Larson (1985) and Emonds (1987) propose different mechanisms to account for English sentences in which a nominal acts as an adverbial but is not introduced by a preposition. We review their analyses and, in light of additional data, show that they are not tenable. Rather, we propose that the noun in such constructions is structurally deficient (hence does not need a preposition for Case) and has an inherent θ- role (hence does not need a preposition to assign one). Although in the same vein as Larson’s and Emonds’s proposals, ours accounts for a wider set of data and is thus preferable. It also has implications for the structure of θ-roles and for lexical entries.
Nominals that modify an event by specifying the location are nearly always introduced by a preposition in English. Consider the following examples:
John ate the apple *(in) the kitchen.
John lives *(in) that place.
Mary is walking *(on) the sidewalk.
As any native speaker of English can verify, the expressions of location in these examples must appear with a preposition. Now consider the following examples:
The place that John lives (in) is expensive.
Mary has lived (in) many places.
While the prepositions in (1) are obligatory, those in (2) are optional. Emonds (1976) first observed that in certain environments adjuncts can appear without prepositions. The goal of this article is to propose an analysis of these bare nominal adjuncts that renders the preposition optional in some of the observed contexts (namely, those related to relative clauses).1
To briefly outline the gist of the article: Larson (1985) proposes that the nouns in bare nominal adjuncts have inherent Case. He argues that certain nouns, such as place, carry inherent Case and thus do not require a preposition. We reject this proposal in favor of Emonds’s (1987) idea that place carries an inherent θ-role, but still needs Case if it is a DP. One reason for this shift is that Larson’s proposal explains the optionality of the preposition in (2), but not the obligatory nature of the preposition in (1b). We argue that the noun place in (1b) is a full DP and thus requires a preposition to satisfy the Case Filter. In (2), place is not a full DP and thus does not require Case. The tacit assumption to be clarified below is that only full DPs require Case. Since the noun place comes with its own θ-role, no preposition is needed if Case is not needed.
The remainder of this article is organized as follows. Section 2 provides background on prepositionless DP adverbs and on the syntax of relative clauses in English. Section 3 covers additional empirical facts about bare nominal adjuncts, some of which have not appeared in the literature before as far as we know. Section 4 presents Larson’s (1985) analysis of bare nominal adjuncts and discusses some shortcomings that it faces. Section 5 presents our analysis. Section 6 summarizes the article and mentions some implications of our analysis for syntactic theorizing.
This section introduces the data on which the current investigation is based. Specifically, we discuss adjuncts that consist of a bare nominal without a preposition (section 2.1). Such forms occur only in a restricted environment, which we introduce here and then explicate fully in section 3. Crucially, the licensing environment of such nominal adjuncts is sensitive to the type of relative clause in which it is found. Thus, we also provide some background on the syntax of relative clauses (section 2.2).
2.1 Bare Nominal Adjuncts
Nominal phrases that are not core arguments are introduced into the clause by a variety of means crosslinguistically, including prepositions, postpositions, and applicatives. It is assumed that these elements contribute either a semantic role—that is, a θ-role—and/or Case to the DP. In English, prepositions serve this purpose. Thus, in the sentence John ate the appleinthe kitchen the preposition in provides both Case and a θ-role for the DP the kitchen. However, Larson (1985: 595) discusses the following examples of bare nominal adjuncts:
I saw John that day/someplace you’d never guess.
John was headed that way.
Max pronounced my name every way imaginable.
Larson, following Emonds (1976), discusses the following core properties of the bare nominal adjunct construction: (a) only a small, restricted set of nouns can appear in this construction, and (b) although the bare nominal adjuncts have the overt form of a DP, they have the distribution of an adverbial PP (or AdvP).2
While the distribution of bare nominal adjuncts appears to be variable and arbitrary on the surface, there is a subset of such constructions with a remarkably static and clearly definable distribution. This subset, which we describe in detail below, opens a window into the organization of the lexicon. Specifically, we show that bare nominal adjuncts are consistently available with a certain set of nouns in bare relative clauses and in that-relative clauses, but with some restrictions. We propose an analysis to account for this descriptive generalization.
