Silent prepositions: Evidence from free relatives

Abstract

Silent prepositions are hypothesized to be prepositions that are phonologically null, but syntactically and semantically contentful. Their existence in the grammar has been argued for on the basis of the behavior of expressions like there, now, and that way, which look like nominals but can (or must) behave as prepositional phrases. We bring further evidence in favor of silent prepositions by studying the syntactic and semantic behavior of free relative clauses introduced by the wh-words where, when, and how. We propose a fully compositional syntactic/semantics analysis for these free relatives that accounts for their puzzling nominal/prepositional distributional and interpretative properties. This analysis is crucially based on silent prepositions. We also briefly discuss how our proposal can be extended to interrogative clauses introduced by the same wh-words, as well as what the licensing conditions and semantic properties of silent prepositions are.

Full text

A slightly different version appeared in: Anna Asbury, Jakub Dotlačil, Berit Gehrke, and Rick Nouwen (eds),
The Syntax and Semantics of Spatial P, pp. 365-385, Amsterdam: Benjamins, 2008.
Silent prepositions: Evidence from free relatives*
IVANO CAPONIGRO
University of California, San Diego
ivano@ling.ucsd.edu
LISA PEARL
University of California, Irvine
pearl@uci.edu
Abstract
Silent prepositions are hypothesized to be prepositions that are phonologically null, but
syntactically and semantically contentful. Their existence in the grammar has been argued for on
the basis of the behavior of expressions like there, now, and that way, which look like nominals
but can (or must) behave as prepositional phrases. We bring further evidence in favor of silent
prepositions by studying the syntactic and semantic behavior of free relative clauses introduced
by the wh-words where, when, and how. We propose a fully compositional syntactic/semantics
analysis for these free relatives that accounts for their puzzling nominal/prepositional
distributional and interpretative properties. This analysis is crucially based on silent prepositions.
We also briefly discuss how our proposal can be extended to interrogative clauses introduced by
the same wh-words, as well as what the licensing conditions and semantic properties of silent
prepositions are.
*We are very grateful to Grant Goodall, Richard Larson, Howard Lasnik, Carson Schütze,
and the participants at the Syntax and Semantics of Spatial P Workshop at the University of
Utrecht for their very helpful comments and suggestions. Any remaining errors are, of
course, our own.

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
2
1. Introduction: Silent prepositions
As noticed by Emonds (1976, 1987), there are a restricted number of phrases in English that
look like NPs, but have the same distribution and interpretation as either NPs or PPs. Larson
(1985) labels these phrases bare NP-adverbs, while McCawley (1988) calls them adverbial
NPs. An example is the string few places that I cared for, which appears to simply be a
(complex) NP. In (1)a, it occurs in a position in which only NPs usually occur (i.e. the
subject position of a predicate like be beautiful), receiving the usual interpretation of an NP
subject (i.e. an individual or a generalized quantifier). However, in (1)b the very same string
occurs in the same position as the bracketed locative PP adjunct in (1)c and has the same
interpretation as that PP.
(1) a. [Few places that I cared for] are really beautiful. NP-like
b. You have lived [few places that I cared for]. PP-like
c. You have lived [PP in [few places that I cared for]].
Adverbial NPs are restricted to three semantic areas: spatial expressions (2), temporal
expressions (3), and manner expressions (4).1
(2) Spatial adverbial NPs
a. You have lived [there]. PP-like
b. [There] is really beautiful. NP-like
c. I went [home]. PP-like
d. [Home] never changes. NP-like
(3) Temporal adverbial NPs
a. John arrived [that day]/[Sunday]/[yesterday]. PP-like
b. [That day]/[Sunday]/[Yesterday] was fantastic. NP-like
(4) Manner adverbial NPs
a. You pronounced my name [that way]/[every way one could imagine]. PP-like
b. [That way]/[Every way one could imagine] was not feasible. NP-like
NPs from other semantic areas do not exhibit the same behavior; they can only behave like
NPs and not like PPs (5).
1 Examples (2)a, (3)a, and (4)a from Larson (1985).

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
3
(5) a.* Jack came [her]/[Lily]/[the person he is in love with]. * PP-like
(cf. Jack came for/with/after [her]/[Lily]/[the person he is in love with].)
b. [She]/[Lily]/[the person he is love with] does not really like him. NP-like
Larson (1985) argues that adverbial NPs are syntactically NPs with the (lexical) property
of self-assigning Case, but Emonds (1987) and McCawley (1988) convincingly show that
Larson’s proposal is empirically and theoretically problematic. Emonds (1976, 1987) and
McCawley (1988) argue that adverbial NPs are NPs that can occur in positions to which no
Case is assigned and behave like PPs in virtue of a silent preposition that takes the NP as its
complement (6).
(6) PP
2
P NP
silent P adverbial NP
In this paper, we bring further support for silent prepositions in the grammar by looking at
the syntactic and semantic behavior of embedded non-interrogative wh-clauses known as free
relatives (FRs). The bracketed FR in (7)a has the same distribution and roughly the same
interpretation as the NP in (7)a’. Yet, in (7)b what looks like the very same FR now occurs in
the same position as the PP in (7)b’ and receives a similar interpretation.
(7) a. [Where you used to live] was really great. NP-like
a’. [NP The little town you used to live in] was really great.
b. I live [where you used to live]. PP-like
b’. I live [PP in [the town you used to live in]].
We will argue that this and another puzzling property of FRs can receive a
straightforward account if silent Ps are available in the grammar and the same structure as in
(6) is assumed. The only difference is that the silent P takes a FR as its complement rather
than an adverbial NP (8).2
(8) PP
2
P CP
silent P FR
The paper is organized as follows. First, we will discuss two puzzles about FRs (§2), and
present a proposal that accounts for them (§3). Then, we will bring further independent
evidence to support the proposal (§4), and discuss potential alternatives and their associated
2 We will discuss later why we assume that a FR is syntactically a CP.

