The syntax-phonology mapping of intonational phrases in complex sentences : A flexible approach

In this paper, we extend to complex sentences the proposal that the notion of clause in Align/Match constraints related to the syntax-prosody mapping of the intonational phrase should be determined in each language (and each construction) by making reference to the highest syntactic phrase whose head is overtly filled by the verb (or verbal material) (Hamlaoui & Szendrői 2015). We propose that while root-clauses have a privileged status from the syntax-toprosody mapping perspective, all clauses are equal in the prosody-to-syntax mapping. In the spirit of the Minimalist Program (Chomsky 2005), we bring in extragrammatical motivation for the proposed mapping principles from parsing and learnability. This allows us to account for the fact that, whereas in many languages like Bàsàá (Bantu) and Hungarian (Finno-Ugric), only root clauses normally map onto intonational phrases, additional intonational phrase edges can be found under the pressure of high-ranked prosodic, processing or information-structural requirements. This is the case with Hungarian embedded foci and Bàsàá embedded topics where, we argue, embedded ι edges are meant to satisfy StressFocus and AlignTopic, respectively. In languages where embedded clauses seem to map onto their own intonational phrases more generally, such as Japanese or Luganda, further independent constraints should be evoked.


Introduction
In a recent paper (Hamlaoui & Szendrői 2015), we engaged with a "central question for [...] any theory of the syntactic-prosodic constituency relation" (Selkirk 2011: 17) how to best characterize the notion of clause in Align/Match constraints related to the syntax-prosody mapping of the intonational phrase. 1 We proposed that the notion of clause should be determined in each construction by making reference to the highest projection of the root clause (see Downing 1970), to which the verbal material (i.e. the verb itself, the inflection, an auxiliary, a question particle) is overtly moved or inserted, together with the material in its specifier (see Szendrői 2001).This gives rise to the following state of affairs.In (1a), a constituent moves to a functional specifier position accompanied by verb movement to the head.By assumption the syntax-prosody mapping principles refer to the highest position of the verb.The whole phrase will, as indicated, map onto an ι.The situation is different if the XP is not accompanied by verb movement (or insertion), as in (1b).In this case, the XP will fall outside the core ι by the proposed syntax-prosody mapping principles.
(1) a. ( ι XP V ... We showed that the state of affairs in (1a) applies to left-peripheral focus movement in Hungarian, illustrated in (2).The focus fronts to a left-peripheral position, accompanied by verb movement (in bold), and as a result the focus ends up as the leftmost constituent within the core ι.This, we show, is the position where main prominence is assigned.As a result, the focus constituent satisfies Reinhart's (1995Reinhart's ( , 2006) ) Stress-Focus correspondence principle, which states that foci must bear main prosodic prominence.
(2) Péter i [ FocP ( ι Marit j szerette k meg t k t i t j .)]Peter Mary-acc loved prt 'It was Mary that Peter started loving.' We also showed that the state of affairs in (1b) applies to zero-coded passive left-dislocation in Bàsàá (Bantu A43), illustrated in (3).Here, the object is located in a specifier position higher than Spec,TP (an inflectional domain-internal TopP), while the verb (in bold) remains in T. As a result, the object falls outside the core ι.We will describe the tonal process that corroborates the prosodic phrasing in (3) in the next section.
(3) ɓ-ɔɔŋgɛ́ɓɔ-ɓá-sô i [ TP ( ι sóɣól j à-ǹ-tɛhɛ́k t j t k ɓɔ́i.)]2-children 2.pro-2.conn-all1.grandfather 1.agr-pst1-see 2.pro 'All the children were seen by the grandfather.'So, we argued that no particular functional head plays a role in the theory of intonational phrasing, rather we must allow for a certain flexibility in the syntax-prosody mapping.Direct reference to some specific functional projection, such as TP or CP will not do.This is because a constituent may be prosodically outside the core ι even though its syntactic position is relatively low, so long as the verb is even lower and, conversely, a constituent may be phrased inside the core ι even if it is in a syntactically high position, so long as the verb also moves high.
In this paper, we would like to extend our flexible proposal to complex clauses.In particular, we would like to address the long noted asymmetry between root and non-root clauses.It was already noted by Downing (1970) (see also Truckenbrodt 2015 for a recent take-up of this idea), that root clauses seem to have a special role for the syntax-prosody mapping.For precisions' sake, let us reproduce Downing's original formulation. (4) Obligatory Boundary Insertion (OBI) (Downing 1970: 31) "Phonological phrase boundaries are inserted as leftmost and rightmost immediate constituents of every root S node [...]." (5) Root sentence (Downing 1970: 30) "A root sentence is any sentence which is not dominated by a predicative sentence." In today's language, the immediate effect of Downing's OBI rule would be that clauses that are not dominated by predicative material form their own ι.This would apply to main clauses, but not to complement clauses, or even adjunct clauses that are part of the highest projection whose head hosts the root verb.Note that prosodic integration of complement clauses (and non-extraposed adjunct clauses) does not seem to be restricted to English.The same is possible and even preferred in Turkish (Kan 2009), Swedish (Myrberg 2010(Myrberg , 2013)), Catalan (Feldhausen 2010), Xhosa (Jokweni 1995) and Durban Zulu (Cheng & Downing 2007).In this paper, we will show that this is also the case in Hungarian and Bàsàá.At the same time, in other languages, even embedded complement clauses seem to form separate ιs.For instance, Ishihara (2014;in press) showed that Japanese embedded clauses form separate ιs.Similarly, Pak (2008) showed that Luganda embedded clauses introduced by a complementizer form their own ιs.More recently, a similar claim was made by Kandybowicz & Torrence (2015) and Kandybowicz (in press) about the Tano languages Krachi and Bono, i.e. that complement clauses also introduce ι boundaries. 2 The authors also argue that the closely related languages Wasa and Asante Twi have prosodically integrated embedded clauses.
So it seems to us that there is considerable cross-linguistic variation in this area.The question that arises is whether the unmarked case is that languages have prosodically integrated embedded clauses or that the default is that all clauses (embedded and matrix) map onto their own ιs.We do not think that there is an a priori case for choosing one approach over the other.Both possibilities strike us as being equally plausible.
In this paper, we adopt the former approach and set ourselves the task to primarily account for those languages where embedded clauses are prosodically integrated.In other words, we try to account for the common denominator amongst languages, i.e. derive those boundaries that are always there: the ι-boundaries corresponding to the root clause.In doing so, we will also attempt to provide some independent motivation for our proposed syntax-prosody mapping principles.In line with the Minimalist Assumption (Chomsky 1995) we try to motivate these principles by requirements placed upon the grammar by extragrammatical cognitive considerations.In particular, we propose that integrated embedded clauses in the syntax-prosody mapping facilitate parsing, while learnability considerations motivate a certain kind of prosody-syntax mapping.For us, then, it is the embedded clause boundaries, in those languages that have them, that require additional explanations.The alternative position would be to assume that ι-boundaries are present at every clause edge.This approach would provide an immediate account for the stateof-affairs in Japanese, Luganda, Huave, Krachi and Bono.The challenge for this approach is to explain why they appear to be missing in languages like English, German, French, Turkish, Wasa, Asante Twi and, we will argue, Hungarian and Bàsàá.
Let us now turn to our proposed mapping constraints and our motivations for assuming them.There is one important difference between Downing's original formulation and standard generalised alignment constraints (McCarthy & Prince 1993).Downing's principles require alignment in only one direction, namely, his formulation is in the direction of syntax-to-prosody, rather than prosody-to-syntax.It is customary to separate syntax-to-prosody and prosody-to-syntax mapping constraints (Cheng & Downing 2009;Selkirk 2009Selkirk , 2011;;Myrberg 2010Myrberg , 2013)), but these proposals do not actually treat them differently.In ( 6) below, we propose that while root-clauses have a privileged status from the perspective of syntax-to-prosody mapping, all clauses are equal in prosody-to-syntax mapping.
The reason why these constraints differ is that they have different motivations.In our view, the syntax-to-prosody mapping constraints are primarily motivated by parsing considerations.We follow Downing's insight that only root clauses but not embedded clauses map onto separate ιs (see (6a) below).In terms of parsing, it seems to us that what helps a listener is that the sense units correspond to intonational units.Rather than marking every single clause including embedded, relative clauses etc., it seems more useful for parsing to map root sentences and/or speech acts onto intonational units.To clarify, we understand a root clause to be any clause that is not dominated by other clauses in the sense that it is neither the complement of a verb, nor does it modify a predicate.
