Flexible syntax–prosody mapping of Intonational Phrases in the context of varying verb height

Abstract

This paper provides new evidence in support of the hypothesis that the syntax–prosody mapping of Intonational Phrases is flexible (Hamlaoui and Szendrői 2015). In the traditional ‘rigid’ approaches, Intonational Phrases are taken to map onto particular syntactic projections. In contrast, in the ‘flexible’ approach, the Intonational Phrase corresponds to the highest projection of the verb (HVP). Accordingly, the ‘flexible’ approach predicts that the HVP should also determine the size of Intonational Phrases in a language where the verb height depends on the utterance type. Our evidence comes from a language of this type, Iron Ossetic (East Iranian). First, we demonstrate that verbs in Iron Ossetic occupy different functional heads in different contexts. Then, based on novel prosodic data, we show that the HVP indeed directly determines the size of Intonational Phrases in clauses with narrow foci and negative indefinites. Additionally, in wh-questions, language-specific mapping constraints come into play.

1. Introduction

The nature of the Intonational Phrase (ι) and its mapping onto syntactic constituents has long been debated. Traditionally, ι is assumed to map onto a clause, but a ‘clause’ in the syntax-prosody literature has been defined (e.g.) as a TP (Zerbian 2006), CP (Truckenbrodt 2005; Henderson 2012), or the complement of Force $^0$ or C $^0$ (Selkirk 2011), to name just a few approaches. The difficulty of identifying the size of ι lies in wide cross-linguistic variation with respect to higher-level mapping of prosodic and syntactic phrases. In a novel type of approach, Hamlaoui and Szendrői (2015, 2017) propose that ι-size is flexible and corresponds to the highest projection that hosts verbal material in a given language, together with its specifier (=HVP, ‘highest verbal projection’). The evidence comes mainly from the prosodic properties of Hungarian narrow focus and Bàsàá (Bantu) zero-coded passives. The advantage of this approach is that it provides a unified, syntax-based account of cross-linguistic variation in ι-size.

A prediction that the flexible ι-mapping hypothesis makes is that the HVP should also determine ι-size in a language where the height of the verb varies with utterance type. We show that, in Iron Ossetic (East Iranian), several projections are available for verb raising, depending on context, which makes it a uniquely suitable testing ground for this prediction. We demonstrate that Iron Ossetic has several discourse projections above the TP that host narrow foci, wh-phrases, and negative indefinites, respectively: [FocP [WP [NegP … ]]]. If these projections are merged, the verb raises to the lowest one with a filled specifier. This analysis correctly derives the fact that, in the surface word order, each of (single) narrow foci, wh-phrases, and negative indefinites must appear immediately preverbally; if cooccurring, they must appear in the order focus ¿ wh-phrase(s) ¿ negative indefinite(s).

Based on prosodic data from an elicitation study, we develop an analysis of Iron Ossetic prosody and show that there are three layers of prosodic constituents above the level of the prosodic word: Phonological Phrase (φ), Intonational Phrase (ι), and Utterance Phrase (υ). φ is the domain of pitch accent assignment and corresponds to smaller constituents that do not include the clausal spine, DPs and PPs. Each φ is assigned a pitch accent, anchored to the stressed syllable in the leftmost prosodic word in the φ; the stressed syllable may be either the initial or the second one, based on vowel quality. The size of ι, we show, is determined by the position of the verb, in accordance with the flexible ι-mapping approach. Within an ι, the realization of a pitch accent on all φs other than the leftmost is suppressed, which serves as the main diagnostic of ι-size.

This paper, therefore, provides further support for the flexible ι-mapping approach, based on a new language type, while also showing that more rigid syntax-prosody mapping approaches cannot account for the same data. At the same time, we show that not all utterance types in Iron Ossetic can be accounted for with the flexible ι-mapping approach alone. While flexible ι-mapping correctly derives the prosodic realization of utterances with narrow foci and negative indefinites, in wh-questions the syntax-prosody mapping constraints are overridden by language-specific alignment constraints that target wh-phrases. ¹

This paper is structured as follows. §2 discusses the approaches to mapping of ι onto syntactic constituents, starting with the ‘rigid’ approaches (§2.1) and proceeding to the flexible ι-mapping hypothesis (§2.2). §3 outlines the relevant aspects of Iron Ossetic grammar: the basic clause structure (§3.1), discourse projections (§3.2), traditional descriptions of Iron Ossetic prosody (§3.3), and recent instrumental findings on stress realization and φ-formation (§3.4). §4 discusses the predictions and aims of the study (§4.1), elicitation materials and methods (§4.2), and the theoretical framework adopted (§4.3). §5 provides a preview of the results and preliminary assumptions (§5.1) and discusses the results of the production study: the contexts accounted for by the flexible ι-mapping hypothesis (§5.2) and those that require additional language-specific assumptions (§5.3). Due to the number of individual contexts investigated, the discussion of the results and an Optimality Theory (OT) analysis for each context are provided in the individual subsections in §§5.1–5.3, followed by the full list of Optimality Theory (OT) constraints used in §5.4. §6 concludes the article.

2. Approaches to ι-mapping

2.1. Rigid ι-mapping approaches

It is an accepted view in the syntax-prosody literature that prosodic constituents are organised into hierarchical units that, on the one hand, systematically reflect syntactic structure and, on the other, are subject to phonological requirements/constraints that are independent from syntax (Selkirk 1978, 1986; Nespor et al. 1982; Nespor and Vogel 1986, among others). Depending on the language, two or three levels of prosodic constituency above the level of a prosodic word are recognised. The smaller one(s) are typically labelled Minor or Major Phrases, or, if there is a single one, Phonological or Prosodic Phrase (φ). The larger ones are Intonational Phrases (ι); additionally, the level of Utterance Phrase (υ) may be recognised (see Shattuck-Hufnagel and Turk (1996) and Selkirk (2011) for an overview). Phonological Phrases are taken to correspond to smaller XPs (Truckenbrodt 1999; Selkirk 2011), or, alternatively, to spell-out domains (Dobashi 2003; Ishihara 2003; Kratzer and Selkirk 2007). There is more variability with respect to the mapping between Intonational Phrases and syntactic constituents: while there is a common understanding that Intonational Phrases correspond to ‘clauses’, different implementations are available, with syntactic, semantic and/or information-structural factors considered primary.

In the earliest syntax–prosody literature, ι was taken to correspond to the syntactic node S, the highest one in the syntactic clause. To account for the prosodic properties of different types of embedded clauses, S was specified as not dominated by a node other than S (Downing 1970; Emonds 1970; Bing 1979; Nespor and Vogel 1986). According to a less syntax-centred view, ι was a semantic/information-structural unit larger than a prosodic word and variable in its extent, not necessarily isomorphic to any syntactic constituent. Accordingly, a single clause could contain one or more ιs (Selkirk 1984). Later, ι was proposed to correspond to the Comma Phrase in syntax, roughly equivalent to a speech act (Selkirk 2005, based on Potts 2005), or more directly to a speech act itself, without addressing its syntactic implementation (Truckenbrodt 2015). In more recent and more syntax-centred work, ι has often been taken as corresponding to CP (Truckenbrodt 2005, 2007; Cheng and Kula 2006; Pak 2008; Henderson 2012), or, less commonly, TP (Zerbian 2006, 2007, based on Northern Sotho, where matrix clauses are analysed as CP-less). In another attempt to account for the prosodic properties of both matrix and embedded clauses, it was suggested that ι corresponds to the complement of C $^0$ in embedded clauses and the complement of Force $^0$ (‘illocutionary clause’; Rizzi 1997) in matrix clauses (Selkirk 2009, 2011). This means that in complex clauses, ι was established as recursive. In a similar vein, it has been argued that ι corresponds to syntactic phases (CP and vP), with the caveat that only non-complement embedded CPs form phases (e.g. non-restrictive relative clauses; Cheng and Downing 2007, 2009).