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
4
problems (§5). Finally, we will briefly mention some open issues and conclude (§6).
2. Two puzzles about free relatives
We focus on two puzzles (among many) which FRs raise that have not received much
attention previously. We argue that they can be accounted for by means of silent Ps.
2.1. Puzzle I: Are FRs like NPs or PPs?
Some FRs can have the same distribution and interpretation as NPs or PPs. These are the FRs
that are introduced by the wh-words where (place), when (time), and how (manner)
(henceforth, w/w/h FRs). The examples from (9)-(11) illustrate this point. (9)a shows that a
w/w/h FR introduced by the spatial wh-word where can behave like an NP, since it can be
replaced and paraphrased with the complex NP the place that this very tree grows (9)a’. On
the other hand, (9)b shows that what looks like the very same w/w/h FR can syntactically and
semantically behave like the complex PP in the place that this very tree grows (9)b’. The
same is true for w/w/h FRs introduced by the temporal wh-word when (10) and the manner
wh-word how (11).
(9) a. Lily adores [FR where this very tree grows [PP __ ]].
a’. Lily adores [NP the place that this very tree grows [PP __ ]].
b. Lily napped [FR where this very tree grows [PP __ ]].
b’. Lily napped [PP in the place that this very tree grows [PP __ ]].
(10) a. Lily dreaded [FR when Jack had to go [PP __ ]].
a’. Lily dreaded [NP the time/moment that Jack had to go [PP __ ]].
b. Lily cried [FR when Jack had to go [PP __ ]].
b’. Lily cried [PP at the time that Jack had to go [PP __ ]].
(11) a. Lily loathes [FR how all thieves work [PP __ ]] – secretly.
a’. Lily loathes [NP the way that all thieves work [PP __ ]] – secretly.
b. Jack works [FR how all thieves work [PP __ ]] – secretly.
b’. Jack works [PP in the way that all thieves work [PP __ ]] – secretly.
Bresnan and Grimshaw (1978: §5) were the first to notice this distribution of w/w/h FRs,
though they only discuss examples of FRs introduced by where and when, without
mentioning FRs introduced by how.
So, we see that w/w/h FRs exhibit the same puzzling behavior as adverbial NPs: they can
have the same distribution and interpretation as either NPs or PPs.

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
5
2.2. Puzzle II: An NP or a PP gap?
The second puzzle concerns the nature of the gap within w/w/h FRs. To the best of our
knowledge, this has not been noticed before. Similar to other wh-clauses like interrogatives
and headed relative clauses, FRs are characterized by a gap in argument or adjunct position.
In (9)-(11) above, we marked and labeled the gaps in the w/w/h FRs and the corresponding
headed relatives. All the gaps are PP gaps, and this is not by chance. It does not matter
whether the whole w/w/h FR behaves like an NP or a PP – only a PP gap is licensed within
the w/w/h FR. Even if the predicate in the w/w/h FR selects for an NP, w/w/h FRs can not yet
license an NP gap – and it does not matter if the whole w/w/h FR behaves like a PP (12)a or
an NP (13)a. Notice that the corresponding headed relatives can license an NP gap,
independently from the properties of their heads (cf. (12)a vs.(12)b and (13)a vs. (13)b).
(12) a. ?* Lily always naps [FR where/when/how Jack despises [NP __ ]].
b. Lily always naps [PP {in the place}/{at the time}/{in the way} that Jack despises
[
NP __ ]].
(13) a. ?* Lily adores [FR where/when/how Jack despises [NP __ ]].
b. Lily adores [NP {the place}/{the time}/{the way} that Jack despises [NP __ ]].
There is an important exception, though. An NP gap can be licensed as the complement of
an overt P in a w/w/h FR that is introduced by where, whether the whole FR behaves like an
NP or a PP (in §4.1, we will discuss similar examples with FRs introduced by the wh-word
when). For example, the w/w/h FR introduced by where in (14)a allows for an NP gap in the
complement position of the P past. Since past can only take NP complements, we can be sure
that we are dealing with an NP gap. The whole FR behaves like an NP. In fact, it can be
replaced and paraphrased with the bracketed complex NP in (14)b. Similarly, the w/w/h FR
in (15)a licenses an NP in the complement position of the preposition through, though it
behaves like a PP, as shown in (15)b.
(14) a. Jack disliked [FR where we just ran [PP [P past] [NP __ ]]] – it smelled funny.
b. Jack disliked [NP the place we just ran [PP [P past] [NP __ ]]] – it smelled funny.
(15) a. Lily lives [FR where we have to fly [PP [P through] [NP __ ]] on our way to
Vancouver].
b. Lily lives [PP in the area we have to fly [PP [P through] [NP __ ]] on our way to
Vancouver].
To sum up, the second puzzle about w/w/h FRs we will address is that only a PP gap
seems to be licensed inside w/w/h FRs, unless an overt P occurs, which then triggers the
licensing of an NP gap as its complement.