Regarding the prosody-to-syntax mapping, one of the primary motivations for alignment between ι and the clause is undoubtedly from infant language acquisition.By now, a substantial body of experimental work exists that shows that the identification of prosodic boundaries is directly used by infants to make generalisations about syntactic structure.Starting with the classic Hirsch-Pasek et al. (1987) study, where it was shown -using the headturn preference procedure -that American infants as young as 7-10 months prefer to listen to speech material showing a coincidence of the typical acoustic cues occurring at clausal boundaries compared to materials in which the coincidence of pauses with other prosodic cues had been disrupted, even if the material was low-pass-filtered.The same effect was demonstrated in various languages (e.g. in German, see Schmitz 2008, or in Japanese, see Hayashi & Mazuka 2002).Nazzi et al. (2010) and Soderstrom et al. (2005) showed that prosodic boundary cues directly support the segmentation of clauses within a passage of sentences.These studies show that infants rely substantially on a correct alignment of ι and clauses.But importantly, what offends them is not if some clauses do not get marked by an ι boundary. 3Rather, their segmentation processes would be significantly hindered by ι boundaries that occur clause-internally.This specifically motivates the kind of prosody-to-syntax alignment constraints that map ι boundaries to syntactic units (i.e.clause).
The specific formulation of our proposed syntax-prosody mapping constraints is given in (6). 4 As (6a) shows, our proposed syntax-to-phonology mapping constraints aim to mark the edges of the root clause by ι edges.In contrast, as (6b) shows, the phonology-to-syntax mapping constraints ensure that ι edges coincide with (any) clausal edges. 5(6) Syntax-prosody correspondences on the clause-level a. Syntax-to-prosody mapping (i) Align-L (HVP, ι) Align the left edge of the highest projection whose head is overtly filled by the root verb, or verbal material with the left edge of an ι. (ii) Align-R (HVP, ι) Align the right edge of the highest projection whose head is overtly filled by the root verb, or verbal material with the right edge of an ι.
3 Although, admittedly if languages did not mark clauses by ι-edges at all, that would also be problematic, so in this sense language acquisition also provides a motivation for (6a-i), and (6a-ii).We thank an anonymous reviewer for pointing this out. 4As Elfner (2012) discusses at length, following and extending discussion by Selkirk (2011), having a framework that uses alignment constraints together with the Strict Layer Hypothesis (SLH) is demonstrably incorrect.In this paper, we are not in a position to explore the potential difference between Match and an SLH-free version of Edge-based Theory.We simply note that all the data we discuss is compatible with a formulation in terms of Match constraints, since none of our accounts hinges on a ranking between Align-Left and Align-Right constraints.We refer the reader to Elfner (2012) for differences between the two approaches and leave this issue open for future research. 5Given (4a), the syntax-phonology mapping must have access to information such as which one is the matrix or root verb.It seems to us that it will be hard to derive the data discussed in this paper, involving complex sentences with no (systematic) ι edge at the edge of the embedded clause, without allowing the syntaxphonology mapping to be sensitive to this kind of information.Also, this information does not seem to us to be qualitatively different from information that is generally assumed to be available to the mapping, such as which projection is functional or lexical, which clause is a CP or a TP, or indeed, for mapping phonological phrases and their prosodic heads, what is a syntactic head or a dependent of the head.These are all very much syntactic notions, some of which seem to be necessary for the mapping to take place appropriately.
b. Prosody-to-syntax mapping (i) Align-L (ι, HVP) Align the left edge of an ι with the left edge of the highest projection whose head is overtly filled by the verb or verbal material.(ii) Align-R (ι, HVP) Align the right edge of an ι with the right edge of the highest projection whose head is overtly filled by the verb or verbal material.
In addition to the proposed mapping constraints, we will use two additional constraints to derive the data we discuss.We propose that both the semantic notion of Speech Act and the syntactic notion of clause, understood as the highest phrase whose head is overtly filled by the verb or verbal material (HVP), are relevant for prosodic phrasing.Truckenbrodt (2015) and Selkirk (2011) proposed that the embedded clauses that form their own ι are those that constitute a speech act.Also inspired by Downing (1970), Truckenbrodt proposes that undominated clauses form their own ι if and only if they constitute a Speech Act (SpA).Selkirk and Truckenbrodt's conclusions strike us as very insightful.So, we agree that there must be an additional mapping constraint that ensures that speech acts correspond to ιs.There are various formulations of this type of constraint in the literature.Selkirk (2011) proposes a Match (Clause) constraint (based on Illocutionary Clauses/Force 0 ), while Truckenbrodt (2005Truckenbrodt ( , 2015) ) proposes a combination of Align (CP) and Wrap (CP).We remain agnostic as to the exact formulation of this constraint, and adopt a short-hand notation SpA-ι for a constraint (or a group of constraints) that is violated whenever a speech act is not contained in a single ι.
(7) SpA-ι Each Speech Act is contained in a single ι.
The other additional constraint we assume regulates recursivity.In line with many existing studies on the syntax-phonology interface (a.o Ladd 1986, 1992, Selkirk 1995, 2011;Wagner 2005; Myrberg 2013), we do not assume the Strict Layer Hypothesis (a.o.Selkirk 1981Selkirk , 1984;;Nespor & Vogel 1986), so recursive ιs can sometimes arise in case of complex syntactic structure.But we will see that recursive phrasing is sometimes dispreferred.We will invoke the constraint *Rec in such cases (Selkirk 1995;Truckenbrodt 1995Truckenbrodt , 1999)), which is given in (8).The existence of this constraint was recently questioned, due to the seemingly systematic presence of recursion in prosodic representation (Elfner 2015;Selkirk & Lee 2015).Note that if we are on the right track, complex sentences with an in situ complement or a low adjunct clause of the type represented in Table 1 (Section 2.2) and 3 (Section 3.2) constitute empirical evidence for the existence of this constraint.
(8) *Rec Selkirk (1995) "Assign one violation mark for every prosodic constituent of type n that contains another prosodic constituent of the same level."It is important to note that by distinguishing the syntax-to-prosody mapping constraints from the prosody-to-syntax ones, we do not propose that they apply at different places in the grammar.What distinguishes them from each other is simply their underlying motivations.All four proposed mapping constraints apply at the same time to determine syntax-prosody correspondences at the level of the clause (in addition to any other mapping constraints such as SpA-ι or *Rec).
In Sections 2 and 3, we will show that the proposed constraints account for the observed prosodic phrasing for Hungarian and Bàsàá complex clauses.To preview, we will show that (i) in situ complement and low adjunct clauses do not form their own ι (9a), (ii) clauses extraposed to a high right-peripheral position give rise to recursive phrasing as in (9b), and (iii) clauses extraposed to a high left-peripheral position give rise to recursive phrasing as in (9c).Given the extragrammatical motivations (parsing and learnability) we provide, it is expected that the proposed syntax-prosody mapping principles should apply universally.This is not to say that they should always be followed on the surface.The proposed mapping principles are violable or soft constraints.In Section 4, we discuss the implications of this for Hungarian and Bàsàá complex sentences with embedded topic and focus.More generally, some sofar unknown constraint X could be evoked in languages like Japanese, Luganda, Huave, Krachi and Bono to ensure the appearance of ι-boundaries aligned with embedded clausal edges, with constraint X ranked higher than *Rec.We are not in the position to make specific suggestions about what this constraint might be in the languages in question, as we simply do not have enough knowledge about these languages.But we would like to stress that the factorial typology that is necessary to account for the cross-linguistic variation between Hungarian/Bàsàá on the one hand and Japanese/Luganda on the other does not involve our proposed mapping constraints, but rather the relative ranking of *Rec and some constraints that favor structures displaying embedded clausal edge marking.In other words, as we will show in detail (see e.g.Table 1 and fn.10), our proposed mapping constraints are equally satisfied by the presence or the absence of an ι boundary at the edge of an embedded clause.
In addition, although we do not discuss such cases in this paper, languages often have prosodically (and pragmatically) marked constructions where the syntax-to-prosody or the prosody-to-syntax mapping principles can be violated.For instance, Hamlaoui and Szendrői (2015)argued that this is the case in Hungarian sentences involving contrastive focus on a universal quantifier, as in (10).
(10) Hamlaoui & Szendrői (2015: 95, ex. 25) A vizsgán mindenki MINDENT megoldott egy óra alatt the exam.oneveryone everything.accprt.solved an hour under 'At the exam, everyone solved EVERYTHING in an hour.' In such constructions, the core ι encompasses the focussed universal despite the fact that the focus fronting is not accompanied by verb movement.However, such constructions are highly restricted (i.e.universal quantifiers are independently banned from the pre-verbal focus position on semantic grounds), pragmatically marked and thus do not constitute a disturbance to the parsing and learnability considerations spelt out above.
To sum up, this formulation of the mapping principles in ( 6) is consistent with the idea that by default, ιs correspond to root clauses and not to embedded clauses, and that ι boundaries are normally aligned with clausal edges.It makes the following predictions: (i) ιs by default correspond to edges of root clauses; (ii) If extra ι edges are inserted, for instance to satisfy information-structural needs or other higher-ranked constraints, they are not inserted randomly: they are licit at both root or embedded clausal edges (or at the edges of speech acts).In the following sections, we will demonstrate that these predictions are borne out for Hungarian and Bàsàá complex clauses.

