TOPIC EXTRACTION FROM ADVERBIAL CLAUSES

This paper offers new data to support findings about Topic extraction from adverbial clauses. Since such clauses are strong islands, they should not allow extraction of any kind, but we show here that if the appropriate conditions are met, Topics of the CLLD kind in Romance can move out of them. We propose that two conditions must be met for such movement to be possible: the first is that the adverbial clause must have undergone topicalisation in the first place; the second is that the adverbial clause is inherently topical from a semantic viewpoint. Contrast with other language families (Germanic, Quechua and Japanese) is provided and the semantic implications of the proposal are briefly discussed.


Introduction
This paper will deal with an empirical problem: the somewhat puzzling behaviour of left-dislocated Topics when they are extracted from the domain of certain adverbial clauses.The core of this problem has to do with the possibility of having such an operation at all, since adverbial clauses are strong islands (cf.Ross 1967, and in the context of topicalisation, see Cinque 1990 and much subsequent work) and should not allow topicalisation of an internal phrase at all.Besides this core problem, there are other unexpected patterns of behaviour in topicalisation out of these adverbial clauses.I propose here that two conditions must be met for this extraction to be felicitous: a syntactic condition and a semantic condition, both of which will be developed below.Since a semantic condition will be proposed, this empirical issue naturally extends into the theoretical domain: what is the format a syntax-semantics interface should have?This issue falls however beyond the scope of this paper and we shall only hope to provide some insights.
As just mentioned, Topic extraction should be impossible in the context of adverbial clauses since they are supposed to be opaque domains for topicalisation.This is generally true, as seen here for causal, concessive and final clauses: 1   (1) a. *A María, porque le contaste esa historia, Juan se enfadó.To María, because CL-her tell.PST.2P that story, Juan get.angry.PST.3P.'María, since you told her that story, Juan got angry.' b. *A María, aunque le cuentes esa historia, Juan se enfadará.To María, although CL-her tell.SBJV.2Pthat story, Juan get.angry.FUT.3P'María, even if you tell her that story, Juan will get angry.'c. *A María, para que le cuentes esa historia, Juan To María, in order for CL-her tell.SBJV.2Pthat story, Juan se callará.be.quiet.FUT.3P.'María, so that you will tell her that story, Juan will stay quiet.' [Spanish] Nevertheless, under the appropriate circumstances, such extraction is well formed, as the following examples show for conditional and some causal clauses: (2) a.A María, si le cuentas esa historia, Juan se enfadará.To María, if CL-her tell.PRS.2P that story, Juan get.angry.FUT.3P.'María, if you tell her that story, Juan will get angry.' b.A María, como le contaste esa historia, Juan se enfadó.To María,since CL-her tell that story, Juan angry.PST.3P.'María, since you told her that story, Juan got angry.' [Spanish] The precise nature of this contrast between different types of adverbial clauses, as well as other unexpected phenomena related to Topic extraction constitute the empirical focus of this paper.
I will therefore present the different problems we have to account for when dealing with Topic extraction from adverbial clauses and will make a proposal attempting to solve them.In order to do so, we shall draw heavily from different Romance languages (Spanish, Italian, Galician, Catalan and Romanian) and make use of conditional and causal clauses.This is so because Romance varieties show the type of Clitic Left Dislocation (henceforth, CLLD) in which we can test these constraints, 2 1 I would like to indicate certain methodological aspects that impact the selection of examples in this work.In the first place, most examples include topicalisation of a dative/indirect object.This is so because in many Romance varieties, the Accusative is unmarked and triggers Hanging Topic readings that are always well formed, therefore invalidating the analysis.On the other hand, using other PPs can sometimes trigger 'aboutness readings' that are usually judged as grammatical regardless of the syntactic structure they are part of, thereby making it impossible to falsify hypotheses.In the second place, unless otherwise indicated, all examples are the product of elicitation and judgments by native speakers of the respective languages.Such elicitation and judgments have been done in a qualitative, not quantitative manner.In the third place, regarding some unconventional notation, I have used CL as the short form of 'clitic' since clitics are crucial in topicalisation analyses and they display a different behaviour from other kinds of pronouns; hence the difference. 2Germanic languages, for instance, do not seem to allow Topic extraction from adverbial clauses under any circumstances, and the kind of Contrastive Dislocation they display may be of a different nature from that found in Romance (see Ott 2011, 2014, Haegeman 2003, 2006, 2007, 2010 and references and conditional and causal clauses are the two types of adverbial clauses where we have found unexpected patterns of behaviour.As a clarification I would like to indicate that whenever I use the words 'Topic' and 'topicalisation' in this paper I will refer specifically to CLLDed Topics, unless stated otherwise.In the same way, when I say that a certain adverbial clause allows topicalisation I will always refer to topicalisation of a phrase from inside the adverbial clause to a peripheral position in the main clause, and not to internal topicalisation which is generally possible in all cases.
The main proposal made in this paper, which I will justify in the following sections, amounts to the following two principles: (3) Feature loss under topicalisation When a phrase undergoes a topicalisation process, 3 it loses syntactic features, therefore becoming more transparent and less prone to act as an obstacle for further syntactic processes.In particular, a Topic will allow subextraction of topical material from inside.(4) Inherent semantic Topic-hood Due to their interpretation (and not just their syntactic position), certain phrases are inherent Topics.As a result, they allow extraction of material for new topicalisation processes as in the case of the feature loss just presented.
As a result of this proposal, the analysis proposed here cannot be done purely on syntactic terms, since doing so would incorrectly predict a similar behaviour for all adverbial clauses, but it must also take the semantics of said adverbial clauses into consideration.That amounts to saying that there are semantic conditions on the output of syntactic computations, which in turn leads us directly to the issue of the format we assume for the interfaces and in particular the syntax-semantics interface.
This paper is structured as follows: in the next section we shall approach the different problematic aspects of Topic extraction out of conditional and causal clauses; a review of previous accounts of this problem follows, including criticism of the shortcomings faced by those proposals; the fourth section is devoted to present a proposal of analysis in terms of a syntactic operation (feature loss) and a semantic condition (on the well-formedness of topicalised phrases); section 5 deals with the extension of the applicability of said principles to languages beyond the Romance family; a conclusion finishes the paper.

