2.1 Theoretical approaches

The responses to questions in (4) are prosodically identical. They show that all questions in (4), each of which selects for a focused constituent of a different size (bracketed), can be felicitously answered with an utterance that carries prosodic prominence/sentential stress (indicated with small caps) on the rightmost word, sweater. Sentential stress is realized as an (intonational) nuclear pitch accent (Ladd 1996). This prosodic ambiguity between focus on a constituent and a sub-constituent is known as focus projection. A nuclear pitch accent on the word sweater may be interpreted as signaling broad focus, VP-focus, or narrow focus on the object.

In contrast with objects, narrow focus on the subject does not lead to such focus ambiguity: the subject carries the nuclear pitch accent, which, intuitively, is incompatible with broad focus.² To illustrate, in (5), only (d) can be felicitously answered with an utterance with the nuclear pitch accent on the subject. As in (4), bracketing marks the focused constituent, and the word carrying the nuclear pitch accent is in small caps.

Numerous analyses of focus projection have been developed. In this section, the existing accounts are presented according to the classification in Arregi (2016).³ The existing accounts of focus projection form two camps: the Default Prosody ones and F-projection ones. All accounts subscribe to some variant of (6):

The two camps diverge with respect to the rules that supplement Focus Prominence. According to the Default Prosody analyses, default principles of prominence distribution in a sentence also govern prominence distribution in a focused constituent – and create focus-projection patterns. Focus projection, thus, is epiphenomenal. In particular, the ambiguity between the responses in (4) results from an ambiguity between ‘contrastive stress’ on the object, sweater, and ‘normal (default) stress’, specified by rule for every sentence (Newman 1946). This intuition also became the basis for the Nuclear Stress Rule (NSR; Chomsky & Halle 1968), according to which sentential stress in a broad-focus utterance (in VO languages) is on the rightmost constituent. In the Default Prosody accounts, some version of the NSR supplements the rule of Focus Prominence, as exemplified in (7) with Jackendoff’s (1972) analysis. As formulated, Default Prominence can also account for the cross-linguistic variation in prominence distribution. For instance, in Basque, it is the preverbal constituent that receives sentential stress, not the rightmost one; focused constituents are also preverbal (Hualde, G. Elordieta & A. Elordieta 1994; Arregi 2002, a.o.).

In (4), according to the Default Prosody approaches, focus projection takes place because in each of broad, VP-, and object focus, Default Prominence assigns sentential stress to the rightmost constituent, sweater, which leads to prosodic ambiguity between nested foci of different sizes. Focus projection simply results from the fact that expanding the size of the focused constituent to the left does not change the position of sentential stress (rightmost). In contrast, in (5), Nini, the focused subject, receives sentential stress. No ambiguity arises in responses (5a–c) because they do not comply with the requirements of Default Prominence for broad, VP-, and object focus, in terms of sentential stress placement.

Jackendoff’s (1972) approach was later disputed (e.g., based on its non-inclusion of factors related to givenness and argument structure; Ladd 1980; Gussenhoven 1983a, a.o.), but the key intuition that default prominence distribution drives focus projection was prominent in subsequent literature. The rule of Default Prominence has been derived from word order (Ladd 1980; Culicover & Rochemont 1983, a.o.), argument structure (Schmerling 1976; Gussenhoven 1983a; Zubizarreta 1998; Büring 2006, a.o.), syntactic depth of embedding (Cinque 1993), height in a syntactic domain (Szendrői 2001), or phase-by-phase computation (Bresnan 1971; Legate 2003; Kahnemuyipour 2004; Adger 2007; Kratzer & Selkirk 2007).

In contrast with the Default Prosody accounts, F-projection approaches calculate relative prominence of adjacent syntactic constituents, based on their structural positions. Focus projection is accounted for with the help of dedicated F-projection rules, which arise from the distribution of pitch accents in a sentence. The latter is rooted in information structure as opposed to the rules of default prominence (Selkirk 1984; 1995; Rochemont 1986; von Stechow & Uhmann 1986). This is illustrated in (8–9), based on the most prominent F-projection account, Selkirk (1984; 1995):

According to the F-projection rules, F-marking can spread ‘horizontally’ from the complement of XP to the head X⁰ (Argument Projection), then ‘vertically’ to the whole XP (Head Projection), and so on, thereby creating a succession of nested foci. This accounts for the ambiguity in (4): when the object is accented, F-projection rules derive different focus sizes, given that an accented object can ‘spread’ F-marking up the syntactic tree. The lack of focus projection in (5) is also accounted for: an accented subject does not license F-marking of the verb because the subject is not its syntactic sister. Accordingly, F-marking is not projected any further than the subject itself (cf. also von Stechow & Uhmann 1986; Selkirk 1995). An additional set of rules governs F-marking in embedded contexts and the interaction of F-marking with givenness.

