Labeling (Reduced) Structures: When VPs Are Sentences

There is a general debate concerning “reduced” clauses, opposing partisans of the hypothesis that all reduced clauses must have a full-clause base and are thus elliptical (Morgan 1973, Merchant 2004) and those who argue that at least some of them are built as reduced as they appear (Barton 1990, Progovac 2013). This opposition is reproduced in the field of acquisition, where the same dilemma arises with respect to early reduced structures produced by children (Radford 1988, Rizzi 1993, Hyams 2005, Guasti 2016).

The first aim of this article is to step into this debate by focusing on a particular kind of “reduced” participial structure in Italian, exemplified in (1), in which the past participle of an unaccusative or passive verb agrees in gender and number with the internal argument. We will argue that this structure should be divided into the five following types:

These five participial structures all present clear evidence of being “reduced” structures: they can host neither tense, nor negation, nor wh-movement, nor focus. As for the internal argument of the verb, its distribution and makeup vary across the different types along a pattern that will play a crucial role in our analysis.

Empirically, this article gives a detailed description of these five structures (some of which have not been described so far). Theoretically, it has two main goals, concerning the debate on reduced structure and the theory of labeling.

To begin with the first goal: The five Italian participial constructions offer a diagnostic that allows us to decide between the ellipsis account and the account that takes them to be originally built as reduced as they appear; it comes down in favor of the latter, at least for the cases under consideration. It is useful to anticipate the logic of our argument. The five reduced structures are composed of at most a participle and its internal argument (though see below for a more precise description and some differences between the types). In addition, they are possible only if the verb is unaccusative or passive. Finally, the structure is so reduced that there is no subject agreement node that can assign nominative to a promoted internal argument, as is normally the case in canonical (unreduced) passive or unaccusative clauses. Therefore, the question arises of how the internal argument is assigned case and, if it is not, how it is licensed. We will claim that accusative or any other case is never (directly) assigned to the internal argument in these reduced structures, and that each of the five structures has its own way to license the internal argument. This is where the five structures diverge, as we will show in detail. The general point we want to make is that, if we are on the right track, this account suffices to explain why the structures are reduced. Assuming Burzio’s Generalization (Burzio 1986), a verb can assign a θ-role to its subject position if and only if it can assign accusative case to its object. Given that accusative is not assigned, the higher part of the structure that hosts the subject simply cannot be projected. Of course, this argument works only if the five structures are built as reduced as they appear. If they were elliptical structures resulting from some kind of truncation, there would be no obvious explanation for why they are reduced the way they are, as Burzio’s Generalization applies to structure-building operations and not to ellipsis-producing ones.

Having established that these structures are literally reduced (as opposed to complete but elliptical) and can be as small as a VP, we turn to our second main goal: to consider why some reduced verbal structures receive illocutionary force, and what the mechanism is that grants a sentential meaning to these fragments. Reflecting on the relation between movement and labeling, we will argue that the agreement-related movement of the internal argument observed in most (but crucially not all) of the five structures activates their sentential meaning through a labeling mechanism. We will discuss this issue by concentrating on the configuration in which two phrases are merged.

The article is organized as follows. After refining some conceptual categories that are needed to analyze these structures, concerning in particular labels, movement, and agreement (section 2), we start describing and deriving the first radically reduced structure under investigation, namely, REDUCED RELATIVE (section 3). In section 4, we address a structure that is minimally different from REDUCED RELATIVE, corresponding to problem solved in English: BARE NOUN REDUCED. In section 5, we focus on DISLOCATED REDUCED, which reveals that topicalization in radically reduced structures licenses a full DP argument that would otherwise be excluded for case reasons. In section 6, we turn to FULLY REDUCED, an even more radically reduced structure in which a participle alone can express propositional content. In section 7, we conclude the survey by addressing so-called absolutive participial constructions, which, as we will argue, come in two varieties, only one of which (UNACCUSATIVE ABSOLUTE REDUCED) is an instantiation of the radically reduced participial clauses that have in common the inability to assign accusative case. In section 8, we provide conclusions, reflecting in particular on some consequences of our results for a general theory of what a sentence is, and where sentential force comes from.

The issue of how syntactic objects get a label, determining their syntactic and interpretational properties, is at the center of much theorizing lately. In most recent versions of the Minimalist Program (Chomsky 2013, 2015, 2019), the leading idea is that labels are simply determined by a principle of Minimal Search: if one feature is structurally prominent within a syntactic object, that feature is promoted as its label.¹ This relation between structural prominence and labels is directly observable in syntactic objects that consist of a simple element (a head) and a more complex one (a phrase), as in (2).

In (2), X is simpler than its complement in an obvious sense, and, as a result, its features are found by Minimal Search and assigned to the entire syntactic object, which is a category with label X. This derives the notion of headedness that has been at the core of phrase structure theory for the last forty years. Notice that this configuration can result from two kinds of operations: when a head is externally merged to a syntactic object, and when a head is internally merged to a syntactic object. In both cases, as is expected if both operations are two facets of the same operation Merge, the simplex object provides the label. The Minimal Search approach has the consequence that movement of a simplex element (a head) can modify the labeling of the hosting object (see Cecchetto and Donati 2015 for details). In Donati and Cecchetto 2011, we argue that the ability of a moving head to relabel a structure is at the core of the notion of relativization. In a nutshell, in simple relative clauses the raising of the nominal head (which, we claim, is always a simple noun) relabels the CP structure as an NP, which in turn is selected by an external determiner. In section 3, we will give another illustration of this “relabeling” movement.

This cannot be the whole story, however, because syntactic objects can also have the form shown in (3); that is, they can contain two complex syntactic objects that are merged together.

Here, neither XP nor YP is prominent, so labeling by structural prominence cannot deliver a label. There is some disagreement on whether unlabeled objects are tolerated in syntax (see Chomsky 2013, 2015, Cecchetto and Donati 2015). Leaving this question aside and assuming for the sake of the argument that labels are necessary, one way to assign a label in the configuration in (3) is proposed by Chomsky (2015): if a feature is shared by a probe and a goal, this feature is promoted to be the label.