The first problem: Topics are extractable, but not always
The first unexpected issue we find when analysing conditional and causal clauses which allow topicalisation is that the latter can only take place when the former are also topicalised.This had already been observed by Etxepare (2002) and Taylor (2007), who claimed that only preposed conditional clauses are transparent for certain therein).This should not mean that the phenomena described in this paper are exclusive to the Romance family: for instance Maki et al. (1999) shows how similar patterns of extraction can be found for different types of conditional clauses in Japanese, which we will discuss later. 3As stated below, this paper assumes that CLLD topicalisation is the result of movement to the sentential periphery.This is done mainly on the ground of the different behaviour shown by Hanging Topics and CLLDed Topics in terms of Case assignment, sensitivity to island constraints and binding properties.For some relevant literature on this issue, see note 8 below.kinds of extraction in Romance varieties.In their analysis, reviewed later on in this paper, the preposed conditional clause has not undergone Topic fronting, but is rather base-generated as an adjunct to the main clause.The issue of whether Topics are adjuncts to the upper functional layers of the sentence or have a dedicated fixed position in the left periphery, while crucial for the general analysis of topicalisation, is not central to the problems under study here.Nevertheless, I shall consider sentenceinitial conditional clauses to have undergone top-icalisation for reasons developed later.In any case, the essential empirical observation is that a sharp contrast exists between CLLD with a resumptive clitic inside the conditional clause when said clause is sentence-initial and when it is sentence-final. 4Moreover, this contrast seems robust across Romance varieties, as exemplified here for Spanish, Italian and Romanian: (5) a.A María, si le cuentas esa historia, Juan se enfadará.[Romanian] (7) a.A María, se le racconti quella storia, Giovanni si arrabbierà.To María, if CL-her tell.PRS.2P that story, Giovanni get.angry.FUT.3P.'María, if you tell her that story, Giovanni will be angry.'b. *A María, Giovanni si arrabbierà se le racconti quella storia.To Mar´ıa, Giovanni get.angry.FUT.3P if CL-her tell.PRS.2P that story. [Italian] The grammaticality of topicalisation when the conditional clause appears at the beginning of the sentence does not seem to be as clear in Romanian5 as it is in Spanish (or Italian), but there is a contrast, nonetheless, in which all of the b sentences are perceived as much more degraded.
The same holds for causal clauses, although with a twist: the complementizer introduc-ing the clause changes when it is sentence-initial as opposed to cases when it is sentence-final.The unexpectedness of the behaviour exhibited by Topics in these examples comes from several considerations.In the first place we face the issue of island violation: if conditional clauses are adjuncts (which is rather uncontroversial), they should not allow subextraction.Therefore, the ungrammaticality of the (b) examples is to be expected, but in that case why don't all (a) examples incur into the same type of island violation?
In the second place, we know from all of the cartographic work that declarative sentences, like the main clause in the cases just presented, have available Topic positions in their left periphery (Rizzi, 1997).It is strange that all of the (b) examples should lack that position.At any rate, the asymmetry between the extraction possibilities of preposed and postposed clauses is unusual and asks for an explanation.
One could assume that well-formed sentences involving a Topic like the ones above have an analysis in terms of base-generation of the Topic in the left periphery.7I will assume, however, that peripheral Topics are the result of syntactic movement, following Boeckx (2003), Rubio (2014) and references therein. 8efore moving on, it should be made completely clear how the topicalisation works only when the adverbial clause is fronted.As we saw in the examples above, Topics can move above the if complementizer to a more peripheral position.Of course, such behaviour is only unexpected when comparing these examples with the ones presented for conditional (5)-( 7) and causal clauses (8).As a matter of fact, what we find here is exactly what we expect to find, both from the viewpoint of a cartographic approach to syntax and from all we know from the theory on locality constraints.The cartographic description of the sentential periphery teaches us that since no projection exists above ForceP -at least no Topic projection, since different authors (see for instance Huang &Liu 2001 andSpeas 2004) have proposed the existence of speaker-oriented projections in the uppermost region of the sentence-, no Topic can ever move to the left of the conditional if.9Moreover, recalling the adjunct status of adverbial clauses, it is also expected that no Topic can be extracted from them.
As in all previous cases, causal clauses fall under the same pattern: ( However, that is not what we find in the data in (5), ( 6) and ( 7), and in fact an approach following that line is incompatible with what we described above.It is therefore difficult to find a syntactic analysis which can account at the same time for these data and the ones before.The main observation stays: Topic subextraction from an adverbial clause must necessarily be preceded by fronting of said clause.

2.2.The second problem: a Dative-Accusative asymmetry
In general, CLLD makes no distinction between the morphological Case or θ-role of the topicalised phrase in terms of the well-formedness of the resulting sentence: (12) a.A María le dimos el regalo.This is again unexpected on two accounts.First, because it contravenes the crucial empirical observation, made in the previous section, that only preposed if-clauses and causal clauses allow topicalisation.Secondly, because of the aforementioned asymmetry between Dative and Accusative Cases, which we do not find in any other instance of CLLD.

