Following Reinhart’s (1981) aboutness view of topichood, I assume that topic-marking sentences have a dual function: first the speaker identifies the topic of the sentence with the topic phrase and second they provide new information about this topic. This section establishes that topicalization, like CLLD, is a topic-marking construction. Specifically, it is shown that topicalization exhibits a number of properties which are also found in CLLD, but not in focus-fronting sentences.

Dimitriadis (1994) shows that topicalization, like CLLD, may express old, discourse-given information. This sets topicalization (and CLLD) apart from focus-fronting sentences which are in general incompatible with old information. The following examples are slightly modified from Dimitriadis (1994: 100). In (5B) and (6B) respectively, CLLD and topicalization target the phrase (to) palto with a salient antecedent (underlined) in the preceding A’s question. The example of topicalization given in (6) involves a bare noun, rather than a definite DP, because definite DP-topics may not occur in topicalization sentences in the first place (see below).

Old information is not an option for the focus-fronted phrase in the sentence in (7B) though, which is clearly infelicitous. The focus-pitch accent is indicated with small caps.

Moreover, CLLD and focus-fronting sentences involve different intonational patterns. Baltazani (2002) shows that in CLLD sentences, which she considers topic-marking constructions, the dislocated topic phrase and the rest of the sentence form separate intonational phrases (Int-P), with a boundary tone in between. Each intonational phrase has its own pitch accent (indicated with small caps), roughly, as in (8).4 The comma (,) after the dislocated phrase denotes the boundary tone that follows the topic phrase.

Focus-fronting sentences are expressed with a different intonational pattern. More precisely there is no boundary tone between the focus-phrase and what follows. The focus phrase receives the nuclear pitch accent, while the words that follow the focus-fronted constituent get de-accented (Baltazani 2002). To my knowledge, there are no studies on the prosody of Greek topicalization. However, it is clear to native speakers that the intonational properties of CLLD (e.g., separate intonational phrases, each with its own pitch accent and lack of de-accentuation of the rest of the sentence) are present in this construction as well, as in (9) (see also Alexopoulou & Kolliakou 2002: 222, for this observation).

Having said that, for a quick diagnostic of the discourse-function (focus vs. topic) of dislocated phrases in the examples below, one can rely on the intonational properties of these two constructions and more specifically on the fact that focus-fronting is associated with de-accentuation of the material that follows the focus-fronted phrase, while de-accentuation does not arise in topicalization sentences at all (Baltazani 2002). I return to this diagnostic below, where I compare between topicalization and focus-fronting with respect to their scope-reconstruction behavior.

These are not the only differences between topic-marking constructions and focus-fronting. In the Appendix, I present further syntactic, semantic and pragmatic properties, listed in Table 1, that keep CLLD and focus-fronting apart. Crucially, as Table 1 demonstrates, topicalization aligns with CLLD, a topic-construction, rather than with focus-fronting.

So far, we concluded that topicalization is a true topic-marking dependency (see Alexopoulou & Folli 2019). Moreover, following Rizzi (1997), I assume that topic-marking is syntactically encoded through a functional projection Top(ic)P in the left periphery. On this view, the dislocated topic (CLLDed or topicalized) phrase targets the specifier position of the TopP, as in (10) below.

Note that, as (10) demonstrates, the derivational component of the grammar does not distinguish between CLLDed and topicalized phrases. In both CLLD and topicalization chains, the topic-XP forms an agree chain with the head of the TopP, the probe. (Chomsky 2001; more on this below). Therefore, from the perspective of narrow syntax, ‘CLLD’ and ‘topicalization’ actually refer to a unique topic-marking syntactic construction, call it Topic-fronting, triggered by a [TOP] feature, which targets the spec,TopP. Along these lines, the terms ‘CLLD’ and ‘topicalization’ are used in a purely descriptive manner referring to the presence or the absence of doubling clitics in Topic-fronting (analogously to terms like ‘gap-relatives’ and ‘resumptive relatives’ often found in the literature), without this however having any consequences on how narrow syntax merge CLLDed and topicalized phrases in the left periphery.5 I return to the derivation of topicalization below.