We illustrate this in (4) with a ϕ-feature configuration. For ease of exposition, we call this mechanism labeling by feature sharing; the mechanism illustrated in (2), by which a lexical item provides the label when it is merged with a phrase, we will call labeling by structural prominence.

In (4), the DP subject and the TP agree in ϕ-features. As a result, the shared ϕ-features are found by Minimal Search and are assigned as the label of the structure. In the same vein, under labeling by feature sharing the category obtained when a wh-phrase agrees and merges with interrogative Comp is assigned the label WhP. This is illustrated in (5). The reasoning can be repeated for other criterial positions in the sense of Rizzi 2006.

Note that labeling by feature sharing and labeling by structural prominence can be seen as two different instances of Minimal Search, the intuition being that in neither case does the internal structure of the two objects that are merged need to be searched. In fact, the category that is promoted to be the label is immediately found either by virtue of being structurally prominent or by virtue of being a feature represented at the upper node of the merged categories. An interesting question that arises is what happens when both ways of determining a label are available in the same configuration—for example, when agreement holds in a configuration involving structural prominence of the type in (2). We will discuss such a case in the next sections (see also below for a comment concerning DPs).

Besides this open question, the labeling-by-feature-sharing proposal faces certain problems, notably with syntactic objects that might or might not host agreement configurations and would accordingly be labeled differently (see Donati to appear for a critical discussion). For example, a structure where the subject has raised to the EPP position and agrees with T (say, in English) would be labeled ϕP (6), while the same structure without a specifier (say, a sentence with an in-situ subject in a pro-drop language) would be labeled TP (7). This is not a welcome consequence, as these two objects seem to have the same syntactic distribution and the same interpretation at the interface.

We tentatively put this problem aside by assuming that there is always a category hosted in what is traditionally referred to as Spec,TP: a DP in English, pro in pro-drop languages like Italian.

Another (potentially more serious) problem arises in the nominal domain. D can agree with the NP when the two are merged; therefore, if labeling by feature sharing were allowed in this case, one would expect the DP to be relabeled ϕP, the same label assigned to the node created by Merge of DP and TP in (4). But obviously the distribution of a DP cannot be equated with the distribution of a clause. Chomsky never discusses this unwanted consequence of the mechanism he proposes, possibly because he implicitly assumes that D can label the structure even in the presence of ϕ-feature sharing, since D is structurally more prominent. In this specific case, labeling by structural prominence provides the label D, while labeling by feature sharing would provide a deviant label and this would stop the derivation. In any case, this is what we tentatively assume for DPs.

Labeling by feature sharing raises a more fundamental question as well: what does it mean for a label to result from an agreement configuration? Traditionally, a label has been conceived as a categorial feature that is already present in the lexical properties of a given category and can be transmitted when this category is merged with another one (this label transmission can be repeated at each occurrence of Merge).² This amounts to describing labeling as a mechanism by which one of the two objects that are merged prevails over the other. However, labeling by feature sharing implies that neither of the objects that are merged “wins.” A comparison with the VP is useful. It is easy to show that the distributional properties of a VP are determined by the verb alone. For example, with verbs that allow a transitive/intransitive alternation, the VP has the same distribution regardless of the presence of the internal argument. In the case of clauses, it is not easy to decide which category determines their distribution (whence the fact that clauses have been referred to in different ways in the generative tradition: S, IP, TP, AgrP, etc.). In an innovative yet controversial move, Chomsky (2019) claims that this terminological indeterminacy is just an indication that clauses are exocentric, so it is hopeless to label them after their alleged head. Still, this cannot mean that they do not have a label, since they have a certain distribution that ultimately results from the label they carry. Therefore, we take the claim that clauses are exocentric to mean that their label does not come from one of the two objects prevailing over the other; rather, it comes from some features they share.

Despite these problems, we think the labeling-by-feature-sharing mechanism has an interesting corollary that is worth exploring: namely, it can be seen as a syntactic marker of the subject-predicate relation. The idea is that a clause is exocentric (therefore labeled by feature sharing) because it corresponds to a relation between two categories rather than being a configuration in which one of the two categories prevails over the other. On the other hand, endocentric categories, like VPs, do not correspond to a relation. For example, there can be a VP with a single verb, which—by virtue of being the only element inside the category—cannot be in relationship with anything.

In this article, we propose to go one step farther in this direction and assume that the existence of the subject-predicate relation is the precondition for the presence of illocutionary force. We explore a consequence of this idea for the domain of reduced structures. More specifically, agreement with the external argument is not the only agreement relation that exists; the other is agreement with the internal argument (possibly realized as “object agreement”). Therefore, if feature sharing is responsible for the establishment of a subject-predicate relation and of illocutionary force, this should be the case for agreement with the internal argument as well. As agreement with the internal argument is realized in a smaller portion of the clause, this means that there should be reduced structures with illocutionary force. This is what happens in most of the reduced structures we will explore, which can be fruitfully analyzed in the light of the labeling-by-feature-sharing hypothesis.

Reduced relative clauses are participial clauses modifying a head noun, as illustrated in (8).

As originally observed by Burzio (1986), reduced relatives are restricted to unaccusative and passive verbs and are impossible with transitive or unergative verbs.

Historically, the term reduced relatives comes from the first analyses proposed for these structures in generative grammar, according to which they were literally elided versions of full relative clauses (see Jacobs and Rosenbaum 1968). At the other extreme is another line of thought, inaugurated by Burzio (1986:150), that identifies reduced relatives with various kinds of small clauses. In Burzio’s analysis, for example, a reduced relative like (8) involves a null PRO, as in (9).

The idea that reduced relatives should be assigned a full-fledged clausal structure assimilated to finite relative clauses is also well-represented. In this view, participial relatives involve a relative operator that is licensed in the specifier position of a functional projection with a complementizer-like functional head, as illustrated in (10).