2.3.The third problem: not as many Topics as one would like
It has been known since at least the onset of the cartographic project that Topics are recursive, that is, if topicalisation is at all possible, it is allowed to take place as many times as one would like.10See for instance this example from Cinque (1990), where four Topics in a row are apparently possible: Di vestiti, a me, Gianni, in quel negozio, non mi ce ne Of dresses, to me, Gianni, in that store, not CL-me CL-locative CL-of them ha mai comprati.have.PRS.3Pnever bought. [Italian] It should be noted that a particular intonation may be necessary to render this kind of examples possible.Even if cases like these examples from Cinque are not universally accepted (not every Italian speaker accepts it, and it's very deviant in Spanish, ??Vestidos, a mí, Juan, en esa tienda, nunca me ha comprado), it is clear that at least two Topics should produce a grammatical sentence in all Romance varieties.
[Catalan] [Spanish] The fact that one Topic is possible with adverbial clauses but more than one is a blocked operation is thus unexpected and also asks for an explanation.However, this contrast (the asymmetry among the examples a, b and c above) is attested across Romance varieties as well, as exemplified here for Italian: (21) *A Maria, questa storia, se gliela racconti, Juan si arrabbierà.To Maria, that story, if CL-herCL-it tell.PRS.2P,Juan get.angry.FUT.3P.

Previous analyses of the issue
The issue of Topic extraction from adverbial clauses seems understudied in the relevant literature.This is natural when one considers that the Condition on Extraction Domains (CED) violation, in which CLLD from conditionals incurs, predicts that no such movement should be possible.A first approach to this matter is Etxepare (2002), in which he made the observation that only preposed conditional clauses are transparent for certain extractions and proposed an analysis for Whmovement out of conditional clauses when they are sentence-initial and appear under certain subcategorising verbs (what he terms 'Stance Verbs', following terminology coined by Cattell (1978)).Later, Taylor (2007) extended the analysis specifically to topicalisation and proposed that it can work for all languages, since the original approach by Etxepare was done for Spanish.
In this section I shall review their analyses, starting from the assumptions they make for it to work and following with the analysis proper.Finally I conclude with a number of criticisms than undermine the validity of the analysis and ask for a new one.They only consider conditional clauses in their approach, but presumably it could also be applied to the kind of causal clauses we have been seeing in this paper.

3.1.Assumptions
For the following analyses to work, two assumptions (plus other two mentioned in the footnote below) must be made.In the first place, sideward movement (Nunes, 2001) is invoked.This is done in order for the moving Topic to be able to escape the adjunct island.In an analysis involving sideward movement, different syntactic trees are built in parallel, and phrasal movement is allowed from one subtree to another (under certain restrictions) before both trees have merged to form a larger syntactic structure.As a consequence, adjuncts are not adjuncts until they join the main tree, and therefore do not behave like islands for sideward movement.
In the second place, an extra head will be crucial at some points in the derivation in order to host the moving element in its specifier.Therefore, an extra functional head F (taken from work by Uriagereka 1995) is assumed to exist under Stance Verbs (in Etxepare's analysis) and in cases of topicalisation (in Taylor's).These two first assumptions are explicitly part of both analyses.11

3.2.Etxepare's analysis
Etxepare ( 2002) is to the best of my knowledge the first proposal to deal with topicalisation out of a conditional clause, but the point of departure is a different issue.The author examines cases in which a subordinate clause may be introduced by a null complementizer in Spanish, 12  [Spanish] That extra F head, as mentioned in the 'assumptions' subsection, provides an extra position for the moved phrase to go through on its way to the periphery (as shown below, this phrase is displaced from a different subnumeration via sideward movement).Although his analysis does not specifically deal with Topics, which is a task undertaken in Taylor's, he uses it to derive the unexpectedness of Wh-movement out of an if-clause: ¿Qué libro crees que si Ricardo lee, abandonará la lingüística What book you.think that if Ricardo reads, he.will.abandonLinguistics de inmediato?inmediately? [Spanish] The analysis then proceeds in a number of steps (which will be replicated in Taylor's analysis for the case of CLLD): a) Two different subnumerations are created for the independent subtrees that will eventually form the whole sentence.Note that the F 0 head available in the second numeration can appear only if the eventual matrix verb being a Stance Verb.The analysis seems to work for the case of Wh-movement out of a conditional clause, but as we shall see, it runs into several problems.One of them is it cannot explain why the conditional clause needs to be topicalised for the resulting sentence to be grammatical.

3.3.Taylor's analysis
Following the lead in Etxepare's work, Taylor's extension into the domain of topicalisation proposes that the extra F head required for the analysis is also present in instances of topicalisation, being the responsible attractor for Topic movement to its specifier.Therefore, the analysis of a well-formed sentence would be as follows: (25) A María, si le cuentas esa historia, Juan se enfadará.To María, if CL-her tell.PRS.2P that story, Juan get.angry.FUT.3P.
[Spanish] e) The Topic can move to its final landing position.This movement is now perfectly legitimate since it does not start from within the adjunct.
We shall now see how a sentence with the conditional clause at the end would crash under this approach.We know the Topic can only escape the conditional clause when the F head is merged (in order to have an available position to move to), but in this case that is too late: since the conditional clause is sentence-final, it must be merged into the main tree at the beginning of the derivation, when the VP is built and thus, it becomes an island at that time.As a result, by the time the extra F head is merged on top of the tree, the Topic is locked inside the conditional clause and trying to escape it would result in an island violation that would result in an ungrammatical sentence.