An advantage of F-projection analyses over Default Prosody ones is that they make more detailed predictions about the correlation between the distribution of pitch accents and information structure, which turn out to be correct. On the other hand, later work showed that F-projection rules wrongly limit F-projection to heads and internal arguments, because F-projection from modifiers is possible as well (Büring 2006). Default Prominence was shown to be indispensable for the correct analysis of focus projection, based e.g. on the difficulty to define the relation between F-marking in a phrase and a particular constituent in it (Jacobs 1988; Schwarzschild 1999; Büring 2006).

2.2 Existing instrumental studies

The judgements about whether responses like those in (4) are indeed ambiguous have been subject to empirical scrutiny. The question guiding these investigations, on English and German, has been whether the prosodic ambiguity between nested foci of different sizes, which has been reported in e.g. Jackendoff (1972) and Selkirk (1984), can be demonstrated instrumentally. Several studies on English found no differences between the prosody of object focus and broad focus, as focus projection accounts predict: in perception experiments, listeners did not reliably distinguish replies to questions eliciting broad or VP-focus from object focus (Gussenhoven 1983b; Birch & Clifton Jr. 1995; Welby 2003). On the other hand, Rump and Collier (1996) and Bishop (2010) found that listeners can reliably tell object focus from broad focus in a perception study. Breen et al. (2010) also concluded that object- and broad-focus utterances differ in gradient phonetic cues, such as mean and maximum F0 values, intensity, and duration. In German, Baumann et al. (2006; 2007) found that speakers differentiate object focus from broad focus with variable patterns of upstep, downstep, and duration.

In different work, it has been shown that focus projection does not obtain in some Romance languages. In Neapolitan Italian declaratives, narrowly focused objects carry a different pitch accent (L+H*), as compared to objects in broad-focus utterances (H+L*) (D’Imperio 1997; 2000). Similarly, in European Portuguese, sentence-final objects bear a falling accent with an early peak (H+L*) under broad focus, but a falling accent with a late peak (H*+L) under narrow focus (Frota 2000; 2012).

2.3 Focus projection in a phrase language

With respect to phrasal/sentential prosody, two language types are commonly recognized: those that mark sentential prominence culminatively, via alignment and identity of pitch accents, anchored to metrically strong syllables, and those that mark it demarcatively, via alignment of prominent syntactic constituents with prosodic boundaries (Jun 2005). Following Féry (2017), the former are called “intonation languages with lexical stress” (henceforth “intonation languages”), while the latter are “phrase languages”. To signal information structure, phrase languages add/delete phrase boundaries at constituent edges. The resulting phrases may but do not have to carry pitch accents; the distribution of pitch accents alone does not necessarily reflect information-structural notions (e.g., focused vs. given) (Skopeteas & Féry 2016).

Phrase languages form two groups (Féry 2017: 269). In those with no evidence for word stress, like French or Korean, pitch accents may be distributed in prosodic constituents larger than a prosodic word, without reference to metrically strong syllables (Vaissière 1983; Jun & Fougeron 1995). In phrase languages with (weakly implemented) lexical stress, like Hindi or Georgian, pitch accents are present, but phrasal tones are predominant. The latter language type is the object of study in this paper (more on the prosodic make-up of Georgian in Section 3.2). In addition to phrase-language status, Hindi and Georgian share another prosodic property: nuclear pitch accent/sentential stress has been explicitly described as absent from Hindi (Bansal 1976: 27) and Georgian (Dzidziguri 1954; Alkhazishvili 1959; Zhghenti 1963; 1965). When asked, Georgian speakers who took part in the present study also reported no consistent intuitions about sentential stress. This is unlike English speakers, whose intuitions about nuclear stress placement converge, as demonstrated by broad agreement about the facts in the literature since at least Newman (1946).