Diagnosing ι edges
Before examining the intonational phrasing of complex clauses, let us first introduce the tonal rule that will be used to diagnose ι edges in Bàsàá. 6 On the surface, Bàsàá displays a 5-way contrast between High (H), Low (L), downstepped High (ꜜH), Falling (HL) and Rising (LH) tones.Underlyingly, it however only distinguishes High and Low tones; the three other tonal specifications are the results of various surface associations of underlying H and L tones (a.o.Dimmendaal 1988;Bitijaa Kody 1993;Hyman 2003;Makasso 2012).
6 Bàsàá is a Bantu language (A43) spoken in Cameroon by over 300,000 speakers (SIL 2005).The present data and judgments stem from the Northern variety of Bàsàá spoken by Emmanuel-Moselly Makasso. 7There are various tonal processes by which Falling tones are derived.Bitjaa Kody (1993) observes that they can either be derived through the doubling/spreading of an underlying tone (i.e.HTS) or through the historical loss of a mora, by re-association of a delinked L to the left.The latter case is illustrated in (i).
(i) Bitjaa Kody (1993: 66) Proto-Bantu *-bédè → -ɓê 'udder' (in Bàsàá): H + (L) → HL We are aware of the fact that the way by which an HL tone is derived determines its ability to simplify (this is however beyond the scope of the present paper).As pointed out to us by Emmanuel Makasso and as corroborated by Lemb & de Gastines (1973), it is likely that the HL on [sô] 'all', which stems from [sónà], is derived by the latter process rather than by HTS, as previously stated in Hamlaoui & Szendrői (2015). 8Bàsàá does not display any type of obligatory sentence final lowering (e.g.sentences ( 13), ( 51) and ( 52) illustrate the fact that sentence-final H tones are found in this language).Declarative sentences with only High tones display a slight declination, but no evidence for the presence of a final L% boundary tone (Makasso et al. 2016).
( 16) ]] 1.grandfather 1.agr-pst1-see 2-children loc 9.house 'The grandfather saw the children at home.' Additionally, in ( 14), the fronted object of a zero-coded passive, in its turn, occupies the specifier position of a topic projection within the inflectional domain, i.e. lower than CP (see Hamlaoui 2014;Hamlaoui & Szendrői 2015 and references therein for more details on this structure).The absence of verb movement from T to Topic is what leads to the lack of prosodic integration of the fronted object into the core-ι formed by the remainder of the sentence.
Let us now turn to the intonational phrasing of complex sentences.