In addition to the difficulty in establishing the syntactic counterpart of ι, some phonological factors, known as eurhythmic constraints, have been recognised as affecting ι-formation (see Elfner 2018 for an overview). The most obvious is phonological weight: heavy syntactic constituents can form higher-level prosodic constituents even if they are not clausal (e.g. Gussenhoven 2004). Among others, ι-formation can also result from the application of the constraint StrongStart, according to which the leftmost prosodic constituent cannot be lower in the prosodic hierarchy than the following one (Selkirk 2011; Elfner 2011, 2012; Bennett et al. 2016).

Despite definitional discrepancies, the notion of ι has proved useful in linguistic theorizing, both with respect to phonological and morphosyntactic processes: it has been argued to be the domain of low-tone insertion in Slave (Na-Dené; Rice 1987) and morphological alternations in K'ichee’ (Mayan; Henderson 2012), to name a few. This, in turn, means that a cross-linguistically valid approach to determining ι-size is called for.

2.2. The flexible ι-mapping approach

Hamlaoui and Szendrői (2015, 2017) propose that accounting for the cross-linguistic variability in mapping of ι onto syntactic constituents is possible if this mapping is not assumed to target a particular syntactic projection. Instead, they argue that ι corresponds to the highest projection that hosts overt verbal material (‘the verb itself, the inflection, an auxiliary, or a question particle’), together with its specifier (HVP). That is, the size of ι is relative and does not rigidly correspond to any syntactic projection (e.g. CP, TP and/or vP), but is determined by the syntactic height of the verb. The proposal is based on the prosodic properties of the Hungarian narrow focus construction, English wh-questions/German V2 clauses, and Bàsàá zero-coded passives. In each of these languages, ι corresponds to the HVP: FocP, CP, and TP, respectively, as schematised in (1), where the ι-edges are represented by curly braces above the syntactic square brackets. There is no restriction on the kind of material that can occupy the specifier of the HVP; for example, it does not have to have a particular information-structural status.

1. (1)

   

These facts are derived with the help of Align constraints, shown in (2). ² The left and right edges of the HVP are aligned with the left and right edges of ι by Align-R/L(HVP, ι). ³ Additionally, the edges of the full ‘illocutionary’ clause (the speech act) are mapped onto the edges of the larger ι by Align-R/L(SA, ι). ⁴ The corresponding prosody–syntax mapping constraints, which ensure mapping of prosodic constituents onto syntactic ones, are low-ranked, and we omit them for the sake of simplicity.

1. (2)

   

To illustrate, let us consider the prosodic properties of narrow-focus constructions in Hungarian, as compared to those of topics. In Hungarian, narrow (identificational, exhaustive) foci appear immediately preverbally. Syntactically, focus-verb adjacency is derived by movement: the narrowly focused constituent moves to Spec,FocP, and the verb raises to Foc $^0$ , as manifested by the fact that detachable preverbs in focus constructions are left behind (Horvath 1986; Bródy 1995; É. Kiss 1998). Prosodically, the narrowly focused constituent receives sentential stress, which has been analysed as targeting the leftmost constituent of an ι (Szendrői 2001, 2003). This means that, in the presence of a narrowly focused constituent, the ι in Hungarian corresponds to FocP, the projection that also houses the verb, which is in accordance with the flexible ι-mapping hypothesis. This is illustrated in (3):

1. (3)

   

In contrast with foci, the movement of topics to the left-peripheral positions is not accompanied by verb movement, as shown by the lack of preverb detachment. The prediction of the flexible ι-mapping hypothesis, then, is that topics should not be part of the ‘core’ ι. This is borne out by the fact that in utterances with topics but not foci, sentential stress targets the preverb+verb complex (Ladd 1996; László 2001; Szendrői 2001, 2003). ⁵ Accordingly, topics in Hungarian are not part of the ‘core’ ι, as shown in (4).

1. (4)

   

Hamlaoui and Szendrői (2015: 6) take multiple topics, if present, to be part of the ‘maximal’ ι, not separated from each other by ι-boundaries, because ‘there does not seem to be any evidence for the presence of intonational phrase boundaries between the topics’. As shown in §5.1, this does not hold for Iron Ossetic, where left-peripheral topics form individual ιs.

3. Iron Ossetic

Iron Ossetic is an East Iranian language spoken in the Central Caucasus, mainly in the Republic of North Ossetia–Alania in Russia, where it has an official status, and in South Ossetia, a breakaway part of Georgia. In Russia, two closely related varieties of Ossetic are spoken, Iron and Digor. Iron speakers are considerably more numerous than Digor speakers, although no precise numbers are available. According to the 2002 census, there were 515,000 Ossetians in Russia. All Ossetic speakers in North Ossetia also speak Russian. The analysis of clausal syntax we adopt here expands the proposal sketched in Borise and Erschler (2021) and draws upon the description in Erschler (2012, 2020).

3.1. Basic clause structure

The neutral word order in Iron Ossetic is SOV, but in actual discourse the word order is largely determined by information structure. Smaller phrases are mostly head-final. Iron Ossetic is morphologically complex, mostly suffixing, with a rich case system, an inventory of aspectual prefixes, and a sophisticated system of pronominal and adverbial second-position clitics (Erschler 2020).

Following Borise and Erschler (2021), we take the clausal spine to be left-branching up to the level of TP, as shown in (5). The finite verb is assembled via head movement through a series of functional heads (v $^0$ , Asp $^0$ ) and raised to T $^0$ . Aspectual prefixes are merged in Asp $^0$ ; their linearization on the left is achieved by means of a diacritic [+prefix]. ⁶ The subject is generated in Spec,vP and raised to Spec,TP.

1. (5)

   

With respect to head directionality, we take the VP to be head-final, because the neutral constituent order is OV (Erschler 2020: 669). The evidence for the head-finality of vP is supplied by the behaviour of complex verbs. Complex verbs are combinations of a nominal part and a light verb that bears tense and agreement markers (e.g. ba-fɐʃtiat kod-ta ‘pfv-delay do-pst.3sg’), as exemplified in (16) and (17) below. The order of elements in such verbs is rigidly nominal part–light verb (Erschler 2020: 656–657). The literature on complex verbs in a number of languages, including Persian and Hindi-Urdu, agrees that the light verb must include v $^0$ or even be the spell-out of it (e.g. Butt and Ramchand 2005; Folli et al. 2005). The order nominal part–light verb can be derived only if vP is head-final.

We know of no direct evidence that would bear on head directionality in AspP and TP. Iron Ossetic lacks auxiliaries or any other items that can be identified as the spell-out of T $^0$ . On the other hand, the CP is head-initial, because a complementiser, if present, always precedes the verb (Erschler 2020: 679–682). Therefore, at some point there must be a switch from the head-finality of lower projections to the head-initiality of higher ones. Given the typologically robust Final-over-Final Condition (FOFC), which prohibits head-final phrases from immediately dominating head-initial ones within the same extended projection (Sheehan et al. 2017: 1), we assume that this switch occurs only once. For the sake of consistency, we assume that all phrases in the inflectional domain (such as AspP and TP) are head-final, and that the phrases in the discourse domain (i.e. NegP and above) are head-initial. Nothing in our analysis hinges on where exactly in the inflectional domain the switch in head directionality occurs.