In Kayne 1994, the functional projection hosting the relative operator is identified with CP; in Siloni 1995, it is identified with DP. The only peculiarity of participial clauses under this view is that they do not contain a tense node.

This revival of the ellipsis approach is partly related to the cartographic framework, whereby structures are defined by dedicated functional projections: the defining feature of relativization is a specialized functional projection hosting an operator (or the raising head in Kayne’s terms): since participial relatives are relatives, they must contain this position.

An obvious problem with Kayne’s proposal is that it does not explain why overt complementizers are systematically banned from participial relatives. This incompatibility not only holds in Romance, but is robustly attested across languages and indeed constitutes a well-established typological generalization (see Doron and Reintges 2007 and references therein). Another problem with assuming a full-fledged structure for participial relatives is that it does not explain Burzio’s (1986) observation that these relatives are only possible with unaccusatives and passives. This is why we assume that reduced relatives do not project the entire structure associated with a full relative (which is later phonologically deleted) but instead are literally reduced.

We will extend to these structures the approach developed in Cecchetto and Donati 2015, which reduces relativization to an application of labeling by structural prominence in a configuration in which the lexical item is internally merged to a phrase. Under this relabeling analysis—so called because the target is relabeled by the lexical item that is internally merged to it—relativization derives from movement of a bare lexical item, no matter whether it happens in a full-fledged structure such as a full, inflected relative clause or a free relative, or in a constituent as small as the VP. All that is needed to build a relative structure under the relabeling approach is a relabeling movement—that is, the movement of a nominal head, which (being structurally prominent in the newly formed syntactic object) labels it.³

As an illustration, consider the structure in (11), containing a participial relative with a passive verb. Compare this with (12), containing a full-fledged relative clause under the relabeling analysis.

In (11), the head of the reduced relative (ragaɀɀo ‘boy’) is external since it precedes the verb, much like the head noun in the full relative in (12). In both (11) and (12), it is the movement of N that “relabels” the structure and provides the external determiner with the NP it selects. This amounts to saying that the derivation in (11) is parallel to the derivation of a full relative except in two ways:

• •

  In reduced relatives, the landing site of N-movement is a position in the VP periphery, whereas in full relatives, it is in the CP area.⁴

• •

  In (11), there is no manifestation of D inside the relative. Participial relatives never contain wh-determiners such as which or the complementizer che, which can be analyzed as a wh-determiner (e.g., Manzini and Savoia 2003).

We take the absence of a D inside the reduced relative at face value, and we assume that in (11) the participle rimproverato ‘scolded’ is merged directly with the bare noun ragaɀɀo ‘boy’. This assumption plays a crucial role in explaining the restriction to unaccusatives, as we will explain. We assume that θ-role assignment takes place configurationally (see Baker’s (1988) Universal Theta Assignment Hypothesis (UTAH); Hale and Keyser 2002); that is, a category (or its copy/ trace) must be in the right environment to receive a θ-role. However, what we say is compatible both with the idea that θ-role assignment is a “filter” at the conceptual-intentional interface and with the idea that it applies when a category is externally merged, possibly because having a θ-role is a feature of the verb/predicate (Hornstein 1999).

Going back to reduced relatives, if the verb does not need to check/assign accusative as in passive and unaccusative constructions, nothing goes wrong: the noun ragaɀɀo ‘boy’ gets a θ-role from the past participle and a case from the main verb conosco ‘(I) know’ after it has moved, relabeled the structure, and merged with the external D. Under this analysis, θ-role assignment is not restricted to DPs, as the past participle assigns a θ-role to the bare noun ragaɀɀo ‘boy’ (notice that there is independent evidence that nouns can receive θ-roles: this happens with adjectives, and participles are a type of adjective). In languages like Italian, bare singular nouns do not get case (DPs do). Consider now an object reduced relative like (13), predicted to be impossible: (13) is a case violation because Gianni does not get case.

Next, consider (14), an ungrammatical participial relative with a transitive active verb.

In principle, in (14) the bare noun professore ‘professor’ should be able to provide the label when it is merged with the structure headed by v (cf. (15)).

However, if v does not provide the label, the configuration for assignment of the Agent θ-role is not created (informally, the noun is not in Spec,vP). Therefore, a θ-role violation occurs and the structure is ruled out.⁵ In other words, the relabeling configuration is incompatible with that for θ-role assignment. Due to this incompatibility, subject reduced relatives with a transitive verb are ungrammatical, as in (14).

Example (16), with an unergative verb, is blocked for the same reason as (15). (16) involves a θ-role violation because v does not provide the label if the noun ragaɀɀo ‘boy’ relabels the structure: in this configuration, θ-role assignment is not available.

There is one aspect of the derivation of reduced relatives that still needs an explanation: the movement of the noun into the edge of the VP. We have shown that this movement is ultimately beneficial, as it leads to the desired (re)labeling configuration—but what is its trigger?

To approach this question, observe that, putting aside radically reduced structures like the ones considered here, the past participle is introduced by an auxiliary (see (17a), where the past participle does not agree with the internal argument and appears in its default masculine singular form). We therefore assume that the past participle is normally (namely, in unreduced structure) licensed by being selected by the auxiliary.⁶

When it is not licensed via selection by the auxiliary, we propose that the past participle can be licensed through morphological agreement with the internal argument and that is what happens in the reduced structures we are considering (see (17b)).⁷ Adopting the Minimalist technology, we assume that in order for the past participle to agree with the internal argument, it must act as a probe looking for a goal (the internal argument) that values its agreement features. In turn, this probing configuration can trigger the movement of the internal argument, which can be internally merged with the probe. This is structurally analogous to the EPP movement of the subject in the finite clause (on participial agreement and its relation to movement, see Kayne 1989, D’Alessandro and Roberts 2008, Belletti 2017).