The ι-phrasing of root and non-root clauses
Relative clauses aside (Makasso 2010;Jenks et al. 2012), little work has been done on embedded clauses in Bàsàá.That-clauses are introduced by the complementizer lɛ, which according to Bassong (2010), is simply located in C. )) The idea that topics have a rightward ι boundary following them would however not be problematic for our approach, as the object would still be located where we expect it to be located, namely immediately outside the core ι as determined by the highest position of the verb, and a prosodic constraint like StrongStart (Selkirk 2011) or EqualSisters (Myrberg 2010(Myrberg , 2013) ) could simply be responsible for it forming an ι of its own rather than a φ.See Section 4.2 for more detail on this issue.Note, however, that in a structure with clausal extraposition, such as (21), the right edge of a main clause fails to block FTS, supporting the view that in topic structures it is the left-edge of the ι that follows that blocks FTS.
As the complementizer underlyingly carries a H tone, it allows us to assess the prosodic structure of a complement clause and more particularly whether it is preceded by an ι-boundary (corresponding to the embedded CP).The phrasing observed in Bàsàá patterns with what is observed in English as well as in Hungarian: a complement clause fails to introduce ι breaks.This is shown in (17), in which the Falling tone on lɔŋgɛ̂ 'well' (here used as an adverb) undergoes simplification and creates a downstep on the complementizer.The sentence in ( 14) thus involves a simple ι, encompassing the entire sentence. (17) [ TP ( ι mɛ̀ ń-sòmból jí lɔŋgɛ́ [ CP ꜜlɛ́ [ TP mbómbó à-ǹ-lɔ.)]]]mɛ̀ń-sòmból jí lɔŋgɛ̀lɛ́mbómbó à-n-ˊ-lɔÌ want to.know well that 1.grandmother 1.agr-pst1-mH-arrive 'I really want to know that the grandmother came.' The same phrasing is observed with adjunct clauses in canonical postverbal position, in ( 18) and ( 19).The adjunct clauses considered here are when-clauses.They are introduced by the prepositional phrase ι ꜜŋgɛŋ 'at hour'.We assume that these clauses attach lower than TP, somewhere in the core-clause (although we have not yet been able to find evidence of this in Bàsàá, in the next Section we give syntactic arguments for the low position of such adjunct clauses in Hungarian).The Falling tone occuring on the last syllable of the monosyllabic verbs arrive and leave (visible here when the verb is in sentence final position, as it is not followed by a H tone) is obtained by the association of a melodic H (noted mH in the glosses) that accompanies certain TAM structures such as the recent past (pst1) and an underlying L tone on the last mora of these verbs (see Makasso 2012Makasso , 2014 and references therein concerning melodic H tones in Bàsàá and more generally in Bantu languages).( ι sóɣól à-ʤɛḱ má-kàlà mɔ-má-só ꜜí ꜜŋgɛŋ Lingom à-ǹ-lɔ.)sóɣól à-ʤɛḱ ma-kàlà mɔ-má-sò íˋŋgɛŋ Lingom 1.grandfather 1.agr-eat 6-doughnuts 6.pro-6.conn-allat hour Lingom à-n-ˊ-lɔ1 .agr-pst1-mH-arrive 'The grandfather ate all the doughnuts when Lingom arrived.'What we see in ( 18) and ( 19) is that the L tone of the root verb and the complement, respectively, delinks and creates a downstep on the first H tone of when, indicating that the adjunct clause does not introduce an ι edge of its own.
The same pattern is found in purpose clauses, as illustrated in (20).They are also introduced by the complementizer lɛ, and we assume that they also attach below TP.As shown by the application of FTS, they do not insert an ι-edge.
(20) ( ι Lingom à-ǹ-lɔ́ ꜜlɛ́ ɓá-tí ɲɛ́ βìʤɛḱ.)Lingom à-n-ˊ-lɔ̀lɛ́ɓá-tí ɲɛ́βiʤɛḱ Lingom 1.agr-pst1-mH-arrive that 2.agr-subj.give1.pro 8.food 'Lingom came so that they give him food.' The ι-phrasing observed in sentences ( 17) to ( 20) is consistent with our mapping constraints in (6): only the highest projection hosting the root verb maps onto an ι.In Table 1, we give an illustration of how our proposed mapping principles account for the intonational phrasing of complex sentences involving an in situ complement (or an adjunct clause attached below T).To aid the reader, we put the syntactic brackets that correspond to the highest position filled by the root verb with bold italics, and the highest position filled by a non-root verb in bold.Alignment with these positions is favoured by some of our proposed mapping constraints (Align-L/R (hvp-ι), Align-L/R (ι-hvp)).We also number the ιs to be able to refer to the specific edges that incur a violation on the mapping constraints.The proposed ranking of our mapping constraints will not be motivated until more elaborate data is considered below.Note that Align-L/R (hvp-ι) and Align-L/R (ι-hvp) (respectively notated hvp-L/R and ι-L/R in the tableaux) are evaluated categorically. 10dditional evidence that embedded clauses do not (necessarily) form their own ι but rather sit outside the ι formed by the core-clause comes from right extraposed clausal subjects (note in passing that preverbal clausal subjects are generally unacceptable in the present variety of Bàsàá, as stated in Hamlaoui & Makasso 2015).In ( 21), the extraposed clause attaches high, i.e. somewhere above TP.The constraint Align-R (hvp, ι) predicts that it is separated from the core-clause by the right-edge of an ι, corresponding to the right-edge of the core-clause.As shown in ( 21), FTS applies between the two clauses.This is important in two respects: first it is consistent with the idea that FTS is not sensitive to ι right edges and, second, it shows that the extraposed clause fails to introduce an ι left edge of its own.111.agr-be.preswell that 1.grandfather 1.agr-pst1-eat 9.pro 'This is good that the grandfather ate it.' The same pattern is observed with extraposed adjunct clauses.In contrast with sentences (17) to (20), whenever an adjunct clause surfaces outside its canonical postverbal position and linearly precedes the main clause, what we observe, as in ( 22), is that FTS fails to apply and thus that the adjunct clause is separated from the main clause by an ι boundary.(22) [ TP [ CP ( ι í ꜜŋgɛŋ Lingom à-ŋ-kɛ] [ TP ( ι sóɣól à-ǹ-lɔ.))]]íˋŋgɛŋ Lingom à-ŋ-ˊ-kɛ̀sóɣól à-n-ˊ-lɔà t hour Lingom 1.agr-pst1-mH-leave 1.grandfather 1.agr-pst1-mH-arrive 'When Lingom left, the grandfather arrived.'As seen in Table 2, Align-L (HVP, ι) favors the presence of a left ι boundary at the left-edge of the root clause which, in ( 22), sets off the adjunct clause from the rest of the sentence.There is no clear evidence in Bàsàá (or in Hungarian) that the adjunct clause is additionally aligned with the right-edge of an ι corresponding to the right edge of the extraposed clause, so candidate b is favoured over candidate e.A phrasing comparable to e is however attested in other languages (e.g.Swedish).Myrberg (2013) argues that the choice between these two candidates is regulated by the ranking of EqualSisters, given in ( 23  The argument against the idea (offered by one of our reviewers) that FTS would be blocked by rightward ι edges can be further strengthened.The sentence in ( 24) additionally shows that FTS also takes place between a HL tone at the end of a relative clause contained in the first object of a ditransitive verb and a H tone at the beginning of the second object.Together with example ( 21), ( 24) casts further doubt on the idea that FTS is generally sensitive to the right edge of ιs rather than to their left edge, in particular if one assumes that all clauses introduce ι edges. 12(24) [ TP ɓá-ń-tí [ CP í-ꜜsóɣól à-ǹ-lɔ] ꜜndáp.]2.agr-pst1-give aug-1.grandfather1.agr-pst1-come 9.house 'They gave the house to the grandfather who came.' 12 According to E.-M.Makasso (p.c.), FTS is obligatory in (24), which contrasts with what is observed in relative clauses that are part of a subject, as in (i).In that case, FTS between the last tone of the relative clause and the subject agreement is optional.A way of accounting for this fact that would be consistent with our findings so far is that whenever FTS is blocked, the subject is actually located in a higher syntactic, perhaps topical, position, crucially, outside the HVP.We leave this issue open for future research.
(i) a. í-ɓ-ɔɔŋgɛ́ɓá-ń-lɔ̂ɓá-ń-tɛhɛ́sóɣól.b. í-ɓ-ɔɔŋgɛ́ɓá-ń-lɔ́ ꜜɓá-ń-tɛhɛ́sóɣól.aug-2-children agr-pst1-come 2.agr-pst1-see 1.grandfather 'The children who came saw the grandfather.'As pointed out by one of our reviewers, our account predicts that if an ι-boundary is to be found in an embedded clause, it might actually correspond to the edge of TP rather than the edge of CP.In order to test this, we would need to observe the behaviour of a complementizer carrying a Falling tone.To the best of our knowledge, no Bàsàá complementizer displays this tonal specification.Evidence for the absence of ι boundary between C and TP is however found in restrictive relative clauses.Bàsàá relative clauses are marked by means of an optional relative operator, which is identical to the near speaker demonstrative (Makasso 2010;Jenks et al. 2012).This demonstrative, which according to Jenks et al. is located under Spec,CP, exhibits a HL tone in all noun classes except for class 9. Whenever the demonstrative pronoun is realized and followed by a H tone, its HL tone simplifies and creates a downstep.This is the case in (25), from Makasso (2010), and (26), adapted from Jenks et al.This shows that no ι boundary intervenes between embedded C and T, as predicted by our mapping constraints.
(27) lí-ꜜwándá lí ꜜlí-βí-ꜜʤɛ́βíʤɛḱ aug.5-friend 5.dem 5.agr-pst2-eat 8.food 'this friend that ate the food.' In sum, what we have shown in this section, through the application of the FTS rule, is that Bàsàá is similar to languages like English, French, Zulu or Xhosa in that only certain types of clauses map onto ιs, i.e. root clauses.Embedded clauses fail to introduce their own ι edges.If they are not dominated by the highest projection hosting the root verb, they can however sit outside the core ι.Let us now turn to Hungarian, which provides further evidence for this asymmetry between root and non-root clauses.