2.2 The Syntax of Relative Clauses
The traditional analysis of relative clauses (stemming from Chomsky 1977) still holds much ground. According to this analysis, a relative clause is right-adjoined to the NP it modifies and identifies the noun through a relative operator that raises to the left periphery of the relative clause. For convenience, we refer to this analysis as the operator analysis. However, a competing analysis tracing back to Vergnaud 1974 holds that the head noun originates inside the relative clause and, in externally headed relative clauses, raises to the left edge of the relative clause. Again for convenience, we refer to this analysis as the raising analysis.
Despite the widespread acceptance of the operator analysis of relative clauses (see Borsley 1997 for a defense of this analysis), the raising analysis has gained much currency (Bianchi 1999, Cecchetto and Donati 2015, Kayne 1994, 2003, 2013). Specifically, the following structures are typical of the raising analysis. First, that-relative clauses in English are derived by movement of the head noun from inside the relative clause to a specifier position adjacent to the determiner, but the head noun and the determiner do not form a constituent.
(4) [DP the [CP [NP book]i [CP [C that] [TP Mary [VP read ti]]]]]
(5) [DP the [CP [DP [NP book]j [DP [D which] tj]i [CP [TP Mary [VP read ti]]]]]]
The crucial difference between these two structures is the size of the nominal phrase that originates inside the relative clause. That-relative clauses contain an NP that raises to the left edge of the relative clause, while which-relative clauses contain a full DP that raises to the left edge of the relative clause. Bianchi provides a wide range of facts in support of these two structures.
Extraposed relative clauses will also be relevant to our discussion. Extraposed relative clauses (including that-relative clauses) behave differently from their nonextraposed counterparts (Bianchi 1999). Crucially, the facts below strongly suggest that extraposed relative clauses are not derived by a Vergnaud-type raising analysis. Consider the data in (6). Example (6a) contains a canonical that-relative clause, while (6b) contains an extraposed relative clause.
The book that I bought yesterday went missing.
The book went missing that I bought yesterday.
The Toronto that I knew 30 years ago no longer exists.
*The Toronto no longer exists that I knew 30 years ago.
The advantage that we took of that poor fool was unforgivable.
*The advantage was unforgivable that we took of that poor fool.
City names in English do not typically appear with a determiner (*the Toronto). Therefore, the appearance of the determiner is unexpected in (6c). Assuming a raising analysis as developed by Kayne (1994) and Bianchi (1999), however, the mystery disappears. The proper noun Toronto originates as the direct object of know and appears linearly adjacent to the determiner as a result of movement. Likewise, the determiner in (6e) is also unexpected as it does not appear in the idiom take advantage. Its presence can be explained the same way. The forms are conspicuously unavailable with extraposed relative clauses, however, suggesting that extraposed relative clauses are not derived by Vergnaud-type movement.
(7) John met a man, and Mary met a woman, who were quite similar to each other.
The antecedent for the relative clause seems to be split between two DPs, [a man . . . a woman]. Needless to say, a satisfactory analysis of hydras is yet to be found more than 45 years after their first description in the literature (but see Cecchetto and Donati 2015 for a recent attempt). Furthermore, given that hydras contain no single antecedent, a raising analysis seems extremely unlikely. To the best of our knowledge, the possibility of a raising analysis of hydras has not been investigated. However, the following data suggest that hydras, along with extraposed relative clauses, are also not derived by Vergnaud-style raising:
*John loves the Toronto and Mary loves the Montreal that they knew 30 years ago.
*John loves the Toronto and Mary loves the Montreal that once resembled each other.
*John despises the advantage and Mary appreciates the care that Bill took of them.
Here, we adopt proposals by Koster (2000) and Sheehan (2011) regarding extraposed relative clauses. Crucially in the present regard, according to these proposals extraposed relative clauses are not derived by raising; rather, the head of the relative clause is a full DP, and the relative clause itself is separate, possibly derived by the traditional operator analysis.
To conclude, we have introduced the core empirical phenomenon under discussion—namely, bare nominal adjuncts, which lack a preposition—and have introduced basic aspects of the syntax of relative clauses pertinent to our analysis.