3.4.A criticism of the analysis
Although the analysis outlined above matches the empirical observation that only sentence-initial conditional clauses are extractable from and cleverly avoids the problem of the island violation by invoking sideward movement, it also faces a series of shortcomings that I shall develop next.

Overgeneration of the sideward movement mechanism
If the sideward movement mechanism works as intended in the analyses by Etxepare and Taylor, there is no reason to suppose that it could not be invoked for other cases.There is nothing among the restrictions applied to sideward movement and adopted in the previous analysis preventing it from applying to different kinds of sentences.The problem is, then, that we could invoke it in order to 'save' any kind of adjunct-island violation, provided that the adjoined embedded clause is sentence-initial.
In other words, it would predict the grammaticality of sentences like the following: (26) *A María, aunque le cuentes esa historia, Juan se enfadará.To María, even if CL-her tell.SBJV.2Pthat story, Juan get.angry.FUT.3P. [Spanish] We would simply follow the same steps we saw for the derivation of A María, si le cuentas esa historia, Juan se enfadará above, and the expectation would necessarily be that the same level of well-formedness should be attained: by merging the aunque ('although') clause at the end, the CLLDed Topic could have moved out of it via sideward movement before it becomes an island.

Superfluous character of the F head for cases of topicalisation
Etxepare's analysis makes use of the extra F head in order to gain a specifier that can work as an intermediate landing site for Wh-movement out of conditional clauses.This extra head is not just a stipulation, since it can be argued to exist under Stance Verbs.As a matter of fact, it is reasonably well known that clauses selected by certain verbs have a richer structure than others.
Extepare's analysis correctly predicts the behaviour for extraction out of conditional clauses under stance verbs, but if it were applied to topicalisation, then the wrong predictions are made.This can be seen when contrasting examples of Whmovement and CLLD under the same conditions, in this case, absence of a matrix stance verb: Taylor's analysis further elaborates on this point by claiming that such an extra F head is necessary in order to contain the unvalued [Topic] feature that must act as an attractor for the Topic, making it move to its specifier.Nevertheless, if we consider Topics to be adjuncts (and there are reasons to believe so, some of which have already been mentioned), no extra head is necessary, since the Topic could adjoin anywhere in the structure.

Stipulation of the initial position of if-clauses
The fact that all conditional clauses are base-generated in these analyses looks like an stipulation which can easily lead to a circular argument.The argumentation seems to be that, for cases in which Topics are extracted, they must be base-generated at the beginning of the sentence, since Topics can be extracted from them.At the same time, the diagnostic to propose that those conditional clauses were sentence-initial all along the derivation is that topicalisation was allowed.Such a hypothesis looks unfalsifiable and we should therefore try to arrive at a better one.
There is an additional downside to the proposal that all conditional clauses, both low and high, are originally merged in the position where they are spelled out.The problem is that such an approach goes against many of the things we know about conditionals.Iatridou (1991) provides tests to defend a low base-generation of many conditional clauses.They take the shape of binding tests in which reconstruction effects show how the conditional must have been generated in a low position for the correct binding interpretation to arise (while avoiding violations of binding principles).Examples like the following show how the if-clause must have generated in the lower layers of the sentence: ( 28 Therefore, we can conclude that topicalisation can also take place from within conditional clauses that have not been generated in the sentential periphery.

Proposal
As a natural result of the shortcomings reviewed in the previous section, the proposal put forward here should at least fulfill two requisites: no appeal to sideward movement and the possibility of having conditional clauses generated below the VP and subsequently raised to their final peripheral position.Therefore, the problems of overgeneration derived from sideward movement allowing us to extract Topics from any kind of adverbial clause and the stipulation of considering conditional clauses to be always base generated in the position where they surface would be overcome.
Let us make now an explicit proposal for the present paper.Since we shall consider conditional clauses to be generated as VP-adjuncts that are raised later on, an important issue must be dealt with: the general observation in syntax is that the more a phrase moves, the more frozen its content becomes.A clear example of this would be derived subjects becoming islands whereas in situ post verbal subjects are transparent for extraction.However, for the analysis proposed in this paper to work, displaced phrases must become more transparent than their base-generated counterparts (since conditional clauses sitting in the lower parts of the sentence are opaque but topicalised phrases become extractable).The process I propose to be at work in these cases is a type of feature loss entailed by topicalisation. (30) Feature loss in topicalisation Phrases that undergo topicalisation lose (part of) their syntactic features, making them less likely to intervene in other syntactic processes and therefore becoming more transparent for further extraction operations. 15 14 She does so, for instance, with examples like the following: If Bill i eats spoiled oysters, he i gets sick, where a co-reference between Bill and he would be impossible under a lower reconstruction analysis of the if-clause. 15I believe further research is necessary to determine to full range of feature loss involved in topicalisation.For the purposes of this paper, I will consider that phrases undergoing topicalisation become inert for other parts of the syntactic tree and do not trigger or prevent other syntactic operations, hence the proposal of a loss of features.
This proposal is not completely new.Rizzi (2004) analyses intervention affects arising among elements of the left periphery in terms of minimality.In that paper, Topics stand out because they are mostly inert for minimality, not giving rise to intervention effects or opacity for other elements of the CP layer of the sentence.In Rizzi's (2004) proposal, the crucial aspect of Topics, and the very reason why they do not suffer from or induce intervention effects of the Relativised Minimality type is precisely because they lack the types of features (Focal, Modifier, Quantificational and so on) that characterize other peripheral elements of the sentence.His treatment of word-order phenomena in the sentential periphery in terms in Relativized Minimality (Rizzi, 1990) effects hinges on features present in elements of the left periphery.His classification of such elements depending of their featural content is as follows: (31) a. Argumental: person, number, gender, case b.Quantificational: Wh, Neg, measure, focus... c.Modifier: evaluative, epistemic, Neg, frequentative, celerative, measure, manner...