Diagnosing ι edges in Hungarian simple clauses
Let us start with a short overview of simplex clauses in Hungarian, but the reader is referred to Szendrői (2001) for more details.Traditionally, Hungarian grammars distinguish so-called neutral (28) and non-neutral sentences (29), with the latter involving a left-peripheral, fronted focal constituent (appearing in small caps).Importantly, focus fronting is accompanied by verb movement, as indicated by the stranded particle (prt) in (29).As a result, our mapping principles in (6) will have the effect that left-peripheral topics, which are not accompanied by verb movement, are phrased outside the core ι, while fronted foci appear as the leftmost φ in the innermost ι.

))]]]]
Emilia from 'The Malay girl escapes from Emilia to Eleonora.'Ladd (1996) proposed a useful diagnostic for the ι in Hungarian.In yes/no-questions, which are syntactically unmarked in Hungarian, the intonational contour (H-)L*H-L% carries the semantic force.It starts with an optional (L)H(L) accent on preverbal elements in neutral clauses and prefocal elements in non-neutral clauses.The L* is anchored on the PRT (or if there is no PRT, then on the verb itself) in neutral clauses (as in Figure 1), and on the focal element in non-neutral clauses (as in Figure 2). 13We assume that this is because these elements constitute the leftmost φ within the core ι.The right edge of the contour is marked by a H-phrasal and a L% boundary tone, which are aligned with the penultimate and final syllables of ι respectively.Word-level and phrase-level stress are always on the initial syllable in Hungarian.In declarative sentences, discourse-new, non-topical phrases have been observed to generally show falling pitch movements (see Genzel et al. 2015 and references therein).In questions, we predominantly observe rise or rise-fall movements (noted L-H(L)).Importantly, no postverbal phrasal accents occur in non-neutral utterances (as seen e.g. in Figure 2).14

Prosody of Hungarian complex clauses
Hungarian CP complement clauses and adjunct clauses associate with a D-type head, az.The function of such elements is unclear, with some arguing that it provides a nominal shell for the CP, which allows case marking the complement Kenesei (1994).This D-like