3 Properties of Bare Nominal Adjuncts
This section describes the relevant facts concerning bare nominal adjuncts in English. We discuss the nouns that can appear in this construction followed by the environments that license them.4
First, only nouns that are semantically bleached, denoting merely a location, can appear as bare nominal adjuncts. In (9a–c), where a bleached noun appears alongside nouns with greater lexical content, only the semantically bleached noun is licit.5
John found a place/*house/*apartment to live.
I’m looking for a spot/*table/*mantel to put the cactus.
Furthermore, only nouns denoting locations are licit in this construction; instruments, no matter how semantically bleached, are not.
John found a *thing/*pen/*marker to write.
John found a thing/pen/marker to write with.
Larson (1985) observes that bare and that-relative clauses modifying the same set of nouns in (3) can appear without the associated preposition in the relative clause; see (11a) and (12a). However, which-relative clauses do not allow this option; see (11b) and (12b).
I saw the place (that) John lives (in).
I saw the place which John lives *(in).
I saw the spot (that) John put the cactus (in).
I saw the spot which John put the cactus *(in).
The place is expensive that John lives *(at).
The spot is dirty that Mary put the cactus *(in).
Also, bare nominal adjuncts cannot appear in hydras.
(14) Bill saw the old place and Mary saw the new place that John lives *(at).
To summarize, bare nominal adjuncts with place are found in that-relative clauses, but not which-relative clauses, and only in nonextraposed relative clauses and nonhydras.
4 Previous Analyses
4.1 Larson’s (1985) Analysis
Larson (1985) develops an analysis in which certain nouns are lexically encoded as being able to assign inherent Case to themselves. He proposes that such nouns possess an abstract feature, [+F]. Other nouns are [−F] and require a standard Case-checking mechanism.
Larson’s analysis of the contrast in (11) runs as follows. That-relative clauses are introduced by a phonologically null operator, which Larson assumes has only the feature [+wh]. In the Government-Binding machinery that Larson employed at the time, a relative clause was assumed to need a Case-marked trace. Larson proposed that this need was filled transitively by the [+F]-marked noun. Consider (15), which gives the structure for (11a) under this proposal. The noun place has the feature [+F]. It transmits this feature to the operator, which is not specified for [±F]. Since the operator and the trace form a chain, the trace now has this feature and is Case-marked.
(15) I saw the place Opi (that) John lives ti.
The which-relative clause in (11b), however, is introduced by an overt wh-operator, which Larson assumes has the feature [−F]. As a result, the [+F] feature of place cannot be transmitted to which, and the trace cannot receive Case in the absence of an appropriate preposition. Crucially, the operator is unmarked with respect to [±F], and thus when the chain (place, Op, t) is formed, the [+F] feature of place spreads throughout the chain. However, the [−F] feature of which blocks the formation of the chain (place, which, t) as this chain would contain contradictory specifications for [±F].
4.2 Problems with Larson’s Analysis
Although Larson’s analysis captures the difference between that-relative clauses and which-relative clauses with respect to the behavior of bare nominal adjuncts, it displays shortcomings that prompt us to propose an alternative analysis. Importantly, there is no way to rule out (16a), (17a), and (18a) under Larson’s approach.
*John lives that place.
John lives in that place.
*That place to appear in Better Homes and Gardens would be surprising.
For that place to appear in Better Homes and Gardens would be surprising.
*It was arranged that place to be cleaned up.
It was arranged for that place to be cleaned up.
Since the noun place is [+F] under Larson’s account, we expect it to be licit without a preposition in those environments where one is typically needed. The fact that the prepositional complementizer is needed for the infinitival subject is particularly troubling—we cannot attribute the ungrammaticality of (17a) to a putative missing semantic contribution of for, since there is none.7
The following paradigm illustrates the same point. Under Larson’s account, we expect the [+F] feature of place to license the DP.
I want that apartment to be cleaned up.
I want that place to be cleaned up.
I would like very much *(for) that apartment to be cleaned up.
I would like very much *(for) that place to be cleaned up.
Furthermore, although Larson’s analysis can capture the difference between that-relative clauses and which-relative clauses, it does not seem to be able to capture the fact that bare nominal adjuncts cannot appear in extraposed relative clauses or in hydras.