Ιt has been argued in previous studies that Greek topicalization and CLLD are in complementary distribution, depending on the referential properties of the left-dislocated topic phrase (Dimitriadis 1994; Alexopoulou & Folli 2019). The dislocation of a referential topic phrase requires CLLD, whereas the dislocation of a non-referential topic results in topicalization (Alexopoulou & Folli 2019). Along these lines, a referential definite topic phrase requires clitic doubling, excluding topicalization.

The same holds for referential or ‘specific’ indefinites which, as (12) (taken from Alexopoulou & Folli 2019: 440–441) demonstrates, necessarily undergo CLLD. Under the specific reading of the indefinite in (12a), mia kokini fusta refers to a certain skirt the speaker has in mind. This reading is promoted by (12b).

By contrast, non-referential topics exclude doubling clitics. The topicalization example in (1), repeated here as (13), involves a non-referential bare noun.

In addition, the following example shows that non-specific indefinites disallow CLLD (Alexopoulou & Folli 2019: 441). The continuation in (14b) suppresses the referential reading of the indefinite topic phrase.

In a recent paper, Angelopoulos & Sportiche (2021: section 5.3.3) argue that in CLLD sentences with a modal verbal form, the indefinite topic phrase may receive a non-specific interpretation. For example, in (15) below the CLLDed indefinite DP may refer to any good dog (‘free choice reading’, in Angelopoulos & Sportiche’s 2021 terms).

These examples seem to challenge the correlation between specificity and clitic-doubling outlined in the previous paragraphs. Ιn the present paper, following the parametrized choice-function approach to specific indefinites (Kratzer 1998), I argue that the ‘free-choice reading’ of the topic phrase in (15) shows that the dog denoted by the specific indefinite covaries with the possible worlds introduced by the modal operator. Evidence for this approach comes from Russian, a language that distinguishes between specific and non-specific indefinites morphologically (see Eremina 2012 and references therein). Eremina (2012) notes that, while –to indefinites in Russian are restricted to wide-scope/specific readings, they may also receive a non-specific interpretation in modal contexts, as in (16) (from Eremina 2012: ex. 47):

It is therefore argued that specific indefinites involve a world variable (an ‘implicit argument’ in Kratzer 1998) which in modal contexts gets bound by the modal-operator with this giving rise to the quasi-non-specific interpretation described above. As a result, I conclude that given an analysis of specific indefinites along the lines of Kratzer (1998), examples with modal operators like (15) do not threaten the correlation between clitic-doubling and specificity in topic-marking sentences.

In short, topicalization is a topic-marking dependency which is restricted to non-referential topic phrases. I return to the distribution of topicalization in more detail in Section 5, where I argue that topicalization is restricted to property-denoting topic phrases (i.e., topics of type <e,t>).

When it comes to scope, syntax goes hand in hand with the semantics. Simply put, α takes scope over β, if α c-commands β at LF. Consider (27) (modified from Alexopoulou & Kolliakou 2002: 222).

(27) involves a displaced existential topic phrase and a universally quantified DP in the post-verbal subject position.6 As Alexopoulou & Kolliakou (2002) observe, (27) admits the inverse scope reading, according to which every musician recommended a different song (∀>∃). Here I argue that the inverse scope reading is derived by total reconstruction of the topicalized phrase, as in (27’) (Hornstein 1995; Lechner 1998).7

It could be argued that the inverse scope reading of (27) is derived by QR of the universal quantifier, rather than by scope reconstruction of the topic phrase. This possibility however is excluded in long-distance topicalization chains as in (28):

Under the standard assumption that QR is clausebound, the narrow scope reading of the topicalized existential QP cannot be derived without total reconstruction.

For comparison reasons, consider now the CLLD sentence in (29) (see Alexopoulou & Kolliakou 2002: 222). Again, the topic phrase is an existential QP, dislocated to the left-periphery. The only difference is that it is now doubled by a clitic.