20
Hamlaoui & Szendrői or rise-fall movements (noted L-H(L)).Importantly, no postverbal phrasal 530 accents occur in non-neutral utterances (as seen e.g. in Figure 2).element either shows up in a position directly left-adjacent to the clause itself (az, hogy ... 'D that ...') or the clause might be extraposed into a left-peripheral or right-peripheral position, with the D-like element appearing in a topic or focus position.If there is A-bar movement across the CP from the embedded clause into the main clause (see (38) in Section 4.1), the D-head must be dropped.Figure 3 illustrates the pitch track of the complex yes-no question in (30), with the main verb megkérdezte 'prt-asked' and its complement clause.As we can see, the initial L* is on the prt-main verb complex, while the right boundary of the ι (marked by H L%) appears at the end of the utterance.Phrasal accents appear throughout the sentence.Thus, it appears that the intonational phrasing is as given at the bottom of Figure 3. 15 As can be seen in Table 3, Align-L/R(HVP, ι) are satisfied by the ι boundaries in front of the main verb and at the right edge of the sentence; SpA-ι is satisfied by the outer ι boundaries 15 Note that as Figure 3 and 4 show, embedded topics are not necessarily aligned with ι edges in Hungarian.
There are two possible accounts for this state of affairs.First, it is possible that in embedded clauses, subjects can occupy a low Spec,vP position, while in the main clause, Spec,TopP must be filled (cf.É. Kiss 1995 on Hungarian as a topic prominent language).This would explain why Align-Topic (Section 4) does not apply in the embedded clause: the subject NP in the embedded clause is not a topic.Another possibility is that Align-Topic is ranked lower than *Rec in Hungarian, giving rise to a non-recursive phrasing and nonalignment for topics inside embedded clauses.We will leave this issue open for future research.

Prosody of Hungarian complex clauses
Hungarian CP complement clauses and adjunct clauses associate with a Dtype head, az.The function of such elements is unclear, with some arguing 14 As a systematic study by Gyuris & Mády (2014: 9) reveals, there is considerable variation as to whether the topic contour is a flat high (most common), a fall or a rise in Hungarian.They did not manage to correlate the variation with a number of factors including contrastivity and NP-type.We therefore think that the falling intonation in Figure 1 is not a good diagnostic for an ι right edge.But even if it was proposed that there is a well-defined subset of topics in Hungarian, displaying the fall contour, and this subset of topics form their own ι, that would not effect the argument made here, which is that the position of topics is at the edge of the core intonational phrase.prt-escaped-q Emilia-to 'Did Lejla ask Eleonora whether the Malay girl escaped to Emilia?' In Figure 4, corresponding to sentence (31), we see the same question in a non-neutral form.
Here the main clause focus occupies the leftmost position in the ι.This is a non-neutral sentence: there is pitch reduction on the elements following the main clause focal element, realized as a high plateau before the final question rise and fall at the end of the embedded clause.This prosodic pattern is consistent with our proposal that complement clauses do not form their own ι in Hungarian. 16As above, the candidate with an additional ι boundary at the edge of the embedded clause, candidate d in Table 3, would have to be promoted by some constraint that is ranked higher than *Rec.Presumably, this would apply to languages like Japanese or Luganda.prt-escaped-q Emilia-to 'Was it Lejla who asked Eleonora whether the Malay girl escaped to Emilia?'An environment where complement clauses do seem to form separate ιs in Hungarian is direct quotation.As shown in Figure 5, corresponding to sentence (32), here the main clause is followed by an obligatory pause, and both the D-like element and the complementizer must be omitted (cf. Figure 3, which illustrates the same utterance with reported speech rather than direct quotation).( 32 Finally, Figure 6, corresponding to (33), illustrates the fact that if the main clause contains a focal element, then accent reduction only affects the main clause in direct quotation, supporting further that the two are independent ιs (cf. Figure 4, where the embedded clause is not a direct quotation, and consequently undergoes pitch reduction).

24
Hamlaoui & Szendrői supporting further that the two are independent ιs (cf. Figure 4, where the embedded clause is not a direct quotation, and consequently undergoes pitch reduction).
( prt-escaped Emilia-to 'It was Lejla who asked Eleonora: "Did the Malay girl escape to Emilia?"' Thus, from these data it seems that the left edge of the embedded clause does not create its own ι, unless the embedded clause constitutes a direct quotation.This is consistent with the proposed syntax-to-prosody mapping principles, in which only the highest projection hosting the root verb maps onto an ι. When it comes to adjunct clauses, a parallel state-of-affairs to Bàsàá obtains in Hungarian.The example (34) (shown in Figure 7) involves an adjunct clause with amíg 'while'.The canonical position of such clauses is either VP internal (often VP-final), as shown in (34), or adjoined to VP with the D-like associate inside the core VP intermingled with the verb's nominal arguments, as in (35).As the phrasing indicates, in both cases, the adjunct clause is integrated into the ι corresponding to the main clause, as expected.17while Mary works 'Does Peter take the children to the museum, while Mary is working?' Let us now consider a third option, namely when the adjunct clause is extraposed to the left-peripheral position. 18 As we see from (36b), topics cannot precede extraposed clauses.This means that the extraposed clause is outside the highest projection whose head is overtly filled by verbal material.It is either in the C-domain or adjoined to whatever projection hosts topics.The intonational contour of such a complex yes-no question is as indicated in Figure 8 (corresponding to (36a)).Here we see that the material preceding the main verb, which bears main stress, is assigned its own phrasal contour.This contour, however, does not signal the presence of an ι.Rather, it is a phrasal contour expressing non-finality.This is also the contour that is associated with contrastive topics (Kálmán 2001: 25).Note also that the contour spreads onto the D-like element and the main clause topic, supporting further the idea that the adjunct clause does not form its own ι.It simply fails to be integrated to the core-ι formed by the root clause. 18The D-like associate can occur in various places, but that is irrelevant for us here.

Extra ι-edges
So far we have seen that in Bàsàá and Hungarian there is generally no ι boundary at the edge of complement and adjunct clauses.It is normally only the root clause whose edges are obligatorily matched by ι edges.In this section, we would like to turn to cases where such an ι boundary seems to be present at the edge of the embedded clause.We will argue that embedded clause boundaries can be marked in Hungarian and Bàsàá under pressure from higher ranked information structural constraints such as Align-Topic and StressFocus.

