Relative Constructions and Unbounded Dependencies
Relative Constructions and Unbounded Dependencies
Relative Constructions and Unbounded Dependencies
1031
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1032
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1033
1 Terminological preliminaries
This chapter deals with what are traditionally called relative clauses. We use the more
general term relative constructions because although it is reasonable enough to call the
underlined construction in [1i] a relative clause, the term is misleading for the type of
construction seen in [1ii]:
[1] i I agree with most of the things that your father was saying. [clause]
ii I agree with most of what your father was saying. [NP]
These two sentences are equivalent. But the phrase what your father was saying in [ii] is
an NP: it corresponds not to the relative clause that your father was saying in [i], but to
the larger NP containing it, the things that your father was saying. And we will see that
there are syntactic as well as semantic reasons for treating what your father was saying in
[ii] as an NP.
We therefore use the term ‘relative constructions’ to cover both the underlined se-
quences in [1], with ‘relative clause’ available as a more specific term applying to cases
like [i]. Often, however, we will talk simply of ‘relatives’, leaving ‘construction’ or ‘clause’
understood.
This section presents an overview of the different types of relative construction that will
be discussed in detail in subsequent sections. The two major dimensions of contrast
yield what we will call formal types and relational types.
The formal types are distinguished according to whether they contain one of the
special relative words who, which, etc., or the subordinator that, or simply a ‘gap’, a
missing constituent.
The relational types are distinguished on the basis of their external syntax, their
relation to the larger construction containing them. The traditional distinction between
restrictive and non-restrictive relative clauses fits in here, but we shall use different terms
and contrast them with two further categories, cleft and fused relatives.
In addition to these major contrasts, we need to invoke the more general distinction
of finiteness: while most relative constructions are finite, infinitivals (and certain minor
types) are also possible under certain conditions.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1034 Chapter 12 Relative constructions and unbounded dependencies
1
Bare relatives are sometimes called ‘contact clauses’.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 2.2 The relational types 1035
modifies, and is often referred to by the term ‘restrictive relative’. The set of boys who
defaced the statue, for example, is smaller than the set of boys; here the information
expressed in the relative clause is an integral part of that expressed by the matrix clause
in that it delimits the set of boys under discussion.
ii It was Kim who wanted Pat as treasurer. (=[2iii])
[cleft]
iii It was Pat that Kim wanted as treasurer.
Example [i] is an ordinary, non-cleft, clause, while [ii] and [iii] are cleft counterparts of
it, and the underlined clauses are the cleft relatives, differing in function and, in certain
respects, their internal structure from integrated relatives. The cleft construction is so
called because it divides the more elementary construction into two parts, one of which
is foregrounded and the other backgrounded. In [ii] Kim is foregrounded and wanted
Pat as treasurer backgrounded, whereas in [iii] the foregrounded element is Pat, with
Kim wanted as treasurer backgrounded. The cleft construction is dealt with in Ch. 16, §9,
and in the present chapter will be mentioned only incidentally.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1036 Chapter 12 Relative constructions and unbounded dependencies
From a syntactic point of view this is the most complex of the four relative constructions.
With the others we can separate a relative clause from its antecedent, but this is not
possible with the fused construction. Compare, for example:
[6] i It would mean abandoning that which we hold most dear. [antecedent + clause]
ii It would mean abandoning what we hold most dear. [fused relative]
These are semantically equivalent (though [i] belongs to very formal style). Syntactically,
that in [i] is antecedent, with which we hold most dear an integrated relative clause
modifying it, but in [ii] what corresponds to that and which combined, so that it is not
possible to separately identify antecedent and relative clause – hence the term ‘fused’.
While the fused relatives in [5] are NPs, those based on where and when are PPs:
[7] i Put it back where you found it.
ii He still calls his parents whenever he is in trouble.2
Because the fused relative construction is so different from the integrated, supplemen-
tary, and cleft relative clause constructions, we will treat it separately, deferring further
consideration of it until §6.
2.3 Finiteness
The great majority of relative constructions are finite, but with integrated relatives we
find infinitivals of the wh type and corresponding ones without a relative word. The
underlined parts of [8] are infinitival relative clauses:
[8] i She found a good place from which to watch the procession.
ii She found a good place to watch the procession from.
There are various other non-finite constructions which bear some resemblance to
relatives, such as gerund-participials and past-participials that modify nouns (anyone
knowing his whereabouts, those killed in the accident): these constructions are discussed
in Ch. 14, §9.
A relative clause, we have said, contains within its structure an overt or covert element
that relates it anaphorically to an antecedent. Other kinds of clause may also contain
anaphoric elements, of course. In I lent Jill my bicycle last week [and she hasn’t returned
it yet], for example, she and it in the second clause are anaphorically related to Jill and
my bicycle in the first. Here, however, the anaphoric relation is incidental: there is no
anaphora in I lent Jill my bicycle last week [and now there’s a bus strike], but we still
have the same syntactic construction, a coordination of main clauses. In relative clauses,
by contrast, the anaphoric relation is an essential feature of the construction. What
distinguishes relatives from other clauses is the specific nature of the anaphoric relation
2
Other terms found in the literature corresponding to our ‘fused relative (construction)’ are ‘free relative’,
‘headless relative clause’, and ‘nominal relative clause’. Terms incorporating ‘clause’ are unsatisfactory for the
reasons we have given. In addition, ‘nominal’ is insufficiently general in that it doesn’t cater for prepositional
examples like those with where or when. And ‘headless’ is misleading in our view since the head of the containing
phrase is not missing but fused with part of the modifying clause.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 3.1 Relativisation 1037
involved. In the central case of the integrated relative, the antecedent is the head that the
clause modifies, and in all cases the anaphoric element itself has distinct properties. In
wh relatives the overt forms who, which, etc., are distinct from the anaphoric forms that
are used in main clauses; they are homonymous with interrogative words, but the latter
are not anaphoric. In non-wh relatives the anaphoric element is a gap, but this too is
distinct from other kinds of anaphoric gap with respect to the positions in which it can
occur and the way it is interpreted.
3.1 Relativisation
The essential anaphoric element in a relative clause we call the relativised element. It is
primarily in respect of this element that the relative differs in form from a comparable
main clause. Consider first the case where the relativised element is subject:
[1] i A letter drew our attention to the problem. [main clause]
ii This is the letteri [whichi drew our attention to the problem]. [wh relative]
iii This is the letteri [that i drew our attention to the problem]. [that relative]
The main clause in [i] has the ordinary NP a letter as subject. In the wh relative the
subject is which, a relative pronoun anaphorically linked to the antecedent letter, as
indicated in [ii] by the identical subscripted indices. In [iii], where that marks the clause
as subordinate, the subject position is empty, but there is still an anaphoric link to the
antecedent letter, which we indicate by attaching the same index to the symbol marking
the gap. The meaning in both [i] and [ii], ignoring the definite article, can be given
roughly as “This is letter x ; x drew our attention to the problem”. This kind of meaning,
with two occurrences of a single variable, is an essential and distinctive feature of all
relative constructions.
We take the antecedent of which and the gap to be letter, not the letter, since the enters
into construction with the whole nominal letter which/that drew our attention to the
problem: it is this, not letter by itself, that is presented as an identifying description that
sanctions the definite article (see Ch. 5, §6.1).
In [1iii], as in almost all cases of subject relativisation in non-wh relatives, that is
non-omissible: there is no bare relative counterpart ∗This is the letter drew our attention
to the problem. But where it is the object that is relativised that is optional, so we have all
three types:
[2] i My neighbour gave me some advice. [main clause]
ii I accepted the advicei [whichi my neighbour gave me]. [wh relative]
iii I accepted the advicei [that my neighbour gave me i ]. [that relative]
iv I accepted the advicei [my neighbour gave me i ]. [bare relative]
As in [1], the main clause in [2i] has an ordinary NP as direct object, whereas the relative
clauses do not. The wh relative [ii] again has which as relative pronoun, while the non-wh
versions simply have a gap in object position. And as before the meaning involves two
occurrences of a variable: “I accepted advice x ; my neighbour gave me x ”.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1038 Chapter 12 Relative constructions and unbounded dependencies
is that while in [2iii–iv] the gap is related directly to the antecedent advice, in [2ii] it is
related indirectly, via which. Example [2ii] can thus be represented as in [3], with the
relative clause having the structure shown in [4]:
[3] I accepted the advicei [whichi my neighbour gave me i ].
[4] Clause
Prenucleus: Nucleus:
NPi Clause
Subject: Predicate:
NP VP
3
In the interrogative case a preposition (of or as to) would often appear after the head noun.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 3.2 Relative words and phrases, and upward percolation 1039
[6] They rejected the idea [that we had advanced]. [relative or content clause]
In the relative interpretation there is a relativised object, a gap, with idea as antecedent. We
had advanced some idea: “They rejected idea x ; we had advanced x”. In the content clause
interpretation, on the other hand, that introduces an ordinary clause (a complement to
the noun idea): the meaning is that they rejected the proposition “We had advanced”.
There are other factors that distinguish relative clauses from content clauses. In NP
structure relatives function as modifiers, whereas content clauses are complements, com-
patible only with a fairly small subclass of noun. Our examples contain suggestion, ques-
tion, idea: others are fact, news, belief, concern, proposal, etc., but innumerable nouns like
cat, boy, health, energy allow relatives but not content clauses. Secondly, content clauses
do not allow the alternation between the wh and that constructions found with relatives:
for example, if we replace that by which in [6] it becomes unambiguously relative. In
content clauses, that marks declarative clause type, while which, who, etc., occur only in
open interrogatives, licensed by such nouns as question.
ii We’ve never met the people [whose house we are renting ].
[complex]
iii We admired the skill [with which she handled the situation ].
In [ii–iii], as in almost all the complex cases, the relative phrase contains more than just
the relativised element. In [ii], for example, the relative phrase in prenuclear position is
the NP whose house, but it is just the genitive determiner within this NP, i.e. whose, that
constitutes the relativised element: it is whose that derives its interpretation from the
antecedent people. The gap in object position in the nucleus is thus not co-indexed with
the antecedent, as it is in [i]. Rather, we have two co-indexed pairs of different extents,
with whose linked to people, and the gap linked to whose house, for it is whose house that
is understood as object of renting. Similarly in [iii], except that the gap is anaphorically
linked to a PP rather than an NP. The anaphoric links can thus be shown as in [8], where
the outer brackets in [ii–iii] enclose the relative clause, and the inner ones the complex
relative phrase:
[8] i I can’t find the booki [whichi he recommended i].
ii We’ve never met the peoplei [[whosei house] j we are renting j ].
iii We admired the skilli [[with whichi ] j she handled the situation j ].
Non-wh relatives do not contain a relative phrase, and consequently there can be no
non-wh relatives matching wh relatives like [7ii–iii]:
[9] i I can’t find the book [that he recommended ].
ii ∗We’ve never met the people [that’s house we are renting].
iii ∗We admired the skill [with that she handled the situation].