Interestingly, now we only get the surface scope reading (∃>∀): there is a unique song which was recommended by every musician. The inverse scope reading of CLLDed indefinites is blocked (Alexopoulou & Kolliakou 2002). Oikonomou et al. (2020; 2022) have experimentally confirmed this observation in recent studies.8

So far, we have seen that topicalized phrases may totally reconstruct to a lower position at LF. In the present paper though, I will argue for a stronger claim, namely that topic phrases in Greek topicalization must undergo total reconstruction, excluding surface scope readings. To show that, we need to examine topicalization sentences in scenarios that keep the surface and inverse scope reading apart. As will be shown below, topicalization sentences cannot describe scenarios that exclude the reconstructed reading. Consider the sentence in (30) with the topicalized negative DP kanena arthro ‘no paper’ and the numeral DP dhio krites ‘two referees’ in the post-verbal subject position. Moreover, to avoid a FOC-accent on the postverbal subject (see fn. 6, example (ii)), which could be argued to be responsible for a wide-scope reading of the subject, (30) involves an adverbial phrase in the sentence-final position. This way the adverbial phrase receives the FOC-accent which is normally assigned to the right edge of the sentence. I return to this issue in more detail in section 5.1.1.

Now imagine that I have submitted three papers for review and each paper was assigned to four referees. In order for any paper to be published it needs to be approved by at least two of the referees. In Figures 1 and 2, the lines linking the papers with the referees denote an approval relation. Figure 1 represents a scenario which is compatible only with the surface scope reading of (30), according to which none of the papers was approved by two referees. That is, no paper is published in this scenario. At the same time, this specific scenario falsifies the inverse scope reading.

On the other hand, Figure 2 represents a scenario which is compatible only with the inverse scope interpretation of (30): there are two referees (2 and 3) for whom it is true that none of them accepted any of the papers. Figure 2 is incompatible with the surface scope reading, because Paper C was accepted by Referee 1 and Referee 4.

It turns out that while (30) can be used to describe the scenario illustrated by Figure 2, it cannot characterize the surface scope context in Figure 1. This means that the topicalized negative DP obligatorily receives a narrow scope reading (2>no). This scope behavior suggests that the topicalized negative DP obligatorily undergoes total reconstruction below the numeral QP.

As expected, the CLLD counterpart of (30), in (31), can only describe the surface scope scenario in Figure 1, where none of the papers was accepted by two referees, hence none of them will be published. Moreover, (31) is false in the inverse scope scenario (Figure 2), where Paper C is approved by two referees (Referee 1 and 4).

In the rest of this section, I will give two more arguments in favor of the obligatory total reconstruction analysis of topicalization.

The second argument involves obviation of Binding Condition C in sentences with late-merge (LM) of adjuncts. It is a well-known fact, first noticed by Freidin (1986) and Lebeaux (1988), that adjuncts do not reconstruct for Condition C, as seen in example (32) reproduced from Heycock (1995: ex32).

On the standard analysis, the adjunct-relative clause is adjoined after the wh-phrase has moved (Lebeaux 1988; Chomsky 1995 among others).9

Crucially, the material that undergoes total reconstruction cannot feed late adjunction, because it is not interpreted in its higher position(s) (see Takano 1995; Heycock 1995; Sportiche 2016).

With this in mind, we can now compare topicalization and CLLD, with respect to total reconstruction. Angelopoulos & Sportiche (2021: ex11a-b) have independently shown that CLLD permits LM of adjuncts. Compare (33) with (34), a CLLD and a topicalization sentence respectively, whose intended meaning is that the president feels very uncomfortable looking at the pictures that incriminate him.

An additional pair ((35)-(36)) is provided, for reasons of concreteness. The intended meaning of these sentences is that Kostas didn’t invite the people who have hurt him.