Another problem facing Larson’s approach is more conceptual. The notion that nouns such as place and spot carry their own Case feature does not seem to be conceptually motivated. These nouns do not seem to have any inherent property on which to base a lexical prespecification for Case. However, they indeed seem to inherently refer to locations. Therefore, we will suggest that such nouns carry their own
This section has discussed problems with Larson’s (1985) analysis of bare nominal adjuncts. We have put forth a proposal, which we spell out in section 5, that captures the basic insight of Larson’s approach that nouns such as place have a lexically specified idiosyncratic property that allows them to appear without a preposition in certain environments. However, we depart from Larson in assuming that this property is Case, suggesting instead that it is a θ-role.
4.3 Emonds’s (1987) Analysis
Emonds (1987) discusses the same kind of data under the umbrella of his Invisible Category Principle, which runs as follows. Essentially, a functional category, X, can appear phonologically empty if the feature or features contained in X appear in the complement of X. Crucially, akin to our proposal below, nouns such as place carry their own [location] feature—the same feature as the null preposition.
The places (that) John lives are expensive.
*The towns (that) John lives are expensive.
[NP, +loc The places] that John lives [PP [P, +loc Ø] [NP, +loc Ø]] are expensive.
*[NP, −loc The towns] that John lives [PP [P, −loc Ø] [NP, −loc Ø]] are expensive.
In (21a), the operator [NP Ø] agrees with the [+locative] feature of the head noun, thereby licensing a null preposition. In (21b), on the other hand, the operator agrees with the [−locative] feature of the head noun, and the null preposition is not licensed.
While Emonds’s analysis does alleviate the conceptual problems that Larson’s analysis faced, it does so at a cost. In Emonds’s analysis, there is no principled way to account for the difference between that-relative clauses and which-relative clauses.9 Also, like Larson’s analysis, Emonds’s cannot account for the lack of bare nominal adjuncts in extraposed relative clauses and hydras. Recall that extraposed relative clauses and hydras have been argued to be derived not by Vergnaudstyle raising but by the standard operator analysis.
We do note, however, that Emonds’s analysis may be on the right track for the kinds of bare nominal adjuncts not addressed here—namely, those not found in relative clauses (see footnote 5). Consider the following example:
(22) The group split up into two. They went [that way] and [to the museum].
The fact that the bare nominal adjunct can be conjoined with PP suggests that a null preposition along the lines that Emonds describes is present. Unfortunately, the converse is difficult to show—namely, that bare nominal adjuncts in that-relative clauses are truly prepositionless since extraction out of a conjoined phrase is ungrammatical (Ross 1967). Furthermore, obligatory preposition stranding with that-relative clauses in English makes the test impossible. Nevertheless, the following example strongly suggests that a null preposition is not available for place:
(23) They have lived [*(in) that place] and [in that apartment].
To conclude, we have noted shortcomings of both Larson’s (1985) and Emonds’s (1987) analyses. We do agree with both proposals, however, that the peculiar nature of bare nominal adjuncts lies in the lexical specifications of individual lexical items such as place. Specifically, we follow Emonds in assuming that the lexical specification is related to θ-roles rather than to Case, although our own proposal departs from the specifics of Emonds’s analysis in minor ways. As we argue in the next section, this lexical specification, coupled with an updated view of the structure of relative clauses, explains the distribution described above.
This section discusses the proposed analysis for bare nominal adjuncts. Crucially, we propose that nouns such as place are lexically prespecified with a θ-role, an idea that we adopt from Emonds (1987). Thus, no θ-role assigner is needed. However, when such nouns appear in a full DP, they must be assigned Case. We assume that bare NPs, though, do not require Case.10
These ingredients lead to the following situation. When a noun such as place appears in a bare NP, it requires neither Case nor a θ-role. Thus, it is only in this situation that the preposition is optional. We now show how this situation holds in the environments discussed above.