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1040 Chapter 12 Relative constructions and unbounded dependencies
In describing the range of permitted complex relative phrases we will invoke the
metaphor of ‘upward percolation’. In [7ii], for example, the relative feature percolates
upwards from the genitive determiner whose to the matrix NP whose house and in [7iii]
it percolates upwards from the NP which to the PP with which. We invoked the same
metaphor in Ch. 10, §7.9, in describing the structure of complex interrogative phrases,
but the phenomenon is considerably more extensive in the case of relatives. Seven types
of upward percolation may be distinguished in relative clauses, five involving percolation
from the element on the right, two from the one on the left:
[10] type percolation from to example
i comp of preposition PP behind which
ii PP NP the result of which
iii PP AdjP prominent among which
iv NP infinitival to refute which
v NP gerund-participial passing which
vi genitive whose NP whose essay
vii determinative which NP which suggestion
We will examine each of these in turn. Where appropriate, we will contrast the ex-
amples with main clause constructions to show the basic, non-relative, form of the
phrase.
ii [behind which Kim was hiding] [Type i applied]
the curtain
iii [which Kim was hiding behind] [Type i not applied]
In [ii] the relative phrase is the PP behind which, with percolation of the relative feature
from the NP which to the PP in which it is complement; the result is that the preposition is
fronted along with its complement. In [iii] the relative phrase is just which, and fronting
this time affects only the complement of the preposition, the latter being left stranded.
The factors favouring, or in some cases requiring, one or other of these structures are
discussed in detail in Ch. 7, §4.1.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 3.2.2 Type II 1041
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1042 Chapter 12 Relative constructions and unbounded dependencies
In general, then, material will not be fronted if fronting it would leave the nucleus
following the relative phrase with too little significant content:
[17] i a. They are striving to explain phenomena [of which we have little or no direct
knowledge].
b. #They are striving to explain phenomena [little or no direct knowledge of which
we have].
ii a. Her first loyalty is to the programme [of which she is director].
b. ∗Her first loyalty is to the programme [director of which she is].
The [b] examples are unacceptable because of the radical imbalance between the content
of the relative phrase in prenuclear position and that of the following nucleus. In [iib]
director is head of the predicative complement, and it is doubtful if upward percolation
of Type ii could ever apply into an NP in predicative complement function, certainly
where the verb is be : we have accordingly marked it as ungrammatical, not merely
infelicitous.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 3.2.4 Types IV and V 1043
3.2.4 Types IV and V: from NP to non-finite (to refute which, passing which)
Type IV: infinitivals – supplementary relatives only
[20] i I felt the need of a better knowledge of Hebrew and archaeology to refute a higher
criticism of the Bible. [main clause]
ii I became disturbed by a ‘higher criticism’ of the Bible, [to refute which I felt the need
of a better knowledge of Hebrew and archaeology]. [Type iv]
This type is rare and very largely confined to purpose adjuncts and catenative comple-
ments that are semantically somewhat similar – e.g. to please whom he had striven so
hard, but not ∗to please whom he had wanted so desperately (only whom/who he had so
desperately wanted to please).
Type iv combines (obligatorily) with Type i when the infinitival is complement of
the preposition in order :
[21] Here is Dr Van Buren, [in order to interview whom Phelps says he was prepared to
fly to Copenhagen].
Type V: gerund-participials – supplementary relatives only
[22] They take a rigorous examination, [passing which confers on the student a virtual
guarantee of a place at the university].
This is again very rare and also highly formal in style – except in the expressions speak-
ing/talking of which/whom, used to indicate the topic of what follows.
ii [whose essay he plagiarised] [Type vi]
the student ∗
iii [whose he plagiarised essay] [Type vi not applied]
Type vi percolation can combine with i, and hence also with ii:
[24] i I hadn’t yet met the people [in whose house I would be staying].
ii She was lecturing on Tom Roberts, [an exhibition of whose work can currently be
seen at the National Art Gallery].
iii You sometimes find yourself unable to describe the physical appearance of someone
[with the very texture of whose thought you are familiar].
The steps involved here are, respectively: vi + i; vi + i + ii; vi + i + ii + i.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1044 Chapter 12 Relative constructions and unbounded dependencies
Determinative which is not itself a phrase, and cannot be separated from the head on
which it is dependent, as evident from [iii]. We take the relativised element in [ii] to be
the whole object NP, which suggestion: this is the phrase whose interpretation is given
by the antecedent (it might be more efficient to hold the meeting on Saturday morning),
though the presentation of it as a suggestion is of course contributed by the relative
clause rather than being inherent in the antecedent itself. The double-variable gloss will
thus be: “I said x (=it might be more efficient to hold the meeting on Saturday morning);
they all enthusiastically endorsed suggestion x”. This is the only type of upward perco-
lation which does not yield a relative phrase that is larger than the relativised element.
Type vii percolation is found only in supplementary relatives. Further examples are
seen in:
[26] i They refuse to support the UN’s expenses of maintaining the UN Emergency Force
in the Middle East as a buffer between Egypt and Israel, and the UN troops in the
Congo, [which expenses are not covered by the regular budget].
ii I may be late, [in which case I suggest you start without me].
iii I will return at 3pm, [by which time I expect this room to be tidy].
iv Both horses, broken and trained by different trainers, were blundering jumpers until
they were seven, [at which age they began to outgrow their carelessness].
v She has to comment on him standing there, and later, when the soldiers march away,
has to tell him not to move yet – [neither of which remarks should be so obtrusive
that the soldiers might notice them, but both of which should be clearly heard by the
audience].
Examples like [i], where the NP concerned is itself an element of clause structure, are
quite rare and formal, verging on the archaic. It is much more usual for the NP to be
complement of a preposition which is also fronted, and the head noun is then predomi-
nantly one of very general meaning such as case or time, as in [ii–iii]. In the last example
the upward percolation involves three steps: vii (which remarks), i (partitive of which
remarks), and ii (the whole NP).
(a) Subject
[27] i A man came to dinner.
ii The mani [whoi came to dinner]turned out to be from my home town.
(b) Object
[28] i a. She received a letter from the Governor. [direct object]
b. This is the letteri [that she received i from the Governor].
ii a. He showed a student the exam paper. [indirect object]
∗
b. The studenti [whomi he showed i the exam paper]informed the police.
Relativisation applies to direct objects but not normally to indirect ones.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 3.3 What can be relativised 1045
The gap in [ii] is functioning as adjunct in its clause, like one day in [i]. A selection
of the major types of adjunct or complement we are concerned with here is supplied
in [32]:
[32] i It was a time in my lifei [wheni everything seemed to be going right i ]. [time]
ii I’ve finally found somewherei [wherei I can work undisturbed i ]. [location]
iii They want to go to the placei [wherei they went last year i ]. [goal]
iv I shall go back the wayi [I came i ]. [path]
v Look at the wayi [he tackled the job i ]. [manner]
vi That’s not really the reasoni [she left him i ]. [reason]
This sort of case is to be distinguished from (d) above: here it is the whole adjunct or
complement that is relativised, whereas in (d) what is relativised is just the NP func-
tioning as complement within a PP. In [30ii], for example, what is relativised is not the
instrumental adjunct itself, but just the complement of with. Case (d) is in fact broader
than (e) since it can apply with virtually the full range of prepositions. For most of the
categories in [32] there is an alternant of type (d): for example, [v] alternates with the
way in which he tackled the job. We look further at such alternations in §3.5.4.
Two extensions of this kind of construction should be noted.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1046 Chapter 12 Relative constructions and unbounded dependencies
Here the infinitival is in extraposed subject function. For further discussion of the con-
structions illustrated in [36], see Ch. 17, §§7.1–3.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 3.5 The formal types: wh, that, and bare relatives 1047
In [ib] the gap is object of the relative clause itself, whereas in [iib] it is object of the
content clause functioning as complement of think – the outer pair of brackets enclose
the relative clause, while the inner pair enclose the content clause embedded within it.
It is by virtue of this possibility that relative clauses belong to the class of unbounded
dependency constructions which we shall be examining in §7.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1048 Chapter 12 Relative constructions and unbounded dependencies
Example [i] belongs to the construction illustrated in [36ib], with relativisation of the
complement of an auxiliary verb. The complement in this case is a predicative, but we
cannot relativise the predicative complement of a lexical verb such as seem in this way:
∗
She thinks he’s a fool, which indeed he seems. The relativised predicative in the which
construction will generally be of the ascriptive type, as in this example: the clause is
concerned with the person’s properties, what kind of a person he was, not his identity.
Who would be impossible here, but is used in the integrated relative clause of [ii], with
be used in its specifying sense: the issue is the identity of the person (I thought he was
person x, but he turned out not to be). Note that in both [i] and [ii] the antecedent is in
predicative complement function as well as the pronoun.
4
Ch. 5 (§16.2.3) also deals with the choice between nominative who and accusative whom.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 3.5.2 Whose 1049
3.5.2 Whose
Used with both personal and non-personal antecedents
The contrast between personal who and non-personal which is neutralised in the
genitive, where whose is the only form. It occurs with both personal and non-personal
antecedents:
[47] i She started a home for women [whose husbands were in prison]. [personal]
ii The report contains statements [whose factual truth is doubtful]. [non-personal]
Alternation with of construction
We have seen (Ch. 5, §16.5.2) that with non-relatives, a genitive determiner characteris-
tically alternates with a construction containing the + post-head of phrase:
[48] i a. The child’s parents were constantly quarrelling.
b. The parents of the child were constantly quarrelling.
ii a. The house’s roof had been damaged in the storm.
b. The roof of the house had been damaged in the storm.
The same alternation is found with whose, except that here we have two versions of the
of construction, one with the of PP in post-head position, one where it is separated from
the head:
[49] i a. a child [whose parents were constantly quarrelling] [genitive]
b. a child [the parents of whom were constantly quarrelling] [post-head of PP]
c. a child [of whom the parents were constantly quarrelling] [separated of PP]
ii a. a house [whose roof had been damaged in the storm] [genitive]
b. a house [the roof of which had been damaged in the storm] [post-head of PP]
c. a house [of which the roof had been damaged in the storm] [separated of PP]
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1050 Chapter 12 Relative constructions and unbounded dependencies
In the post-head of PP construction the relative phrase is the whole NP, so that there
is upward percolation of Type ii, from PP to NP. In the separated of PP version the
relative phrase is just that of PP, with Type ii percolation not applying.
With non-relatives the genitive alternant is more likely with personal nouns than with
non-personals, and this general tendency applies with relatives too, where it may well
be strengthened by the morphological resemblance between whose and who: the great
majority of instances of whose have personal antecedents. In [49i], therefore, the genitive
alternant [a] is much the most likely of the three. With non-personal antecedents one
or other of the of constructions will often be preferred, but it must be emphasised that
genitives like [49iia] are completely grammatical and by no means exceptional.5 One
genre where they occur very readily is scientific writing: examples like a triangle whose
sides are of equal length are commonplace.
Where
[51] i She wanted to see the housei [wherei she had grown up].
ii They met in the journalists’ clubi , [wherei he went every Sunday afternoon].
iii She often climbed the knoll behind the missioni , [from wherei she could look down
on roofs and people].
Where takes locative expressions as antecedent; within the relative clause it functions as
adjunct of spatial location, goal complement, or complement of a locative preposition.
A ‘double-variable’ representation of [i] is “She wanted to see house x ; she had grown
up in x”: the “in” component is contributed by where together with its spatial location
function, with the antecedent determining the value of the variable x. In [ii] we under-
stand “to x”, with the “to” component derivable from the goal function. And in [iii]
we have “from x ”, with the “from” component overtly expressed. Analogously for the
examples below.
5
It is interesting to note that a number of usage manuals feel it necessary to point out that relative whose can have
a non-personal antecedent: there are apparently some speakers who are inclined to think that it is restricted
to personal antecedents.
6
In traditional grammar, all these are classified as adverbs; in the present grammar, we take why as an adverb
and the others as prepositions: see Ch. 7, §2.4.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 3.5.3 Other relative words 1051
When
[52] i It happened at a timei [wheni I was living alone].
ii In those daysi , [wheni he was still a student,] he used to babysit for us.
iii He left college in 1982 i , [since wheni I’ve only seen him twice].