After this background discussion, we can conclude that in reduced relatives the noun moves into the edge of the VP by virtue of entering a probe-goal configuration with the past participle. If this is true, an interesting question arises. The configuration resulting from the movement of the internal argument N is a case of Internal Merge of a lexical item; therefore, it is a case of labeling by structural prominence. This is what happens in reduced relatives, where a VP is relabeled and becomes an NP, later selected by an external determiner, as illustrated in (11). However, the movement of the internal argument N, being associated with agreement, creates a feature-sharing configuration. Therefore, given what we said in section 2, we expect that this configuration could also be a case of labeling by feature sharing. If so, the resulting structure should also be interpretable as a subject-predicate relation, that is, a root sentence. In other terms, we predict the configuration where the N moves to the VP edge to be a potential case of labeling ambiguity. Labeling by structural prominence leads to a nominal output (a reduced relative). Labeling by feature sharing should lead to a clausal output. This prediction is borne out, as we will illustrate in the next section.

In the previous section, we crucially assumed that the internal argument of the participial is a bare noun. This hypothesis played an instrumental role in providing an input to the relabeling movement that makes reduced relatives possible. But is there any overt evidence that unaccusative participials can select bare NPs? We believe there is, and that the relevant case is instantiated by another radically reduced structure that is minimally different from reduced relatives and crucially involves a bare NP. This section will focus on this second structure, which we will call BARE NOUN REDUCED for ease of reference. The distribution of this reduced structure, which is productive in spoken interactions, is constrained in a manner reminiscent of the constraints affecting reduced relatives. Some examples are given in (18)–(20), in question-answer contexts (for simplicity, in the glosses we indicate only the morphological agreement between the past participle and the internal argument).

The answers in (18)–(20) are interpreted as declarative clauses, and as such they are legitimate answers to the preceding questions. However, they are very reduced, the only overt elements being the past participle and its internal argument, which, crucially, is a bare NP. Neither the auxiliary nor the external argument is overtly expressed.

BARE NOUN REDUCED is not restricted to question-answer pairs, as long as the context is rich enough. For example, imagine that Gianni meets a friend, who welcomes him with a wide smile. She might react to his puzzled look by uttering (21).

On the other hand, BARE NOUN REDUCED cannot be embedded, either as a complement (22) or as an adjunct clause (23).

We can conclude that BARE NOUN REDUCED is rigidly a root phenomenon. Now, granted that no tense or inflection is possible, the next question is how rich its structural endowment can be. For one thing, note that it allows neither a focalized element (24), nor negation (25), nor a wh-element (26)–(27), pointing toward a radically restricted structural span.⁸

A striking observation is that despite being very reduced, BARE NOUN REDUCED structures can have a full force specification. They can be declarative (as in the examples discussed up to now), interrogative (28), or exclamative (29).

Summarizing our observations so far, all this strongly suggests that BARE NOUN REDUCED is a smaller-than-vP structure (assuming that Spec,vP is where the external θ-role is assigned; see below). We propose that it is a VP. This conclusion can be reinforced by testing one of its corollaries: if BARE NOUN REDUCED is a VP, it should be able to host more than one nominal expression provided that case is not a problem. (30) and (31) bear out this prediction.

These examples confirm that recipient arguments introduced by a preposition are perfectly acceptable in BARE NOUN REDUCED. Of course, their optionality (30) or obligatoriness (31) depends on the selection properties of the verb: mettere ‘put’ is obligatorily ditransitive (32), while restituire ‘return’ is not (33), independently of their occurrence in BARE NOUN REDUCED ((32) and (33) are full sentences).

Notice that the recipient argument in (30) is not a bare NP, but a regular DP, and that this goes with the presence of a preposition assigning case.

Overall, these data confirm that BARE NOUN REDUCED structures have all the properties associated with full bare VPs. On the other hand, they have a full force specification.

Turning now to what BARE NOUN REDUCED can consist of, note that all the examples discussed so far obey very strict constraints: (a) the participial needs to be passive or unaccusative, and (b) the internal argument can only be a bare NP. Constraint (a) is confirmed by the ungrammaticality of the answers in (34) and (35), which contain an unergative and a transitive verb, respectively.

As for constraint (b), the internal argument crucially cannot be a full DP (36)–(37).

Notice that the expressions in (36) are not ungrammatical, but they can only be interpreted as DPs. In our account, they derive from the relabeling movement of the NP and External Merge of a D, as described in section 3 and summarized in the derivation in (38). In other words, they are reduced relatives.

The expressions in (37) are not ungrammatical either, but they cannot receive a propositional interpretation as BARE NOUN REDUCED can.¹⁰ They can only be interpreted as absolutive participial constructions, which typically require a continuation in the form of a main clause (e.g., Risolto il problema, andammo a festeggiare ‘Once we solved the problem, we had a party’). We will return to this construction in section 7. What is important now is to emphasize that (37a–c) cannot have the interpretation of BARE NOUN REDUCED, namely, that of an autonomous clausal constituent.

How can we derive the correlation between full force specification and the presence of a bare NP? We believe this arises straightforwardly from the hypothesis that the relevant structure corresponds to a simple VP hosting at its edge the internal argument that underwent a movement resulting in agreement with the past participle. Consequently, the resulting VP is labeled by feature sharing and is interpreted as a clause. As for the restriction to a bare NP, it follows directly if a bare NP can be the recipient of a θ-role but does not need case: under Burzio’s Generalization, it can be happy within a root VP whereas a DP cannot.

Turning now to word order, we have illustrated that in BARE NOUN REDUCED the bare NP precedes the past participle, although the former is the object of the latter. Why is the in-situ version of BARE NOUN REDUCED, which reflects VO order, impossible?

We believe that this is the case because the reduced structure receives a subject-predicate interpretation, and ultimately a sentential meaning, only if it hosts a local Agree relation activating ϕ-feature sharing. Again, this Agree relation, which is associated with the movement of the internal argument to the left of the participial, is reflected in morphological agreement.

To be completely clear: we think that BARE NOUN REDUCED is a case of labeling by feature sharing involving agreement with the internal argument in gender and number. It adds to the case of labeling by feature sharing involving agreement in person and number, proposed by Chomsky.