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
6
3. Proposal
The two puzzles just discussed can be solved by giving a syntactic and semantic analysis of
w/w/h FRs that relies on silent Ps. Let us start with a preliminary syntactic assumption: FRs
introduced by bare phrasal3 wh-words (who, what, where, when, how) are plain CPs with a
moved wh-phrase in their specifiers. This is just for the sake of simplicity, since the syntax of
FRs is an open issue that goes largely beyond the purposes of this paper.4 Nonetheless,
nothing crucial in our proposal hinges on this assumption. Any analysis in which the
wh-word in a FR is base-generated in the gap position and moved to a higher position will
suffice.
Our proposal then can be articulated in two main claims. First, a whole w/w/h FR (i.e. a
CP) can occur as the complement of a silent P. When the w/w/h FR is the complement of a
silent P, the whole w/w/h FR looks as if it is a PP. Note that this is because the w/w/h FR is
embedded within an actual PP, whose head is silent (16)a. When the w/w/h FR is not the
complement of a silent P, then it behaves like an NP, just as all other FRs introduced by bare
phrasal wh-words do (16)b. This accounts for the first puzzle, where w/w/h FRs behave
sometimes as NPs and sometimes as PPs.
(16) a.
PP
3
P CP
silent P 5
FR
b.
CP
5
FR
Our second claim is that the wh-words where, when, and how, which introduce
w/w/h FRs, are syntactically and semantically NPs that can only be base-generated as the
complement of a possibly silent P (17).
(17) CP
3
wherei/wheni/howi C’
6
PP
3
P NP
(silent) P t
i
3 “Bare” excludes FRs introduced by wh-ever (whoever, whatever, wherever, whenever),
“phrasal” excludes FRs introduced by wh(-ever) + XP (e.g. I’ll drink what(ever) beer you
drink).
4 See Grosu (2003) and van Riemsdijk (2005) for an overview of the open issues of the
syntax of FRs, the proposals have been made, and their problems.

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
7
The wh-word that is base-generated as the sister of P moves to the Spec of CP for
independent reasons, specifically whatever the reasons for wh-movement in English are. The
wh-word leaves behind an NP trace, which is an NP gap. Note that this NP gap is
perceptually identical to a PP gap when the P is silent. If the P is overt as in (14) and (15)
above, it becomes clear that at least where is an NP. Also, according to our analysis, silent Ps
are always stranded. This is due to independent semantic reasons that will be discussed
below. This accounts for the second puzzle, in which FRs seem to license PP gaps and NP
gaps.
Let us now look at the detailed syntactic derivations of two of the previous examples in
order to make concrete how our proposal works. The syntactic tree in (18)b includes the CP
of a w/w/h FR that behaves like an NP, since it occurs in the complement position of a
predicate like adores (which selects for an NP complement). The w/w/h FR contains a PP
with a silent P and a wh-trace as its complement, resulting from the wh-movement of where
to the Spec of CP of the w/w/h FR. Notice that the fact that P is silent makes the NP gap
perceptually indistinguishable from a true PP gap like in the wh-interrogative clause In which
areas does this tree grow [PP __ ]?
(18) a. Lily adores [FR where this very tree grows].
b. IP
6
Lily VP
3
V CP
adores 3
NP C’
4 3
wheren C IP
3
NP I’
4 3
this very tree I VP
3
VP PP
4 3
grows P NP
e t
n
The next example (19)a contains the same w/w/h FR as in (18), but this time the w/w/h FR
specifically behaves like a PP. In fact, the FR looks as if it was occurring where a PP, but not
an NP, can occur, that is as an adjunct of the predicate napped. But our tree in (19)b shows
that this apparent puzzling behavior can once again be accounted for as a perceptual illusion.
The CP of the w/w/h FR occurs in the complement position of a silent preposition P2.

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
8
Therefore, the whole w/w/h FR still behaves like an NP, as in (18)b . However, PP2, which is
made of its head P2 and the CP of the w/w/h FR as its complement, is perceptually
indistinguishable from just the CP of the w/w/h FR, since P2 is silent. The internal structure
of the w/w/h FR with the silent preposition P1 in (19)b is identical to (18)b above.
(19) a. Lily napped [FR where this very tree grows].
b. IP
6
Lily VP
3
V PP
2
napped 3
P
2 CP
e 3
NP C’
4 3
wheren C IP
3
NP I’
4 3
this very tree I VP
3
VP PP
1
4 3
grows P1 NP
e
1 t
n
Let us now move to the semantics of w/w/h FRs. This is relevant for our purposes for at
least two reasons. First, we want to make sure that the syntactic analysis just suggested is
compatible with a semantic analysis that captures the semantic intuitions we discussed
earlier: w/w/h FRs should be semantically equivalent to either NPs or PPs. Second, we need a
detailed semantics for w/w/h FRs, since our argument to support obligatory stranding of
silent Ps is crucially based on the semantic derivation.
Following the semantic analysis for FRs in Caponigro (2004), we pursue the idea that
w/w/h FRs denote an individual, like non-quantified NPs. How do w/w/h FRs end up
denoting an individual? Let us go step by step from the bottom up, following the schema in
(20).