When takes a temporal expression as antecedent; it generally functions as an adjunct of
temporal location within the relative clause, but it can also appear as complement to a
temporal preposition such as since.
While
[53] i From 1981 to 1987 i , [whilei his uncle lived with them,]she had a full-time job.
%
ii He wrote most of his poetry during the yearsi [whilei he was in Paris].
Relative while is mostly found in the fused construction, but it can occur in supplementary
relative clauses and, for some speakers, in integrated ones. The antecedent denotes a
period of time, and while can be replaced by when or during/in which (time).
Why
Relative why is used in a very narrow range of constructions – integrated relatives with
reason as antecedent:
[54] i That’s the main reasoni [whyi they won’t help us].
ii There was no reasoni [whyi he should stay at the dance any longer].
iii I can’t see any reasoni [whyi you shouldn’t have a little fun].
The majority of examples are of the types shown in [i–ii]: either the specifying be
construction, where it is a matter of identifying reasons, or the existential construction,
where we’re concerned with the existence of reasons. Why alternates with for which, as
in the attested example The Physical Training and Recreation Act of 1937 deals with the
acquisition of playing fields, which may not be absolutely the reason for which an authority
would wish to acquire property. For which, however, is comparatively rare and formal: it
could not idiomatically replace why in ordinary examples like [54].
Whence
[55] i He sent his son with the papers to another congressman’s housei , [whencei they were
spirited to a governor].
ii But this means that the Taniyama-Shimura conjecture is truei , [whencei it follows
that Fermat’s Last Theorem is true].
Whence belongs to formal style, serving in its primary sense to express spatial source,
as in [i]. The “from” meaning can be incorporated in whence or expressed separately
by the preposition from, which is obligatorily fronted. This use is in general somewhat
archaic, though it is still found in journalistic writing. Whence is also used for logical
source, normally in supplementary relatives, as in [ii]; this is the relative counterpart of
the most common use of hence.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1052 Chapter 12 Relative constructions and unbounded dependencies
archaic and rare, though whereby and to a lesser extent wherein and whereupon are still
regularly used:
[56] i His Lordship might make an orderi [wherebyi each side would bear its own costs].
ii Size segregation occurs when a powder is poured into a heapi , [wherebyi the larger
particles run more easily down the slope of the heap].
iii Try to imagine a marketi [whereini the majority consistently wins what the minority
loses].
iv She told him his essay was incoherenti , [whereuponi he tore it up and stormed out
of the room].
In integrated relatives whereby is equivalent to by which, with by having approximately
the “means” sense: typical antecedent nouns are agreement, arrangement, mechanism,
method, plan, proposal, scheme, service, suggestion, etc. In supplementary relatives it can
also occur with a clause as antecedent, as in [ii]. Wherein is equivalent to in which. Where-
upon means approximately “immediately after which”; it is found only in supplementary
relatives whose antecedent is a clause (or larger).
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 3.5.4 Wh vs non-wh constructions 1053
[59] i They won’t register companies [whose directors are undischarged bankrupts].
ii It’s a burden [of which they will never be free ].
Where upward percolation is optional, as in [ii], there will be an alternant with a simple
relative phrase (It’s a burden [which they will never be free of ]) and if relevant con-
ditions are satisfied this will have a non-wh counterpart (It’s a burden [they will never be
free of ]).
7
The restriction to wh relatives does not apply when where is complement to stranded at : the hotel where/(that)
we stayed at last year. Where . . . at seems to be a blend between where and which . . . at ; note that with in we
can have which but not where : the hotel which/∗where we stayed in last year.
8
Some non-standard dialects differ; hence the line !It ain’t what you do, it’s the way how you do it in a rock ‘n’
roll song.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1054 Chapter 12 Relative constructions and unbounded dependencies
This case covers the compound determinatives (anything, everything, nothing, something)
and non-personal fused determiner-heads all, much, most, few, little, some, any, etc. There
is a preference for the non-wh type here, but of varying strength, with everything which,
for example, significantly better than ?all which.
(i) Complexity
Increasing the distance between the relative clause and the head noun, notably by adding
other post-head modifiers, favours the wh type (just as, within the non-wh type, it favours
that over a bare relative). Thus a material of great tensile strength and very remarkable
electroconductive properties which has been widely used in the aviation industry is preferred
over the version with that in place of which.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 3.5.5 Non-wh relatives: presence or absence of that 1055
obtain a grammatical that relative. However, the converse does not hold: under certain
very limited conditions, the subordinator cannot be omitted from a that relative without
loss of grammaticality.
The relativised element is subject
That cannot normally be omitted if the relativised element is subject of the relative clause:
[67] non-subject subject
i a. The car [that I took ]was Ed’s. b. The car [that hit us] was Ed’s.
ii a. The car [I took ]was Ed’s. b. ∗The car [ hit us]was Ed’s.
The [a] cases represent the default: that can be omitted from [ia] to produce the gram-
matical [iia]. In this example the relativised element is object, but any other non-subject
would similarly allow omission of that: He’s not the man (that) he was a few years ago
(predicative complement); I can’t find the book (that) you asked for (complement of
preposition); He’s the one (that) they think was responsible for the first attack (embedded
subject), and so on. In all these cases that precedes the subject, but when the subject itself
is the relativised element, and hence missing, that must be retained, as in [b].
The prohibition on dropping the that with relativised subjects is associated with the
need to distinguish the subordinate relative clause from the matrix predicate. Since hit
us in [ib] immediately follows the car, there is nothing to stop the listener construing
hit us as the main clause predicate, with the car as its subject: that prevents such a
misconstrual by explicitly signalling the start of a subordinate clause. This is not to
suggest that there would always be a danger of misconstrual if that were omitted from
clauses with a relativised subject. In ∗We didn’t take the number of the car hit us, for
example, the car is complement of of and hence not a possible subject for a predicate
hit us. The grammatical restriction preventing subject relativisation with bare relatives
covers a wider range of cases, but the point is that it includes those where that serves a
role in aiding perception of the structure.
Some varieties of English do allow that to be omitted from clauses with relativised
subjects under certain conditions:
[68] i ?It was my father [ did most of the talking]. [it-cleft]
ii ?There’s someone at the door [ wants to talk to you]. [existential]
iii !Anyone wants this can have it.
Most such cases are clearly non-standard, like [iii]. The status of [i–ii], where the relative
clause functions within an it-cleft and existential construction respectively, is less certain:
they fall at the boundary between very informal and non-standard. Note that the position
of the relative clause in [i–ii] is such that it could not be misconstrued as predicate of
the matrix clause.
That not omissible when not adjacent to the subject
A second, less important, exception to the optionality of that is seen in examples like:
[69] I found I needed a file [that only the day before I had sent to be shredded].
That is needed here to mark the beginning of the subordinate clause: without it there
would be the potential for the following adjunct to be misconstrued as belonging in the
matrix clause. Bare relatives always have the subject in initial position.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1056 Chapter 12 Relative constructions and unbounded dependencies
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 3.5.6 That as a subordinator (not a relative pronoun) 1057
It would not only cover the ground of all the simple ‘wh’ words put together, as shown in
[i–v]: it would also appear in a variety of constructions where no ‘wh’ word could replace
it, as in [vi–viii]. Particularly important here is the cleft construction shown in [vii], and
in [70iii–iv] above. Note that, leaving aside the disputed case of the relative construction,
there is no pro-form in English that takes as antecedent such complements and adjuncts as
to you (in the sense it has in [71vii]), with considerable misgivings, in order to avoid this kind of
misunderstanding, and the like. Instead of postulating a pro-form with such an exceptional
range of use, we are saying that that relatives do not contain any overt pro-form linked to the
antecedent: they simply have an anaphoric gap, like bare relatives.
(c) Finiteness
That relatives are always finite, as are the declarative content clauses introduced by that. Note,
then, that we cannot insert that into non-wh relative infinitivals like a knife to cut it with – cf.
∗
a knife that to cut it with. If that were a pronoun this would be a special fact needing
explanation, but under the subordinator analysis it is exactly what we would expect, given
that that is a finite clause subordinator.10
(d) Omissibility
As we have noted, that can be regarded as very largely omissible in relative clauses in the same
way as in declarative content clauses. The conditions under which omission is prohibited are
not the same in the two cases (those for content clauses are given in Ch. 11, §3.1), but in both
they have it in common that they are related to the need to mark explicitly the beginning of
a subordinate clause under certain structural conditions. And in both cases, moreover, that
is more readily omitted in simple structures than in complex ones. There is no pro-form in
English that is systematically omissible under remotely similar conditions.
9
There are non-standard regional dialects of English in which that’s does occur, as in the man that’s leg was
broken. We do not believe that such examples necessitate a pronoun analysis for the dialects concerned, and
certainly they do not establish this analysis as valid for all dialects.
10
The force of this argument is diminished by the fact that which can’t occur here either: we have a knife with
which to cut it, not ∗a knife which to cut it with. The absence of ∗a knife with that to cut it is then already covered
under point (b). Nevertheless, the analysis of that as a finite clause subordinator does provide a very general
account of why the only type of bare relative that can’t be expanded by means of that should be the infinitival
one.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1058 Chapter 12 Relative constructions and unbounded dependencies
(b) Syntax
An integrated relative clause usually functions as modifier within the structure of an NP.
Those in [1ia/iia], for example, are constituents of the NPs every student who had lent
money to the victim and the necklace which her mother gave to her. Note that every student
and the necklace do not themselves constitute NPs in these examples.
The syntactic structure of sentences containing supplementary relatives is less clear:
the relative clauses are only loosely incorporated into the sentence. In [1ib/iib] Jill and
the necklace constitute NPs by themselves, but the supplementary relatives do not com-
bine with them to form larger NPs. We suggest in Ch. 15, §5.1, that the antecedent +
relative clause here is a special case of a supplementation construction, which is dis-
tinct from a head + dependent construction. The supplement is in construction with an
anchor (in this case the antecedent), but does not combine with it to form a syntactic
constituent.
(c) Meaning
The content of an integrated relative is presented as an integral part of the meaning of
the clause containing it, whereas the content of a supplementary relative is presented as a
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 4.1 Major syntactic differences 1059
separate unit of information, parenthetical or additional. We will see that there can be a
range of reasons why the content of a relative should be presented as integral to the larger
message, but our initial examples illustrate two very obvious cases. In [1ia] dropping the
relative would drastically change the meaning: I would be saying that they wanted to
interview every student, not just those who had lent money to the victim. And [1iia]
implicates that there was more than one necklace, so if the relative were dropped it would
be unclear which one I was referring to. The supplementary relatives here, by contrast,
can be omitted without affecting the meaning of the remainder. Example [1ib] says that
they interviewed Jill, and it would still say that if we dropped the relative. Example [1iib]
says that the necklace was still in the safe, where I assume the necklace I’m referring
to is identifiable in the context, and again the same would hold if the relative were
dropped.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1060 Chapter 12 Relative constructions and unbounded dependencies
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 4.1 Major syntactic differences 1061
The integrated relative construction is recursive: an integrated relative can combine with
its antecedent to form a larger unit which is antecedent for a second integrated relative. In
[i], for example, you wear combines with its antecedent ties to give ties you wear and this
is then the antecedent for the second relative, that your sister knits for you. This kind of
recursion is known as stacking. It is limited to the integrated construction: antecedent +
supplementary relative cannot serve as antecedent for a second supplementary relative,
as illustrated in [ii].