These properties of phrase languages – pitch accents not used to mark information structure and lack of sentential stress – make them difficult to discuss from the perspective of focus projection, because, as shown in Section 2.1, it is precisely these notions that are used to model focus projection. While the existing accounts of focus projection do not make explicit cross-linguistic predictions, such predictions can be formulated. In fact, the two types of approaches discussed in Section 2.1 make different predictions about the availability and/or prosodic expression of focus projection in phrase languages. According to the Default Prosody approaches, exemplified in (7) above, the expression of default prominence is not explicitly tied to the notions of pitch accent or nuclear stress: rather, the main idea is that the default principles of prominence distribution govern focus projection. Viewed from the cross-linguistic perspective, this means that the phonetic expression of focus projection may be subject to cross-linguistic variation. Accordingly, Default Prosody approaches give rise to the DF (= Default Prosody for Focus) Hypothesis:

In contrast, the canonical F-marking approach, Selkirk (1984; 1995), explicitly defines F-marking as a function of accenting, as was shown in (8). This stems from Selkirk’s (1984; 1995) analysis of sentential prosody, according to which placement of pitch accents (as opposed to other tonal targets, e.g. boundary tones) is of primary importance both for prosodic structure building and for focus marking. This is because pitch accents, stress-aligned, signal prominence in languages like English. Therefore, it is not surprising, that, in Selkirk’s account, a fundamental link between pitch accent placement and F-marking is also established.⁴ Selkirk’s account was not meant to be English-specific (see, e.g., Kenesei & Vogel 1989 on Hungarian), but it is clear that it can only make predictions for intonation languages – i.e., those that utilize pitch accents in an English-like way. Accordingly, it gives rise to the FP (= Focus Projection) Hypothesis:

What would focus projection in a phrase language look like? The key insight behind Focus Prominence in (6), which all focus projection accounts recognize, is that focused constituents should be prosodically prominent. In intonation languages, prosodic prominence is expressed with pitch accents (Cohen & ’t Hart 1967; Bolinger 1970, a.o.). In contrast, phrase languages, in marking prosodic and information-structural categories, use boundary tones as an alternative to canonical prominence marking (Féry 2013). Therefore, the DF Hypothesis predicts that phrase languages may signal focus projection with prosodic boundaries. Concretely, the same realization of the direct object (with respect to the prosodic boundary that it is aligned with) under narrow focus and in broad focus would be consistent with focus projection. Similarly, different realizations of subjects under narrow focus and in broad focus, with respect to the same prosodic parameters, would also be consistent with focus projection. The FP Hypothesis, in turn, excludes phrase languages from those in which focus projection may be found.

Phrase languages have not been systematically studied from the point of view of focus projection – except Hindi, for which some, albeit indirect, evidence is available.⁵ In Hindi, the existence and placement of word stress and the distribution of pitch accents are debatable, whereas phrase-final boundary tones are prominent. The literature shows that focus is marked by being right-adjacent to a prosodic boundary, which is not expressed in F0 but in other cues, like glottal stop insertion and the direction of cliticization (Patil et al. 2008; Féry 2010; Féry, Pandey & Kentner 2016). Though the literature does not discuss it explicitly, the reported F0 contours that span the constituents of interest from the perspective of focus projection – narrowly focused subjects and objects, and those in broad-focus contexts – exhibit a pattern that is consistent with focus projection, with the help of tonal targets. Specifically, objects in [SOV]_F and SO_FV are intonationally identical: they have rising contours of the same shape and F0 height. In contrast, the rising contours on subjects in S_FOV have significantly higher final F0 values than those on subjects in [SOV]_F (and are accompanied by strong post-focal pitch target suppression). This pattern of focus marking is consistent with focus projection, and with the DF Hypothesis. The availability of such patterns in some phrase languages calls for the investigation of other languages of the same profile, such as Georgian.