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
9
(20) CP
2
3
δ | λXιx[X(x)] CP1 | λx1 [WH(x1) . Q(x1)]
3
wh-1 | λXλx1 [WH(x1) . X(x1)] C’ | λx1Q(x1)
3
C | λx1 IP | Q(x1)
6
t
1 | x1
The wh-trace within IP translates into a variable that is bound at by the set-formation
operator, the lambda operator (which we assuming to be under C just for the sake of
simplicity). Therefore, C’ denotes the set of individuals that have the property Q (i.e. the set
λx1Q(x1)) or, equivalently, the set of individuals that make the proposition Q(x) expressed by
the IP true.
The next step is to combine this set with the semantic contribution of the wh-word. But
what is the semantic contribution of where, when, or how? These words behave like set
restrictors: they apply to a set of individuals and return a subset. In particular, where applies
to a set of individuals and returns all and only the individuals that are locations, when all and
only instants or situations, and how all and only manners. When one of these wh-words
applies to the set of individuals λx1Q(x1), it returns the subset λx1 [WH(x1) . Q(x1)], i.e. the
set of individuals that have both the property Q and the property WH, the restriction
conveyed by the wh-word (see Caponigro 2004 for further details and arguments).
So at the level of CP1, w/w/h FRs denote the set of locations, instants/situations, or
manners that satisfy the property Q. But, according to our syntactic analysis, w/w/h FRs,
which so far denote sets of individuals (semantic type: <e,t>), occur in the argument positions
of heads (verbs or preposition) that select for an individual-denoting expressions (semantic
type: <e>). It appears we have a so-called type mismatch. Partee (1986), Chierchia (1998),
and Dayal (2004) have argued that type-shifting rules are made available by the grammar to
fix type-mismatches. Among those, iota (ι) applies to a set P and returns its maximal
individual (21).
(21) iota (ι): P → ιxP(x) (<e,t> → <e>)
The empirical evidence for a type-shifting rule like iota originally comes from the
crosslinguistic behavior of nominals, in particular bare plurals and bare singulars (Chierchia,
1998; Dayal, 2004). Caponigro (2004) – developing a suggestions in Jacobson (1995) – adds
further support coming from FRs. He syntactically encodes iota by means of a silent operator
δ, adjoined to CP of FRs. We did the same in (20). The set of individuals denoted by CP1 in
(20) type-shifts to its maximal (plural) individual when it combines with δ at the level of the
adjoined CP2.
In (22) and (23) below, we repeat the syntactic trees in (18) and (19) respectively and add
their logical translations in accordance to the analysis we just proposed. PPs are treated as
extra arguments of the predicate they modify for sake of simplicity. Also, no detailed
semantic analysis is given for PPs, since it is not relevant for our purposes. Any semantics for

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
10
Ps would suffice as long as Ps select for an individual-denoting complement. Finally, we
ignore the deictic nature of nominal this very tree for the sake of simplicity and we translate it
as the simple individual-denoting expression tvt.
(22) IP | adores′(ιx [place′(x) ∧ grow′(in′(x))(tvt)]))(lily)
Lily adores CP
2 | ιx [place′(x) ∧ grow′(in′(x))(tvt)]
3
δ | λXιx[X(x)] CP1 | λx1[place′(x1) ∧ grows′(in′(x1))(tvt)]
2
NP C’| λx1[grows′(in′(x1))(tvt)]
where1 | λXλx1[place′(x1) . X(x1)] 2
C IP
1 | grows′(in′(x1))(tvt)
λx1
this very tree grows PP1
|
in′(x1)
2
P1 NP
e1|in′ t
1 | x1
(23) IP | napped′(in′(ιx [place′(x) ∧ grow′(in′(x))(tvt)]))(lily)
Lily napped PP
2 | in′(ιx [place′(x) ∧ grow′(in′(x))(tvt)])
3
P2 CP
2 | ιx [place′(x) ∧ grow′(in′(x))(tvt)]
e2 |in′ 3
δ | λXιx[X(x)] CP1 | λx1[place′(x1) ∧ grows′(in′(x1))(tvt)]
2
NP C’| λx1[grows′(in′(x1))(tvt)]
where1 | λXλx1[place′(x1) . X(x1)] 2
C IP
1 | grows′(in′(x1))(tvt)
λx1
this very tree grows PP1
|
in′(x1)
2
P1 NP
e1|in′ t
1 | x1
To sum up, we have argued that w/w/h FRs are very similar both syntactically and
semantically to the FRs introduced by the bare phrasal wh-words who and what.
Syntactically, they are wh-CPs (or any other phrasal structure that would better account for
the property of FRs); semantically, they denote an individual. The crucial difference is the
structure of the wh-words that introduce w/w/h FRs: where, when, and how are always
base-generated as the complement of a possibly silent P. In addition, whenever the w/w/h FR