(e) Non-declaratives and question tags found only with supplementary relatives
[10] i He said he’d show a few slides towards the end of his talk, at which point please
remember to dim the lights.
ii It may clear up, in which case would you mind hanging the washing out?
iii She may have her parents with her, in which case where am I going to sleep?
iv I didn’t get much response from Ed, who seemed rather out of sorts, didn’t he?
Relative clauses mostly belong to the default declarative clause type, but with supplemen-
tary relatives other clause types are possible. Those in [i–iii], for example, are respectively
imperative, closed interrogative, and open interrogative. And declaratives can have ques-
tion tags attached, as in [iv]. These constructions are quite impossible with integrated
relatives.
Analysis
The antecedents of integrated relatives are sub-phrasal, parts of a phrase. In the great
majority of cases, the antecedent of an integrated relative is a nominal, and the relative
clause combines with it to form a larger nominal, as in the following structure that we
propose for [7ii]:
[11] Clause
Subject: Predicate:
NP VP
Det: Head:
D Nom
Head: Mod:
N ClauseREL
With supplementary relatives, on the other hand, the antecedents are full phrases, such
as NPs, or larger constituents, such as clauses, and the relative clause does not function
as a dependent of the antecedent. The structure we propose for [1ib] is as follows (and
for [1iib] see Ch. 15: [52ii] of §5):
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1062 Chapter 12 Relative constructions and unbounded dependencies
[12] Clause
Subject: Predicate:
NP VP
Consider these structures in the light of some of the syntactic differences noted above.
Non-declaratives
The non-declarative clauses in [10] provide evidence for the view that supplementary relatives
are not dependents of a head. Except for the relativisation, the structures are like those found
in main clauses, not clauses in dependent function – compare At that point please remember
to dim the lights; In that case would you mind hanging the washing out? ; and so on.
Stacking
The structure proposed for integrated relatives predicts the possibility of stacking. In [9i] ties
is a nominal which combines with you wear to form a larger nominal ties you wear, and this in
turn combines with that your sister knits for you to form the further nominal ties you wear that
your sister knits for you, which enters into construction with the determiner those. In [9ii],
however, the supplementary relative who has no qualifications does not form a unit with its
antecedent, so Max, who has no qualifications is not a possible antecedent for the second who.
Quantification with no, any, every
The structure for [7ii] is given as [11]; the antecedent for the integrated relative is candidate,
not no candidate – this is why the relative clause is not interpreted as “No candidate scored
40% or more”.11 In [7i], with a supplementary relative, the antecedent is no candidate, yet
this NP has no reference – so there is no referent for the relative pronoun who. NPs of this
kind can no more serve as antecedent for a relative pronoun than they can for a personal
pronoun – compare the incoherence of #I have no money; it’s on the desk, if no money is taken
as antecedent of it.
Proper names
In [6i] the who of the supplementary relative has as antecedent the NP Sue Jones. This refers
to the person of that name, and who refers to her too: it is coreferential with its antecedent.
In [ii] Sue Jones is a nominal, not an NP, and as such does not refer; the integrated relative
combines with it to form a larger nominal which is not a proper name and hence (given that
it is count singular) it requires a determiner.
Definite descriptions
Consider, finally, the necklace examples in [1ii]. In the supplementary case [iib] (The necklace,
which her mother gave to her, is in the safe) the antecedent is the necklace : this is marked as
definite, indicating that the description necklace is assumed to be sufficient in the context
to identify the referent. In the integrated case [iia], the antecedent is necklace ; the relative
clause combines with this to form the nominal necklace which her mother gave to her, and
11
A more complex case is seen in Nobody who scored 40% or more was ever failed. Here there is grammatical
fusion of the determiner and the head (see Ch. 5, §9.6), but semantically the negative is again not part of the
antecedent.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 4.2 Meaning and use 1063
the indicates that this nominal provides an identifying description of the referent. The rel-
ative clause thus forms part of the identifying description in the integrated case but not in
the supplementary: hence the implicature in the former but not the latter that there is some
other necklace from which the one being referred to needs to be distinguished.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1064 Chapter 12 Relative constructions and unbounded dependencies
Note similarly that the deviance of [7i] is matched by that of ∗No candidate – he or she
scored 40% or more – was ever failed.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 4.2 Meaning and use 1065
The narrator is three weeks short of eighteen and is saying that his father took it for
granted that during those three weeks he would continue to do as his father directed.
The relative clause here belongs to our integrated class: it cannot be omitted or spoken
on a separate intonation contour and allows that as an alternant of who (albeit somewhat
less favoured). Yet it does not serve to distinguish this father from other fathers of the
narrator: he has only one father. The reason for presenting the content of the relative
clause as an integral part of the message is not, therefore, that it expresses a distinguishing
property but that it explains why the father took it for granted that the son would do as
he was told.
Compare similarly:
[19] i He sounded like the clergyman he was.
ii She had two sons she could rely on for help, and hence was not unduly worried.
Both underlined clauses are bare relatives and hence necessarily integrated. But we do
not understand he was in [i] as distinguishing one clergyman from another: it conveys
that he was a clergyman, and an obvious reason for presenting this as an integral part
of the message is that sounding like a clergyman when you are one is significantly
different from sounding like a clergyman when you are not. In [ii] it could be that she
had more than two sons (in which case the relative would be serving a distinguishing
role), but an at least equally likely context is one where she had only two sons. In this
context the property expressed in the relative clause does not distinguish these sons
from other sons she has, but is an essential part of the reason for her not being unduly
worried.
The relative clause in [19ii] is embedded within an indefinite NP, and here it is very
often the case that the crucial factor differentiating the integrated and supplementary
constructions has to do with what we are calling information packaging rather than with
whether the relative restricts the denotation of the antecedent. Consider:
[20] i She had two sons(,) who were studying law at university(,) and a daughter(,) who
was still at high school.
ii A: Have you been to Paris? B: Yes, often: I have a brother who lives there.
iii I’ve been talking to one of the porters, who says the train may be an hour late.
Example [i] could be spoken/punctuated equally readily with integrated or supple-
mentary relatives in a context where she has just two sons and one daughter. On the
supplementary reading the primary information being imparted is that she had two
sons and a daughter: the information given in the relative clauses is supplementary,
secondary. On the integrated reading, by contrast, the content of the relatives is part
of the main information. In [ii] a supplementary reading would be incoherent even if
B has only one brother. It would involve presenting “I have a brother” as the primary
message, whereas it has in fact no relevance by itself in the context of A’s question:
the crucial point is that the brother lives in Paris, since this explains B’s having fre-
quently been there. Example [iii] has a supplementary relative, dividing the message
into two separate pieces of information. But if we replace one of the porters by, say,
a guy it would be much more natural to have the relative integrated. This is because
“I’ve been talking to a guy” is less likely to be considered worth presenting as a self-
contained piece of information: the crucial information will be that concerning the train’s
delay.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1066 Chapter 12 Relative constructions and unbounded dependencies
13
There is one case where only the postposed position is possible – the case where we have ‘split antecedents’
(Ch. 17, §1.3): There’s a boy in Group B and a girl in Group E who have asked to be on the same team.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 5 Infinitival relative clauses 1067
Integrated relatives may have infinitival form, with or without a relative phrase.
Wh type infinitivals
The most obvious kind of infinitival relative clause is illustrated in:
[1] i I’m looking for an essay question with which to challenge the brighter students.
ii She is the ideal person in whom to confide.
iii The best place from which to set out on the journey is Aberdeen.
This construction is limited to somewhat formal style. It is found only with integrated
relatives, and is subject to the following severe structural restrictions:
[2] i The relative phrase must consist of preposition + NP.
ii There can be no expressed subject.
The first restriction excludes examples like ∗She’s the ideal person whom to invite and
∗
I’m looking for an essay question which to challenge the brighter students with (where the
preposition is stranded rather than being part of the relative phrase). Condition [ii] rules
out ∗She’s the ideal person in whom for you to confide, and the like. There is no evident
explanation for the first restriction, but the second is predictable from the properties of
wh relative clauses and infinitivals taken together: infinitivals allow subjects only when
introduced by the subordinator for, but this cannot occur in wh relatives since both it
and the relative phrase require to be in initial position.
Non-wh infinitivals
Infinitival relatives without a relative phrase allow a considerably wider range of struc-
tures:
[3] i She’s the ideal person [( for you)to confide in ].
ii I’ve found something interesting [(for us)to read ].
iii A systems analyst wouldn’t be such a bad thing [(for her)to be ].
iv That is not a very good way [(for him) to begin ].
v You’re not the first person [ to notice the mistake].
The relativised elements here are respectively complement of a stranded preposition,
direct object, predicative complement, manner adjunct, and subject. Except in the latter
case a subject can be optionally included, preceded by the subordinator for.
Where the relative clause is within an NP functioning as object or complement of a
preposition, there is overlap with an infinitival adjunct of purpose. Compare:
[4] i He found a video [for the kids to watch]. [relative]
ii He got it [for the kids to watch]. [purpose adjunct]
iii He got a video [for the kids to watch]. [ambiguous]
In [i] the infinitival is a relative with a meaning close to that of the finite relative that
the kids could watch. A relative interpretation of this kind is excluded in [ii] because it
does not permit modification by a relative clause (cf. ∗He got it that the kids could watch);
[ii] has, rather, a purposive interpretation: “He got it in order that the kids could watch
it”. This interpretation is not possible in [i] because finding is non-agentive and therefore
does not allow purpose adjuncts. In [iii] the conditions for both constructions are met;
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1068 Chapter 12 Relative constructions and unbounded dependencies
it can be construed in either way, though there is little effective difference in meaning
between them.
Modal meaning
Infinitival relatives characteristically have a modal meaning comparable to that expressed
in finites by can or should. Here’s something interesting for you to read, for example, is
comparable to Here’s something interesting that you can/should read. This modal mean-
ing is indeed what makes relatives like those in [4] semantically so close to purpose
infinitivals. Where the matrix NP is definite there is very often some explicit or implicit
evaluative modification, such as ideal in [1ii] or best in [1iii].
Infinitivals where the relativised element is subject have a somewhat wider range of
interpretations than others, allowing non-modal as well as modal meanings:
[5] i She’s obviously the person to finish the job. [modal]
ii She was the first person to finish the job. [non-modal]
Example [i] is like the non-subject examples considered above: we understand “best,
most appropriate” and “should” (“the person who should finish the job”). But [ii] has
no such modal meaning, being equivalent simply to the first person who finished the job.
Nominals containing relatives with this kind of interpretation usually contain a modifier
such as only, next, last, or one of the ordinals first, second, etc.
Classification
An initial illustration of the range of constructions belonging to the fused relative category
is given in [1]:
[1] simple series ·EVER series
[NP]
i a. I spent what he gave me. b. I spent whatever he gave me.
ii a. I gave him what money I had. b. I gave him whatever money I had.
iii a. I’ll go where you go. b. I’ll go wherever you go. [PP]
On one dimension we have a contrast between the simple series and the ·ever series, the
latter being marked by a relative word ending in ·ever. Cutting across this is the major
category contrast: the fused relatives are NPs in [i–ii], PPs in [iii]. And within the NP
category we have a further distinction according as the relative word is a pronoun, as in
[i], or a determinative, as in [ii].
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 6.1 Fused relatives as phrases, not clauses 1069
The fused relative is equivalent not to the relative clause which we hold most dear but to
the NP containing it, that which we hold most dear. Compare similarly:
[3] i The dog quickly ate the scraps that I’d left on my plate.
ii The dog quickly ate what I had left on my plate.