Hungarian embedded foci
In Hamlaoui & Szendrői (2015), we adopted earlier proposals and argued that certain information structural constraints play a role in determing the prosodic status of certain elements.This is the case of the constraint in (37), noted StressFocus in Table 4.
(37) Focus Rule or Stress-Focus Correspondence Principle (Reinhart 1995(Reinhart , 2006;;Szendrői 2001Szendrői , 2003) ) "The focus of a clause is a(ny) constituent containing the main stress of the Intonational Phrase, as determined by the stress-rule." In particular, we argued that in Hungarian, syntax is responsible for placing the focal element into a position that will satisfy the constraint in (37) -the left-edge of a clause -while conforming with the default syntax-prosody mapping constraints and the default prosodic stress rules.In this language, information-structural requirements are primarily satisfied by non-canonical syntactic constructions with regular syntax-prosody mapping (and default stress).This was illustrated in (29), repeated below for convenience.
(29) [ TopP ( ι A maláj lány k [ FocP ( ι Eleonórá-hoz j menekül i [ vP el [ VP t i t i t k t j the Malay girl Eleonora-to flees away Emília elöl.))]]]]Emilia from 'The Malay girl escapes from Emilia to Eleonora.'As for complex clauses, we predict that foci in Hungarian can target the edge of the root clause, even if they originate in the embedded clause, as in (38).This is because our syntax-prosody mapping principles place the left edge of the core ι at the left edge of the highest overt position of the root verb.Such long focus constructions are indeed possible (in long focus movement the D-like associate of the embedded clause must be omitted, as stated in Kenesei 1994).Comparing Figure 9 (corresponding to (38)) to focus movement in the main clause of a complex yes-no question, as in Figure 4 reveals that their prosodic structure is identical.Emilia-to 'Was it the Malay girl that Lejla said that she escaped to Emilia?'However, presumably due to processing difficulties, it is a dispreferred option for many speakers (at least in the case of Hungarian, as shown in Gervain 2002Gervain , 2005;; no large-scale study on this topic exists yet on topicalization in Bàsàá).Even for those speakers that allow it, long focus movement is often only possible with certain matrix predicates, such as the bridge verb mondta 'said ', as in (38).For those speakers and those predicates that disallow long focus movement, an alternative construction is evoked.In this alternative construction, embedded foci target the left edge of the embedded clause, accompanied by verb movement inside the embedded clause.See (39) and Figure 10.Note that in Hungarian, a left-peripheral focus may occur in an embedded sentence so long as the D-like associate of the embedded clause also occurs Note that in Hungarian, a left-peripheral focus may occur in an embedded sentence so long as the D-like associate of the embedded clause also occurs in a prominent position, either focused itself, or in a contrastive topic position, but in the latter case, some other element must occur in the matrix focus position.As shown in (40) (a and b), this holds for bridge verbs, like mond 'say', stress-requiring main clause predicates like utál 'hate' and verbs that have prefixes, like the factive verb megbán 'regret'.
(40) a. Péter azt mondta/utálta/bánta meg, hogy Marit választottuk Peter d-acc said/hated/regretted prt that Mary-acc selected be a bizottságba.prt the committee-to 'What Peter said/ hated/ regretted was that we selected MARY to the committee.'b.Azt, Péter mondta/utálta/bánta meg, hogy Marit választottuk d-acc Peter said/hated/regretted prt that Mary-acc selected be a bizottságba.prt the committee-to 'As for the statement that we selected MARY to the committee, it was PETER who said/ hated/regretted it.' We would like to propose that in these cases, StressFocus is satisfied by the selection of a prosodic structure that displays an extra ι boundary at the left-edge of the embedded clause (see Table 4).While this is an extra boundary, whose presence is only justified by the need to satisfy StressFocus, note that it does not violate any of our syntax-prosody or prosody-syntax mapping constraints.This is because our syntax-prosody mapping constraints would only be violated by the absence of an ι-boundary at the edge of the root clause, and because the position of the extra ι boundary satisfies our prosody-syntax mapping constraints (i.e.Align-L/R (ι, HVP)), as it is at the edge of the XP that contains the highest position the embedded verb overtly occupies.Thus, a welcome corollary of our proposal is that although it is possible to have extra ι boundaries to satisfy information-structural needs, the position of such boundaries is regulated by the prosody-syntax mapping constraints.As a result, these extra ι edges are preferably realized at existing syntactic clausal edges.In other words, given the prosody-syntax mapping constraints we propose, it is not an option to leave topics/foci completely in situ and satisfy the information-structural constraints by inserting extra ι boundaries somewhere inside the embedded clause. 19Such boundaries would always be dispreferred compared to boundaries that align with some syntactic clausal edge.
So, if focus movement is short, then the main stress and corresponding L* accent falls on the embedded focus, as indicated by the pitch track in Figure 10.This is consistent with the fact that a sentence like the one in (40a) answers a question such as 'Who was it that Peter said/hated/regretted that we selected to the committee?'.The material involving the main clause and the complementizer associate with a rising pitch contour that is characteristic of contrastive topics in Hungarian (Kálmán 2001: 25).This is a non final contour, which cannot occur independently.This suggests that the phrasing of examples with embedded focus is as in (41).
(41) ( ι azt bánta meg, hogy ( ι Marit választottuk be a bizottságba?))d-acc regretted prt that Mary-acc selected prt the committee-to 'What he regretted was that we selected MARY to the committee?'Such a phrasing satisfies both SpA-ι and Align-L/R (HVP, ι), and since no topic occurs in the matrix clause, Align-L/R (ι, HVP) are also satisfied by all the boundaries.
Interestingly, there is an alternative intonation available to yes-no questions with embedded foci.As the pitch track in Figure 11 shows, in such cases the question contour is in fact repeated.Indeed, this intonation feels somewhat marked and seems to emphasize the question.We take the presence of the two question contours to show that there are two independent ιs.But crucially, the right-edge of the higher ι boundary coincides with the last element of the first clause.We believe that this boundary, as well as the lefthand boundary of the second ι, is there to satisfy SpA-ι.This is consistent with the intended interpretation that the matrix and the embedded clauses constitute two separate speech acts.This comes at the cost of violating Align-R (ι, HVP) and Align-L (ι, HVP), respectively.The other boundaries satisfy all the contraints.So, the phrasing is as given in (42).In fact, the exact syntactic analysis of such an example is unclear.It is possible that the embedded clause is extraposed to a right-peripheral position, indicating afterthought. 20If so, the phrasing does not even violate Align-R (ι, HVP), only Align-L (ι, HVP).
(42) ( ι azt bánta meg,) ( ι hogy Marit választottuk be a bizottságba?)d-acc regretted prt that Mary-acc selected prt the committee-to 'What did he regret?Was it that we selected Mary for the committee?' To sum up, we have shown how the syntax and prosody of Hungarian complex clauses with an embedded focal constituent can be explained by the interaction of StressFocus and our proposed mapping constraints.In doing so, we do not follow proposals according to which foci directly introduce their own prosodic edges (Kanerva 1990;Frascarelli 2000;Kenesei 2009).In our view, at least in languages with a stress system, like Hungarian, foci simply tend to occur at the edges of prosodic domains, here ι, because this is where sentence stress, i.e. the head of ι, is assigned.In addition, we argue against theories that link the position of foci to specific (rigid) syntactic positions (including proposals invok-20 Extraposition to the left is also possible in Hungarian with the same intonation as in Figure 8.Here too, the interpretation seems to be that the extraposed clause clarifies the exact referent of the D-like element, i.e.Interestingly, there is an alternative intonation available to yes-no questions with embedded foci.As the pitch track in Figure 11 shows, in such cases the question contour is in fact repeated.Indeed, this intonation feels somewhat marked and seems to emphasize the question.We take the presence of ing the notion of syntactic phases, e.g.Bošković 2014).The syntactic position of foci is crucially dependent on the position of the verb in our proposal, precluding a rigid syntactic treatment in terms of designated positions (cf.Rizzi 1997).In our proposal, the position of the verb determines the size of the clause for the purposes of the syntax-prosody mapping of ιs, which in turn has repercussions for the position of the focus, as the focus needs to occupy a position where the head of ι can be assigned by the grammar.