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
11
behaves as if it was a PP, the whole w/w/h FR itself is the complement of another silent P.
4. Further evidence
In this section, we present further evidence in favor of our analysis. We show that silent Ps –
either within a w/w/h FR or as its sister – can be replaced with overtly realized Ps in some
cases (§4.1). We give further support to our proposal that where, when, and how are NPs,
rather than PPs, by showing that Ps usually take an NP as their complement, rather than a PP
(§4.2). We then show that our proposal about the syntactic and semantic nature of w/w/h FRs
has larger empirical coverage by extending it to wh-interrogatives that are introduced by
where, when or how (§4.3). Finally, we show that the same kinds of restrictions on licensing
silent Ps that we noticed for adverbial NPs earlier (§1) are found with w/w/h FRs as well
(§4.4), which supports the claim that these two constructions are similar and should receive a
similar analysis.
4.1. Overt Ps
We have argued that the wh-trace within a w/w/h FR is always the complement of a possibly
silent P and that the whole w/w/h FR is the complement of another silent P whenever the
w/w/h FR behaves as if it was a PP. Though these two Ps are silent most of the time, there are
cases where one or both of them are overt. In (24), we see that the P that takes the wh-trace
of where as its complement can be overtly realized, in this case as the P past. In (25), we see
that the P that takes the whole w/w/h FR introduced by where as its complement can be
overtly realized, in this case as the P near. In (26), both the P sister to the wh-trace and the P
sister to the w/w/h FR are overtly realized as through and near, respectively.
(24) Jack disliked [FR wherem we just ran [PP [P past] [NP tm]]] – it smelled funny.
(25) Lily lives [PP1 [P1 near] [FR wherem we had dinner [PP2 [P2 e] [NP tm]] last night]].
(26) Lily lives [PP1 [P1 near] [FR wherem we have to fly [PP2 [P2 through] [NP tm]] on our way
to Vancouver]].
In (27), we see that w/w/h FRs introduced by the wh-word when can also occur as the
complement of over Ps like from or to, while (28) shows that the trace of the wh-word when
can be the complement of an overt P as by.
(27) Lily was sick [PP1 [P1 from] [FR whenm Jack arrived [PP2 [P2 e] [NP tm ]]]] [PP3 [P3 to]
[FR whenp he left [PP4 [P4 e] [NP tp ]]]].
(28) Lily’s schedule can’t accommodate [whenm Jack needs the car [PP [P by [NP tm ]]]].

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
12
Finally, (29) shows that a whole FR introduced by the wh-word how can occur as the
complement of an overt preposition like on.5
(29) Lily knew that Jack was about to get upset based [PP1 [P1 on] [howm he was looking at
her [PP2 [P2 e] [NP tm ]]]]
We take these cases where the P is overt as further evidence in favor of the existence of
the two distinct P heads we are arguing for, even when these P heads are not phonologically
realized.
4.2. Complements of P must be NPs
We just saw that where, when, and how can be base-generated as complement of Ps like past,
through, or near. These Ps can take an NP complement, but not a PP one, as shown in
(30)-(32).
(30) a. past [NP the house]
b. * past [PP at the house]
(31) a. through [NP the grass]
b. * through [PP on the grass]
(32) a. near [NP the store]
b. * near [PP in the store]
Since where, when, and how can be base-generated as the complement of Ps like past,
through, or near it follows that they must be NPs, rather than PPs. We take this as
independent evidence in favor of our conclusions that the categorical status of where, when,
and how is NP, rather than PP.
5 We have not been able to find any example in which the trace of how can occur as the sister
of an overt preposition. Although we do not have an account for this asymmetry of how with
respect to where and when, we note two possibly related facts. First, a headed relative clause
with the nominal way as its head, which is the most natural paraphrase for a FR introduced by
how, can be optionally followed by the complementizer that or, crucially, by the relative
pronoun which preceded by the preposition in. No other preposition can occur between way
and which.
(i) Lily knew that Jack was about to get upset based [[on] [the way [(that)/(in/*by/*with
which) he was looking at her]]].
Second, the P in can never occur as the sister of the trace of where, when, and how. If for
whatever reason in is the only overt preposition in English that is compatible with very
general manner expressions like way, but at the same time is incompatible with where, when
and how, it follows that no overt preposition can ever take the trace of how as its
complement.

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
13
4.3. Wh-interrogative clauses
Our claim that where, when, and how are always base-generated as the complement of a P
would look suspiciously restricted if it was true just for w/w/h FRs rather than for any
wh-clause introduced by those wh-words. Wh-interrogatives introduced by where behave
exactly like the w/w/h FRs we saw in §4.1 (33)-(34). Their wh-word is base-generated as the
complement of a P that can also be overt. Huang (1982:536) makes a similar point.
(33) Wherem did we just run [PP [P past] [NP tm]]?
(34) Wherem do we have to fly [PP [P through] [NP tm]] on our way to Vancouver?
Also, if no overt P is present, wh-interrogatives introduced by where, when, and how only
license what looks like a PP gap on the surface (35). This is the same behavior that we
observed in w/w/h FRs (§2.2).
(35) a. * Where/when/how did Lily despise [NP __ ]?
b. Where/when/how did Lily sleep [PP __ ]?
4.4. Semantic restrictions on silent P licensing and adverbial NPs
At the very beginning of this paper (§1), we mentioned that adverbial NPs in English look
like NPs, but can behave like NPs or PPs. We also noticed adverbial NPs are restricted to
three semantic areas: spatial expressions (36), temporal expressions (37), or manner
expressions (38). NPs from other semantic areas do not exhibit the same pattern (39).
(36) Spatial adverbial NPs
a. You have lived [there]. PP-like
b. [There] is really beautiful. NP-like
c. I went [home]. PP-like
d. [Home] never changes. NP-like
(37) Temporal adverbial NPs
a. John arrived [that day]/[Sunday]/[yesterday]. PP-like
b. [That day]/[Sunday]/[Yesterday] was fantastic. NP-like
(38) Manner adverbial NPs
a. You pronounced my name [that way]/[every way one could imagine]. PP-like
b. [That way]/[Every way one could imagine] was not feasible. NP-like
(39) a. * Jack came [her]/[Lily]/[the person he is in love with]. *PP-like
(cf. Jack came for/with/after [her]/[Lily]/[the person he is in love with].)
b. [She]/[Lily]/[the person he is love with] does not really like him. NP-like