3.2. Discourse projections

Ossetic has a well-articulated left periphery, which houses several types of constituents, including topics, narrow foci, wh-phrases, and negative indefinites (Erschler 2012, 2020). The latter three constituent types share the following property: descriptively, each of them must appear in the immediately preverbal position (in the absence of another element with the same requirement). Details of the distribution and co-occurrence requirements of the left-peripheral constituents are provided below.

Negative indefinites in Iron Ossetic must appear immediately preverbally, as shown in (6a)–(6b); if there are several, all surface as a cluster, left-adjacent to the verb, as in (6c). No material can intervene between the negative indefinites and the verb, or between adjacent negative indefinites, as in (6d): abon ‘today’ cannot be inserted in any of the positions where it appears in angled brackets. The exponent of sentential negation is in complementary distribution with negative indefinites in negative sentences: that is, in the presence of a negative indefinite, no exponent of negation is used, but in the absence of a negative indefinite, the exponent of negation is obligatory.

1. (6)

   

In a similar fashion, a wh-phrase in a wh-question must surface immediately preverbally. If there are several wh-phrases, they form a unit that is left-adjacent to the verb, as in (7a). No material can separate the wh-phrases from each other or from the verb, as shown in (7b) and (7c).

1. (7)

   

Finally, narrowly focused constituents also appear immediately preverbally. This applies to constituents modified by ‘only’, as in (8), or, in responses to wh-questions, the constituent corresponding to the wh-phrase in the preceding wh-question, as in (9). ⁷

1. (8)

   

1. (9)

   

If elements that require immediately preverbal placement co-occur, their order is strictly focus ¿ wh-phrase(s) ¿ negative indefinite(s). Topicalised constituents precede the resulting preverbal complex; non-topical material may also follow the verb. This is illustrated for wh-phrase(s) ¿ negative indefinite(s) in (10), focus ¿ negative indefinite(s) in (11), and focus ¿ wh-phrase(s) in (12). ⁸

1. (10)

   

1. (11)

   

1. (12)

   

To account for the order of the preverbal elements and their properties, we propose that the clausal architecture switches from head-final to head-initial in the discourse projections above the TP, as shown in (13). Here, foci, wh-phrases, and negative indefinites are housed in a sequence of dedicated discourse projections. For NegP in Digor Ossetic, this was proposed in Erschler and Volk (2011: 149).

1. (13)

   

If these projections are merged, we propose that the verb raises to the head of the lowest discourse projection with a filled specifier; cf. a somewhat similar treatment of Turkish by Akan and Hartmann (2019). In accordance with the Bare Phrase Structure approach (Chomsky 1994, 1995), we assume that discourse projections that house no overt material are not projected. Examples with syntactic bracketing are provided in (14).

1. (14)

   

That the verb indeed undergoes movement to a discourse projection in these contexts is supported by the positioning of the constituents that the verb raises past; for example, subjects and temporal (i.e. TP-level) adverbs:

1. (15)

   

We assume that NegP and WP have identical structures, with a single head and the possibility for multiple specifiers, if multiple wh-phrases or negative indefinites are present. This assumption is based on the fact that neg-phrases and wh-phrases are subject to identical ordering restrictions: no superiority constraints are attested, but animate arguments must precede inanimate ones:

1. (16)

   

1. (17)

   

Furthermore, it has been shown that the exponent of sentential negation nɐ is a phrase rather than a head (Erschler and Volk 2011). The complementary distribution of the negative marker with negative indefinites, as illustrated in (6), is accounted for if we assume that sentential negation is spelled out in Spec,NegP as a last resort when the specifiers of NegP would otherwise remain empty. If, under the alternative assumption, negative indefinites occupied the specifiers of separate (iterated) negative projections, the complementary distribution between negative indefinites and sentential negation would be much harder to explain. Based on this and the overall parallelism between the distribution and behaviour of negative indefinites and wh-phrases, we conclude that multiple wh-phrases are also merged in multiple specifiers of a single functional head. The fact that no material can intervene between multiple wh-phrases or multiple negative indefinites follows from the multiple specifier analysis.

Finally, evidence for the verb raising to the head of the lowest discourse projection with a filled specifier comes from word order: no adverbs can intervene between a constituent in the specifier of the lowest discourse projection and the verb, as was shown in (6d), (7c), (8b), (8c), (9b) and (9c). If the verb had stayed in the TP after the merger of the discourse projections, we would expect TP-level adverbials to intervene between the verb and the constituents in the discourse projections. This does not take place. ⁹

3.3. Prosody: traditional descriptions

Traditional literature on Iron Ossetic describes the prominent role of prosodic phrasing in the language, closely connected with word stress and the way stress is rendered intonationally. In a lexical word, stress targets the first or second syllable, which together comprise the ‘stress window’. The exact location of stress depends on vowel quality (Bagaev 1965; Isaev 1959; Dzakhova 2010). Iron Ossetic has ‘strong’ (S) and ‘weak’ (W) vowels: /a, e, i, o, u/ and /ɐ, ə/, respectively. Stress targets the initial syllable if the first vowel is strong (ŚS: rálizən ‘to run away’, χábar ‘news’; ŚW: ráʒmɐ ‘forward’, sólpə ‘ladle’), and the second syllable if the first vowel is weak (WẂ: kɐʃtr ‘young’, ʃɐnkk ‘lamb’; WŚ: bɐláʃ ‘tree’, χɐdón ‘shirt’). ¹⁰ Personal names, regardless of vowel quality, are stressed on the second syllable.

In connected speech, stress is described as assigned within a larger prosodic constituent: a so-called ‘prosodic group’, as opposed to a prosodic word. Within a prosodic group, only the stress on the leftmost word is intonationally expressed; other words are described as ‘stressless’ (Abaev 1924, 1939; Bagaev 1965; Isaev 1959; Testen 1997). The nature and intonational expression of what is described as stress in a prosodic group have not been discussed in the grammars, but the important insight that comes from the traditional literature is that the distribution of stresses allows for identifying prosodic groups.

Prosodic grouping and the corresponding assignment of the intonational expression of stress applies to a number of contexts, which may be divided into ‘nominal’ and ‘verbal’ ones. The nominal contexts include combinations of nouns and their modifiers, and nouns and postpositions (DPs and PPs). The verbal contexts include combinations of sentential negation/negative indefinites, wh-phrases or narrowly focused immediately preverbal constituents and verbs, as well as combinations of more than one of the above and verbs (Abaev 1939). The verbal contexts may include second-position clitics and certain particles, which surface between the preverbal constituent and the verb and are also included in the prosodic group. Any other material is described as placed outside the prosodic group.

3.4. Stress and φ-formation

As an OT analysis of stress placement in Iron Ossetic, we adopt the proposal put forward in Borise and Erschler (2022), according to which a prosodic word in Iron Ossetic contains a binary iambic foot, under a moraic (as opposed to syllabic) analysis: each foot corresponds to two morae. This is enforced by Ft-Form=I and Ft-Bin constraints (Prince 1980; René 1989; Prince and Smolensky 1993). Feet are left-aligned in a prosodic word. This is derived via Align-Ft-L and Parse-syll (Hayes 1980; Halle and Vergnaud 1987; McCarthy and Prince 1993; Prince and Smolensky 1993). The constraints are defined in (18), and the tableaux deriving word stress placement in the four stress-window types are provided in (19)–(22). We adopt the following constraint ranking: Align-Ft-L $\gg$ Ft-Bin $\gg$ Parse-syll; the ranking of Ft-Form=I with respect to the other constraints is undetermined. Justification for the ranking is provided in the context of individual tableaux. Note that syllables with strong vowels are taken to be heavy/bimoraic (S), and syllables with weak vowels are taken to be light/monomoraic (W).