d. Topic
What sets Topics apart is precisely the absence of syntactic features on them, which would fit with an approach involving this type of feature loss I have proposed here.
There is a second advantage to the suggestion that topicalised phrases undergo feature loss: it makes predictions regarding other kinds of topicalised phrases that can become transparent for extraction.An example of this are relative clauses without antecedent.As is well known from analyses on the locality of movement in syntax, relative clauses as complements of a DP are strong islands.
(32) *A Pedro, conocemos al espía que lo traicionó.To Pedro, know.PRS.1P.PL the spy who CL-him betrayed.'To Pedro, we know the spy who betrayed him.'Nevertheless, Topic extraction is possible in cases when the relative clause is sentence-initial due to its not having an antecedent: (33) a. Quien quiera visitar a la abuela, que lo diga ya.whoever want.SBJV.3svisit grandma, COMP CL-it say.SBJV.now 'Whoever wants to visit grandma should say so now' b.A la abuela, quien la quiera visitar, que lo diga ya. to grandma whoever CL-her want.SBJ visit, COMP CL-it say.SBJV now [Spanish] It has been suggested that this condition does not involve feature loss at all, and should be rather considered a condition on adjacency: as long as the CLLDed Topic and the adverbial clause remain together in the left periphery, the resulting sentence is well formed.Nevertheless, if that were true, contrasts such the following would be impossible: ( If the condition for the well-formedness of Topic extraction from adverbial clauses were mere adjacency, the separation of the Topic from its adjacent adverbial clause would result in ungrammaticality.However,what example (34) shows is that once the conditional clause has been topicalised, the Topic extracted from within can keep on moving up the tree to the matrix clause of the sentence without a severe degradation in judgments.If the conditional clause has not been topicalised, though, such movement is impossible.
The remaining problem, as mentioned before in this paper, is that not all kinds of adverbial clauses that undergo sentence fronting become transparent for extraction (and in fact, this was mentioned as the main issue with Taylor's analysis), so feature loss seems necessary but not sufficient.The crucial question is, therefore, what the difference is between the kinds of adverbial clauses that allow topicalisation from within and those which do not.I repeat here some of the contrasts presented at the beginning of the paper.
(35) a. *A María, porque le contaste esa historia, Juan se enfadó.Causal clauses introduced by porque ('because') behave differently from those introduced by como ('since') in two respects, which I claim to be related.In the first place, porque-clauses cannot appear naturally in sentence-initial position, while como-clauses are only allowed in sentence-initial position.This fact has to do with the intrinsic topical character of both conditional and como-introduced causal clauses.By their very nature, conditional clauses imply that the information contained in the ifclause be known in the communicative context.In the same fashion, como-introduced clauses imply that the knowledge of their content is shared among speakers.In other words, the types of adverbial clauses that allow topicalisation from within are intrinsically topical from a semantic viewpoint.This was the second part of the proposal outlined in the Introduction to his paper: (36) Inherent semantic Topic-hood Due to their interpretation (and not just their syntactic position), certain phrases are inherent Topics.As a result, they allow extraction of material for new topicalisation processes as in the case of the feature loss just presented.
A few more words about the way in which the terms 'inherent Topic-hood' and 'presupposition' are used are needed before moving forward.By calling the information contained in Topic phrases (and by extension, in conditional clauses) presuppositional, I refer to the classical characterisation of the topic-comment system as one in which the former contains shared or old information (cf.for instance Lambrecht 1996, where the same distinction between presuppositional versus new information is applied to because-and since-causal clauses; cf. also Erteschik-Shir 1997 on old and new information in Topics and Foci).The notion of inherent Topics from a semantic and syntactic viewpoint is built both on the previous idea of presuppositional information contained in Topics and on cross-linguistic studies linking conditional sentences and Topics, most notably Haiman (1978Haiman ( , 1986)), where it is shown how languages like Vietnamese and Hua use the same morphological mark for conditional clauses and for Topics.Ecuadorian Quechua, discussed later in this paper displays the same pattern, to the extent that one can define a conditional sentence in that language as an adverbial clause with a Topic marker (Cole 1982).
By formulating the semantic criterion in terms of an interface condition, we do not just avoid the problems encountered by the previous analyses, but also hopefully provide some insight into the way we can use interfaces to approach linguistic problems in a vein following the Minimalist Program (Chomsky 1995 and subsequent work).We can now formulate a two-condition description of the conditions under which Topic extraction out of an adverbial clause is possible. (37) Conditions for Topic extraction out of an adverbial clause Topics are free to move out of an adverbial clause, therefore ignoring its adjunct-island status if: a. Said adverbial clause is topicalised in the first place, and b.Said adverbial clause is an inherent Topic from a semantic point of view.