These are not of course fully equivalent since [i] contains the lexical item scraps, but in
[ii], no less than in [i], the object of ate denotes something concrete, a physical entity.
Clauses, by contrast, denote abstract entities: propositions, events, and so on. These
points demonstrate the semantic likeness between the fused relatives and NPs, but there
is also strong syntactic evidence for analysing these constructions as NPs.
(c) No extraposition
[6] a. What she suggests is unreasonable. b. ∗It is unreasonable what she suggests.
Like ordinary NPs, fused relatives do not occur in the extraposition construction. Here
too they differ from clauses: compare That we should have to do it ourselves is unreason-
able and It is unreasonable that we should have to do it ourselves.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1070 Chapter 12 Relative constructions and unbounded dependencies
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 6.2 Fused relatives vs open interrogatives 1071
Open interrogatives, whether main clauses (e.g. What did she write?) or subordinate (what
she wrote) normally express what we have called variable questions (Ch. 10, §4.5). The propo-
sitional content of such questions contains a variable (“She wrote x”), and the answers specify
values of the variable (She wrote the preface ; She wrote a textbook on phonetics, etc.). We have
also analysed integrated relatives as containing variables, but here the variable is anaphorically
bound to an antecedent. In the earlier no candidate who scored 40% or more, for example, we
have an analysis along the lines of “no candidate x [x scored 40% or more]”, i.e. “no candidate
x such that x scored 40% or more”. In the case of fused relatives the antecedent and pronoun
are not syntactically discrete, but we still have linked occurrences of the variable in the inter-
pretation, e.g. for [11i] “I liked the x such that she wrote x”. Both relative and interrogative thus
contain the “she wrote x” component: in the relative case, the variable is bound to an an-
tecedent, whereas in the interrogative case the value of the variable is to be given in the answer
to the question.
Consider the following further examples in the light of this account:
[12] i The dogs wouldn’t eat what she gave them. [fused relative]
ii I told him what she gave them. [open interrogative]
iii I told him what she suggested I tell him. [ambiguous]
Again, the fused relative is roughly equivalent to an NP containing antecedent + integrated
relative, e.g. the food which she gave them, so we might analyse [i] as “The dogs wouldn’t
eat the x such that she gave them x”. There is again no ambiguity here because eat cannot
take clausal complements. Example [ii] can be glossed as “I told him the answer to the
question ‘What did she give them?’ ” – i.e. “I told him the value of the variable in ‘She gave
them x’ ”.
Tell can take NP complements, as in I told him the news, but the things you can tell are
distinct from the things you can give, so there is no fused relative interpretation “# I told him the
x such that she gave them x”. However, if we change the example to remove this incompatibility,
we can get an ambiguity with tell , as in [12iii]. The interrogative interpretation matches that
for [ii]: “I told him the answer to the question ‘What did she suggest I tell him?’ ” – i.e. “I
told him the value of the variable in ‘She suggested I tell him x’ ”. And the fused relative
interpretation is “I told him the x such that she suggested I tell him x”.
The difference can be brought out by imagining the case where she suggested I tell him
that his offer would have to be raised. In this scenario the interrogative interpretation of
[12iii] is equivalent to I told him that she suggested I tell him that his offer would have to be
raised (and I thereby implicitly distance myself from this evaluation of his offer), while the
fused relative interpretation is equivalent to I told him that his offer would have to be raised
(i.e. the value of x in “I told him the x such that she suggested I tell him x” is “his offer would
have to be raised”).
We have focused above on the semantic difference between the constructions. We now
turn to the syntactic differences.
(a) NP vs clause
We have shown that fused relatives (other than the prepositional ones introduced by
where, when, etc.) are NPs; interrogatives, however, are not: they are clauses. The points
made in §6.1 above concerning agreement, subject–auxiliary inversion and extraposition,
preposition fronting, and adjective complementation can therefore be applied to the
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1072 Chapter 12 Relative constructions and unbounded dependencies
iii a. I wonder what she wrote. b. I wonder who spoke last.
[interrogative]
iv a. I know what he offered. b. I know which he offered.
Conversely the ·ever series of forms occur freely in fused relatives, but they are generally
not permitted in interrogatives:15
[15] i a. He accepted what/whatever she offered.
[relative]
b. He planted roses where/wherever there was enough space.
ii a. He didn’t tell me what/∗whatever she offered.
b. He went to see where/∗wherever there was enough space. [interrogative]
14
This last point is of only limited value as a distinguishing test because the stranded preposition construction is
often strongly preferred or else the only option even in the subordinate interrogative construction (cf. Ch. 7,
§4.1), as in I can’t imagine what he’s getting at / ∗at what he’s getting.
15
We ignore here cases where ever (often written as a separate word) has a quite different sense, like that of on
earth : I can’t imagine what ever he was thinking about. There is, however, one type of interrogative where the
·ever forms are found, namely interrogatives functioning as exhaustive conditional adjuncts, as in He won’t
be satisfied, whatever you give him. This construction is discussed (and contrasted with the fused relative) in
Ch. 11, §5.3.6.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 6.3 Syntactic analysis 1073
[16] i a. A: Jill gave him something last night. B: What?
[interrogative]
b. Jill gave him something last night, but I don’t know what.
ii a. ∗Jill gave him something last night, but he lost what. [fused relative]
b. ∗Jill gave him a book last night, but he lost the book which. [integrated relative]
In [ia] what is equivalent to interrogative What did she give him?, while [iia] shows that
relative what she gave him cannot similarly be reduced to what. Analogously in [ib/iib].
Head:
Nom
Modifier:
ClauseREL
Head-Prenucleus: Nucleus:
NPi Clause
Subject: Predicate:
NP VP
Predicator: Object:
V GAPi
Case
The pronoun what in [17] is simultaneously head of the whole NP and object (in
prenuclear position) in the relative clause. In constructions with personal who and
whoever, the pronoun has to satisfy the case requirements of both the relative clause and
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1074 Chapter 12 Relative constructions and unbounded dependencies
16
There are also archaic variants of the ·ever series with ·so : whosoever, whatsoever, etc.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 6.4 Relative words in the fused construction 1075
In [iii] it is a matter of selecting one edition from an identifiable set of editions, whereas
there is no such feature in [iv]. Fused which(ever), although a determinative, not a
pronoun, can function as fused determiner-head, as in:
[22] [Whichever of these two finished ahead of the other]would betheundisputed financial
leader of the tour.
Three of the simple items, who, which, how, are virtually restricted to the particular
use of the fused relative that we call the free choice construction; we will look at this
first, and then turn to other uses, taking the ·ever and simple forms separately.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1076 Chapter 12 Relative constructions and unbounded dependencies
18
In the coordination case we often find the ·ever word on its own: Disturbed by this telephone call or whatever,
she walked out into the night. We take this to be a simple NP, though it might alternatively be regarded as a
fused relative with something like “it was” understood.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 6.4 Relative words in the fused construction 1077
%
We will not change how we use future contracts during the term of this Prospectus ; %I don’t
like how it looks.
What
The most frequent type of fused relative NP has what as fused head. It is found in
non-referential NPs, like whatever, but it also readily occurs in referential NPs:
[27] i I’ll do [what I can]to help you. [non-referential (non-specific)]
ii They seem pleased with [what I gave them]. [referential]
Determinative what (leaving aside the free choice construction) is restricted to non-
count or plural NPs and has a paucal meaning that can be reinforced by little or few and
is inconsistent with much or many :
[28] i This will further erode [what (little)economic credibility the government has left].
ii [What (few)mistakes she had made]were all of a minor nature.
The specifying be construction
Simple forms of the unbounded dependency words are commonly found within a matrix
clause containing be in its specifying sense, and here it is by no means a straightforward
matter to distinguish between the fused relative and subordinate interrogative construc-
tions. Compare, for example:
[29] i What caused the trouble was a faulty switch.
[fused relative]
ii A faulty switch was what caused the trouble.
iii That’s who I meant. That’s not how to do it.
[interrogative]
iv He’s not who she thinks he is.
Example [i] belongs to the pseudo-cleft construction (Ch. 16, §9.3), which is normally re-
versible, yielding in this case [ii]. There is no doubt that these involve fused relatives. Note, for
example, that we can have subject–auxiliary inversion (Was what caused the trouble a faulty
switch?) and that preposition fronting is completely impossible (What I’m referring to is his
intransigence, but not ∗To what I’m referring is his intransigence).
The examples in [iii–iv] are not reversible (cf. ∗How to do it is not that, etc.), and there
are grounds for saying that the underlined expressions are interrogative clauses even though
they can be paraphrased by such NPs as the person I meant, the way to do it, the person that
she thinks he is. Note first that who is found here, but not elsewhere in fused relatives, other
than in the free choice construction. In particular, it cannot occur in the pseudo-cleft: ∗Who
caused the trouble was your brother.19 Similarly, the infinitival of how to do it is not possible
in fused relatives. Conversely, the one item that appears in relatives but not interrogatives,
while, is not found here. That was while we were in Paris, for example, is quite different from
[29iii]: it does not serve to identify the period during which we were in Paris but locates ‘that’
within this period.
A further difference between [29iii–iv] and the pseudo-cleft is that only the latter can
incorporate an integrated relative (cf. the discussion of [10i]). Compare What she left me
that I treasure most is this little music-box with ∗That’s who she recommended who has the best
qualifications (“the person she recommended who has the best qualifications”).
19
Why (which has no counterpart in ·ever) appears freely in the interrogative construction, as in This is why I’m
leaving, but is marginally possible in the pseudo-cleft: Why I’m leaving is that/because there’s no opportunity to
use any initiative. It does not occur elsewhere in fused relatives.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1078 Chapter 12 Relative constructions and unbounded dependencies
Finally, and most decisively, the specifying be construction of [29iii] allows preposition
fronting: That is precisely for what it was designed; That is exactly against what we should be
fighting now. This distinguishes them clearly from the pseudo-cleft construction and indicates
that they must be interrogatives.
The likeness in meaning between the relative and interrogative constructions of [29] is
consistent with the different syntactic analyses we have proposed. For [i] we have “The x such
that x caused the trouble was a faulty switch”: a faulty switch gives the value of the variable
defined in the fused relative. And in That’s who I meant the subject that gives the value of the
variable in “He meant person x”, the propositional content of the embedded question – i.e.
it gives the answer to the question “Who did he mean?”20
The prepositions when, where, while
[30] i [When it rains]they play in the garage.
ii We must put it [where no one will be able to see it].
iii They insisted on talking [while I was trying to get on with my work].
These have paraphrases containing noun + integrated relative: On occasions when it
rains, they play in the garage ; We must put it in a place where no one will be able to see it;
%
They insisted on talking during the time while I was trying to get on with my work.
While differs from the other fused relative words in having no ·ever counterpart and
in having no interrogative use.
Fused relatives or preposition + content clause?
An alternative analysis of examples like those in [30] is to treat when, where, and while as
prepositions that take content clauses as complements – like before, whereas, although, etc.
There are certainly some cases where the latter is the preferable analysis:
[31] i [When they weren’t home at six o’clock]I began to get worried.
ii Let me know [if and when you need any help].
iii [Where the British Empire was established with musket and gunboat,] America’s empire
has been achieved with the friendly persuasion of comedian and crooner.
iv [While I don’t agree with what she says,] I accept her right to say it.