1. (18)

   

In ŚS stress windows, the candidates with both strong vowels parsed into a foot, (19b) and (19c), fatally violate Ft-Bin, because the feet in them contain four morae. Candidate (19d), with the initial vowel unfooted, fatally violates Align-Ft-L. The winning candidate, (19a), violates the lower-ranked Parse-syll only. In terms of constraint ranking, (19b) would win over (19a).

1. (19)

   

Similarly, in ŚW stress windows, Ft-Bin is fatally violated by (20b) and (20c), which have trimoraic feet. (20d), with the initial vowel unfooted, fatally violates Align-Ft-L. The winning candidate, (20a), again violates Parse-syll only. Like (19), (20) illustrates the Ft-Bin $\gg$ Parse-syll ranking: under the opposite ranking, (20b) would win over (20a).

1. (20)

   

In WẂ stress windows, Ft-Bin is responsible for excluding candidate (21b), in which the foot contains only one mora, and Align-Ft-L excludes (21d), where the foot is not left-aligned in the prosodic word. Candidate (21c), which is not iambic, fatally violates Ft-Form=I.

1. (21)

   

Finally, in WŚ stress windows, (22d) fatally violates Align-Ft-L, (22b) incurs a fatal violation of Parse-syll, and (22c) of Ft-Form=I. The winner, (22a), violates Ft-Bin but still fares better than its competitors. WŚ stress windows show that Align-Ft-L is ranked above Ft-Bin. If the opposite were the case, (22d) would be the winner instead of (22a).

1. (22)

   

Borise and Erschler (2022) also show, based on a production study, that DPs of all sizes in broad-focus declaratives in Iron Ossetic consistently map onto prosodic constituents, φs, as illustrated in (23). This is ensured by Align-L/R(DP/PP, φ) and Align-L/R(φ, DP/PP) constraints, listed in (24). The signature property of a φ is a single pitch accent, anchored to the stressed syllable in the leftmost prosodic word. This is ensured by Align-L(Hd-PrWd, φ) (based on Prince and Smolensky 1993), provided in (25). Therefore, the distribution of pitch accents allows for tracking the size of φs; these results provide an instrumental validation to the existing descriptions of Iron Ossetic.

1. (23)

   

1. (24)

   

1. (25)

   

4. Current study

4.1. Predictions and aims

The syntactic facts in §3.1 and §3.2 show that if the discourse projections are merged, the verb in Iron Ossetic may be found at different heights in the clause. The prediction of the flexible ι-mapping hypothesis, then, is that the size of ι will vary, depending on verb height. Based on the traditional descriptions of Iron Ossetic prosody, this is indeed the case, with the expression of ‘stress’ marking the left edges of ‘prosodic groups’, in the contexts that we identify as containing the discourse projections. This has not previously been verified instrumentally, which means that the current study was also largely exploratory.

Therefore, the aims of the study were the following: to (a) verify instrumentally the traditional accounts of the formation of ‘verbal’ prosodic groups (i.e. those including verbs and negative indefinites, wh-phrases or narrowly focused constituents), (b) recast them in terms of Autosegmental-Metrical Theory, (c) provide an Optimality Theory account of the syntax–prosody interaction, and (d) test the predictions of the flexible ι-mapping hypothesis.

4.2. Materials and methods

The study targeted the contexts described in the literature as triggering ‘verbal prosodic grouping’, as discussed in §3.3. The elicitation materials consisted of 68 pre-constructed utterances in Iron Ossetic, which fell into the groups in (26). The number of test utterances per condition was dictated by the number of possible components that can affect phrasing: one or two negative indefinites in (26a); one or two wh-phrases of different complexities, with or without negative indefinites in the same wh-question, in (26b); and varying syntactic complexity of narrow foci, either accompanied by negative indefinites or not, in (26c). The stimuli were constructed by the authors and checked with a native speaker who did not participate in the study.

1. (26)

   

The utterances were presented one at a time on a computer screen. Participants were instructed to familiarise themselves with the utterance and pronounce it using natural intonation. The examples intended to elicit focus were preceded by a wh-question (for context). Thirteen speakers of Iron Ossetic took part in the study (8 male, 5 female; age range 20–60; mean age 36.8; median age 35). All participants came from North Ossetia and had a complete or in-progress university degree. The recordings were made in Vladikavkaz, Russia, in January of 2019. The data were recorded with a head-worn Shure SM10A microphone and a Marantz PMD 620 recorder, at a sampling rate of 44.1 kHz and 16 bits per sample, in a quiet room. The recordings were manually annotated in Praat (Boersma and Weenink 2021). Where applicable, quantitative F0 data was collected with ProsodyPro (Xu 2013).

Examples that illustrate individual clause types in §5.1 and §5.2 represent typical productions, as uttered by most or all speakers in our sample. We take them to be representative intonational renditions of each utterance type. Interspeaker variation, where applicable, is mentioned in the context of individual examples.

4.3. Theoretical framework and scope of the results

For the purposes of the prosodic analysis, we adopt Autosegmental-Metrical (AM) theory (Liberman 1975; Bruce 1977; Pierrehumbert 1980). According to the AM theory, the tonal contour consists of a sequence of pitch targets aligned with specific hosts in the prosodic structure, and transitions between them (interpolation). The values of pitch targets are high (H) or low (L), and there are several types of pitch targets: pitch accents, which align with metrically strong syllables (e.g. H*, L*), and boundary tones, which align with edges of prosodic domains (e.g. %H, L%). Complex pitch targets consist of two tones. In a complex pitch accent, the main pitch target, aligned with the stressed syllable, is asterisked, with a leading or trailing tone preceding or following it (e.g. L+H*, L*+H) (for later refinements and critiques of tonal alignment within complex accents, see e.g. Grice 1995; Arvaniti et al. 2000; Atterer and Ladd 2004; Dilley et al. 2005; Barnes et al. 2012). Smaller prosodic units, such as prosodic words, are grouped into larger prosodic units, such as Prosodic Phrases and Intonational Phrases. Pitch accents are assigned within smaller prosodic units, while all types of prosodic units can carry initial and/or final boundary tones.

To the best of our knowledge, no AM analysis of Iron Ossetic has so far been proposed. Borise and Erschler (2022) take the first step towards a systematic account by demonstrating that in neutral broad-focus declaratives, each φ in Iron Ossetic carries a complex pitch accent consisting of two tonal targets, L and H. The L portion is invariably associated with the stressed syllable in the leftmost word of a φ (the first or the second syllable, depending on vowel quality, as discussed above), and the H portion is realised on the post-tonic syllable. The exact alignment of the rise from L to H is shown to be determined by the quality of the stressed vowel: ‘strong’ stressed vowels can carry a low or rising tonal contour, while ‘weak’ ones carry a low tone only. Borise and Erschler (2022) propose that the tonal alignment is determined by the mora count of the stressed vowel, as introduced in the context of stress assignment above: strong stressed vowels correspond to two morae, and weak ones correspond to one. The two morae of strong stressed vowels can accommodate a low plateau or rise in F0, whereas weak stressed vowels can accommodate only a single low tone. Accordingly, Borise and Erschler (2022) label the two rising pitch accents L+H* and L*+H. The intuition behind these labels is that, in L+H*, the starred tone H* is primary, in that it appears on both the stressed and the post-tonic syllable, and in L*+H, L* is primary, because this is the only tone aligned with the stressed syllable. Strong, stressed vowels can carry either accent, but weak vowels can only carry L*+H.