3.1 Relevant syntactic properties

Georgian allows for considerable flexibility of word order, including both OV and VO in broad-focus declaratives (Vogt 1971; Skopeteas & Fanselow 2010). Following the majority of authors (Pochkhua 1962; Nash 1995; Harris 2000, a.o.), I take OV to be underlying, and VO to be derived by short, semantically vacuous verb movement; see Skopeteas & Fanselow (2010) and Borise (2019) for evidence. Georgian exhibits split case marking. In the present and imperfective tenses, subjects carry nominative and objects carry dative; in the aorist, “active” (transitive and unergative) subjects carry ergative, and “inactive” (unaccusative) subjects and objects carry nominative. The structural position of the subject co-varies with case-marking: nominative subjects are generated in Spec, VoiceP, ergative ones in Spec, vP, and dative ones in Spec, ApplP; they receive case in situ and do not raise to Spec, TP. The subject, therefore, is never the most deeply embedded constituent; this will be relevant for the discussion in Section 6. Nominative and dative objects are generated within the VP (Legate 2008; Nash 2017; Thivierge 2019). The fact that, in a transitive clause (OV or VO), the object is the most deeply embedded constituent will also become relevant in Section 6. Focused constituents appear immediately preverbally, as shown in (12). This holds for all types of narrow foci: those in replies to wh-questions, contrastive foci in corrective replies, and those modified by only and even.⁶

Syntactic evidence suggests that preverbal narrow foci are found in situ, and adjacency with the verb results from “altruistic” movement of the intervening material (Borise 2019).⁷

3.2 Prosodic properties

The prosody of Georgian is relatively well-studied. The properties of word stress have long been debated, but there is consensus that (i) in di- and trisyllabic words stress is initial, and (ii) in longer words, in addition, secondary stress may occur on the antepenult or penult (Selmer 1935; Robins & Waterson 1952; Alkhazishvili 1959; Tevdoradze 1978). Experimental evidence demonstrates that Georgian has fixed initial stress, cued by greater duration as compared to other syllables/vowels (Vicenik & Jun 2014; Borise & Zientarski 2018; Borise 2019).

Many rules of Georgian phrasal prosody have been established, from the realization of neutral statements to that of questions and narrow focus (Tevdoradze 1978; Bush 1999; Müller 2005; Skopeteas et al. 2009; 2018; Skopeteas & Fanselow 2010; Skopeteas & Féry 2010; 2014; 2016; Asatiani & Skopeteas 2012). Vicenik & Jun (2014) provide a comprehensive Autosegmental-Metrical analysis of Georgian prosody, establishing the levels of prosodic phrasing and inventory of F0 targets. Unless noted otherwise, the rest of this section summarizes their analysis.

Autosegmental-Metrical (AM) theory is one of the main approaches to prosodic analysis, developed in Liberman (1975), Bruce (1977), and Pierrehumbert (1980).⁸ The main tenet of the AM theory is that intonation can be modelled as a sequence of pitch targets, aligned with specific hosts in the abstract prosodic structure, and transitions between them (interpolation). The values of pitch targets can be high (H) or low (L). These labels are not absolute and take into account the speaker’s pitch range and the surrounding pitch targets. There are several types of pitch targets. Pitch accents are anchored to stressed syllables (e.g., H*, L*), and boundary tones are aligned with edges of prosodic domains (e.g., H%, L%). Complex pitch targets consist of several tones (typically two). In a complex pitch accent, the main pitch target, aligned with the stressed syllable, is asterisked, with leading or trailing tones preceding or following it (e.g., L+H*, L*+H). Another pitch target, phrase accent, has been analyzed either as a boundary tone for medium-level prosodic phrases, or as a distinct pitch target found between the last pitch accent and the final boundary tone (Bruce 1977; Pierrehumbert 1980; Ladd 1983; Grice, Ladd & Arvaniti 2000). Smaller prosodic units, such as prosodic words, are grouped into larger prosodic units, such as prosodic phrases and intonational phrases. In a given language, two or three levels of prosodic phrasing are distinguished (Shattuck-Hufnagel & Turk 1996). Pitch accents are assigned within smaller prosodic phrases, while all types of prosodic phrases can carry initial and/or final boundary tones.

According to Vicenik & Jun (2014), the smallest prosodic phrase in Georgian is an Accentual Phrase (AP). By default, an AP spans a single lexical word (plus postposition/discourse particle, if present). In an all-new, broad-focus declarative utterance, each AP, except for the right-most one, carries a rising F0 contour. Vicenik & Jun (2014) analyze it as a low pitch accent L* on the initial (stressed) syllable of the AP and a high final boundary tone on the final syllable, Ha (‘a’ stands for ‘AP’). This intonational contour is illustrated in Figure 1; see also Skopeteas and Féry (2010) and Skopeteas, Féry & Asatiani (2009: 112).⁹

Figure 1

Typical broad-focus intonation in Georgian.