The null hypothesis is that an illocutionary force can be associated with any ϕP structure, no matter whether the latter is complete (a tensed clause) or reduced (a participial clause). This explains why a structure as reduced as BARE NOUN REDUCED can be interpreted as a sentence and has illocutionary force.

As mentioned at the end of section 3, given the analysis we are advocating, the only difference between BARE NOUN REDUCED and REDUCED RELATIVE amounts to labeling. This minimal difference is schematized in (40a–b).

In both cases, only a VP is projected. Within this VP, the unaccusative verb assigns a θ-role to its sole argument, which—being a bare NP—does not need case. This NP moves to the edge of the VP as the result of the participial being probed for agreeing ϕ-features. It is only then that the two derivations diverge: if the shared ϕ-features provide the label, as in (40a), the resulting object is clausal; if the N provides the label, as in (40b), the output is a nominal expression that can merge with a D and get case from it.

We conclude this section with an observation concerning the syntax-semantics interface. As we have stressed and as its name explicitly indicates, the construction we called BARE NOUN REDUCED must contain a bare NP, as opposed to a full DP. Still, the bare noun gets a definite interpretation. For example, by uttering Missione compiuta ‘Mission accomplished’, a speaker says that a specific mission has been accomplished, and by uttering Nonna guarita ‘Grandma recovered’, the speaker is talking about a specific person (say, their own grandma). This means that bare nouns in BARE NOUN REDUCED, although syntactically impoverished, are interpreted as if they were DPs. Interestingly, this mismatch between syntax and semantics extends to proper names in Italian. Although proper names cannot be introduced by an article in standard (normative) Italian, in many regional varieties they can be (with further differences, such that an article may or even must introduce a feminine proper name in northern and central Italian regions, but not in the south, while the possibility of using an article with masculine proper names is restricted to some northern regions). However, proper names cannot be introduced by an article in BARE NOUN REDUCED even in those varieties that allow or force this option in other syntactic contexts; see (41)–(43), which are representative of the varieties (say, Milanese or Florentine) where the article is fully acceptable before a feminine proper name.

In the present article, we cannot investigate why the internal argument can be interpreted as definite in BARE NOUN REDUCED despite being syntactically bare; however, we suspect that this might be related to whatever mechanism allows Italian proper names to be interpreted referentially even when they lack a determiner (see Longobardi 1994, Chierchia 1998).

The two radically reduced constructions that we have considered so far (REDUCED RELATIVE and BARE NOUN REDUCED) contain an NP (as opposed to a DP). In our analysis, this is linked to the main properties of the constructions: namely, their reduced nature and the restriction to unaccusative and passive verbs. In a nutshell, no DP is present because these structures are smaller than vP; therefore, accusative case cannot be assigned to a full DP.

We now turn to another type of radically reduced structure, which has very similar properties but contains a DP: DISLOCATED REDUCED, illustrated in (44).

A defining property of this construction, illustrated in (44), is that the DP is right-dislocated, as these sentences would be ungrammatical without the intonational break signaled by the comma.

The analogies between DISLOCATED REDUCED, on the one hand, and REDUCED RELATIVE and BARE NOUN REDUCED, on the other, are obvious. First, these structures are reduced, as no auxiliary introduces the past participle. Second, they are incompatible with active and unergative verbs.

Third, they are too small to contain sentential negation.¹¹

Despite being reduced, DISLOCATED REDUCED—like BARE NOUN REDUCED—are sentences, as shown by the fact that they can have interrogative (47) and exclamative (48) force in addition to declarative.

Interestingly, the left-dislocated version is acceptable as well, although it becomes fully natural only with a continuation that allows a contrastive topic interpretation (i.e., the sentence is grammatical only with the intonation break signaled by the comma).

Given these analogies, the analysis of dislocated reduced should differ only minimally from the one we proposed for BARE NOUN REDUCED. In particular, we maintain that these structures are as small as VP, and we extend to them the labeling-by-feature-sharing analysis. In particular, we propose that the internal argument (here, a DP) moves to the periphery of the VP as the goal probed by the past participle. Once it is in this position, labeling by feature sharing takes place and the structure is labeled ϕP, thus being identified as a sentence, much like BARE NOUN REDUCED. It is at this point that the derivations of DISLOCATED REDUCED and BARE NOUN REDUCED differ, because the derivation of the former involves a further step, namely, the topicalization of the internal argument.

A question arises, though. Given the absence of v, accusative case cannot be assigned. How then, without a case, can the topicalized DP be licensed?

We answer this question by capitalizing on the fact that the dislocated DP in (44) (and in the other examples presented so far) must be interpreted as a topic. First, we must briefly discuss the relationship between topichood and structural case. The best languages for investigating this interaction are those in which both topic and case features are morphologically expressed. Japanese and Korean are such languages. We use Japanese to illustrate the basic pattern, which displays a curious incompatibility: when a subject or an object that would normally be marked by the nominative morpheme -ga or the accusative morpheme -o is dislocated and is marked by the topic particle -wa, the case marker disappears (case neutralization). (50) shows case neutralization with a direct object.

Importantly, case neutralization is not due to a superficial morphological filter that prevents the topic marker from associating with case particles in general, because -wa can cooccur with the genitive case marker -no (51) and with the dative case marker -ni (52).

What these data suggest is that a topic phrase can exploit a licensing mechanism that is in competition with structural case assignment. If the ultimate reason why case is needed is visibility (Chomsky 1986), we are led to conclude that a dislocated nominal can remain caseless because it becomes visible by receiving the information structure status of a topic.

An independent piece of evidence suggesting that topics do not receive structural case is the pattern observed in languages with a nominative/absolutive alignment system, like Berber and Cushitic varieties (Comrie 2013). In these languages, the subject is morphologically marked as nominative, while absolutive is the elsewhere case (it is associated to all nonsubject nominals). Interestingly, when a subject is topicalized, it receives the unmarked absolutive case (see the oral communications by Orin Gensler quoted in Satzinger 2018). This suggests that, in languages in which an elsewhere case is available (or a topic marker is present), a topic does not receive structural case.