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
14
W/w/h FRs show interesting similarities with adverbial PPs. As we have seen, w/w/h FRs
can also behave like NPs or PPs. In addition, w/w/h FRs are introduced by wh-words that
carry the same semantic feature as adverbial NPs: location/space (where), time/situation
(when), or manner (how). This contrasts with FRs introduced by other bare phrasal wh-words
like who (animate) and what (inanimate) that do not behave like adverbial NPs. In particular,
they can occur where NPs can occur, but cannot occur where PPs can occur. As an example,
(40)a shows that a FR introduced by what can occur as the complement of a predicate like
adore, which selects for an NP complement, as shown (40)a’. However, if the very same FR
occurs as the adjunct of an intransitive predicate like work in which only PPs are allowed, as
shown in (40)b’, then the result is unacceptable, as shown in (40)b. The examples in (41)
make a similar point for FRs introduced by who.
(40) a. Lily adores [FR what Jack despises]. NP-like
a’. Lily adores [NP the things Jack despises].
b. * Lily works [FR what Jack despises]. * PP-like
b’. Lily works [PP [P on] [NP the things Jack despises]].
(41) a. Lily won’t marry [FR who the king chooses]. NP-like
a’. Lily won’t marry [NP the person the king chooses].
b. * Lily will dance [FR who the king chooses]. * PP-like
b’. Lily will dance [PP [P with] [NP the person the king chooses]].
We take these similarities between w/w/h FRs and adverbial NPs as an indication that a
similar syntactic and semantic analysis for both constructions is justified. As we briefly
mentioned in §1, an analysis based on silent Ps has been convincingly put forth for adverbial
NPs (Emonds, 1976, 1987; McCawley, 1988). We take this to further support our analysis of
w/w/h FRs based on silent Ps. In particular, FRs introduced by who and what lack any spatial,
temporal, or manner feature; therefore, they cannot occur as the complement of a silent P.
The reason that FRs introduced by who and what lack the relevant features is because the
wh-words that introduce them and act as a set restrictor lack them as well. This makes the
prediction that who and what should never be base-generated as the complement position of a
silent P or, equivalently, FRs introduced by who or what should never license a PP gap. This
prediction is borne out. A FR introduced by who or what can license an NP gap in the
complement of the P to (42)a, but cannot license a PP gap within what looks like the same FR
except for the omitted P (42)b. As expected, a FR introduced by where can easily license a
PP gap in the very same environment (42)c.
(42) a. Lily doesn’t like [FR who/what Jack goes to [NP __ ] every Friday night].
b.* Lily doesn’t like [FR who/what Jack goes [PP __ ] every Friday night].
c. Lily doesn’t like [FR where Jack goes [PP __ ] every Friday night].
5. Problems with alternative accounts
In this section, we discuss potentially alternatives to our proposal and where they are

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
15
problematic.
5.1. Against an ambiguity account
A simple alternative to our proposal could be to assume that where, when, and how are
syntactically ambiguous: they are listed in the lexicon as both NPs and PPs. There are two
issues with this approach, however. First, we would not be dealing with the idiosyncratic
behavior of just one word, but with three lexical items that are members of the same class
(wh-words) and exhibit very similar syntactic behaviors (introduce wh-clauses, license both
NPs and PP gaps, etc.). Also, we have observed the same pattern in other languages with
w/w/h FRs like Italian (although a careful crosslinguistic investigation is needed). This kind
of (crosslinguistic) systematic ambiguity appears more like restating the generalization than
an actual explanation.
Second, an ambiguity approach would end up making at least two incorrect predictions in
this specific case. If where, when, and how were listed in the lexicon as both NPs and PPs,
then they should be able to license either NP or PP gaps in the clause they introduce. But this
is not the case, since they license only PP gaps (§2.2). Also, obligatory matching between the
syntactic category of the gap and the whole FR would be expected. In particular, if a PP gap
is licensed within a w/w/h FR, then the whole w/w/h FR should always behave like a PP.
This is also not the case. While where, when, or how only license a PP-gap within w/w/h FRs,
the whole w/w/h FR can behave either like an NP or PP, depending on the matrix predicate
(43).
(43) a. Lily adores [FR [where/when/how] Jack sleeps [PP __ ]]. NP-like FR
b. Lily sleeps [FR [where/when/how] Jack sleeps [PP __ ]]. PP-like FR
5.2. Against a pied-piping analysis
Another alternative approach could be based on optional pipe-piping of silent Ps. Like ours, it
would assume that where, when, and how are always base-generated as the NP complement
of a possibly silent P. Unlike ours, it would also assume that whenever the whole w/w/h FR
behaves like a PP, this is due to the entire wh-PP moving to the Spec of CP, rather than just
the wh-word. In other words, this approach would replace the stranding of the silent P in the
w/w/h FRs with its pied-piping. There would be no need to postulate another silent P, which,
according to our proposal, takes the whole w/w/h FR as its complement. In (44), we see the
syntactic derivation of the example we discussed in (19) according to the pied-piping
approach: it is the whole wh-PP [e where]m that moves to Spec of CP, leaving behind the PP
trace tm.