In [i] we cannot gloss when as “at the time at which” since the temporal adjunct function
within the subordinate clause is pre-empted by at six o’clock. In [ii] when is co-ordinated
with if, which quite clearly takes a content clause as complement. Example [iii] illustrates a
use of where that has been bleached of the basic locative meaning: it indicates contrast, like
whereas. Moreover, it would conflict with the meaning to posit a location adjunct within the
subordinate clause: the sentence does not say that America’s empire was established in the
same place as the British Empire. Example [iv] is similar: while here is used for contrast and
its meaning does not involve temporal duration.
On the other hand, we find places where a fused relative analysis is required. The clearest
cases are with where and while in such examples as:
[32] i I put the key [where I always put it], in the top drawer.
ii It was fun [while it lasted].
20
Another case where an interrogative has an interpretation similar to that of a relative construction is
illustrated in There’s an article in the weekend magazine on how to grow orchids. We might instead say on
the way to grow orchids, with an NP containing a relative clause. But how to grow orchids must be an interrog-
ative, by virtue of the how and the infinitival. And again the meaning in fact fits the interrogative analysis: the
magazine article is concerned with answering the question “How to grow orchids?”
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 7 Unbounded dependency constructions 1079
Here the subordinate clauses must contain complements of goal and duration respectively
because of the complementation requirements of the verbs put and last. Thus I always put it
and it lasted are not themselves structurally complete, and could not occur as complement of
a preposition. They must have a gap in final position that is linked to where and while : wherei
I always put it i and whilei it lasted i . In the light of examples [31–32], we conclude
that both the fused relative and the preposition + content clause analyses are needed; some
examples require just one, whereas others are consistent with either.
The outer brackets enclose the relative clause, while the inner ones enclose the nucleus.
The nucleus contains a gap in the position of object of the verb recommended, and this
gap is linked to the relative phrase which in prenuclear position. The relation between the
gap and which is comparable to that between an anaphoric pronoun and its antecedent
– between, for example, which and its antecedent book. Which derives its interpretation
from book, and the gap derives its interpretation from which: a component of the meaning
of all three examples is “she recommended x”, where x is some book. We will say, therefore,
that the gap is anaphorically linked to which, i.e. that which is antecedent for the gap.
This relation between the gap and which is a dependency relation. Semantically,
the gap derives its interpretation from which, as we have just seen. And syntactically
which requires an associated gap: the object of recommended cannot be realised by an
ordinary NP – compare ∗This is the book which she recommended ‘War and Peace’.21 The
dependency relation between the gap and its antecedent is unbounded in the sense that
there is no upper bound, or limit, on how deeply embedded within the relative clause
the gap may be. In [1i] the gap is object of the topmost verb in the relative clause. In
[1ii] it is object of the verb that heads a clause embedded as complement to the topmost
verb (think). In [1iii] there are two layers of clause embedding: the recommend clause is
complement in the say clause, and the latter is complement in the think clause. And there
is no grammatical limit on how many such layers of embedding there can be. Adding
a third might give, for example, the book which I think you said Kim persuaded her to
21
The dependency relation between a gap and its antecedent is not to be equated with that of a dependent to a
head. Dependent and head are functions within a syntactic construction, and the gap is not a dependent of
which in this sense. The gap and which are related anaphorically, not as functions within a construction.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1080 Chapter 12 Relative constructions and unbounded dependencies
recommend. And further layers still can be added without loss of grammaticality even
though they may result in stylistically undesirable complexity.
A second unbounded dependency construction is the open interrogative, illustrated
in:
[2] i Whati [does he want i ]?
ii Whati [do you think he wants i ]?
iii Whati [do you think she said he wants i ]?
The gap represents the object of want and is anaphorically linked to the interrogative
phrase what in prenuclear position. This relationship indicates that the question concerns
the object of want: the meaning contains the component “he wants x”, and answers to the
question supply a value for the variable x. Again, the dependency relation between
the gap and the interrogative phrase is unbounded: the examples show the want clause
progressively more deeply embedded, and again there is no grammatical limit as to how
many layers of embedding are permitted.
In the light of these examples we may define an unbounded dependency construction
as follows:
[3] An unbounded dependency construction is one which sanctions within it an
anaphoric gap, with no upper bound on how deeply embedded the gap may be.
Not all unbounded dependency constructions are of this kind, however. In preposing,
the prenuclear position is filled by a phrase or clause that can also occur in a canonical
clause construction. Compare:
[5] i The other chaptersi [she wrote i herself ].
ii The other chaptersi [I think she wrote i herself ].
iii The other chaptersi [I think she said she wrote i herself ].
The other chapters is an ordinary NP, functioning as object in the canonical She wrote the
other chapters herself, but in [5] it is in an unbounded dependency relation with a gap.
These examples illustrate the main preposing construction, with the preposed element
in prenuclear position within a clause. It is also possible for the preposed element to
22
These words are often referred to as ‘wh words’; the category, however, is obviously not unique to English, and
we prefer to use a more general term.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 7.1 Definition and taxonomy 1081
These are just like the wh relatives in [1] above, except that they contain no relative
phrase in prenuclear position. The gap is thus related directly to the nominal book,
rather than indirectly, via the relative pronoun which. This construction still satisfies the
definition given in [3]: the relative clause can contain an anaphoric gap that is embedded
indefinitely deeply within it.
Another construction of this type is the comparative clause:
[8] i Kim made more mistakesi than [Pat made i ].
ii Kim made more mistakesi than [I think Pat made i ].
iii Kim made more mistakesi than [I think you said Pat made i ].
Comparative clauses function as complement to a preposition (than, as, or like); the gap
is within the comparative clause while the antecedent is outside. Comparative clauses,
however, differ in significant ways from other unbounded dependency constructions
with respect to the kind of gap allowed and the way it is interpreted: we examine them
in detail in Ch. 13, §2, and will pay no further attention to them here.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1082 Chapter 12 Relative constructions and unbounded dependencies
embedded within the hollow clause, the embedded clause must be non-finite, like the
hollow clause itself. Compare:
[10] a. The problemi is too difficult [to expect a ten-year-old to be able to solve i ].
b. ∗The problemi is too difficult [to expect [that a ten-year-old could solve i ]].
While [i] is fully acceptable, [ii] is ungrammatical. This is because the hollow clause
(enclosed within the outer pair of brackets) contains a finite clause within it (enclosed
within the inner brackets), and the gap is inside this finite clause. The same constraint
applies to infinitival relatives and infinitival open interrogatives, and we accordingly
include these too in the set of minor unbounded dependency constructions.
Summary taxonomy
Unbounded dependency constructions may be classified in terms of the above distinc-
tions as follows:
[11]i major constructions
ia Prenuclear antecedent
iai Contain unbounded wh relatives (finite), open interrogatives (finite),
dependency word exclamatives
iaii No such word preposing in clause, preposing in concessive PP
ib External antecedent non-wh relatives (finite), comparatives
ii minor constructions
iia Prenuclear antecedent infinitival wh relatives and open interrogatives
iib External antecedent hollow clauses, infinitival non-wh relatives
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 7.2 Gaps and antecedents 1083
The inner brackets in [ii] and [iv–viii] enclose the constituent within which the gap is
located: a PP in [ii], content clause in [iv], an NP in [v–vii], and an NP-coordination
in [viii]. Examples [v–viii] are ungrammatical because the gap does not have one of the
functions permitted by rule [12]. They can be corrected by making the gap conform to
[12]:
[14] i Whose cari did you borrow i ?
ii How many applicationsi did they receive i ?
[complement of verb]
iii How serious a problemi will it be i ?
iv Whoi have they shortlisted i in addition to Kim?
In addition to rule [12], certain more specific conditions apply:
Gaps not normally allowed in indirect object function
As we observed in Ch. 4, §4.3, one of the main syntactic differences between indirect
and direct objects is that gaps are more or less excluded from the former function. The
qualification ‘more or less’ is needed because there is some variation with respect to
acceptability judgements on clauses with indirect object gaps, but for the most part
there is a clear difference between the acceptability of direct and indirect object gaps.
Compare:
[15] i a. This is the CDi [she got me i last Christmas].
b. ∗He’s the onei [she got i that CD last Christmas].
ii a. The copies [he sold me i ]were defective.
b. ∗The personi [he sold i them]seemed satisfied.
iii a. How muchi do you owe them i ?
b. ?How many peoplei do you owe i more than $50?
In each pair the gap is direct object in [a], indirect object in [b]. Most verbs that take
indirect objects also occur in an alternative construction with direct object + PP com-
plement, and this construction can be used to express the meanings of the [b] examples:
He’s the onei she got that CD for i last Christmas, and so on. (The prepositional con-
struction will also often be preferred over a ditransitive one with indirect object + gap
in direct object function – e.g. the storyi that he was reading i to his children, over the
storyi that he was reading his children i .)
Gaps in subject function
As we saw in §3.4, it is necessary to distinguish between an immediate subject (i.e. the
subject of the topmost verb in the construction) and an embedded subject (the subject
of a clause embedded within the unbounded dependency construction).
Embedded subject gaps are permitted only in bare content clauses, i.e. declaratives
without the subordinator that (cf. Ch. 11, §3.1). Compare:
[16] i He’s the mani [they think [ i attacked her]]. [bare declarative]
ii ∗He’s the mani [they think [that i attacked her]]. [expanded declarative]
iii ∗He’s the mani [they wonder [whether i attacked her]]. [closed interrogative]
With immediate subjects we can have a gap in Type ib constructions (with external
antecedent), but not in Type ia (with prenuclear antecedent). Compare:
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1084 Chapter 12 Relative constructions and unbounded dependencies
ii This is the copyi [whichi is defective].
[no gap]
iii Who signed the letter?
In [i] that is the subordinator in prenuclear position, and the subject in the nucleus is
realised by a gap anaphorically linked to the antecedent copy : this is a Type ib construc-
tion, with the antecedent of the gap external to the relative clause. (As we noted in §3.5.5,
the subordinator that is generally not omissible when the gap is in immediate subject
function.) In [ii–iii], by contrast, the subjects are realised not by a gap but by relative
which and interrogative who: in the absence of convincing evidence to the contrary, we
take the structure to be the same as that of the canonical clauses This copy is defective and
Kim signed the letter. Note, moreover, that the preposing construction does not allow
preposing of an immediate subject. Compare:
In [i] we have a gap in object position, but there is no gap, no preposing in [ii], where
she is in its canonical position.
Hollow clauses
In the hollow clause construction the gap can only be complement of a verb or preposi-
tion: see Ch. 14, §6.
[19] i a. Have you seen the booki [I got i from the library]?
b. Where’s the booki [I got i from the library]?
ii a. Their proposali was hard [to accept i ].
b. We found their proposali hard [to accept i ].
In [i] the bracketed clauses are non-wh relatives with the gap in object function. The
antecedent is the nominal book, which is head of an NP, and this NP can occur in any NP
function: it is, for example, object in [ia], subject in [ib]. In [ii] we have hollow clauses
with the gap in object function. The antecedent is the NP their proposal, and again the
function of this NP does not need to match that of the gap: in [iia] it is subject, while in
[iib] it is object.