The judgments are subtle and subject to variation.10 However, there is a clear pattern regarding the availability of counter-cyclic merge of adjuncts in the pairs illustrated above. This pattern speaks in favor of the hypothesis pursued in this section: topicalization must undergo total reconstruction and LM is blocked for all speakers, whereas CLLD does not undergo total reconstruction and LM of relative clause adjuncts is possible, at least for some speakers.11

One more piece of evidence in favor of the total-reconstruction analysis of topicalization comes from anaphor binding. To begin with, in the topicalization sentence (37), the topic phrase contains an anaphor which must be locally bound by a DP-antecedent in satisfaction of Binding Condition A (see Anagnostopoulou & Everaert 1999, on the syntactic behavior of Greek anaphor eaftos).

In (37) the topic phrase reconstructs below the postverbal subject, to its external-merge position, providing the anaphor with a local c-commanding antecedent (o Kostas). However, if I am right that topicalization undergoes total reconstruction, a more complicated behavior is expected in long-distance dependencies.

The long-distance topicalization chain in (38) involves movement of the topic to the left periphery of the matrix clause (CP1), through the left periphery of the embedded clause (CP2). This chain consists of (at least) three copies of the topic phrase. On the assumption that topicalization mandatorily undergoes total reconstruction, only the lower copy will be interpreted at LF.

Given that Greek anaphors must be bound locally, we obtain the following predictions regarding anaphor reconstruction in topicalization.12 The antecedent in the embedded clause is in a local relation with the reconstructed topic phrase, and anaphor binding proceeds without problems (39a). By contrast, in (39b), the antecedent in the matrix clause is too far away from the reconstructed anaphor, violating Condition A (see Huang 1993; Heycock 1995).

In the following I provide the relevant minimal pairs. Compare (40) with (41):13

The data in (40)–(41) confirm the predictions that are laid out in (39), lending further support to the obligatory total reconstruction analysis of topicalization.14

Returning to the WCO-sensitivity of topicalization (see (24)), recall that the left-dislocated topic phrase may not bind into the subject from its surface position. I would like now to argue that the total reconstruction analysis of topicalization offers a direct explanation to these WCO effects. As we have seen above, the dislocated topic phrase undergoes total reconstruction below the subject as in (24’). This means that neither of the copies of the topic phrase can bind the intervening pronoun, since the lower copy does not c-command the subject-internal pronoun, while the c-commanding higher copy is not interpreted at all. The intended reading, under co-indexation of the topic phrase and the intervening pronoun, is thus ruled out.

To summarize, Greek topicalization is a topic-marking strategy that involves A’-movement of a non-referential topic phrase to the spec,TopP. Moreover, the moved topic phrase obligatorily undergoes total reconstruction which means that all, but the lowest copy of the topic phrase, are neglected by LF:

(42) raises a question with respect to the topic-interpretation of topicalized phrases, though. How do they receive a topic-interpretation, if the semantic component has only access to the low copy which, according to (42), is not associated with the topic-marking projection (TopP) in any direct way? If we consider the syntactic derivation of topicalization more carefully, we will see that this problem is only apparent. As explained above (see (25)), movement of topicalized phrases is mediated by an Agree chain which involves a [Top]-feature. In short, the low copy of the topicalized phrase bears an interpretable topic-feature, as in (43):

Given (43), I argue that the assignment of the topic-interpretation in topicalization is guided by the topic-feature on the low copy of the topicalized phrase.

An additional issue that arises for the total reconstruction analysis, concerns the Criterial Freezing Principle proposed by Rizzi (2006 and later works). According to a reviewer, the topic phrase in (43), which A’-moves to satisfy a topic-criterion (in the sense of Rizzi 1997) should be frozen in the criterial position spec,TopP, without being able to move higher or reconstruct lower. It should be noted that the main assumption that underlies the freezing principle is that movement, as a last resort operation, is driven by certain interface requirements. In particular, A’-movement satisfies the scope-discourse properties of the moving phrase, which are relevant to the semantic interface (see Rizzi 2006: 99, following Chomsky 2001). However, while for instance the scope and discourse requirements of wh-phrases in questions coincide, as both are met at spec,CP (under the assumption that wh-phrases introduce an existential quantifier which takes scope over the ‘question nucleus’), this is not the case in the topicalization chains examined in this paper. The scope position of topic-phrases in topicalization is their base-position while their discourse-position is the left-peripheral spec,TopP. Given this ambivalent nature of the semantics of topicalized phrases, it is not clear what their criterial position actually is. Here I propose that the total reconstruction analysis (and the total reconstruction operation in general) can be reconciled with the Criterial Freezing Principle if we assume that discourse properties (i.e., Top, Foc, Q, etc.) determine the criterial/freezing position for overt movement, while scope properties determine the freezing position for reconstruction, where in this case ‘freezing’ means that a copy in a scope-position cannot be deleted at LF.