Finally, the hypothesis that a topic does not need structural case can shed light on a very solid crosslinguistic generalization for which, as far as we know, no systematic explanation has ever been proposed.¹² We refer to the distribution of clitic left-dislocation (CLLD) as opposed to that of clitic doubling. CLLD, in which a category (including an object DP) is topicalized and a clitic surfaces clause-internally, is very widespread (no Romance variety that lacks it has been reported, to the best of our knowledge). It is very tempting to analyze CLLD as the result of topicalizing the left-dislocated category out of a clitic-doubling configuration. However, while CLLD is crosslinguistically widespread, its alleged source—clitic doubling—is rare. Jaeggli (1986), building on work by Kayne (1975), proposed a case theory explanation for the distribution of clitic doubling, which he called Kayne’s Generalization. According to this generalization, a clitic and a full DP cannot coexist if the clitic absorbs a structural case because, if it does, the double DP cannot get case and the Case Filter will be violated. Clitic doubling is thus only allowed if the Case Filter violation is obviated due to a preposition-like element that can assign a second case to the double, as in Rioplatense Spanish ((53), the preposition-like element being a) or Romanian ((54), the preposition-like element being pe; see Dobrovie-Sorin 1990). Borer (1984) proposes a similar case explanation for another instance of clitic doubling made available by a preposition, this time in Modern Hebrew.)¹³

If clitic doubling is very restricted because it leads to a Case Filter violation, and if, as the Korean and Japanese data suggest, a topic phrase does not need structural case, we do have an explanation for why CLLD is allowed across the board: the double in does not need case; therefore, the case is left available for the clitic, which can absorb it.

After this discussion concerning the relationship between case and topichood, our strategy for analyzing DISLOCATED REDUCED should be clear. Only in DISLOCATED REDUCED is an internal argument DP allowed in an environment in which it cannot receive accusative, namely, in a root VP that is not selected by v. The reason is that in DISLOCATED REDUCED, the internal argument DP is dislocated to a topic position where it does not need case. The topichood of the DP in DISLOCATED REDUCED is confirmed by the incompatibility of this construction with quantifiers like ‘all’ that resist topicalization in general, as shown in (55).

Interestingly, the same quantifier is perfectly possible in BARE NOUN REDUCED. This is expected, since no such topicalization occurs in that construction.¹⁴

In this section, we consider yet another radically reduced structure. In a sense, it is the most radically reduced of them all because it surfaces as a simple past participle, as the question-answer pair in (57) and the attested example in (58) illustrate.

We call the structure exemplified by (57) and (58) FULLY REDUCED. (57B) is interpreted as a declarative clause although the only overt element is the past participle partito ‘left’. Similarly, in (58) the main clause introducing the adverbial clause consists only of the past participle licenɀiati ‘fired’.

The difference between FULLY REDUCED and the other reduced structures we have considered so far is that in FULLY REDUCED the internal argument is dropped, although it is interpretively and syntactically present (e.g., the dropped argument acts as controller for the subject of the infinitival clause in (58)). Just like the other reduced structures, FULLY REDUCED is allowed only with passive and unaccusative verbs.

In FULLY REDUCED, the past participle agrees with a contextually given internal argument. While in (57) the makeup of the past participle is compatible with its default value (singular, masculine), (62) and (63) show that in FULLY REDUCED the past participle is inflected for gender and number.

FULLY REDUCED is not restricted to question-answer pairs, as long as the dropped argument is contextually salient. For example, imagine a context in which Gianni enters his office and notices that the desk next to his has been emptied out. A colleague might react to his puzzled look by uttering (64).

Sentential negation is not allowed in FULLY REDUCED.

Although ne-extraction is possible from the internal argument of a transitive verb (66), it is not possible in FULLY REDUCED (67).

Despite being very reduced (in fact, it typically contains only a past participle), FULLY REDUCED has a full force specification. It can be declarative as in the examples discussed so far, but it can also be interrogative (68a) or exclamative (68b).

The analogies between DISLOCATED REDUCED and FULLY REDUCED are striking and call for a unified analysis. As for FULLY REDUCED specifically, it is pretty clear that this is a case of topic drop.¹⁷ We propose that the dropped topic is the DP that surfaces in DISLOCATED REDUCED. In other terms, the input configuration of the FULLY REDUCED construction in (69a) is the DISLOCATED REDUCED sentence in (69b).

There is independent evidence supporting the topic drop analysis. It is fairly uncontroversial that an expletive cannot be a topic. This can explain why FULLY REDUCED is ungrammatical with meteorological verbs (70b): by assumption, it involves topic drop; however, the only potential topic is the null expletive subject, and expletives cannot be topics.

Absolutive past participle clauses are yet another reduced structure that is clearly reminiscent of the reduced structures we have analyzed so far. Just like the other structures, (a) they contain an agreeing participle and no further inflectional head and (b) they overtly show only an internal argument. The difference is that an absolutive past participle clause cannot be a sentence. That is, it cannot have an illocutionary force on its own; it cannot be declarative, interrogative, or exclamative. It has only an adverbial distribution, somehow juxtaposed to the main clause—hence the traditional term absolutive. In line with the labels we have applied so far, we will call it ABSOLUTE REDUCED.

Some examples are given in (71)–(73).

Interestingly, (71) and (72) display a further resemblance to the reduced structures examined so far: they contain respectively a passive (71) and an unaccusative (72) verb. As confirmation, note that minimally different sentences with an unergative verb are indeed ungrammatical (74).

However, example (73) is different: the verb here is neither passive nor unaccusative, but transitive (as Belletti (1990) originally noted). This is confirmed by the fact that there is an implicit agent, which is obligatorily controlled by the main clause subject. (75) confirms that the implicit agent is obligatorily controlled, as we are forced to assign it the weird interpretation in which the cat opens the book (as opposed to the more plausible interpretation in which the opener is Maria).