Most pertinently for the current purposes, Borise and Erschler (2022) show that in neutral broad-focus contexts, each φ carries a rising pitch accent, with the F0 peak reached on the post-tonic syllable. We find that the same applies to topicalised φs in our data. In contrast, we find that the pitch accents carried by the leftmost φs in the ‘core’ ιs in our data – such as the ιs in the context of narrow foci, wh-phrases, and neg-words – are monotonal H*s aligned with the stressed syllables themselves. Therefore, we tentatively assume that the distinction between the bitonal rising and monotonal high pitch accents might be rooted in information structure: rising pitch accents seem to mark given/familiar/topical material, while monotonal high pitch accents mark new constituents. Put differently, the constituents outside of the core ι carry bitonal rather than monotonal accents. The one exception to this is the wh-word savɐr ‘which’, which often carries a rising rather than high pitch accent, in contrast with other wh-phrases. This, in fact, fits well with the hypothesis that bitonal pitch accents are correlated with givenness, due to the given or D(iscourse)-linked status of ‘which’ (Pesetsky 1987, 2000). The relevant examples are discussed in §5.2.2 and §5.3.2.

Because it is not the aim of this paper to provide a description of the intonational phonology and the full tonal inventory of Iron Ossetic, we leave other issues pertaining to the pitch accent types for future research. The contrasts between L+H*, L*+H and H* are largely irrelevant for our current purposes and have been introduced to facilitate visual recognition of the pitch accents in the figures. What is important is the presence or absence of an accent on a particular constituent, not the type of accent. Visually, the main difference between L+H* and L*+H is the presence or absence of rise in F0 on the stressed syllable. The difference between L+H* and L*+H, on the one hand, and H* on the other is the location of the F0 peak: it is on the post-tonic syllable in the case of the bitonal accents and on the stressed syllable in the monotonal accent. However, the type of pitch accent and the exact alignment of its subparts are not important for the argument at hand.

5. Results

5.1. Preliminary assumptions and preview of the results

The prosodic phrasing of the constituents occupying the discourse projections in Iron Ossetic is correctly predicted by the flexible ι-mapping hypothesis: the size of ι corresponds to the projection that hosts the verb in a given context. In addition to the ‘core’ ι, Hamlaoui and Szendrői (2015) discuss ‘maximal’ ιs, which encompass full syntactic sentences (see also Selkirk 2011; Ito and Mester 2012, 2013). In the absence of evidence for recursion of prosodic categories in this context in Iron Ossetic, we refrain from adopting the notion of maximal ι and take full sentences to map onto Utterance Phrases (υ), which carry final boundary tones, L%. υs are not discussed further; we take them to be derived by undominated constraints Align-L/R(SA, υ), parallel to (2c) and (2d), and Align-L/R(υ, SA) constraints. Recursive ιs are found only in the contexts of multiple wh-questions and are discussed separately in §5.3.2. A ‘core’ ι corresponds to the HVP, which is derived by Align-L/R(HVP, ι) (defined in (2a) and (2b)) and Align-L/R(ι, HVP) constraints. Of these, Align-L(HVP, ι) plays the most important role.

While φ-formation and marking, described in §3.4, are not the primary focus of this paper, φs play an important role in the current analysis as the domains of pitch-accent assignment. An ι in Iron Ossetic may consist of one or more φs. If there is more than one φ, a pitch accent is realised only within the leftmost φ and suppressed on all others. Therefore, the main diagnostic for ι formation is the lack of pitch accents on non-initial φs. This is derived with the constraint Align-L(Hd-φ, ι), shown in (27), which assigns a violation whenever a φ other than the leftmost one in the ι carries a pitch accent. It also penalises ιs that carry more than one pitch accent, because that amounts to having more than one head φ.

1. (27)

   

One of the main differences between the Iron Ossetic and Hungarian facts, as described in Hamlaoui and Szendrői (2015), is that multiple topics in Iron Ossetic behave as separate prosodic constituents, in that each topic carries its own pitch accent. Accordingly, we propose that each topic in Iron Ossetic forms its own ι, each of which is a sister to the ι formed by the HVP, as schematised in (28). ¹¹ The pitch accents in (28) are represented as X*, as their actual values may vary.

1. (28)

   

The reasoning for this analysis of the prosody of topics in Iron Ossetic is twofold. First, phonetically, the final syllable of a topic receives a degree of final lengthening that is (less than but) comparable to that found on the ι-final constituent at the right edge of the utterance, and greater than the lengthening received by the focused constituent (ι-medial). This can be demonstrated by comparing the durations of final syllables in the same words when (i) topicalised (i.e. at the right edge of the topic ι), (ii) focused (i.e. forming a φ that is not adjacent to an ι-edge), and (iii) utterance-final (i.e. at the right edge of the core ι). In our sample, the words that occur in all three positions include majrɐmbonǝ ‘Friday-loc’, bɐgɐnǝ ‘beer’, and Alan (personal name). The results are provided in Table 1.

Table 1. Average duration of final syllables in different positions; standard deviations are provided in brackets.

Second, from the theoretical standpoint, treating topics as ιs complies with the Strict Layer Hypothesis. Accordingly, we adopt an existing constraint that applies specifically to topics and maps them onto ιs (Frascarelli 2000; Feldhausen 2010), as in (29). ¹² Additional constraints, needed for accounting for more complex contexts, are introduced later in this section, together with the relevant examples. The full list of OT constraints used is provided in §5.4.

1. (29)

   

5.2. ι-formation determined by the HVP

In this section, we show that the size of ι in the contexts that involve one or multiple negative indefinites, a single wh-phrase, or a focused constituent, corresponds to the HVP – i.e. NegP, WP or FocP, respectively – to the exclusion of the topicalised material further to the left.

5.2.1. Negative indefinites

As described in §3.2, negative indefinites in Iron Ossetic are obligatorily left-adjacent to the verb. If there are multiple negative indefinites, they cannot be separated from the verb or from each other by other material. We propose that, syntactically, the presence of negation warrants the merger of NegP above TP, and negative indefinites occupy the specifiers of NegP. Obligatory adjacency of the negative indefinite(s) and the verb follows from the fact that the verb complex – that is, the complex head consisting of V $^0$ , v $^0$ , Asp $^0$ , and T $^0$ – head-moves to Neg $^0$ , as shown in (30):

1. (30)

   

Based on this syntactic configuration, the prediction of the flexible ι-mapping hypothesis is that the left edge of NegP, which contains the verb and negative indefinites, regardless of their number, corresponds to the left edge of ι. This prediction is borne out, as shown in Figure 1 for a single negative indefinite, and in Figure 2 for multiple ones, with the glosses, translations, and prosodic structure provided in (31a) and (31b), respectively. The OT account of the proposed phrasing is provided in (32) below.

1. (31)

   

Figure 1. Realization of the utterance in (31a) (speaker M1, stimulus pt1_1).

Figure 2. Realization of the utterance in (31b) (speaker F2, stimulus pt1_2).