Implementing the proposal with respect to the three problems
With this proposal in mind we shall see now how to approach all three problems described before.Let us remember that the first problem was the violation of the island constraint itself.The proposal suggested in this paper is essentially formed in order to deal with this problem, and as a consequence, the first problem is immediately solved.Let us recall that the issue consisted the fact that Topics are able to escape conditional and causal clauses when (and only when) they are sentenceinitial.For cases such as the following, if only one of the two criteria for Topic extraction is met, the resulting sentence is ill-formed.
Let us see how that works for the core examples in this paper: (38) A María, si le cuentas esa historia, Juan se enfadará.To María, if CL-her tell.PRS.2P that story, Juan get.angry.FUT.3P.
Regarding the first criterion, whether the adverbial clause is topicalised, the answer is yes.Regarding the second criterion, whether this type of adverbial clause is inherently topical, the answer is yes as well.Therefore, the resulting sentence is wellformed.
Nevertheless, other kinds of adverbial clauses fail to meet one or both criteria and render ill-formed sentences.
( Example (39a) fails to meet the second criterion, that is, the concessive adverbial clause headed by aunque ('although') does not contain presuppostional information and is not inherently topical as a result.In example (39b), neither of the criteria is met: the adverbial clause is not fronted and besides, being a causal clause headed by porque ('because'), its information is not considered to be presuppositional and does allow topicalisation of an internal phrase.
There is a kind of concessive clause, however, which unlike examples like (39a), does not yield such deviant sentences when a Topic is extracted from within.They are headed by aun ('even') and can be seen in examples such as the one that follows in Spanish: (40) ?A María, aun contándole esa historia, Juan se enfadó.To María, even telling.CL-her that story, Juan get.angry.PST.3P.'Even though we told that story to Mar´ıa, Juan got angry.'For this kind of concessive clause we should assume that there is inherent Topichood for it to allow extraction of a Topic.As a matter of fact, we have hints that this is the case.For instance, if we force the concessive clause to be finite (unlike the gerund example above), then the complementizer aun must be followed by the conditional complementizer si ('if'), pointing to the fact that these sentences may actually be closer to conditional than concessive clauses: (41) ?A María, aun si le contamos esa historia, Juan se enfadará.To María, even if CL-her tell.PRS.1P.PL that story, Juan get.angry.FUT.3P.'Even if we told that story to Mar´ıa, Juan will get angry.'By appealing to the two criteria, we can explain the whole range of extraction from adverbial clauses we have seen along this paper.We can now implement a final test in order to see the extent of the validity of the proposal.What we would need to do would be to find an adverbial clause which is identical in all respects to the ones we have examined but failing to meet one of the criteria.I shall illustrate this test with central conditional clauses.There is a difference between two sorts of conditional clauses regarding both their semantics and their internal syntactic structure.
This distinction can be found in Haegeman (2007) as a particular case of the general distinction between central and peripheral adverbial clauses, which translates into premise and event conditionals in the field of if-clauses.Event conditionals are those in which the if-clause sets the condition that triggers the event in the main clause (42).Premise conditionals are those in which the if-clause is a prerequisite for the speech contained in the main clause, and they are often echoic in nature ( 43). (42) If it rains, we will get terribly wet.( 43) If (as you say) it is going to rain this afternoon, why don't we stay at home?
This distinction has a reflect in the behaviour of Topics in conditional clauses, and besides, is not confined to conditional clauses, since other types of adverbial clauses seem to enter into the same pattern.In general, Haegeman assumes that peripheral adverbial clauses are not actually embedded within a matrix clause, and therefore their sentential peripheries are like those of root clauses, thus explaining why they display Root Clause Phenomena such as topicalisation.However, central adverbial clauses are actually embedded under a matrix clause and have an impoverished periphery that cannot accommodate peripheral phenomena.We can illustrate this contrast with two uses of while, one as a temporal complementizer and one as a contrastive marker that structures the discourse: (44) a. *Mary listened to the radio while the dinner she was preparing.b.While your book they are using in two courses, mine they haven't even or-dered for the library.
All cases of extraction from conditionals in this paper so far belong to the first class.However, if we take a look at the second kind, premise conditionals, it seems that by all accounts they pattern like any other type of embedded adverbial and do not allow topicalisation of any sort from their domain.
(45) a.A María, si le cuentas esa historia Juan se enfadará.Why would that be the case?By making use of the two criteria I have proposed here, we can find an answer.Premise conditionals contain a prerequisite for the speech act in the main clause, but not necessarily previously shared information, which is the condition that must be met to allow Topic subextraction.Therefore, premise conditionals cannot fulfill the semantic criterion and as a result, do not allow extraction of an internal Topic despite being superficially identical to the conditionals we have studied in this paper.
A clear case in which this prediction is borne out can be seen when contrasting ifconditionals with in case-conditionals, since the latter do not presuppose shared information between speakers and therefore should not allow topicalisation of an internal argument.If we recall the three problems outlined in section 2, the second one stemmed from the apparent asymmetry between the behaviour of dative and accusative arguments when they were topicalised from an adverbial clause.I repeat here the contrast in the case of conditional clauses: (47) a. *A María, Juan se enfadará si le cuentas esa historia.[Spanish] This problem was derived from an unexpected observation on two grounds: not only because it shows an apparent asymmetry between the two kinds of DP, but also because it contravenes the data and the analysis proposed above in terms of a large Topic, formed by the adverbial clause and the CLLD to its left, as the only wellformed alternative when topicalising from within an adverbial clause.
What I propose is that in fact both kinds of Topics are ill-formed, and that the availability of extraction of the direct object DP is only apparent.I propose that it is in fact a Hanging Topic,17 not receiving but the default Case Hanging Topics receive.If this is the case, we would expect it to be base-generated directly in its position, and not to be constrained by the kind of locality restrictions that would affect CLLD generated via movement.
If they are indeed Hanging Topics, we have another three predictions: they should be able to be paraphrased by "Speaking of..."; they should not accept being a) The topicalisation of the conditional clause would take place just like in the previous examples in this paper: b) The first topicalised element from the conditional clause would be adjoined to its maximal projection, thus filling the only available position for adjuncts in that maximal projection.
c) The second Topic cannot move to the same maximal projection, and following Villalba (2000), can only adjoin to the the previously adjoined Topic, therefore producing a banned To wrap up this section, I would like to emphasize how the proposal tries to avoid the main problems of previous analyses of Topic extraction out of islands: by proposing interface conditions on the well-formedness of Topic extraction from adverbial clauses instead of appealing to sideward movement, we can avoid the overgeneration of ungrammatical sentences and we maintain one established analysis of conditional clauses on the hand and of the featureless character of sentential Topics on the other.