However, the external argument cannot be realized lexically (76).

There are two possible approaches here. One is to claim that ABSOLUTE REDUCED is a single construction and that it is different from the reduced structures considered so far in that it is not restricted to specific types of predicate. We do not think this is the right way to go, as unergatives are never allowed, so there are restrictions on predicate types. Another approach is to assume that there are two types of ABSOLUTE REDUCED: one corresponding to (71) and (72), where only the internal argument is licensed, and one corresponding to (73), where both arguments are licensed. We will call them UNACCUSATIVE ABSOLUTE REDUCED and TRANSITIVE ABSOLUTE REDUCED, respectively.

We will deal with these two structures in turn. First, though, we need to answer a fundamental question, which is relevant for both subtypes of ABSOLUTE REDUCED: why are they absolutive? After all, the other structures we have considered here have an illocutionary force—that is, they are sentences (as opposed to clauses) despite being very reduced—so the issue cannot be neglected. We believe we can answer this question: in ABSOLUTE REDUCED, unlike in BARE NOUN REDUCED and DISLOCATED REDUCED, the internal argument stays in argument position rather than moving past the past participle. We illustrate this in (77).

We have argued that the movement of the internal argument is the crucial device that allows labeling by feature sharing in DISLOCATED REDUCED and BARE NOUN REDUCED and that it is instrumental to attributing illocutionary force to these structures: when the internal argument moves, the structure is labeled ϕP, and in turn it can be interpreted (like any other ϕP) as a sentence. On the other hand, in ABSOLUTE REDUCED the internal argument does not move. Therefore, the structure is not labeled ϕP, it has no illocutionary force, and it can only be interpreted as an adverbial clause.¹⁸

Let us start from those absolutive clauses that contain unaccusatives, like (71)–(72). Here, there is no reason to postulate that the structure goes beyond VP: no accusative case is licensed, since the verb is unaccusative; no external argument is projected; and the participle does not seem to move, as confirmed by the impossibility of enclisis (78b).

So, the hypothesis is that UNACCUSATIVE ABSOLUTE REDUCED is another instance of the bare VP structure that stands alone as a radically reduced structure. However, a question arises in relation to this structure, concerning the makeup of the internal argument: given that the verb is unaccusative and that by hypothesis no case assigner is projected, how can the full DP be licensed here as opposed to what happens in other radically reduced structures? In the other radically reduced structures, the case problem is fixed either because the internal argument is an NP that does not need case (REDUCED RELATIVE and BARE NOUN REDUCED; see (79) and (80)) or because the internal argument DP (being a topic) does not need structural case (DISLOCATED REDUCED; see (81)). But UNACCUSATIVE ABSOLUTE REDUCED (see (82)) has a DP, and it is not dislocated. How can it get case?

We build on our previous claim that topicalization represents an alternative way to ensure visibility to a caseless DP. Now, suppose the entire UNACCUSATIVE ABSOLUTE REDUCED is in topic position here. This makes sense if we consider that it is dislocated, separated by an intonational break from the main clause, in a position reminiscent of that of a hanging topic. Furthermore, the informational status of UNACCUSATIVE ABSOLUTE REDUCED is that of introducing background information against which the informational content of the main clause is understood. Given the topic-like status of UNACCUSATIVE ABSOLUTE REDUCED, the hypothesis we would like to entertain is that the internal argument in (82) receives a topic marking thanks to long-distance agreement with the past participle—namely, the category that labels the participial clause (in traditional terms, the “head”). The proposal is that topichood can be shared via agreement.¹⁹ The form assumed by a pronoun in UNACCUSATIVE ABSOLUTE REDUCED appears to confirm this hypothesis.

In (83), the pronoun cannot exhibit the typical accusative form it takes in case-marked positions. Rather, the only possible form is io, which, besides being nominative, is crucially the form that shows up in (hanging) topics, as in (84).

Summarizing so far, UNACCUSATIVE ABSOLUTE REDUCED is yet another instance of the bare VP structure containing an unaccusative participial agreeing with its internal argument that we have seen at play in the reduced structures explored in this article. The only detail specific to this case is that here agreement is not coupled with movement of the internal argument to the edge of the structure. As a result, the structure does not get labeled by feature sharing and it does not get any sentential meaning.

We now turn to a minimally different case: TRANSITIVE ABSOLUTE REDUCED. Building on observations by Belletti (1990, 2017), we can show that TRANSITIVE ABSOLUTE REDUCED has a richer structure than UNACCUSATIVE ABSOLUTE REDUCED. In particular, the participle is in a higher position, as shown by the possibility of cliticization illustrated in (85a). As (85b–e) illustrate, cliticization is impossible in the other reduced structures considered so far (including UNACCUSATIVE ABSOLUTE REDUCED; see (78b)).

The possibility of cliticization (more precisely, enclisis) also shows another important peculiarity of this participal construction: the verb is able to assign accusative case to the internal argument. This is of course not possible in the other reduced structures, which are all unaccusatives. That the verb assigns accusative case is confirmed by sentences like (86), where the form of the strong pronoun is accusative ((86) contains a coordination because a strong pronoun is more natural in such a context).

Finally, as noted earlier, the verb here takes an external argument, which is controlled by the subject of the main clause, although it cannot be lexically expressed. Following Belletti (1990), we can reasonably assume that it is PRO.

All this suggests that the structure of TRANSITIVE ABSOLUTE REDUCED is larger than that of the simple VPs corresponding to the other reduced structures analyzed here. This structure has the following properties: (a) it triggers verb movement, which in turn explains enclisis; (b) it assigns accusative case; (c) it introduces an external argument; but (d) the latter cannot be lexically expressed. For present purposes, since we focus here on structures as reduced as VPs, we assume that TRANSITIVE ABSOLUTE REDUCED projects a vP and that the subject position is occupied by PRO.²⁰

If the structure is a vP, the impossibility of an overt external argument is explained, since nominative case cannot be assigned due to the absence of the T layer. Still, one can ask why the overt external argument cannot be licensed by the mechanism of indirect topic marking that we introduced for UNACCUSATIVE ABSOLUTE REDUCED. If this mechanism were possible, we would expect (87) to be grammatical, which it is not. Why so?