The key property of an AP is the presence of one pitch accent, on the initial syllable, and a final boundary tone. This holds independently of the information-structural status of an AP.¹⁰ An utterance-initial AP may also carry a falling F0 contour, H* La; this is analyzed as semantically vacuous variation by Skopeteas & Féry (2016). Similarly, Vicenik & Jun (2014) take the H* La tonal pattern to indicate that the AP is semantically/syntactically related to the next one, but not to encode a distinct information-structural notion. As shown in Section 3.3 and in the results of the current study, the H* La contour is also used to mark subject focus. Note that H* may take a shape of a peak or high plateau (Vicenik & Jun 2014).

A prosodic constituent larger than an AP is an Intermediate Phrase (ip). Optional ip-formation in Georgian applies to two or three syntactically/semantically related APs, such as a noun and an adjective/demonstrative. Obligatory ip-formation is found in certain focus contexts; more on this in Section 3.3. Each of the APs within an ip carries its own pitch accent. The H* La tonal pattern is used on the non-ip-final APs, to signal the relationship between the ip-internal APs. The ip-final AP carries an L* pitch accent and a H- or L- final boundary tone, as shown in Figure 2. The ip-boundary tone overrides that of the final AP. The largest prosodic constituent, an Intonational Phrase (IP), corresponds to a clause and carries a final low boundary tone L%, as shown in Figure 1 and Figure 2.

Figure 2

A realization of an utterance containing an ip lamazma kalbat’onma ‘a beautiful lady’.

The existing descriptions demonstrate that the distribution and functions of pitch accents in Georgian differ from those in typical intonation languages in several ways. First, each AP (i.e., in most contexts, each lexical word) obligatorily carries a pitch accent. Second, the choice of a pitch accent (e.g., H* or L*) is not determined by information structure: instead, it tracks semantic/syntactic units, or may be semantically vacuous. These characteristics of Georgian make it a phrase language – i.e., one that primarily relies on boundary tones and phrasing as opposed to stress-aligned pitch accents to signal information-structural notions (Skopeteas & Féry 2010; 2016; Féry 2017). Note that the value of the boundary tone often co-varies with the value of a pitch accent (L* Ha, H* La).

With respect to syntax-prosody mapping, Georgian has been subject to conflicting analyses. This is likely because, as a rule, every lexical word forms an AP, and no tonal events consistently mark syntactic constituents larger than a word but smaller than a clause, which obscures the correspondence between prosodic and syntactic structure (other than optional ip-formation). Skopeteas & Féry (2010) propose that broad-focus SOV utterances are phrased as (SOV) or (S)(OV), while SVO ones are phrased as (SV)(O), based on the derived syntactic status of the object. In contrast, Skopeteas & Féry (2016) argue for (SO)(V) as the default. Because a full account of the interaction between the syntactic and prosodic structures in Georgian is still to be established, no such account in focus contexts is offered here.

3.3 Prosodic realization of focus

The prosodic realization of focus in Georgian has received some attention in the literature. There is a consensus that foci may be associated with (rising-)falling F0 contours or may be realized with the default rising contour instead. The reports of what conditions the variation differ, and so do the available phonological accounts.

One of the earliest experimental studies that targeted the realization of focus in Georgian was conducted by Alkhazishvili (1959). He suggested that a Georgian sentence can be divided into “subject” and “predicate” phrases, which correspond to topic and focus/comment, respectively. The “predicate” includes the verb and the immediately preverbal focused constituent, while the “subject” includes any other material in a clause. Crucially, the two phrase types were shown to have different prosodic realizations. Within “subjects”, each word carried a rising contour, while “predicates” carried a (rising-)falling contour. Falling contour notwithstanding, Georgian speakers perceived “predicates” as more prominent than “subjects”.