A preliminary comparison with other language families
In this section I will show that the two criteria used to analyse topicalisation from adverbial clauses in Romance varieties are not universal by comparison with Germanic languages, Japanese and Ecuadorian Quechua.This is not meant to be an in-depth investigation concerning the behaviour of those languages, but rather a preliminary exploration of the flexibility of the criteria presented in this paper, since languages seem to be restrictive for extraction from adverbial clauses to different degrees.We saw before how Germanic languages are very restrictive and do not allow extraction of Topics from an adverbial clause under any circumstances.However, they do allow movement of a Topic to the left periphery of an adverbial clause if said clause is of the appropriate kind, following the distinction between premise conditionals and event conditionals seen in examples ( 43) and ( 42).This is illustrated here for English (data from Haegeman 2003Haegeman , 2006Haegeman , 2007, where similar judgments are reported for German and Dutch).
(52) a. *If these exams you don't pass, you won't get the degree.
b.If with all these precautions you don't succeed, you will have to try again next week.(53) a. *Mary listened to the radio while the dinner she was preparing.
b.While your book they are using in two courses, mine they haven't even ordered for the library.
Japanese does not seem to allow any topicalisation in the context of adverbial clauses. 18it is harder to know whether these topicalisations are internal or external, due to the linear order of words in that language (data from Haegeman 2007 and Tomoko Kawamura (p.c.)).Since the conditional clause must precede the main clause and the wa-marked Topic must appear to the left of the clause, more research would be necessary to determine whether the Topic is moving to an internal position within the adverbial clause (and as a result, Japanese would ban such internal topicalisation in a similar way to Germanic) or trying to move outside the conditional clause (and as a result Japanese bans this topicalisation because it involves violation of an island constraint).( 54 Ecuadorian Quechua, on the other hand, seems to be much more flexible when it comes to extracting Topics from adverbial clauses.19Cole (1982) argues that Ecuadorian Quechua, unlike other varieties of Quechua, even allows A'-movement like wh-movement even from strong islands, as exemplified in (55), against analyses of this type of movement like Cinque (1990).

Conclusions
This paper shows the precise conditions under which topicalisation can be performed out of one type of strong island, namely Adjunct Islands.This range of data provided here adds new cases to those studied by Etxepare (2002) and Taylor (2007) by incorporating causal clauses and by showing the range of unexpected patterns of behaviour identified in Topic extraction.These patterns were classified under three different problems, and a proposal has been put forward in order to approach them.The proposal, in terms of a syntactic (feature loss) and a semantic condition (necessity of information contained in the adverbial clause to be old) tries to overcome the shortcomings found in previous analyses of this issue, which in our view overgenerated ungrammatical cases and were at odds with the data, while finding some problems of its own.The main one identified in this paper is the need to deal with the aforementioned semantic condition in a seemingly purely syntactic phenomenon.
Besides, we have seen how different languages fare with respect to Topic movement out of adverbial clauses, showing how different languages have different degrees of flexibility in this issue.Several venues of research remain open for the future.The issue of whether Japanese does allow topicalisation in conditional clauses, and whether such topicalisation is internal to the clause or movement outside remains open.The matter of defining the exact format of interface conditions like the one presented in this paper continues to be very much open in general linguistic research.Finally, the two criteria presented in this paper should be tested against more languages and structures to see to which extent they work.

Carlos Rubio Alcalá Universidad de Investigación en Tecnología Experimental Yachay Tech
Hacienda San José s/n, Urcuquí (Ecuador) crubio@yachaytech.edu.ec+593 6299 9130 Ricardo lee qué libro] L=[ FP F 0 [ IP abandonará la lingüística de inmediato]] M=[ pro crees] b) The conditional clause is not an island at this point since it exists in a parallel tree and has not yet become an adjunct to the matrix clause.Therefore, the complement qué libro ('what book') of the verb leer ('read') can move, according to the rules of sideward movement, to the specifier of FP.N=[ FP [qué libro] F 0 [ IP abandonará la lingüística de inmediato] ] c)The conditional clause is added on top of the structure via pair merge.N=[ FP [si Ricardo lee] [ FP [qué libro] F 0 [ IP abandonará la lingüística de inmediato]] d) Everything is embedded under the matrix clause.The presence of the null complementizer head F does not prevent an overt complementizer que to coappear with it.O=[ pro crees [ C ' F 0 +que [ CP si Ricardo lee] [ IP abandonará la lingüística de inmediato] ] ] e) Finally, Wh-movement of the normal sort takes place, moving the Wh-phrase to the upper part of the tree.O=[ [Qué libro] pro crees [ C ' F 0 +que [ CP si Ricardo lee] [ IP abandonará la lingüística de inmediato] ] ] a) In the first step of the derivation, different subnumerations are built from elements of the lexicon.I shall label them M (for 'Main Clause') and C (for 'Conditional Clause').M=[F 0 , Juan, se enfadará] C=[si, le, cuentas, esa, historia, a, María] b)Two independent syntactic trees are constructed following both subnumerations.M=[ FP F 0 [ TP Juan se enfadará]] C=[ CP si [ TP le cuentas esa historia a María]] c) Sideward movement takes place, moving the Topic from on subtree into the other.Two of the assumptions previously made are important now: that an F head exists in order for it to host the Topic and that since the conditional clause has not yet merged into the main-clause tree, it is not yet an adjunct and therefore, not an island.M=[ FP a María F 0 [ TP Juan se enfadará]] C=[ CP si [ TP le cuentas esa historia a María]] d) The conditional clause adjoins the other tree, thus becoming an island at this point in the derivation.M=[ FP [ CP si [ TP le cuentas esa historia a María]] [ FP a María F 0 [ T P Juan se enfadará]]] The central question is of course, what makes the (a), (b) and (c) examples different from their (d) and (e) counterparts.A particularly good case study is that of the contrast between causal clauses which allow extraction and those which don't.