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
16
(44) a. Lily napped [FR where this very tree grows].
b. IP
6
Lily VP
3
V CP
napped 3
PP
m C’
1 3
P NP C IP
e where 3
NP I’
4 3
this very tree I VP
3
VP PP
4 tm
grows
This approach faces at least two major problems. First, it would predict that the silent PP
within the w/w/h FR and the whole w/w/h FR should always be interpreted as the same kind
of PP, since their interpretation would depend on the very same silent P. This prediction is
not borne out. The w/w/h FR in (45)a is interpreted as basically equivalent to the bracketed
complex NP in (45)b, the only relevant difference being that all Ps are overt in (45)b. In
particular, the PP gap in (45)b results from the movement of the PP at which, which is headed
by the overt P at. Yet, the whole complex NP occurs as the complement of a PP that is
headed by a different P: to. Crucially, the two Ps are not identical and cannot be interpreted
equivalently. If we try to interpret the clauses this way, at would also precede the whole
complex NP and the result would be semantically different from the intended meaning of
(45)a – as well as unacceptable, as shown by (45)c. Thus, a pied-piping approach with only
one silent P as in (45)d fails. In contrast, our approach can easily handle these facts, since
two independent silent Ps would be present in the syntactic/semantic representation, as shown
in (45)e.

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
17
(45) a. Lily just went [FR where Jack stayed last year on vacation].
b. Lily just went [PP to [NP the place [PP at which] Jack stayed [PP tj] last year on
vacation]].
c. * Lily just went [PP at [NP the place [PP at which] Jack stayed [PP tj] last year on
vacation]].
d. Lily just went [FR [PP[P e] [where]]j Jack stayed [PP tj] last year on vacation]].
e. Lily just went [PP [P e1] [FR [NP where]j Jack stayed [PP [P e2] [NP tj]] last year on
vacation]].
A similar point can be made with overt Ps. In (46)a, we see a w/w/h FR with a stranded
overt P to. Notice that to cannot be pied-piped (46)b (at least not overtly) which is already
unexpected under a pied-piping approach. Besides that, the w/w/h FR can be paraphrased
with a complex NP whose internal P is to but whose sister is a different P: in/at (46)c. The
interpretation that the pied-piping approach would predict is once again unacceptable (46)d
and the reason is again the lack of a second P as a sister of the whole w/w/h FR (46)b. This
contrasts with our analysis (46)e.
(46) a. Lily lives [FR wherem Jack is about to go [PP to [NP tm]]].
b. * Lily lives [FR [PP to where]j Jack is about to go [PP tj]]].
c. Lily lives [PP in/at [NP the place wherem Jack is about to go [PP to [NP tm]]].
d. *Lily lives [PP to [NP the place wherem Jack is about to go [PP to [NP tm]]].
e. Lily lives [PP [P e] [FR [NP where]j Jack is about to go [PP [P to] [NP tj]]]].
Finally, even when the w/w/h FR is preceded by an overt P, the overt P is never pied-piped
with the wh-word from within the FR. It is always part of the matrix clause taking the
w/w/h FR as its complement. The sentence in (47)a can only have the bracketing in (47)b,
according to which near is base-generated in the matrix clause and does not form a
constituent with where. This is the only bracketing that is compatible with the interpretation
the sentence receives, which we tried to paraphrase in (47)c. If near had been pied-piped with
where (47)d, then we would expect the interpretation of the sentence to be like (47)e, which
is not the case.
(47) a. Jack lives near where we had dinner last night.
b. Jack lives [near [FR where we had dinner last night]].
c. Jack lives [near [the place at which we had dinner last night]].
d. Jack lives [FR [near where] we had dinner last night]].
e. Jack lives [near [the place near which we had dinner last night]].
The second incorrect prediction of the pied-piping analysis concerns the semantic
derivation that this approach would require. The wh-PP would move to Spec of CP and leave
a PP trace in the gap position within the w/w/h FR, rather than an NP trace. As we saw in §3,
an NP-trace is translated into a variable whose value ranges over individuals (semantic
objects of type <e>). When the set-formation operation over that variable (λ abstraction)
applies, it returns a set of individuals (type <e,t>). On the other hand, a PP-trace translates
into a variable whose value ranges over more complex semantic objects, functions whose