Prenuclear antecedents inherit function of the gap
The situation with antecedents in prenuclear position is quite different. These elements
are located within the unbounded dependency construction itself, and thus do not have
a function outside it. Because they fall outside the nucleus the only function that can be
assigned directly to them is that of prenuclear dependent. This is shown in the following
tree diagram for the preposing construction The others I know are genuine, corresponding
to canonical I know the others are genuine.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 7.2 Gaps and antecedents 1085
[20] Clause
Prenucleus: Nucleus:
NPi Clause
Subject: Predicate:
NP VP
Predicator: Comp:
V Clause
Subject: Predicate:
GAPi VP
We could not label the others as subject, for it clearly does not stand in the subject rela-
tion to the clause of which it is an immediate constituent. Nevertheless, it is under-
stood as subject of are genuine, just as it is in the canonical counterpart I know the
others are genuine. Notice, moreover, that the verb form are agrees with the others –
again, just as it does in I know the others are genuine. We will regard these prenuclear
antecedents, therefore, as taking on the function of the associated gap. In a secondary,
or derivative, sense, that is to say, the others is subject of the content clause whose head
(predicate) is are genuine. This information is retrievable from the tree diagram as it
stands: the secondary function of the others is that of the co-indexed gap. Similarly in
The othersi I haven’t yet read i we will say that the others is, in this derivative sense, object
of read.
The same applies with constructions where the prenuclear element consists of or
contains an unbounded dependency word, as in:
We will say that which in [i] is object of wrote and who in [ii] is subject of wrote. This
is of course what is said in traditional grammar too: our concern here has been to show
how that kind of statement can be reconciled with the tree diagrams that are used to
represent syntactic structure in this book. Which and who are not labelled object and
subject directly, but are treated as inheriting this function from the gap with which they
are co-indexed.
[22] The syntactic and semantic properties of the antecedent must normally match
those of expressions which in other constructions can occur as overt realisations
of the gap function.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1086 Chapter 12 Relative constructions and unbounded dependencies
In the simplest cases, the antecedent expression itself can elsewhere realise the function
of the gap. Illustrations of this situation are provided by preposing and hollow clauses:
[23] i a. To Kimi [they gave a bicycle i ]. [preposing]
b. They gave a bicycle to Kim.
ii a. Kimi is very hard [to please i ]. [hollow]
b. It is very hard to please Kim.
In [ia] the PP to Kim is antecedent for the complement gap in the nucleus, and this PP
can itself realise the same complement function, as shown in [ib]. Similarly in [ii]: the
gap in [a] is object of please and the antecedent for this gap, the NP Kim, can elsewhere
fill that function, as in [b]. These examples may be contrasted with the following:
[24] i ∗To Kimi [they bought a bicycle i ].
ii ∗That he comes home so latei is very hard [to enjoy i ].
In [i] the preposed complement contains the wrong preposition: we need for Kim, to
match They bought a bicycle for Kim. In [ii] the antecedent is a content clause but enjoy
does not license a complement of that kind: we need an NP, such as his novels, to match
I enjoy his novels.
Compare, again, the following examples of the it-cleft construction:
[25] i It was that jari [that she says she put the key in i ]. [NP ∼ NP]
ii It was in that jari [that she says she put the key i ]. [PP ∼ PP]
∗
iii It was that jari [that she says she put the key i ]. [NP ∼ PP]
∗
iv It was in that jari [that she says she put the key in i ]. [PP ∼ NP]
In [i] the antecedent is an NP, and this is the category needed to realise the gap function,
object of the preposition in. In [ii] the antecedent is a PP, which can realise the function of
goal complement in the put clause, as in She put the key in that jar. The other examples
are ungrammatical because the antecedent fails to meet the requirements of the gap
function: compare ∗She put the key that jar and ∗She put the key in in that jar.
Condition [22] is formulated in terms of matching rather than identity: there is
no requirement that the antecedent expression itself should be able to realise the gap
function. Three very general cases where it can’t are illustrated in:
[26] i Every booki [we have consulted i ] ignores this problem. [non-wh relative]
ii That’s not the reason [whyi [he did it i ]]. [wh relative]
iii i Don’t be so hard [to please i ]. [hollow clause]
The bracketed clause in [i] is a non-wh relative of the integrated type. As explained
in §4.1, the antecedent is the nominal book, not the sequence every book: the sentence
doesn’t say that we have consulted every book. A nominal as such cannot realise the
gap function, which requires a full NP: ∗We have consulted book. The antecedent can
nevertheless be said to satisfy the matching requirement in that it can realise the gap
function if an appropriate determiner is added to make it into a full NP: We have read
a book.
The outer brackets in [26ii] enclose a wh relative clause, and here the relative phrase
is required to occupy initial position, so relative why could not occur within the nucleus
as a realisation of the gap function. The matching requirement is satisfied, however, in
that why is a reason expression and non-relative expressions of that kind can realise the
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 7.2 Gaps and antecedents 1087
gap function, as in He did it for that reason. The same applies, of course, to other relative
expressions.
In [26iii] the antecedent for the gap in the hollow clause is not overtly expressed. But
it is understood, by virtue of being subject of an imperative, as you, and this can realise
the gap function: It is hard to please you. In Pati wants i to be hard [to please i ],
the antecedent (the subject of the be clause) is likewise missing, but this time it is
recoverable from the superordinate want clause.
Mismatches
There are a number of constructions where the matching requirement [22] is not strictly
observed. They are illustrated in [27], but as all are dealt with elsewhere in the book only
a summary commentary is needed at this point.
[27] i Whoi [did you give it to i ]?
%
ii He always chose those [whomi [he thought i were most vulnerable]].
iii [Whati [I’m hoping i ]] is that nobody will notice my absence.
iv What on earthi [do you want i ]?
v That no one realised such action might be illegali [I find i surprising].
vi That they’ll give him a second chancei [I wouldn’t gamble on i ].
vii [Brilliant advocatei [though she is i ],] she’s unlikely to win this case.
Examples [27i–ii] show that the inflectional case of prenuclear interrogative and
relative who does not always match that of pronouns in the position of the gap. Compare
the nominative who of [i], with accusative them in You gave it to them, and the accusative
whom of [ii] with the nominative required in He thought they were most vulnerable (see
§3.4 above, and Ch. 5, §16.2.3).23
Fused relative what in [27iii] is an NP, but hope does not license an NP complement:
compare ∗I was hoping some respite. Hope takes declarative content clause complements,
and the presence of such a content clause following the fused relative is apparently
necessary for what to be admissible: compare ∗What I was hoping was a little peace and
quiet. The fused relative in [iii] is subject within a pseudo-cleft clause (see Ch. 16, §9.3),
and the same extended use of what is found with a few other verbs in pseudo-clefts.
Compare, for example, What we decided was to interview all the candidates. Although
decide does license NP complements, they don’t stand in the same semantic relation to
it as what does here – compare The weather will decide the outcome, but not #We decided
an interview.
Example [27iv] is an open interrogative. Unlike relative phrases, interrogative phrases
are not in general required to occupy initial position – compare And so you want what,
exactly? (cf. Ch. 10, §4.5). Interrogative phrases containing emotive modifiers such as on
earth, the hell, ever, etc., however, can only occur initially, hence not in the position of
the gap in [iv]: ∗And so you want what on earth?
The remaining examples in [27] are preposings. In [v] the function of the gap is
that of object in a complex-transitive clause. The preposed content clause could not
23
In constructions with an external antecedent the case of the antecedent will be determined by its function
within its own clause, which is independent of the function of the gap. Compare Hei is hard to get on with i
(where he is subject and hence nominative) and I find himi hard to get on with i (where he is object and
hence accusative). In [27i–ii], however, who is in prenuclear position, so its case does depend on the function
of the associated gap.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1088 Chapter 12 Relative constructions and unbounded dependencies
occur in post-verbal position: instead of ∗I find that no one realised such action might
be illegal surprising we need the version with extraposition I find it surprising that no one
realised such action might be illegal. In [vi] the preposed content clause could not replace
the gap because the latter is complement of the preposition on, which does not license
complements of this category: it requires an NP (see Ch. 11, §8.3, for further discussion
of this very marginal type). Finally, [vii] has preposing of a predicative complement NP
from within a concessive PP. Here there is a more systematic departure from the form
found in non-preposed position, with the latter requiring an indefinite article: Although
she is a brilliant advocate, she’s unlikely to win this case.
Combinations of unbounded dependency constructions
It is possible for certain unbounded dependency constructions to combine in such a way
that the gap in one is the antecedent in the other. In the following, for example, an open
interrogative is combined with a cleft relative:
[28] i Which jari was it i [that she says she put the key in i ]?
ii In which jari was it i [that she says she put the key i ]?
These are the open interrogative counterparts of the declarative clefts given in [25i–ii] above.
The first gap has the interrogative phrase as its antecedent, and itself serves as antecedent
for the second gap. In both examples the matching requirement is satisfied. In [i] the gap in
the put clause is object of in, and hence requires an NP antecedent: this requirement is met
because the gap in the be clause has the overt NP which jar as its antecedent. Similarly in [ii]
the gap in the put clause requires a PP antecedent, and this requirement is satisfied because
the gap in the be clause has an overt PP as its antecedent. Interrogative counterparts of the
ungrammatical [25iii–iv] will thus be ungrammatical too:
[29] i ∗Which jari was it i [that she says she put the key i ]?
ii ∗In which jari was it i [that she says she put the key in i ]?
The antecedent for the gap in the put clause is the PP headed by in: we have enclosed it in
square brackets and co-indexed it with the gap in the relative clause. This PP contains a gap
with the interrogative phrase as antecedent. What makes the sentence ungrammatical is that
the antecedent for one gap contains another gap within it, so we have two gaps with different
interpretations – “which jar” and “in which jar”. Thus while the antecedent for a gap may
itself be a gap, as in [28], it cannot merely contain a gap.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 7.3 Location of gaps 1089
the value of x in the proposition ‘You insisted that we need x’ ”, and the intended meaning
of [ii] is similarly “I told her the value of x in the proposition ‘That we need x is agreed’ ”.
This meaning can in fact be expressed by means of the extraposition construction: I told
her what it is agreed that we need.
The structure for the interrogative clause in [31i] is as follows:
[32] Clause
Prenucleus: Nucleus:
NPi Clause
Subject: Predicate:
NP VP
Predicator: Comp:
V Clause
Marker: Head:
Subordinator Clause
Subject: Predicate:
NP VP
Predicator: O:
V GAPi
The boxes enclose points in the tree that lie on the path from the top down to the gap, and
grammaticality depends on the function and category labels that occur on this path. The
deviance of [31ii] is due to the fact that the path to the gap passes through a constituent
with the form of a clause and the function of subject: that we need is subject of is
agreed. Note that we are concerned here with the path to the gap, not with the gap itself.
We noted in §7.2 that (under restricted conditions) the gap can be subject, as in Whoi
[do you think i wrote the letter]? : what is not admissible is for the gap to be part of a
larger clause that is functioning as subject.24
In this section we will review a range of constituent types, examining whether or not
they may occur on the path leading to the gap. Before we start, however, two general
points should be made. In the first place, while the status of [31i–ii] as respectively
well-formed and deviant is quite clear, there are many intermediate cases where the
status is uncertain. Secondly, while we can confidently say that [31ii] violates a rule of
grammar, the acceptability of examples may also be affected by semantic considerations.
Compare:
[33] i That’s a subjecti [that Steven Jay Gould wrote a book about i ].
ii #That’s a subjecti [that Steven Jay Gould despises a book about i ].
24
Constituent types that do not allow gaps within them are often called ‘islands’.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1090 Chapter 12 Relative constructions and unbounded dependencies
These have the same grammatical structure, differing only lexically, with [i] having
write and [ii] despise as the verb of the relative clause. But they differ significantly in
acceptability: [i] is clearly acceptable, while [ii] is very unnatural. This difference has a
semantic explanation. The relative clause combines with the antecedent subject to form a
nominal that denotes a class of subjects. In the case of [i], this class has some coherence:
to say of some subject that Steven Jay Gould wrote a book about it points to a selection
of significant topics in areas like evolutionary biology, geology, palaeontology, etc. The
class denoted by the nominal in [ii] has no such coherence. What would have to be true
of a subject in order for it to be an x such that Steven Jay Gould despises a book about x ?