Skopeteas, Féry, and Asatiani (2009) show that Georgian word order is conditioned by information structure, and narrowly focused constituents either occupy the immediately preverbal position, or may surface postverbally. Skopeteas and Féry (2010) investigate the prosody of S_FVO, SO_FV and SVO_F, and show that, in all contexts, the narrowly focused constituent has significantly greater duration than the corresponding constituent in a string-identical broad-focus utterance. Though not tested for significance, there is a trend in their data that resembles the one discovered in the current study: on average, a narrowly focused subject is 53 ms longer than a subject in broad-focus conditions (335 ms vs. 282 ms), while this difference is smaller for narrowly focused preverbal objects (35 ms; 314 ms vs. 279 ms). With respect to F0, objects in SO_FV contexts were found to have the same rising F0 contour as objects in broad focus; this was also replicated in the current study.¹¹ In contrast, subjects in both SOV and S_FVO conditions were found to have both rising and falling contours, with no obvious conditioning factor. These results were not corroborated in the current study, where the falling contour on narrowly focused subjects contrasted with the rising one on subjects in broad focus. Vicenik and Jun’s (2014) findings are somewhat different. In their data, most focused constituents carried a (rising-)falling contour, analyzed as an (L)H* pitch accent and an La boundary tone. Some focused objects, in addition, were realized with the usual declarative contour, L* Ha, which is in accord with Skopeteas and Féry (2010) and the current findings.

In a phonological analysis, Skopeteas & Féry (2010) conclude that preverbal narrowly focused subjects are separated from the following material, (S_F)(VO), based on the frequent absence of a high boundary tone between V and O (but not the realization of S_F itself). Preverbal narrowly focused objects are taken to be prosodically grouped with the verb, (S)(O_FV), based on the low height of the boundary between S and O. In contrast, Asatiani & Skopeteas (2012) conclude that all preverbal focused constituents are prosodically grouped with the verb, and separated by a prosodic boundary from the preceding material, X(Y_FV), based on the presence of a prosodic boundary to the left of the focused constituent. The same conclusion is reached by Vicenik and Jun (2014). Finally, Skopeteas and Féry (2016) reanalyze some of their earlier findings: they conclude that all foci are separated from the rest of the clause when clause-initial and in the presence of postverbal material, (S_F)(VO) and (O_F)(VS), and integrated with the preceding material if clause-medial and in the absence of postverbal material, (SO_F)(V) and (OS_F)(V). The conflicting phonological analyses of phrasing are mainly driven by theoretical considerations, and stem from selecting different acoustic cues as decisive for marking prosodic boundaries. The current study does not contribute to this debate, and instead adopts Vicenik and Jun’s (2014) approach to prosodic phrasing (see Section 3.2).

One of the reasons for the reported variability in the data (which gave rise to incompatible phonological accounts) is likely to be methodological: the existing studies are based on ‘lab speech’ instead of (semi-)spontaneous production. Skopeteas, Féry, and Asatiani (2009: 110) used read speech, and instructed participants “to put emphasis on the information under question”. Skopeteas and Féry (2010; 2014; 2016) asked the participants to memorize the responses to the experimental questions. This is problematic, because the prosodic characteristics of e.g. read speech differ considerably from those of spontaneously produced speech (Lieberman et al. 1985; Howell & Kadi-Hanifi 1991; Nakamura, Iwano & Furui 2008). The conclusions drawn from lab-produced speech may not be applicable to (semi-)spontaneous speech, the mode of language production that is, arguably, most used by speakers. Methodological differences are likely to contribute significantly to the differences between the current results and those reported in previous studies. These two issues – the reported variability in the data and the need for more naturally produced data – provided further motivation for the current study.

3.4 Predictions for Georgian

According to the DF Hypothesis, a focus-projection pattern in a phrase language like Georgian would rely on a language-specific distribution of prosodic prominence – more specifically, a boundary tone-based alternative to prosodic prominence (Féry 2013). In Georgian, this is the distribution of AP-final boundary tones. A focus-projection pattern would be exhibited by a narrowly focused object and an object in broad focus that are identical with respect to the tonal contour and the type and height of the boundary tone. Given that there is a consensus that, in broad focus, the object carries the default L* Ha contour, the DF Hypothesis predicts that L* Ha would appear on narrowly focused objects too, with the same parameters of Ha. On the other hand, if a narrowly focused subject and a subject in broad focus carry different tonal contours and boundary tones, this would also be consistent with focus projection. Subjects in broad focus are predicted to carry the default L* Ha, but the realization of narrow focus on subjects is hard to predict: based on the literature, they may carry (L)H* La or L* Ha. In turn, according to the FP Hypothesis, languages that do not utilize pitch accents to mark prominence should not exhibit focus-projection patterns. Based on what we know about the prosody of Georgian, the FP Hypothesis predicts that F-projection approaches would have nothing to say about Georgian, since whatever patterns can be found there would fall outside the purview of the F-marking approaches.

