subset of all constituent types, mostly NPs" (p. 5) to
be extracted.
Personally, I found these first 50 pages pretty dense which is not to say
that they were particularly difficult to follow. On the contrary, the style
of writing is rather pleasant, detailed yet entertaining. But with such a
wealth of information and ascribed (variance in) properties, it is easy to
persuade the reader simply by force. It should be noted that not all data
are as clear as Postal makes them, but the fast succession of data makes it
difficult to immediately look through and form an opinion on one's own.
While appealing to some degree--after all, having A1-, A2- and
B-extractions with these (three) sets of properties is really not too
bad--I for one am not utterly convinced that this book offers the last word
on extractions.
Looking through chapter 1, most of which remarking already on the content
of chapter 2, I stumbled over example (8) on p. 4. This set of data,
(apparently) illustrating the syntactic differences between 'determine' and
'tell' (both in the general use of 'determine'), supposedly serves the
purpose to show why we need this distinction, and the relevance of the AC.
This example is explicitly not used in chapter 2, the more detailed
discussion, and when I first read it and consulted with native speakers I
thought I knew why: it's really not that clear (if true at all), a
characteristic that many of Postal's data share.
(8) a. We could easily determine/*tell it.
b. What we could easily determine/tell t was that Mike was a spy.
c. The first thing that we could determine/tell t was that Mike was
a spy.
d. That Mike was a spy we could easily determine/*tell t.
e. That, which I wish we had been able to determine/*tell t sooner,
is surprising.
f. It was that which we could immediately determine/*tell t.
Assuming the judgements in (8) as given, Postal's argument would run like
this. The verb 'determine' but not 'tell' may take a pronominal argument
(8a). A non-pronominal argument may be extracted, as in the specificational
pseudocleft (8b) or the restrictive relative clause extraction (8c). The
difference in (8a) suggests that 'tell' gives rise to an AC, suggesting
that anything extracted from here must be B-extraction; on the same token,
'determine' allows pronominal objects, hence must be a verb that allows
A-extraction contexts. The extraction differences in (8d-f) show that while
A- and B-extraction are fine in some instances, they are bad in others--and
this they are systematically, i.e. all instances of extraction in (8d-f)
are predicted to be grammatical for (at least one type of) A-extraction,
while ungrammatical for B-extraction. This, of course, conforms to the
generalization that every possible context for B-extraction is also a
possible context for an A-extraction, but not vice versa. Now, while for
many contexts this may really be the case, if one does not agree (at all)
with the contrast depicted in (8d,e), the force of the argument gets
somewhat diminished.
Postal sounds very convinced by these/his (?) judgements which, if they
were true, would make the case very strong from the start. However, my own
uneasiness about (8d,e) was confirmed by my two consultants, one of whom
comes from an area geographically very close to Postal's. In any case, I
dare to object to the ungrammaticality of 'tell' (in the 'determine'
meaning) in (8d) and (8e), and cases like this persist throughout the book.
Be it as it may, I believe I have summarized the main proposal of chapter 2
(with a little bit of help from chapter 1): islands are not relative as in
Cinque (1990) but absolute; constructions that apparently extract out of
islands involve prior extraction of a null resumptive pronoun; we have to
distinguish between extractions that require an RP (B-extractions) and
those that do not (A-extractions); of the latter, we have two types:
A1-extractions with RP and A2-extractions without RP. To use somewhat more
familiar terminology, A1-extractions include question extraction,
restrictive relative clause extraction, pseudoclefting, negative
NP-extraction, exclamatory extraction; A2-extractions include comparative
extraction, extraction associated with 'the more...the more' plus some
types of relative-like extractions; B-extractions include nonrestrictive
relative clauses, topicalization, clefting.