(
55) a. May-pi-taj Marya ka-jpi Juan ruwana-ta randi-rka.where-LOC-inter Marya be-ADV Juan poncho-ACC buy-PST 'Where did Juan buy a pongho when María was?' b.Ima-ta-taj Juzi apamu-shka-manda Ilina wasi-manda what-ACC-INTER José bring-nmlz-because Elena house-from llugshi-rka leave-PST 'What did Elena leave because José brought?'This lack of restrictions for certain types of movement extend to Topic extraction from adverbial clauses, as shown in the following examples.They have in common with the examples from Romance we have reviewed in this paper that the adverbial clause must be sentence-initial for it to allow extraction of a Topic.However, unlike Romance languages, Ecuadorian Quechua allows topicalisation from other types of adverbial clauses beyond conditional and causal clauses.For instance, concessive and b. *Wagra-ta-ka, Ilina wasi-manda llugshi-rka, Juzi apamu-shka.cow-ACC-TOP, Elena house-from leave-PST, José bring-nmlz Taking the flexibility of Topic extraction to the other end of the spectrum, Ecuadorian Quechua seems to need only to meet the syntactic criterion of fronting an adverbial clause in order to allow extraction of a Topic from the adverbial clause.It does not seem to care about the semantic content of the adverbial clause to prevent topicalisa-tion from taking place.
To María, Juan get.angry.FUT.3P if CL-her tell.PRS.2P that story.'To María, Juan will get angry if you tell her that story.' b.Esa historia, Juan se enfadará si se la cuentas a María.That story, Juan get.angry.FUT.3P if CL-her CL-it tell.PRS.2P to María.'That story, Juan will get angry if you tell it to María.'To Maria, Giovanni get.angry.FUT.3P if CL-her tell.PRS.2P that story.'To Maria, Giovanni will get angry if you tell her that story.' b. ?Questa storia, Giovanni si arrabbierà se la racconti a Maria.That story, Giovanni get.angry.FUT.3P if CL-it tell.PRS.2P to Maria.'That story, Giovanni will get angry if you tell it to Maria.' Juan get.angry.PST.3P because CL-her tell.PST.2P that story.'To María, Juan got angry because you told her that story.' That story, if CL-her CL-it tell.PRS.2P to Mar´ıa Juan get.angry.FUT.3P.c. *A María, esa historia, si se la cuentas, Juan se enfadará.To María, that story, if CL-her CL-it tell.PRS.2P,Juan get.angry.FUT.3P.
However, we can see how it seems that conditional clauses only allow one Topic extracted from inside: observing that the presence of a moved constituent is necessary for a null complementizer to be legitimate:In order to provide an analysis of that movement, Etxepare proposes the presence of a phonologically null functional head F under Stance Verbs 13 such as say, think, claim or believe which would be absent (or unselected) under Non Stance Verbs, which would therefore prevent movement of the previously shown kind: ) a. *His i mother gets upset if every boy i is late.b. *If every boy i is late, his i mother gets upset.c.Every boy i gets upset if his i mother is late.d.If his i mother is late, every boy i gets upset.e.*John scolds his i mother if every boy i is late.f.*If every boy i is late, John scolds his i mother.g.John scolds every woman i if her i son is late.h.If her i son is late, John scolds every woman i .i.Every boy i gets upset if John scolds his i mother.j.If John scolds his i mother, every boy i gets upset.k.*His i mother gets upset if John scolds every boy i .l. *If John scolds every boy i , his i mother gets upset.However, it is also true that Iatridou presents examples in which the if-clause seems to have been generated in the preposed position. 14Therefore, it could still be maintained that there are cases for which the base-generation of sentence-initial ifclauses could explain the availability of Topic extraction.Nonetheless, such argument can be put aside if we can find an instance of topicalisation out of a sentence-initial conditional clause which in turn shows the kind of binding effects associated by Iatridou with displaced conditionals.It turns out we can in fact find such examples: To María, since CL-her tell.PST.2P that story, Juan get.angry.PST.3P.
) a. *Sono youna zasshi-wa anata-ga yomeba, yasai ga suki That kind of magazine-TOP you-NOM read if, vegetables like ni naru.become.'If you read that kind of magazines, you'd come to like vegetables.' b. *Keiko-ni-wa sore-o iu naraba, Keiko-no koibito-ni Keiko-DAT-TOP that-ACC say if Keiko-GEN boyfriend-DAT mo itta hou ga ii desu yo.too say-PST preferable be PRT.'If you say that to Keiko, it would be better to say it to her boyfriend too.'