This section established that topicalization must undergo total reconstruction, nevertheless a deeper explanation for the LF-representation of this construction is still missing. It is not clear yet why the topic phrase in a topicalization sentence must totally reconstruct. Below I will attempt to draw a correlation between this behavior of topicalization and the distribution of this construction, after showing that topicalization is restricted to topic phrases of a specific semantic type.15

As mentioned in Section 2.2, Alexopoulou & Folli (2019) argue that topicalization is restricted to non-referential topic phrases (while CLLD involves referential topics). In this section my aim is to capture the distribution of topicalization in semantic/type-theoretic terms. Specifically, I will show that Greek topicalization is restricted to property-denoting (<e,t>) topic phrases (44).16

The argument proceeds in two steps: (a) in 5.1.1, I examine certain cases of topicalization (with topic-marked bare nouns, weak indefinites, CPs, quantificational phrases) focusing on the semantic type of the topic phrase in each case. It is shown that these topics have been independently analyzed as phrases of type <e,t>; (b) in 5.1.2, I provide independent evidence for the property type analysis from existential constructions.

In a recent work, Poole (2017; 2022) concludes that the syntax-semantics mapping for movement chains does not allow traces ranging over properties (see also Sauerland 1998: section 6.1). This claim is schematically given in (76):25

How then is movement of phrases of type <e,t> represented at LF? In these cases, the moved phrase must be interpreted in its base-position, through what is here called total reconstruction, as in (77) (see Poole 2017; 2022).

Crucial to Poole’s claim is the observation that certain syntactic positions can only host property-denoting phrases. The best-known case is the pivot of an existential construction, already introduced above (McNally 1998, among others).

Heim (1987) notes that (78) also constrains traces created by movement, therefore movement chains that leave a trace of type e must not be able to target the post-copular position in existential sentences.26 Having said that, the existential construction is expected to affect sentences which are otherwise scopally ambiguous.

For instance, it has been observed that when it comes to existential constructions, as in (79), the amount wh-question must scope below the modal should, i.e., it can only receive the reconstructed interpretation (should > how many) (Rullman 1995; Cresti 1995).

(78) requires that the pivot in (79) must be filled with an <e,t>-type element. This forces the restrictor of the wh-phrase to reconstruct deriving the narrow scope interpretation (Romero 1998):

Note that if the Grammar allowed <e,t>-type traces, the dislocated wh-phrase would be able to be interpreted in its surface position with a wide scope reading, binding an <e,t>-type variable, contrary to what we actually find. Poole concludes that <e,t>-traces do not exist; thus the wh-phrase must reconstruct into its base position.27

The present subsection tried to establish two points: (i) property traces do not exist; (ii) when a moved phrase may not leave an individual-denoting trace, it is forced to totally reconstruct. These points become relevant in the next section, where I discuss the reconstruction behavior of Greek topicalization.

Greek topicalization is a movement dependency which is interpreted through total reconstruction. However, in order to examine the source of total reconstruction in topicalization, let us take a step back: (81) demonstrates movement of a topic phrase to the left periphery without the total reconstruction step. This allows us to explore all the analytical options for the LF-representation of such a movement chain.

Given the semantic type of the moved topic, there are at least three possible LF-representations for (81) which should be considered:

In what follows I will argue that (82a) and (82b) are blocked for independent reasons. Specifically, I will show that these LF-representations cannot be translated by standard semantic mechanisms; hence Greek topicalization resorts to total reconstruction (82c).