In Figure 1, the negative indefinite nikɐmɐj ‘from no one’ carries a pitch accent. Given that the F0 peak is aligned with the stressed syllable, ni, in a ŚW stress window, we label it H*; this is a typical pitch accent that negative indefinites carry in our data. There are no other pitch accents further to the right, the only other pitch target being the final boundary tone L%. Lack of further pitch accents is a hallmark of ι-formation. The left-peripheral topics abon ‘today’ and Alan carry their own (rising) pitch accents, typical of topics. All participants produced the same intonational realization of this example.

Figure 2 shows that, in a sequence of negative indefinites, only the leftmost one carries a pitch accent. Here, there is an H* on the stressed syllable ni in niʧi ‘no one’, the leftmost negative indefinite, but not on nikɐmɐj ‘from no one’ or the verb. This was the case for all our participants: they consistently contrasted the tonal realization of examples (31a) and (31b).

These prosodic phrasing facts are predicted by the flexible ι-mapping hypothesis, given the syntax of negative indefinites: the negative indefinites, no matter their number, occupy the specifiers of the NegP projection, with the verb raising to Neg $^0$ and thus becoming the HVP, as shown in (30). Only the leftmost negative indefinite carries a pitch accent, which is aligned with the left ι-edge. The constraints that derive the ι-formation are provided in (32), based on the example in (31b). The syntactic constituent corresponding to HVP is contained in square brackets in the input of the tableau. The constraints in (32) are unranked with respect to each other.

Starting from the bottom of the tableau in (32), failure to phrase the topic separately results in a fatal violation of AlignTopic for candidate (32e). Excluding the leftmost negative indefinite from the core ι leads to a fatal violation of Align-L(HVP, ι) for (32d). Candidates (32c) and (32b), in which a head φ (i.e. one that bears the pitch accent) is not aligned with the left ι-edge, are excluded by Align-L(Hd-φ, ι).

1. (32)

   

The OT analysis of an utterance with a single negative indefinite would work in a similar fashion, except that the configurations in candidates (32b)–(32d) would not be relevant (due to there being only one negative indefinite). Constraints AlignTopic and Align-R(HVP, ι) are omitted from subsequent tableaux for the sake of simplicity.

5.2.2. Wh-phrases

Like negative indefinites, wh-phrases in Iron Ossetic appear in the immediately preverbal position, as discussed in §3.2. ¹³ We propose that wh-phrases move to the specifiers of a dedicated projection, WP, which is merged above the TP in wh-questions, and that the verb complex head-moves into W $^0$ , in a parallel manner to the syntax of negative indefinites, as shown in (33). The evidence for that comes from the impossibility of any intervening material (other than negative indefinites) between the wh-phrase and the verb in W $^0$ . ¹⁴

1. (33)

   

The prediction for wh-phrases, then, is the same as for negative indefinites: the left edge of WP, which contains the wh-phrase and the verb, should be aligned with the left edge of ι. This prediction, too, is borne out, as shown in (34) and Figure 3.

1. (34)

   

Figure 3. Realization of the wh-question in (34) (speaker F5, stimulus pt2_25).

In Figure 3, the stressed syllable mɐ in the WẂ stress window in the wh-word kɐmɐ ‘who.all’ is aligned with a peak in F0, which we analyse as the pitch accent H*. There are no further pitch targets to the right, until the final boundary tone L%, which shows that the wh-phrase and the verb are combined into an ι. The topicalized constituents, abon ‘today’ and Madina, carry their own (bitonal) pitch accents and are outside of the core ι. Figure 3 also demonstrates that wh-phrases, in contrast to negative indefinites, are the locus of two high pitch targets: in addition to the stress-aligned pitch accent, they also carry an initial high boundary tone %H. In Figure 3, it is realized as an F0 peak on the unstressed initial syllable kɐ in kɐmɐ ‘who.all’. %H appears only on ιs that include wh-phrases. Anticipating the discussion in §5.3.2, the presence of %H contributes to the special prosodic behaviour of more complex wh-questions – multiple wh-questions and those that also include negative indefinites – which is unexpected from the point of view of the flexible ι-mapping hypothesis.

In (35) and Figure 4, a wh-question with a heavier wh-phrase, savɐr wɐjgɐnɐʤǝ binojnag ‘which seller's spouse’, is shown. Despite the weight, it carries only a single pitch accent, anchored to the wh-word savɐr ‘which’. As mentioned in §4.3, savɐr is unlike other wh-phrases in that it can be realized not only with a monotonal but also with a bitonal pitch accent: in our data, eight speakers realised it with the former, and four (M1, M2, M3, F3) with the latter. ¹⁵ Monotonal H* is realized as an F0 peak on sa, the stressed syllable in the ŚW window in savɐr, while in the bitonal realization, the peak in F0 is reached on the post-tonic syllable, vɐr. In Figure 4, the bitonal realization is provided: vɐr carries the H* part of the pitch accent. The initial syllable, sa, is aligned with %H, which overrides the L part of the pitch accent.

1. (35)

   

Figure 4. Realization of the wh-question in (35) (speaker F3, stimulus pt2_20).

To sum up, the left edge of WP, which hosts the wh-phrase and the verb, corresponds to the left edge of ι, as predicted by the flexible ι-mapping hypothesis. This is shown in the tableau in (36). Here, similarly to the examples with negative indefinites, misalignment of the left ι-boundary and the left edge of the WP, as in (36c), is penalised by Align-L(HVP, ι), and anchoring the pitch accent to any constituent other than the leftmost one in the core ι, as in (36b), is excluded by Align-L(Hd-φ, ι).

1. (36)

   

5.2.3. Preverbal focus

The last constituent type that requires immediately preverbal placement in Iron Ossetic is narrow focus. We propose that, syntactically, the adjacency between the focused constituent and the verb results from movement of the focused phrase into the specifier of FocP, accompanied by movement of the verb to Foc $^0$ , in a similar manner to the derivation of the discourse projections provided in the previous sections. This is shown in (37).

1. (37)

   

The flexible ι-mapping hypothesis makes the same predictions about the prosodic behaviour of preverbal foci as it did for negative indefinites and wh-phrases: the left edge of the discourse projection that attracts the verb (in this case, FocP) should align with the left edge of ι. This prediction is also borne out, as shown in (38) and in Figures 5–7.

1. (38)

   

In Figures 5 and 6, the narrowly focused constituents, lɐgʷən gɐdətɐ ‘bald cats’ and majrɐmbonə ‘on Friday’, respectively, carry a pitch accent, with no pitch accents further to the right. This fits with the definition of ι in Iron Ossetic. The F0 peaks in pitch accents on focused constituents are reached within the stressed syllable: gʷən in the WẂ stress window in lɐgʷən ‘bald’, and maj in the ŚW stress window in majrɐmbonə ‘Friday.loc’. Therefore, we label them H*. The narrowly focused constituent in each of the examples is preceded by topical constituent(s), external to the core ι, each of which carries its own pitch accent.

Figure 5. Realization of (38a) (speaker F5, stimulus pt3_21).

Figure 6. Realization of (38b) (speaker F3, stimulus pt3_27).

There is also an alternative realization of narrow focus, shown in Figure 7. Here, the pitch accent on the focused constituent is shaped like a high plateau instead of a peak. This realization is often accompanied by increased duration of the stressed syllable in the focused constituent (maj in Figure 7). We did not find a consistent contextual difference between the two focus realizations and, provisionally, also label the plateau realization H*. ¹⁶ Among our participants, the peak realization was somewhat preferred by the female speakers, and the plateau type by the male speakers. The focused constituent in (38a) received seven peak realisations (from 3 male and 4 female speakers) and six plateau realisations (from 5 male and 1 female speaker); in (38b), the focused constituent received six peak realisations (from 3 male and 3 female speakers) and seven plateau realisations (from 5 male and 2 female speakers). Most (10/13) speakers (the exceptions being M4, F4 and M7) produced (38a) and (38b) with the same realization of H*.