The F0-based prosodic parameters investigated here, therefore, are tonal targets (boundary tones and combinations of boundary tones with pitch accents) on Georgian objects and subjects, when narrowly focused and in broad focus. Specifically, the identity of these tonal targets (as part of the global F0 contour) and the relative height of boundary tones are investigated. The choice of these parameters is motivated by the phrase-language status of Georgian. The evidence from typologically similar Hindi, introduced in Section 2.3, confirms that these F0 parameters can create focus-projection patterns.

The other prosodic parameter investigated here is the duration of the stressed syllable. An increase in duration of the stressed syllable as a cue for narrow focus has been documented for English (Xu & Xu 2005), Dutch (Cambier-Langeveld & Turk 1999), Swedish (Heldner & Strangert 2001), German (Baumann et al. 2007; Kügler & Genzel 2009), and Arabic (De Jong & Zawaydeh 2002), among others. The results from Hindi, a phrase language similar to Georgian, are incomplete: duration of stressed syllables/vowels is greater in contrastively focused words than in words in broad focus (Genzel & Kügler 2010; Puri 2013), but no information is available for non-contrastively focused constituents. In Georgian, a phrase language with non-phonemic stress, stressed syllables are, nevertheless, reliably cued by duration (Borise 2019). This gives rise to the following question: does duration of stressed syllables participate in creating focus-projection patterns in Georgian (i.e., differentiate narrowly focused subjects, but not objects, from their counterparts in broad-focus contexts)? But before this question can be answered, a more general question needs to be resolved: does duration of stressed syllables increase under (contrastive and non-contrastive) narrow focus in Georgian?

Whether the duration of the stressed syllable exhibits systematic focus-projection patterns has not been tested. But, under the Default Prosody approach, it is reasonable to hypothesize that, in the absence of prominence-marking by pitch accents, and with boundary tones only serving as an alternative to prominence-marking, other language-specific cues may participate in marking prominence. Consequently, they also may participate in focus projection. Given that duration cues stress in Georgian, it is a good candidate for also marking phrasal prominence. A pattern consistent with focus projection would then be one where the duration of the stressed syllable does not differ between narrowly focused objects and those in broad focus, whereas the stressed syllable of a narrowly focused subject has greater duration than that in a subject in broad focus.¹²

Duration of the stressed syllable (as opposed to e.g., intensity, vowel quality, etc.) was selected for investigation for several reasons. First, it has been shown for Georgian that duration marks prosodic prominence (stress). Second, duration is a consistent and robust cue, commonly investigated when parameters of individual syllables (as opposed to full words) are compared. Finally, targeting the duration of the stressed syllable will allow for evaluating Georgian in the context of languages above in which the duration (but not other parameters) of the stressed syllable in different focus contexts has been investigated.

The acoustic parameters, investigated on nouns in OV and SV contexts, are schematized in Table 2. ‘X* Xa’ in Table 2 refers to the combination of a pitch accent and boundary tone. The narrow focus condition, in the discussion of the results, is further subdivided into contrastive (e.g., S_CF) and non-contrastive (e.g., S_F) focus.

Table 2

The set-up of the acoustic parameters tested.

5.1 F0 properties

5.1.1 Global F0 contour

In the subject dataset, broad-focus utterances (SV, XSV, and SVO) carried the typical ‘default’ F0 contour. Each lexical word formed an AP: utterance-initial and -medial APs received an L* Ha or, rarely, (L)H* La realization, and final ones an L* L% realization. In line with the literature (Skopeteas & Féry 2016), the L* Ha/ (L)H* La variation utterance-initially seems semantically vacuous. All subjects in broad focus carried the L* Ha contour (n = 83; boldfaced in Table 6). This is illustrated in (18) and Figure 4.

Table 6

Tonal realizations of subjects in broad-focus utterances vs. utterances with subject focus (non-contrastive and contrastive).

Tonal events	S, broad focus	SF	SCF
L* Ha	83	8	0
H* La	0	58	2
LH* La	0	7	1

Figure 4

Tonal realization of a broad-focus utterance (subject dataset).