Let us start from the representation with the property-denoting trace in (82a), where the λ-binder inserted by Predicate Abstraction ranges over a property-type variable giving rise to a derived predicate of properties (<<e,t>,t>) at the CP level. Then the derived predicate is saturated by the dislocated topic phrase. This is the analysis followed by É. Kiss & Gyuris (2003) for the Hungarian QP-topicalization sentences. However, in the previous section I argued, following Poole (2017; 2022), that property-denoting traces are not available in natural languages. This essentially blocks the LF-representation in (82a) where topicalization is mapped onto a property-denoting trace (T_<e,t>).

It should be noted that traces of higher type (e.g., T_<<e,t>t>) are widely assumed to be implicated in the process of semantic reconstruction (e.g., Cresti 1995; Rullman 1995; Lechner 1998). Specifically, semantic reconstruction refers to the fact that a trace of a higher type allows the moved element to scope in its base position. On the other hand, syntactic reconstruction posits an actual copy in the base position of the moved phrase at the LF-level. Crucially, as Romero (1998) and Fox (1999) highlight, semantic reconstruction does not lead to a violation of Binding Condition C, which applies at LF. Thus, reconstruction for Condition C has become a standard diagnostic for syntactic reconstruction (vs. semantic reconstruction) (e.g., Keine & Poole 2018; Lechner 2019). Against this background, the fact that Greek topicalization systematically exhibits Condition C reconstruction effects (see (19)) argues against semantic reconstruction or property-denoting traces.

Having excluded property traces, I proceed to the LF-representation with λ-abstraction over individual variables (x_e), as in (82b) repeated here as (83):

λ-abstraction over x_e derives a predicate over individuals (<e,t>). Above, I omitted the last step of the derivation of (83) in which the derived predicate combines with the dislocated topic phrase. As the reader might have noticed, both constituents are of type <e,t>, thus none of them can apply to its sister constituent, by Function Application. But the derivation may proceed by Predicate Modification, an operation that takes two sets and returns their intersection (Heim & Kratzer 1998). Therefore, the composition of the dislocated topic phrase (topic) with the derived predicate (indicated here as comment) results in a property-denoting constituent:

The problem with (84) is that it fails to assign to [TopicP] – a sentential constituent, in the sense that all the arguments of the verb are saturated – a propositional semantic type (cf. (75b)).28

The proposition-type of [TopicP] is independently justified by the fact that this constituent may be selected by the complementizer oti (‘that’), as in (85). Following Kratzer (2006), Moulton (2015) and others, I assume that the complementizer takes a proposition and returns a predicate (see section 5.1.1, for relevant evidence). Therefore (82) is not a possible LF-representation for Greek topicalization sentences.

Moreover, there is vast empirical evidence showing that Greek topicalization is not mapped onto individual variables. First, I made clear in Section 4 that a characteristic property of Greek topicalization is that the topic phrase is restricted to narrow scope readings, that is, it is interpreted within the scope of subject QPs, modals and negation. This property of topicalization clearly contradicts the LF-representation in (83) which predicts a wide scope reading for dislocated topic phrases.

An additional argument against the view that the copy of a topicalized phrase translates into an individual variable at LF comes from the existential construction, introduced in Section 5.1.2. Recall that the existential impersonal verb ehi requires a property-denoting pivot. When movement targets the pivot of the existential construction it must leave a property denoting phrase, as in (84):

Against this background, the fact that Greek topicalization may freely target the pivot of the existential construction (as illustrated by (87)) is one more piece of evidence that Greek topicalization does not involve individual-type traces.29

To conclude, out of three theoretically possible LF-representations for Greek topicalization (82a-c) two are blocked by the Grammar. More specifically it has been shown that the LF-representations with an individual or a property denoting trace are excluded for independent reasons. As a consequence, Greek topicalization resorts to total reconstruction which is the only well-formed – hence interpretable by the semantic component – LF-representation.