Turning to chapter 3 next, its main purpose seems to me to demonstrate that
the Coordinate Structure Constraint (CSC)--which was first proposed by Ross
(1967) and has since become known as the most stable and robust of all
island constraints--really holds. Here he turns to interesting data from
English casting doubt on the strength of the CSC provided by Lakoff (1986)
and others. The main tool for rebuking a weakening of the CSC comes from
the previous analysis of extractions and the role of RPs in licit
extraction from islands. Other cases Postal rejects are shown to not
involve coordination in the first place, hence taking the consequence for
the CSC out of the argument.
To briefly go over the three types of construction, consider first what
Postal calls "A-$" (where '$' shall represent the Greek letter 'sigma'
which in turn stands for 'scenario'), following Lakoff's (1986)
classification, after presenting a basic background for coordinate
structures and the CSC.
It is well known that coordinate structures form islands for extraction.
Thus, no proper subconstituent of a conjunct may be extracted, unless it
happens across-the-board. It thus comes as a surprise that cases like (9)
are grammatical, where the extractee does not have a corrssponding gap in
each conjunct (Postal's (11) and (13), p. 56).
(9) a. the cheese which Frank drove to the store, bought t, went home,
and gave t to Greta
b. The cheese which Frank went to the store but didn't buy t later
spoiled.
As (9b) shows, this type of apparent CSC-violation is not limited to the
conjunctor 'and' and as Postal shows in the following the recursive
possibilities of A-$ are as in standard coordinate structures, 'and' is
subject to the same constraints on (non)appearance in some conjuncts, and
due to the existence of ATB-extraction these cases must involve true
conjunction (Lakoff 1986, Postal: 56-57).
But Postal finds a number of (convincing) counter-arguments that seem to
suggest that cases such as (9) are not exactly of the same type as any
coordinate structures of which the CSC clearly holds. Among these, we find
that 'and' cannot be replaced by 'or' in A-$ as well as a bunch of
properties already noted by Ross (1967)--who incidentally attributes their
recognition to Lakoff (Postal: 58)--such as a constraint on the main verb
of the second conjunct to be non-stative or various tense restrictions.
Moreover, A-$ are only well-formed with VPs and they cannot contain the
quantifier 'both'.
After carefully teasing the different properties apart, Postal discusses
his selective islands again, this time with relevance to the CSC and then
accounts for A-$ conjuncts as selective islands. As such, the CSC can be
rescued to hold across-the-board (in the non-technical meaning) and the
apparent violations follow from the by now well-established account that
extraction from selective islands involves RPs and hence adhere to the same
constraints set by Ross (1967). (Needless to say that A2-extractions are
not permitted in A-$.)
Lakoff's (1986) "B-$" (from Goldsmith 1985) look slightly different, were,
however, argued by him to also involve true coordination and are
illustrated in (10) (Postal's (100) on p. 77):
(10) a. How many courses can we expect our graduate students to teach t
and (still) finish a dissertation on time?
b. How much can you drink t and still stay sober?
Postal then worries over evidence in favour or against a coordination
analysis as for A-$ but concludes this time that B-$ are not instances of
true coordination. Instead, Postal analyses the 'conjuncts' in these
constructions as adjuncts, so the CSC still holds.
The "C-$" cases, lastly, follow a similar path yet a different line of
reasoning. In a nutshell, "[i]f they are true coordinates, C-$s
counter-exemplify the CSC. Since there are good reasons for doubting their
coordinate status, though, at this stage there is little basis for thinking
they genuinely threaten that principle" (p. 92).
In sum, Postal defends the CSC as few people have ever since Ross (1967)
(see also fn. 3 on p.184). He is very aware of the limitations in this
chapter and admits to have pursued "an extremely limited goal" (p. 92). I
leave it at that, then.
Lastly, in chapter 4, the longest chapter, Postal is concerned with
establishing Right Node Raising (RNR) as an extraction on a par with all
types of L-extraction considered throughout the book. Here he takes issues
with critical views of the RNR as an extraction proper (such as
Wh-extraction, topicalization etc.). Postal further explores theoretical
tools from a specific framework, namely the Slash Category notation from
Generalized/Head-Driven Phrase Structure Grammar (GPSG and HPSG,
respectively). He concludes that RNR and Slash category approaches under
current conceptions are incompatible despite initial appeal.