5.1 The Syntax of Relative Clauses
Recall that the raising analysis, particularly as developed by Bianchi (1999), holds that there is a distinction between that-relative clauses and which-relative clauses in terms of the category of the nominal in the head position. Crucially, the head of a that-relative clause is an NP, while the head of a which-relative clause is a DP. Under standard conceptions of Case theory, only a DP is assigned Case. Thus, we propose the derivations in (25a–b) for the that-relative clause in (24a) and the which-relative clause with the bare noun place in (24b).
the place that John lives
the place which John lives *(in)
[DP the [CP [NP place]i [CP [C that] [TP John [VP lives ti]]]]]
[DP the [CP [DP [NP place]j [DP [D which] tj]i [CP [TP John [VP lives [PP in ti]]]]]]]
In (25b), a full DP appears in the embedded position inside the VP lives. This DP requires Case, so the preposition must be present. In (25a), however, only an NP is present inside the embedded VP. A bare NP does not require Case, so no preposition need be present. Furthermore, since place is an inherently locative noun, it does not require the locative thematic contribution of the preposition for interpretation.11
5.2 The Structure of θ-Roles
Recall that our proposal holds that nouns such as place have their own θ-role, which we have labeled
Mary is walking to that place/store/house.
Fred came from that place/store/house.
The θ-roles involved here are
the place that Mary is walking *(to)
the place that Fred came *(from)
This follows from our proposal that nouns such as place bear an inherent
Next, note that θ-role can potentially conflict. Specifically, in (28) the noun place bears an inherent
(28) Mary is walking to that place.
We now offer a tentative suggestion regarding the structure of θ-role, the precise details of which we leave open to further research. Following Hornstein (1999), we assume that θ-role are features, which are checked. We implement this as follows. We assume that probes and goals can be prespecified (Béjar 2003, Béjar and Rezac 2009). Thus, we assume the following goals for some of the nouns mentioned earlier:
Assuming that θ-role are indeed morphosyntactic features, we follow much recent work on the geometry of such features (Cowper 2005, Harley and Ritter 2002, McGinnis 2005) and propose the partial geometry in (30).
Now, how do these mechanisms allow examples such as (28) to survive? Assume that the preposition assigns a31), which seeks an appropriate goal. In this case, the goal, too, is lexically specified. The probe values the goal as shown with the dashed line, and the DP goal is understood as a
With an ordinary DP, not lexically prespecified with a structured θ-role goal, the derivation proceeds as in (32). The probe seeks an appropriate goal, apartment, and values the uninterpretable θ-role as shown with the dashed line.
We have examined a class of bare DP adjuncts in English and have provided an updated analysis building on Larson’s (1985) original proposal. Crucially, we have adopted Larson’s idea that nouns such as place are lexically marked as containing a special feature that allows them to exist in certain environments without a preposition, but we have argued that this feature is not Case, as Larson proposed, but a θ-role. This change was motivated by certain empirical and conceptual shortcomings in Larson’s analysis that the current approach remedies.
In summary, we have argued that nouns such as place are lexically prespecified with aVergnaud (1974) and further developed by Kayne (1994) and Bianchi (1999), we have assumed that the relativized noun in such clauses is a bare NP (or some higher XP in the extended nominal hierarchy) and does not require Case. It is exactly in this situation that the preposition is not obligatory. As a bare NP, the nominal place does not require Case. It is lexically prespecified with a θ-role, so it does not need a θ-role assigner. The relativized nominal in which-relative clauses and extraposed that-relative clauses, on the other hand, is a full DP that requires Case. Hence, the preposition is required.
We end by briefly discussing the implications of our proposal for current syntactic theorizing. The role of the lexicon in grammar has diminished dramatically since Kratzer’s (1996) claim that the external argument is not selected by the verb. Since that time, it has been proposed that lexical roots themselves are essentially devoid of grammatical information—that is, they possess no formal features (the so-called Chomsky-Borer conjecture; Borer 2005). Moulton (2014), however, has argued that at least some roots must include some grammatical information. If correct, our approach adds to the discussion of exactly how bare roots can be, by arguing that at least some lexical roots are prespecified with θ-role. Thus, the facts discussed here present a problem for a strong version of the Chomsky-Borer conjecture.
We wish to thank the audiences at the 2015 annual meeting of the LSA in Portland, OR, and the Syntax Project meeting at the University of Toronto for helpful comments and suggestions. We are also grateful to Tim Stowell, whose encouragement helped us pursue this line of research. All errors are our own.