Bàsàá embedded topics
Here, we would like to propose that the same type of general interface requirement that applies to foci applies to topics.This requirement is given in ( 43).
(43) Align-Topic Align the left or right edge of a topic with the left or right edge of an ι.
Our primary goal here is to account for the position of topical constituents rather than their prosodic category (i.e.whether they form an ι of their own, as they seem to do in some languages, or whether they can also constitute a prosodic category lower in the prosodic hierarchy.We come back to this issue subsequently).In our view, topics primarily acquire prosodic prominence by being located at clausal edges, and this is so because this is where intonational phrase edges are naturally inserted. 21This will give rise to the following typology: In (44a) and (44b), the topic is external to the core ι.These configurations arise when the topic is structurally higher than the highest position occupied by the verb.In particular, (44a) arises when the topic is left-peripheral, as in zero-coded object dislocation in Bàsàá, or in sentences with left-peripheral topics in Hungarian and Sotho (Zerbian 2006).(44c) arises if topics are right-peripheral, as in Aghem (Polinsky & Hyman 2009) or French Lambrecht (1994).( 44b) and (44d) are configurations where the topic is internal to the core ι.We showed in Hamlaoui & Szendrői (2015) that (44b) arises in German V2 clauses, where topics, even contrastive topics, can fill the leftmost φ within the core ι.The configuration in (44b) is also found in Durban Zulu (Cheng & Downing 2007).The configuration in (44d) is to the best of our knowledge unattested.This is because many languages place main prominence on the final phonological phrase of the core ι.In such languages, this position is actually the natural position for focus and leads topics to satisfy (43) from one of the three other possible positions.In addition, it seems to us that right-peripheral topics are much rarer than left-peripheral ones.So, a combination of left-peripheral main prominence (like Hungarian) and right-peripheral topic (like Aghem) would be necessary to potentially allow for the configuration in (44d), both of which seem independently rare.
21 This preference for edge positions of topics might be related to the serial position effect (Ebbinghaus 1913), that is, the fact that the first and last items in a list are better recalled than items in medial positions.This is generally explained by primacy and recency effects respectively, that is, the fact that items that come first in a list are stored and retrieved from the long term memory and items that come last are present in the short term memory (Murdock 1962).Prosodic prominence acquired by the contiguity with an ι-edge (signifying a non-medial position in the speech stream) could potentially facilitate processing and help with both the storage and retrieval of the file cards (in Reinhart's 1982 view of topics) under which information is saved.This however extends the scope of the present paper considerably and we leave it open for further research.
Again, like in Hungarian, what we generally observe in Bàsàá is that, aside from clauses that constitute their own Speech Act, only root clauses come with their own ι boundaries.Additional ι boundaries can sometimes be found, for instance to accommodate embedded topics in Bàsàá.Even these boundaries will satisfy the prosody-to-syntax constraint that ensures that ι boundaries mark the edges of the highest projection that host overt verbal material (rather than being inserted in a clause-internal position).
2013: 75) "Sister nodes in prosodic structure are instantiations of the same prosodic category."

Figure 1 :
Figure 1: Neutral yes-no question in Hungarian.

Figure 2 :
Figure 2: Non-neutral yes-no question in Hungarian.

Figure 2 :
Figure 2: Non-neutral yes-no question in Hungarian.

Figure 3 :
Figure 3: Neutral complex yes-no question in Hungarian.

Figure 4 :
Figure 4: Non-neutral complex yes-no question in Hungarian.

Figure 3 , 400 Figure 4 :
Figure3, which illustrates the same utterance with reported speech rather 571 than direct quotation).572

Figure 5 :
Figure 5: Complex neutral yes-no question in Hungarian (direct quotation).

Figure 7 :
Figure 7: Hungarian yes-no question involving an adjunct clause.
ι [ TopP [ CP Amíg Mari dolgozik], ([ addig i ) [ TopP Péter j [ vP ( ι el-viszi a while Mary works (d) Peter prt-takes the gyerekeket a múzeumba t i t j ?]]]])) children-acc the museum-to 'While Mary is working, does Peter take the children to the museum?' b. *Péter amíg Mari dolgozik, addig el-viszi a gyerekeket a Peter while Mary works d prt-takes the children-acc the múzeumba?museum-to 'While Mary is working, does Peter take the children to the museum?' 39

Figure 10 :
Figure 10: Yes-no question with an embedded focus in Hungarian.

Figure 10 :
Figure 10: Yes-no question with an embedded focus in Hungarian.

Figure 11 :
Figure 11: Yes-no question with an embedded focus in Hungarian (alternative intonation).

Figure 11 :
Figure 11: Yes-no question with an embedded focus in Hungarian (alternative intonation).

Table 1 :
Complex sentence with in situ complement.

Table 2 :
Complex sentence with left extraposed complement or adjunct.
wrapping the whole sentence, which constitutes a Speech Act.The left edge of this latter boundary violates Align-L(ι, HVP), but the other boundaries satisfy Align-L/R(ι, HVP).16 23

Table 3 :
Complex sentence with in situ complement clause and a sentence-initial topic.FocP Lejla k kérdezte i [ vP meg j [ VP t i t j t k Eleonórától [ CP hogy [ TopP a [ vP elmenekült k -e [ VP t k t l Emíliához?]]]]]]]girl ) [ TopP Lejla [ vP megkérdezte k [ VP t k Eleonórát:[ CP [ TopP "A maláj lány j Did the Malay girl escape to Emilia?"' Finally, Figure6, corresponding to (33), illustrates the fact that if the main clause contains a focal element, then accent reduction only affects the main clause in direct quotation, 33) [ FocP Lejla m kérdezte k [ vP meg l [ VP t k t l t m Eleonórát: 34) ( ι [ TopP Péter j ( ι [ vP el-viszi i Does Peter take the children to the museum, while Mary is working?' (i) [ TopP Péter k [ FocP kétszer j vitte i [ vP [ vP el [ VP t i t j t k a gyerekeket a parkba]] [ CP amíg Mari There were two occasions that Peter took the kids to the park while Mary was working.'

Table 5 :
Complex sentence with topic in the in situ embedded clause.