To conclude this review, the reader will have noted that I reinterpreted
the book's title for the purposes of this review. Rather than giving equal
space to all "three investigations of extraction," I concentrated on the
three types of extraction (A1-, A2- and B-extraction) for the purpose of
exposition, while only briefly laying out the main claims of the other two
investigations (CSC and RNR). The chapter on the extraction types is very
detailed and highly informative with an interesting proposal, a proposal
which finds its way into the other two parts as well. The chapter on
coordination is a great introduction to the issues in general and certain
tricky cases in particular. The chapter on RNR is very nice from a
minimalist perspective as arguments for any direction concerning these
still tricky constructions are brought into the discussion. Anyone with
only remote interest in Kayne's (1994) framework is advised to read this
chapter carefully, regardless of certain framework-specific discussions
(which should be useful to any open-minded theoretical syntactician these
days). Speaking of open-mindedness, as appealing as Postal's general
non-committal stance may seem, it is unfortunate that his general
anti-Chomskyan attitude comes through at times and it is obvious that not
too many recent (minimalist) treatments of certain issues were considered.
Especially Appendix C is entertaining to read and should boost young
scholars' confidence to deal with reviewers' comments. If I'm in any
position to say so, I encourage everyone who's interested in syntactic
theory to read this book carefully and admire the methodology at least in a
general sense.
Acknowledgements:
I'm grateful to Jennifer Graham and Alan Munn for discussion and evaluation
of data.
Bibliography:
Chomsky, Noam. 1981. Lectures on Government and Binding. The Pisa Lectures.
Dordrecht: Foris.
Cinque, Guglielmo. 1990. Types of A-bar Dependencies. Cambridge, Mass.: MIT
Press.
Goldsmith, John. 1985. A principled exception to the Coordinate Structure
Constraint. In W.H. Eilfort, P.D. Kroeber and K.L. Peterson (eds.), CLS 21.
Part 1, Papers from the General Session. Chicago Linguistics Society.
Chicago, Ill.: University of Chicago.
Kayne, Richard S. 1984. Connectedness and Binary Branching. Dordrecht: Foris.
Kayne, R.S. 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press.
Lakoff, George. 1986. Frame semantic control of the Coordinate Structure
Constraint. In A.M. Farley, P.T. Farley and K.-E. McCullough (eds.), Papers
from the Parasession on Pragmatics and Grammatical Theory. Chicago
Linguistics Society. Chicago, Ill.: University of Chicago.
Obenauer, Hans-Georg. 1984. On the identification of empty categories. The
Linguistic Review 4, 153-202.
Postal, Paul M. 1974. On Raising. Cambridge, Mass.: MIT Press.
Rizzi, Luigi. 1990. Relativized Minimality. Cambridge, Mass.: MIT Press.
Rizzi, Luigi. 1992. Argument/adjunct (a)symmetries. Proceedings of NELS 22.
Amherst, Mass.: GLSA, 365-381.
Ross, John R. 1967. Constraints on Variables in Syntax. Doctoral
dissertation, Massachusetts Institute of Technology. [Appeared 1986 as
Infinite Syntax! Norwood, N.J.: Ablex.]
Biographical information:
Kleanthes Grohmann is currently a graduate student at the University
of Maryland, College Park. I work on syntactic theory and comparative
syntax, concentrating on left-peripheral issues in a variety of languages,
with a strong emphasis on German(ic). I'm currently working on my PhD
thesis entitled "Prolific Peripheries: A Radical View from the Left." For
more information, visit my web-site at http://www.wam.umd.edu/~grohmann/ling.
**************************************
Kleanthes K. Grohmann
ZAS, Jaegerstr. 10/11
D-10 117 Berlin
phone: ++ 49-30-20 192 561
fax: ++ 49-30-20 192 402
email: grohmann@zas.gwz-berlin.de
http://www.wam.umd.edu/~grohmann/ling
|