1 We use the term nominal to remain agnostic between the labels NP and DP for now, until we come back to the issue.
These bare nominal adjuncts were called bare NP adverbs by Larson (1985), as his discussion predated the DP hypothesis. We merely update the terminology here to reflect current understanding of the structure of nominals. Heidi Harley (pers. comm.) also mentions that the label adverb may not be the best since many of the PPs/DPs in question have an obligatory flavor to them. For instance, the sentence John lives in Toronto is qualitatively different from John lives. We abstract away from the property of obligatoriness here and use the label adjunct for convenience.
2 A reviewer asks about other nouns such as way, day, and time, as they can also appear in bare nominal adjuncts. The empirical facts for these nouns are quite different from those for place, however, so they will have to wait for future research.
3 A reviewer asks us to clarify the circumstances in which a bare NP can appear in the position of a DP. Beyond the details explicated by Bianchi (1999), we have little more to add regarding the syntactic and semantic consequences of her proposal. An in-depth discussion of the syntactic environments that allow an NP to surface in positions otherwise occupied by a DP would take us too far afield. The same reviewer also asks about bare NP plurals (as in Mary tunes pianos). We would suggest that such nominals contain a null determiner (hence constituting a full DP). Again, an in-depth discussion of this issue would take us too far off track.
4 One environment where bare nominal adjuncts are licit is following certain quantifiers. We provide examples here but leave analysis to future research.
I have lived many/few/?some places in LA.
*I have lived many/few/some apartments in LA.
5 One must be careful to control for the reading in which the infinitival clause modifies the superordinate VP rather than the noun. The following sentence has both of these readings:
John is looking for a place to eat.
= John is looking for [a place in which he can eat]
= John is [looking for [a place] [in order to eat]]
We are interested only in reading (ia) here since it is only in this structure that the noun in question is an argument of the verb/preposition in the relative clause. Observe that in sentences with a structure paraphrasable as in (ib) there is no restriction on the type of noun since it is not an argument of the verb/preposition in the relative clause.
John is looking for a pen to write.
≠ John is looking for a pen with which to write.
= John is looking for a pen in order to write.
John is looking for a restaurant to eat.
≠ John is looking for a restaurant in which to eat.
= John is looking for a restaurant in order to eat.
6 A reviewer points out that all our extraposed examples are built on definite DPs. We note this fact but do not believe it affects our analysis.
7 The following paradigm crucially underscores the lack of a semantic contribution by the prepositional complementizer for:
I would like very much for John to give Mary the book.
I would like very much for Mary to be given the book by John.
I would like very much for the book to be given to Mary by John.
8 Evidence for this claim comes from the fact that the operator and the head noun agree in number: the place that John says is/*are cheap, the places that John says *is/are cheap. Since the two agree in number, Emonds argues it is not a leap to expect that they agree for all features, including crucially [±location].
9 A reviewer suggests that Emonds might be able to capture the difference between that- and which-relative clauses by taking advantage of the difference in feature composition between these two words. This suggestion still does not explain why extraposed relative clauses and hydras behave differently, however.
10 For reasons of space, we do not defend the claim that bare NPs do not require Case. As a reviewer notes, however, Cecchetto and Donati (2015) make a similar argument for reduced relative clauses with past participles.
11 A reviewer notes that our proposal predicts that the following example should be grammatical:
I saw the place (that) John is happy.
We have consulted several English speakers about (i), and most judge it to be either grammatical or only slightly degraded. All the speakers we consulted judged (ii) to be ungrammatical, however.
I saw the apartment/house (that) John is happy.
Given these facts, we conclude that (i) is indeed grammatical, as our proposal predicts.
The few speakers who reported that (i) is ungrammatical also found (iii) ungrammatical.
I saw the place that John lives.
This judgment is likely due to a difference in the lexical specification of the θ-role of the noun place.
The fact that some speakers feel that (i) is slightly degraded may be due, we suspect, to the argument/adjunct distinction noted by Heidi Harley (pers. comm.) in footnote 1. Crucially, place is arguably an argument in (24a), but an adjunct in (i). We leave this observation to future research.