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
18
semantic type we call <f> for convenience. When the set-formation operation over a variable
of type <f> applies, it returns a set of PP-functions of type <f,t>. Iota, the crucial
type-shifting operation that turns a set of individuals into its maximal individual (21), cannot
apply this time since there is no set of individuals to begin with. Therefore, the whole w/w/h
FR would be left denoting a set of PP-functions, while the matrix clause would require an
expression denoting what a PP denote, i.e. just a single PP-function. In conclusion, the FR
would be expected not to have the same interpretation as a PP and, therefore not even the
same distribution. This is contrary to what the data show.
We have argued that silent Ps are always stranded. This is not particularly problematic for
a language like English that allows for preposition stranding extensively. But what about
those languages (actually, the majority) that have w/w/h FRs but do not allow for preposition
stranding (for instance, Italian)? Unfortunately, we are still far from a deep understanding of
preposition stranding and why it is so rare across languages. Until then, preposition stranding
cannot be used to make a point against or in favor of our proposal. For instance, suppose that
the ban on preposition stranding turns out to phonological in nature (maybe due to the
clitic-like nature of Ps in many languages). Then, it would not be surprising that silent Ps are
always stranded, since they lack any phonological content by definition.6
6. Some open issues and conclusions
There are a few open issues that we believe are particularly relevant and deserve further
investigation. First, we saw that silent Ps are licensed only as sisters of phrases that carry
spatial, temporal, or manner semantic features (adverbial NPs, wh-words, or FRs). It would
be interesting to better understand the nature of this restriction and give it a principled
explanation.
It may turn out that this open issue is related to another: the semantic content of silent Ps.
More specifically, the semantic contribution of silent Ps is “minimal”: they are always
interpreted as semantically equivalent to Ps like to, in, at, or on whose semantic contribution
is often fully recoverable from the semantic contribution of the predicate and the
complement. Yet, semantic “lightness” and recoverability are necessary but not sufficient
conditions for the licensing of silent Ps. For instance, the semantic contribution of the silent
Ps in the w/w/h FRs in (48)a and (49)a, and in the wh-interrogative in (50)a is fully
recoverable: something very close to the P from, as shown in (48)b, (49)b, and (50)b.7 While
from seems semantically “light”, none of those silent Ps in these examples is licensed.
(48) a. * Lily really does not like [FR where Jack is coming e].
b. Lily really does not like [FR where Jack is coming from].
6 Thanks to Richard Larson for pointing out the issue of obligatory stranding.
7 Thanks to Grant Goodall for suggesting this kind of example.

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
19
(49) a. * Lily is coming [PP e [FR where Jack is coming from]].
b. Lily is coming [PP from [FR where Jack is coming from]].
(50) a. * Lily wonders [INT where Jack is coming e]?
b. Lily wonders [INT where Jack is coming from]?
To conclude, we have argued in this paper that if silent Ps are assumed in the grammar, we
can account not only for adverbial NPs – which had already been suggested – but also for the
puzzling syntactic/semantic behavior of a subclass of FRs. In particular, we have argued that
a silent P can take a FR introduced by where, when, or how as its complement and that where,
when, and how are always base-generated as the complement of a possibly silent P. Silent Ps
have allowed us to account for two puzzling properties of these FRs: their NP/PP-like
behavior and the restrictions of the internal gap they license. In addition, silent Ps have
allowed us to account for unexpected similarities between these FRs and adverbial NPs. We
believe there is strong evidence and good reason to assume silent Ps as a component of the
grammar. Continued investigation, especially crosslinguistic, would further illuminate their
syntactic and semantic properties.

Ivano Caponigro – Lisa Pearl Silent prepositions: Evidence from free relatives
__________________________________________________________________________________
20
REFERENCES
Bresnan, J. and Grimshaw, J. 1978. The syntax of free relatives in English. Linguistic Inquiry
9.331-391.
Caponigro, I. 2004. The semantic contribution of wh-words and type shifts: Evidence from
free relatives crosslinguistically. Proceedings of Semantics and Linguistic Theory (SALT)
XIV, R. Young (ed.), 38-55. Ithaca, NY: CLC Publications, Cornell University.
Chierchia, G. 1998. Reference to kinds across languages. Natural Language Semantics
6.339-405.
Dayal, V. 2004. Number marking and (in)definiteness in kind terms. Linguistics and
Philosophy 27.393-450.
Emonds, J. 1976. A transformational approach to English syntax. New York: Academic
Press.
Emonds, J. 1987. The Invisible Category Principle. Linguistic Inquiry 18: 613-632.
Grosu, A. 2003. A unified theory of ‘standard’ and ‘transparent’ free relatives. Natural
Language & Linguistic Theory 21: 247-331.
Huang, J. 1982. Logical relations in Chinese and the theory of grammar. PhD dissertation.
Massachusetts Institute of Technology, Boston.
Jacobson, P. 1995. On the quantificational force of English free relatives. Quantification in
Natural Languages, E. Bach, E. Jelinek, A. Kratzer, and B. Partee (eds), 451-486.
Dordrecht: Kluwer.
Larson, R. 1985. Bare-NP adverbs. Linguistic Inquiry 16: 595-621.
McCawley, J. 1988. Adverbial NPs: Bare or clad in see-through garb? Language 64:
583-590.
Partee, B. 1986. Noun Phrase interpretation and type-shifting principles. J. Groenendijk, D.
de Jongh, and M. Stokhof (eds), Studies in Discourse Representation Theory and the
Theory of Generalized Quantifiers, 115-143. Dordrecht: Foris.
van Riemsdijk, H. 2005. Free relatives: A syntactic case study. The Blackwell Companion of
Syntax, M. Everaert and H. van Riemsdijk (eds). Oxford, UK: Blackwell.