Figure 7. Realization of (38b) (speaker M1, stimulus pt3_27).

The prosodic phrasing in clauses with narrow foci also adheres to the predictions of the flexible ι-mapping hypothesis, as shown in the tableau in (39). As before, Align-L(HVP, ι) is responsible for the alignment between the left ι-edge and the left edge of FocP, and Align-L(Hd-φ, ι) ensures the realization of the pitch accent on the leftmost constituent in the ι.

1. (39)

   

Next, let us consider those cases where more than one discourse projection is merged. One such combination is FocP and NegP, in those examples where the verb is immediately preceded by a negative indefinite, itself preceded by a narrowly focused constituent: focus ¿ negative indefinite(s) ¿ verb; other word order permutations are not allowed. According to the syntactic analysis in §3.2, these contexts are derived by movement of the verb to the head of the lowest discourse projection with a filled specifier (here, Neg $^0$ ), as shown in (40). Accordingly, the prediction of the flexible ι-mapping hypothesis is that the left edge of ι should be aligned with the left edge of NegP, as the HVP, and the focused constituent should be phrased separately, as it is not part of the HVP.

1. (40)

   

The prediction is borne out, as shown in (41) and Figure 8 for an utterance that contains a narrowly focused constituent and two negative indefinites: ¹⁷

1. (41)

   

Figure 8. Realization of (41) (speaker M6, stimulus pt3_18).

In Figure 8, the first negative indefinite, niʧi ‘no one’, carries an H* pitch accent (F0 peak aligned with the stressed syllable ni in an ŚS stress window), and there are no pitch accents further to its right, neither on the second negative indefinite nor on the verb. This means that the negative indefinites and the verb form an ι, to the exclusion of the narrowly focused constituent. The focused constituent, alanǝl ‘Alan-sup’, is phrased separately, which is manifested by a stress-aligned L+H*, with a rise throughout the stressed and post-tonic syllables (la and nǝl, respectively). Note that the bitonal pitch accent on alanǝl is typical of material external to the core ι and different from the realisation of focus within the core ι in more simple contexts discussed above. The left-peripheral topic carries its own pitch accent. This is the realization that most (10/13) participants produced; the remaining three (speakers F1, F4 and F5) included the focused constituent into the core ι; we leave the factors that might condition this variation for future research.

To recap, the prosodic properties of these more complex contexts also straightforwardly follow from the flexible ι-mapping hypothesis. The OT analysis is provided in (42). Like in the preceding, less complex contexts, Align-L(HVP, ι) penalises the candidates in which the left boundary of the core ι does not correspond to the left edge of the HVP, (42b)–(42d). Similarly, Align-L(Hd-φ, ι) penalises the candidate with the pitch accent realised not on the leftmost constituent of the ι, (42c).

1. (42)

   

5.3. ι-formation determined by language-specific factors

The flexible ι-mapping hypothesis successfully accounts for the behaviour of simple wh-questions (i.e. those with a single wh-phrase and no other discourse projections merged). In contrast, the behaviour of more complex wh-questions – multiple wh-questions and wh-questions that include negative indefinites – cannot be explained by the constraints we have so far introduced. Instead, we propose that the prosodic phrasing in these constructions is rooted in the mapping requirements of wh-phrases of Iron Ossetic that are independent from and override the mapping constraints of the flexible ι-mapping hypothesis.

5.3.1. Wh-questions with negative indefinites

As discussed in §3.2, wh-questions in Iron Ossetic may also include one or more negative indefinites: in such constructions, the word order is strictly wh-phrase ¿ negative indefinite(s) ¿ verb. Syntactically, wh-questions of this shape are parallel to the focus ¿ negative indefinite(s) ¿ verb constructions in (40): the verb raises to Neg $^0$ , the negative indefinite(s) occupy the specifier(s) of NegP and the wh-phrase is in Spec,WP, as illustrated in (43).

1. (43)

   

Accordingly, the flexible ι-mapping hypothesis predicts that such constructions should be prosodified in a similar way to constructions in (40), as schematised in (44):

1. (44)

   

However, the phrasing in (44b) is only marginally attested. Instead, based on the distribution of H*, the ι in these constructions, in the overwhelming majority of our examples, includes not only the negative indefinite but also the wh-phrase, as shown in (45).

1. (45)

   

Figure 9 illustrates the prevailing realization of (45b): here, neither of the negative indefinites carries H*s, which means that they are not at the left edge of ι. Instead, the wh-word kɐmɐn ‘who.dat’ carries the H* pitch accent on the second syllable (as well as %H on the initial syllable), which means that the core ι includes the wh-phrase, both negative indefinites and the verb. Most speakers (10/13) produced this pattern; only speakers M1, F2, and F3 placed kɐmɐn outside of the core ι, as in (44b). Notably, the prevailing pattern is not predicted by the flexible ι-mapping hypothesis.

Figure 9. Realization of (45b) (speaker F5, stimulus pt2_38).

We propose that the prosodic behaviour of wh-phrases, as revealed by the wh-questions with negative indefinites, is due to a mapping constraint that targets wh-phrases and overrides the requirements of the flexible ι-mapping hypothesis. According to this constraint, introduced in (46), the left edge of the specifier of WP must be aligned with the left edge of the core ι (the precise formulation of this constraint, referring to the specifier of WP as opposed to the maximal projection of WP, will be relevant in the discussion of multiple wh-questions in §5.3.2). ¹⁸

1. (46)

   

While the constraint in (46) is language-specific, there is, in fact, robust phonetic evidence for a prosodic boundary aligned with the left edge of the occupant of Spec,WP – i.e. the wh-phrase: the %H boundary tone, introduced in the context of simple wh-questions in §5.2.2. ¹⁹ The realization of polysyllabic wh-phrases demonstrates that this target is distinct from H*, which is aligned with the second or third syllable of a wh-phrase, depending on the location of stress. This is shown in Figure 10, which provides averaged results for the F0 contours that span disyllabic wh-phrases in our data, of WẂ and ŚW stress window types (ŚS and WŚ types were not attested). The WẂ dataset includes wh-words kɐmɐ ‘who’, kɐmɐn ‘to whom’, and sɐmɐn ‘why’ ( $n=91$ , from all speakers), and the ŚW dataset is based on the realization of the wh-word savɐr ‘which’ ( $n=65$ , from all speakers). ²⁰ Figure 10 also includes the F0 values of the third syllable (the initial syllable of the following verb), to illustrate the subsequent drop in F0. To account for the pitch range difference, the results are shown separately for male and female speakers.

Figure 10. Averaged F0 contours on disyllabic wh-phrases preceded by left-peripheral constituents, according to stress window type. On the x-axis, ticks correspond to syllable boundaries: first (0-1), second (1-2), and third (2-3) syllables.

Wh-words of both stress window types present evidence for a high F0 target on the initial syllable. In the ŚW condition, the H*-part of the stress-aligned L+H* is realised on the second, post-tonic syllable, and the high target on the initial syllable is %H, which overrides the L-part of the pitch accent. In the WẂ context, H* is realised on the stressed (second) syllable itself, due to the second syllable being the rightmost one in a φ. The ŚW and WẂ stress windows, therefore, are similar in that in both, the stress-related F0 peak is realized on the second syllable. In both, we also see another, even higher F0 peak on the initial syllable, which is independent of stress. We take it to be %H. %H is present both in topicless wh-questions, in which the wh-phrase is utterance-initial, and in wh-questions that include topical constituents to the left of the wh-phrase. ²¹ %H is unique to wh-question contexts in Iron Ossetic: ŚW and WẂ stress windows in non-wh-contexts do not carry %H.