We assume that what is missing in TRANSITIVE ABSOLUTE REDUCED is the crucial mechanism of agreement that, we argue, is the vector of topic transmission: in (87), the participle does not share any feature with the external argument (it agrees with the internal argument). As a result, its topichood cannot be transmitted to the external argument DP, which is not licensed.

An alternative analysis that would make (87) grammatical needs to be excluded: namely, promoting the external argument to a topic position (the type of licensing we assumed for DISLOCATED REDUCED). This (in fact illicit) derivation is illustrated in (88).

We tentatively assume that what goes wrong is that (88) involves a topic (Gianni) extracted out of a topic (the absolute clause) and that the “topic out of a topic” configuration is not acceptable at least for Italian, for reasons we cannot explore further here. Independent evidence for this conclusion is given in (89). While the PP a scuola ‘to school’ or the entire embedded clause can be topicalized separately ((89a–b) are fully acceptable), topicalizating the PP out of the topicalized embedded clause is very marginal (89c).

The definition of (root) clauses has been at the center of a great tension in syntactic theorizing. On the one hand, it has a clear core, traditionally identified with a relation—namely, between a subject and a predicate (Williams 1980, Cardinaletti and Guasti 1995). On the other hand, much investigation has shown that sentences are typically associated with a number of functional projections that go beyond expressing the subject-predicate relation. These are involved in case assignment, agreement, tense, force, topic, focus, and potentially many other semantic features identified in the cartographic approach (e.g., Rizzi 1997, Cinque 1999). Furthermore, the Minimalist approach, which aims at reducing syntax to recursive application of the simplest operation Merge and its associated labeling operation, highlights the question of how the richness of the clausal skeleton and its fixed hierarchy can be derived from simple operations.

Here, we offer a small contribution to this vast question by building on the idea that Internal Merge of two complex syntactic objects that share ϕ-features can create an exocentric phrase, labeled by those ϕ-features, which in turn is identified with predication and receives a sentential interpretation. Chomsky (2019) has proposed this for the node created when the subject is merged to the rest of a finite clause, the node previously identified as TP or AgrP. Here, we assume that the same can happen at the level of VP, if the internal argument agrees with a past participle and is internally merged to its edge. Although the internal makeup and the size of the resulting exocentric structure varies, it can be a sentence, either complete or reduced, but still a structure with its own illocutionary force. So, we believe we have shown that having an illocutionary force, and a related simplified left periphery, can be dissociated from the vast cartography of complex tensed clauses and can result from the labeling-by-feature-sharing operation when it applies to a structure as small as VP. If we are right, there could be two ways of building a sentence: either by labeling a structure through simple ϕ-feature sharing, possibly yielding a bare root reduced structure, or by piling up functional projections probing for criterial positions of various kinds, producing full-fledged clauses.

We can now go back to the question we started with. Are reduced clauses full structures where ellipsis applies or are they created as such? At least for the radically reduced sentences we considered, we have given evidence for the latter approach: the structures are literally reduced. If so, how can they coexist with full sentences? Notice that the reduced structures discussed here are all productive and robustly attested in dialogues, and do not even seem to be subject to significant regional variation. Thus, they deserve and require an account in syntactic terms. Most of them are nevertheless largely confined to spoken, though not necessarily informal, interactions; this is because they require a rich and specific context to be used felicitously, as what is not expressed grammatically must be inferred pragmatically.²¹ A radically alternative analysis, according to which these constructions encode no structure beyond simple lexical items and require only pragmatic inference, does not seem viable in light of the syntactic properties we have explored here. A purely pragmatic account could not explain why the hearer is able to infer an illocutionary force from the context, but not tense, negation, or anything else. More importantly, the restriction to certain classes of verbs (unaccusatives, passives) clearly shows that syntax is at play here.

In fact, reduced structures in adult grammar are most likely overwhelmingly outnumbered by complete sentences, because in the latter the simple predicative core is enriched by the semantic and pragmatic information conveyed by the cartographic structure. However, the frequency and availability of these reduced structures in adults’ language contrasts with what is observed in toddlers, who—in the first syntactic phases—typically produce bare root forms lacking specification for tense or agreement and not including a left periphery. Interestingly, children learning Italian do not use root infinitives in this phase as English learners do; rather, they use participials of the kind illustrated in (90).

While a systematic survey of the morphosyntax of these bare participials is still lacking (see Franchi 2006, Hyams and Schaeffer 2008, Moscati and Tedeschi 2009), notice that the examples above look quite similar to the reduced structures we have explored here: they include unaccusative or passive verbs, and the internal argument occupies a preverbal position where it agrees with the participial. It is tempting to analyze this type of early production as the child grammar equivalent of the radically reduced sentences described here: a VP in which two phrases are merged and display agreement. Under this view, the reduced structures produced by adults could be considered a sort of fossil from this first syntactic phase.

This view is consistent with the idea that the growth of grammar involves the growth of the clausal spine starting from a VP. However, VP can get a clausal interpretation by the same mechanism (namely, exocentric labeling by feature sharing), which is operative in adult grammar “one step up”—namely, at the T level. Still, the application of the mechanism at the VP level remains as a fossil to be employed when the utterance conditions favor it.

We leave this extension to child grammar as a speculation to be further explored. We also leave to future research the identification of analogous reduced structures in other languages, which are expected to exist if they result from a mechanism as fundamental as the one we have been assuming (but see Halm 2021 for interesting converging evidence based on radically reduced sentences in Hungarian).

We thank two anonymous Linguistic Inquiry reviewers and Eric Reuland for very useful comments that led us to refine our analysis.