Someone, at some time in history, has to have written a book about x that Gould despises
for some reason (it is badly written, or was plagiarised, or has annoyingly pretentious
page design, or is full of mistakes, or whatever reason there might be). The subject in
question could be shoes, ships, sealing wax, cabbages, or kings. In other words, there is
no sensible characterisation of a class of subjects in [ii] at all, and as a result the example
seems anomalous.
Let us turn now to the review of constituent types. In the examples, we use one pair
of square brackets to delimit the constituent in question, and another to delimit the
unbounded dependency construction if it is less than the whole sentence. Antecedents
are underlined if they contain more than the one word that carries the subscript index.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 7.3 Location of gaps 1091
(See Ch. 10, §7.12, for discussion of such ambiguities in open interrogatives.) In [iv] the
content clause is complement of an adjective rather than a verb.
25
In ∗There are words or termsi in this Guide [that you may not be sure [what j theyi really mean j ]] (taken with
minor and irrelevant modification from an Australian government publication) the personal pronoun they
is used instead of a gap linked to the antecedent words or terms. Pronouns used in place of a gap in relative
clauses are known as ‘resumptive pronouns’. In some languages they represent a regular feature of relative
clause formation, but in English they are ungrammatical, as evident from their inadmissibility in simpler
constructions like ∗words or terms [whichi you may not understand themi ].
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1092 Chapter 12 Relative constructions and unbounded dependencies
(g) PP
[40] i Some of usi he wouldn’t even speak [to i ].
ii This is the knifei [you should cut the tomatoes [with i ]].
?
iii What dayi will you not be able to return the book [until i ]?
∗
iv Here is a list of the objectionsi [that they went ahead [despite i ]].
∗
v You pay mei , I’ll do it [if i ].
∗
vi It was this proposali [that they sacked me [because I criticised i ]].
In [i–iv] the gap is complement of a preposition and has an NP as antecedent. This
results in what is called a stranded preposition – a transitive preposition with the com-
plement missing but understood. It is a very common construction except in formal
style: see Ch. 7, §4.1. Stranding is most generally permitted when the PP is in comple-
ment function, as in [i]. With PPs functioning as adjunct, acceptability depends on the
semantic type of the adjunct and the particular preposition; for example, instrumental
with strands easily, whereas until is fairly resistant to stranding, and with despite it is
excluded.
PPs do not permit gaps linked to a finite clause antecedent, as illustrated in [40v].
The preposing here must apply to the whole PP, not just the complement: If you pay mei ,
I’ll do it i . Nor do they permit a gap within a finite clause that is complement of the
preposition, as we see from [vi]. Again it can be corrected by having the gap in place of
the whole reason PP: It was because I criticised this proposali that they sacked me i .
(h) NP
NPs accept gaps considerably less readily than VPs. Gaps cannot occur as or within mod-
ifiers in NP structure (see subsection (i), Modifiers, below). Complements are normally
either PPs or clauses, and we will consider these two cases in turn.
PP complements
[41] i Of which institutei did you say they are going to make him [director i ]?
ii To which safei is this [the key i ]?
iii He knows little about any of the companies [in whichi he owns [shares i ]].
iv I can’t remember [which countryi she served as [prime minister of i ]].
v What kinds of birdsi have you been collecting [pictures of i ]?
vi It’s a topici [that I’d quite like to write [a book about i ]].
vii ∗It’s a topici [you should read [my philosophy tutor’s book on i ]].
In [i–iii] the gap itself functions as complement and has a PP as antecedent. In [iv–vi] the
gap is complement of the preposition, yielding a further case of preposition stranding.
The NP in [iv–vi] is indefinite, and this construction is clearly acceptable except in formal
style. Where the NP is definite, however, and especially where it has a genitive determiner,
acceptability is generally very much reduced, as in [vii].
Clausal complements
∗
[42] i That it was my faulti I emphatically reject [the insinuation i ].
∗
ii How the accident happenedi they haven’t begun to address [the question i ].
∗
iii How muchi did the secretary file [a report that it would cost i ]?
∗
iv He’s someonei [I accept your contention that we should not have appointed i ].
v How many staffi did he give you [an assurance that he would retain i ]?
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 7.3 Location of gaps 1093
Examples [i–ii] have the gap itself in complement function, but this time – in contrast to
[41] – the result is very clearly ungrammatical. Preposing must apply to the whole NP:
The insinuation that it was my faulti I emphatically reject i and The question how the
accident happenedi they haven’t begun to address i .26 In [iii–v] the gap is within the
declarative content clause functioning as complement in NP structure. In general, this
construction is of low acceptability. There is, for instance, a very sharp difference between
[iii–iv] and comparable examples where the clause is complement of a verb: How muchi
did the secretary report that it would cost i ? and He’s someonei [that I agree we should
not have appointed i ]. However, the construction is by no means wholly excluded. It
is most acceptable in examples containing collocations of light verb + noun such as
give an assurance, make the claim, hold the belief, etc., which have essentially the same
meaning as the verbs assure, claim, and believe respectively (cf. Ch. 4, §7). Thus [v] does
not differ appreciably in acceptability from How many staffi did he assure you that he
would retain i ?
The examples in [42] involve content clauses; with infinitival complements gaps are
more generally admissible:
[43] i Whati had Dr Harris secretly devised [a plan to steal i ]?
ii It is not clear [which felonyi he is being charged with [intent to commit i ]].
(i) Modifiers
[44] i That’s the cari [that I’m saving up [to buy i ]].
ii Which monthi are you taking your holidays [in i ]this year?
∗
iii It’s this riveri [that I want to buy a house [by i ]].
∗
iv List the commoditiesi [that you have visited countries [which produce i ]].
Gaps occur very much less readily in modifiers than in complements. One type of
modifier where they are unquestionably allowed, however, is an infinitival clause of
purpose in VP structure, as in [i]. Example [ii] shows a gap inside a PP functioning as
modifier of temporal location, but we noted in (g) above that the stranded preposition
construction has a quite strong preference for PPs in complement function. Modifiers
in NP structure very strongly resist internal gaps, as illustrated in [iii–iv]. In [iii] the
gap is in a PP modifying house, while that in [iv] is in a relative clause modifying
countries.
(j) Subjects
[45] i They have eight children [of whomi [five i ]are still living at home].
ii ∗They have eight children [whoi [five of i ]are still living at home].
iii ∗Whati would [to look at i too closely] create political problems?
Gaps are almost wholly excluded from occurring within a subject. The main exception
is the construction shown in [i], where the gap is complement within the subject NP
and has a PP as antecedent. Examples like [ii–iii] are completely ungrammatical; in [ii]
the gap is within a PP dependent in the subject NP, while in [iii] it is within a clause
functioning as subject. The clause in this example is infinitival, but the same prohibition
26
Examples like Why he did iti I have no idea i are acceptable, but here we take the interrogative clause to be a
complement in the structure of the VP, not the NP: see Ch. 4, §6.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1094 Chapter 12 Relative constructions and unbounded dependencies
applies to finite clauses, as seen in the example used in the introduction to this section,
[31ii].27
(k) Coordinates
We saw in §7.2 that a gap cannot itself function as a coordinate (cf. [13viii]), but there
are also constraints on the occurrence of gaps within coordinates. Compare:
[46] i Who was the guyi [that [Jill divorced i ][and Sue subsequently married i ]]?
ii ∗Who was the guyi [that [Jill divorced Max] [and Sue subsequently married i ]]?
In general, a gap can occur within a coordinate element only if a gap with the same
antecedent occurs in all other coordinates in the coordination construction. In [i], for
example, each of the two coordinates (enclosed by the inner sets of brackets) con-
tains a gap in object function with guy as its antecedent. The sentence presupposes
that there was some guy x such that Jill divorced x and Sue subsequently married x.
Example [ii] is ungrammatical because the gap figures in one coordinate but not the
other.
There are certain conditions, however, under which this constraint is relaxed:
[47] i There are some lettersi [that I must just [go downstairs] [and check i over]].
ii What is the maximum amounti [I can [contribute i ] [and still receive a tax
deduction]]?
iii He has built up a high level of expectations, [whichi he must [either live up to i ]
[or suffer a backlash]].
These are cases of asymmetric coordination, i.e. cases where the coordinates are not of
equal status from a semantic point of view (see Ch. 15, §§2.2.3–4). This is reflected in
the fact that such coordinations have approximate paraphrases where one coordinate
is replaced by an adjunct. Compare I’ll go downstairs and check them over with I’ll go
downstairs to check them over ; I contributed $1,000 and still received a tax deduction with
Although I contributed $1,000, I still received a tax deduction; He must either live up to
these expectations or suffer a backlash with If he doesn’t live up to these expectations, he will
suffer a backlash. Note that in each case the gap appears in the coordinate corresponding
to the adjunct in the paraphrase. We pointed out at the beginning of this section that
the acceptability of gaps in various locations is not determined by purely grammatical
factors, and the contrast between [46] and [47] is a clear instance where a grammatical
constraint is overridden by semantic factors.
27
The following is a rare attested example of a gap within an infinitival subject, showing that the constraint is not
absolute: The eight dancers and their caller, Laurie Schmidt, make up the Farmall Promenade of nearby Nemaha,
a towni [that [to describe i as tiny] would be to overstate its size].
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
§ 7.5 Parasitic gaps 1095
[48]
Which of the two instrumentsi will this piecej be easier [to play j on i]?
The brackets enclose the hollow clause, and the underlining marks the antecedents of the
two gaps, one functioning as object of the verb play, the other as object of the preposition on.
One plays pieces of music on instruments, as reflected in a main clause such as Kim will play
the sonata on this piano. The NP containing the noun piece will thus be the antecedent for the
gap which is object of play, and the NP containing the noun instruments will be antecedent
for the gap which is object of on.
It will be noted from the diagram that the first antecedent is linked to the second gap, and
the second antecedent to the first gap: the pair with the j index is said to be nested between
the pair with the i index. The dependency relations are required to be nested one within the
other in this way. It is not possible for them to cross each other, as in:
[49]
Here the second gap cannot be replaced by a personal pronoun. There is thus nothing
parasitic about the second gap here: it is required by the rules for coordination. As
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013
1096 Chapter 12 Relative constructions and unbounded dependencies
explained in Ch. 15, §2.1, a distinctive property of coordination is that such processes as
relativisation must apply ‘across the board’: if relativisation applies within one coordi-
nate it must apply within all. This is what accounts for the difference in grammaticality
in [46], where [i] satisfies the across-the-board requirement and [ii] violates it.
Parasitic and across-the-board gaps can combine, as in the following attested example,
where the parasitic gap is marked by an initial subscript ‘p’:
[52] Fairbanks reached for a towel, a clean one and not the scarcely crumpled onei [that
Comore himself had [used i ] [and left i thriftily on the ledge below the mirror
rather than consign p i to the linen basket]].
The outer brackets enclose a relative clause within which there is a coordination func-
tioning as complement of the perfect auxiliary have. The two coordinates, enclosed by the
inner pairs of brackets, each have an ordinary gap as object (of used and left respectively),
but in addition the second coordinate has a parasitic gap in the adjunct headed by rather.
Again, the parasitic gap could be replaced by the personal pronoun it, but the ordinary
gaps could not.
Downloaded from https:/www.cambridge.org/core. Boston University Theology Library, on 04 Jun 2017 at 01:29:10, subject to the Cambridge Core terms of use,
available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316423530.013