Another constraint that plays an active role in the prosody of wh-questions, as demonstrated by more complex wh-questions, is Wrap-WP, (47), modelled after a general Wrap-XP constraint (Truckenbrodt 1995, 1999) and a more specific Wrap-CP (Truckenbrodt 2005). The insight behind this is that the whole WP constituent should be contained within the same ι.

1. (47)

   

The last active constraint in the formation of more complex wh-questions is NoRecursion (Truckenbrodt 1999; Ito and Mester 2013), defined in (48):

1. (48)

   

We propose that the left ι-boundary that precedes the wh-phrase, as evidenced by the presence of %H, overrides the formation of the left ι-boundary that results from alignment with HVP. This is achieved by ranking Wrap-WP higher than the syntax–prosody mapping constraint Align-L(HVP, ι). In the tableau in (49), we also show Align-L(Spec,WP, ι) as a high-ranking constraint, together with Wrap-WP; the evidence for this is provided in §5.3.2. Finally, NoRecursion, which penalises recursive ιs, is ranked below Wrap-WP but above Align-L(HVP, ι); the evidence for this is also provided in §5.3.2. The constraints in (46)–(48) do not affect prosodic phrasing in simple wh-questions (i.e. those that involve a single wh-phrase and no other discourse projections) but determine the formation of more complex wh-questions, such as those involving negative indefinites.

The OT derivation of the phrasing in (45b) is provided in (49). Here, the high-ranked Wrap-WP penalises candidate (49d), in which the WP – the wh-phrase and the rest of the clause to the right – do not form an ι. NoRecursion bans candidate (49c), which includes recursive ιs. As before, Align-L(Hd-φ, ι) bans the realisation of the pitch accent on a constituent other than the leftmost one in the core ι in (49b). Although the winning candidate, (49a), incurs a violation of Align-L(HVP, ι), it is not fatal.

1. (49)

   

5.3.2. Multiple wh-questions

The constraints in (46)–(48) also play an important role in the prosodic shape of multiple wh-questions. According to the syntactic analysis proposed here, multiple wh-phrases occupy multiple specifiers of WP, as shown in (50). If prosodic phrasing in wh-questions were governed by the standard syntax-prosody mapping constraints alone, multiple wh-phrases and the verb would form an ι, as was the case for multiple negative indefinites in §5.3.1.

1. (50)

   

Instead, in multiple wh-questions, the left edge of each wh-phrase is aligned with an ι-edge, marked by %H. This is shown in (51) and Figure 11. Figure 11 also demonstrates that each of the wh-words carries its own %H and H* (the visible portion of L+H*; recall that savɐr ‘which’, in contrast with other wh-phrases, often carries a bitonal pitch accent). ²² Furthermore, the wh-phrases that are not immediately preverbal in multiple wh-questions, unlike topics, do not receive final lengthening. Accordingly, we take multiple wh-questions to be prosodified as nested ιs as opposed to sister ιs. This is ensured by ranking Align-L(Spec,WP, ι) and Wrap-WP above other constraints (most importantly, NoRecursion), which means that recursive ιs are found only in the context of multiple wh-questions in Iron Ossetic. The example in (51) also includes a negative indefinite to demonstrate that our proposal successfully accounts for these even more complex cases.

1. (51)

   

Figure 11. Realization of the wh-question in (51) (speaker F3, stimulus pt2_39).

The pattern shown in Figure 11 was produced by most (10/13) participants. However, speakers F2 and M7 excluded both wh-phrases from the core ι and placed H* on nikʷǝ ‘never’; speaker M6 included both wh-phrases and the negative indefinite in the core ι. We do not provide an account of these minority patterns.

The OT analysis of multiple wh-questions is provided in (52). In candidate (52d), failure to align each Spec,WP with a left ι-edge is fatal. In candidate (52c), the right ι-boundary after the first wh-phrase leads to a fatal violation of Wrap-WP. Candidate (52b), which contains three recursive ιs, including one aligned with the left edge of the HVP (NegP), incurs two violations of NoRecursion, the second one being fatal. The winning candidate, (52a), incurs a single violation of NoRecursion, thus winning over (52b). Even though (52a) also violates Align-L(HVP, ι), it fares better than its competitors.

1. (52)

   

To recap, the phrasing facts in complex wh-questions demonstrate that the formation of ι in Iron Ossetic has two sources. In the default scenario, the size of ι is determined by the standard syntax–prosody mapping constraints. In wh-questions, ι-formation is governed by dedicated higher-ranked constraints, which is demonstrated by more complex wh-contexts: those that involve multiple wh-phrases and/or negative indefinites.

5.4. Full list of OT constraints used

For the convenience of the reader, (53) lists all the constraints introduced in this paper and (54) provides the ranking relationships among them that can be established on the basis of our data.

1. (53)

   

1. (54)

   

6. Conclusions

The mapping of ι onto syntactic constituents has long been a matter of debate, with most existing approaches assuming that there is a particular syntactic projection that the ι maps onto. This leads to wide variation in analyses, both between languages and between studies. The flexible ι-mapping hypothesis (Hamlaoui and Szendrői 2015, 2017) is an attempt to provide a unified, cross-linguistically valid analysis of ι-mapping by dispensing with the notion that ι corresponds to a specific syntactic projection and, instead, taking it to map onto the highest projection that hosts the verb/verbal material (HVP). This approach was originally developed for a set of languages that vary with respect to the structural height of the HVP: Hungarian and Bàsàá. To the best of our knowledge, the flexible ι-mapping hypothesis had not been tested on a range of constructions within a single language that vary with respect to verb height.

Iron Ossetic provides a unique testing ground of this sort, because, as we demonstrate, the HVP in this language varies between TP, NegP, WP, and FocP, depending on utterance type. Then, based on instrumental prosodic data, we show that the prediction of the flexible ι-mapping approach that the size of ι co-varies with the height of HVP is borne out in Iron Ossetic. This applies to the prosody of utterances that contain negative indefinites, narrow foci, and single wh-phrases. Given that these elements are housed in specifiers of different syntactic projections and attract the verb to the head of the projection they occupy, more rigid approaches to ι-formation, which equate ι-size to a particular XP, would not be able to account for the Iron Ossetic data. In turn, the Iron Ossetic facts provide support for the flexible ι-mapping approach.

This paper also demonstrated that the constraints governing flexible ι-mapping may be overridden by high-ranking language- and construction-specific constraints. In Iron Ossetic, these are Align-L(Spec,WP, ι) and Wrap-WP, which, together with NoRecursion, ensure the placement of the left ι-boundary at the left edge of each Spec,WP and penalise the insertion of the left ι-boundary at the left edge of the HVP. These constraints apply to the prosody of wh-questions, and their contribution becomes apparent in the more complex ones (multiple wh-questions and wh-questions that also include negative indefinites). The non-HVP-aligned ι-boundary in wh-questions carries a high initial boundary tone %H.

In sum, the current analysis of Iron Ossetic strengthens the case for the flexible ι-mapping approach. Further research will show whether it can be used to provide a unified account of some of the phenomena described in the literature, in which ι is taken to map onto a variety of different syntactic projections (i.e. CP or TP).