In contrast, most narrowly focused subjects (total n = 68/76; boldfaced in Table 6) received the (L)H* La realization. This is illustrated in (18) and Figure 5. Note that the verb carries a shallow falling contour, analyzed as H* (H+)L L%, containing a phrase accent (H+)L, anchored to (antepenult+)penult, in accordance with Vicenik & Jun (2014) and Borise (2017). The H*/LH* variation was found in both contrastive and non-contrastive subject foci and did not seem meaningful. The summary of the tonal realizations of subjects is provided in Table 6.

Figure 5

Tonal realization of a subject-focus utterance (subject dataset).

1. (18)

1. (‘What happened last night?’/‘Who had dinner last night?’)

1.  

1. Gasul
2. last
1. ɣame-s
2. night-dat
1. Mariami/Mariami_F
2. Mariami.nom
1. sadil-ob-d-a.
2. dine-sf-sm-ipfv.3sg
1. ‘Mariami/Mariami_F had dinner last night.’

In the object dataset, the realization of broad-focus utterances (SXOV, SOV, XSOV, and OV) adhered to the same generalizations as that described for broad-focus sentences in the subject dataset. It is exemplified in (19) and Figure 6. In contrast with subject foci, however, the majority of object foci (18/24; boldfaced in Table 7) received the same realization as objects in broad focus, L* Ha; this is shown in (19) and Figure 7. There was more variability in the object dataset, though, with 6 focused objects and, surprisingly, 4 objects in broad-focus contexts carrying the contour typical of subject foci, H* La. It should be noted that 3/6 of these contours in the narrow-focus condition and 2/4 in the broad-focus condition come from the same speaker, F4. The summary of the tonal realizations of objects is provided in Table 7.

Figure 6

Tonal realization of a broad-focus utterance (object dataset).

Table 7

Tonal realizations of objects in broad focus utterances vs. utterances with object focus.

Tonal events	O, broad	OF
L* Ha	21	18
H* La	4	6
LH* La	0	0

Figure 7

Tonal realization of an object-focus utterance (object dataset).

1. (19)

1. (‘What did grandma clean yesterday morning?’/‘What happened yesterday morning?’)

1.  

1. Bebia
2. grandma.nom
1. guʃin
2. yesterday
1. dilas
2. morning
1. samzareulo-s_F/samzareulo-s
2. kitchen-dat
1. a-lag-eb-d-a.
2. ver-clean-sf-sm-ipfv.3sg
1. ‘Grandma cleaned the kitchen_F/kitchen yesterday morning.’

5.1.2 Final boundary tone (Ha/La)

A summary of the F0 realizations of subject-final syllables, S_–1, in the subject dataset is provided in Table 8 and Figure 8. The F0 values are converted into semitones (st), given that the participants included males and females, and calculated with reference to the mean F0 of each speaker. In addition to the main object of comparison – the values for broad-focus subjects and narrowly focused ones (contrastive and non-contrastive) – they also include the values for subjects in other contexts for which the data were available (contrastive and non-contrastive object focus, and VP-focus). Here, the mean values of S_–1 are considerably lower in contrastively and non-contrastively focused subjects than in subjects in broad focus (and other contexts).

Table 8

Summary of the F0 values of the subject-final syllable (S_–1) in the subject dataset, by focus type.

Focus type	n	Mean F0 (S–1), st	SD (S–1), st
SF	95	0.449	2.735
SCF	34	–0.74	2.535
broad	83	2.404	2.338
VPF	73	2.477	3.151
OF	14	2.089	2.508
OCF	7	4.136	1.043

Figure 8

F0 values of subject-final syllables (S_–1) in the subject dataset, by focus type.

Figure 9 shows averaged F0 contours over SV strings in the subject dataset. It also demonstrates that the boundary tone on S_–1 is considerably lower on contrastively and non-contrastively focused subjects than on subjects in broad focus (and other contexts). The qualitative difference in F0 shapes that span narrowly focused subjects and those in broad-focus utterances is also apparent: the former carry a falling F0 contour, and the latter have a rising one. To allow for better visualization, the continuation of the F0 contour on the following verb is also included.

Figure 9

Averaged F0 contours in SV strings in the subject dataset, smoothed at 0.2. On the x-axis, ticks mark syllable onsets.

The quantitative analysis of the S_–1 F0 values was performed using a linear mixed-effects model, with the lmer function in the lme4 package (Bates et al. 2015) for R (R Core Team 2020); the lmerTest package (Kuznetsova, Brockhoff & Christensen 2017), which calculates Satterthwaite’s approximation to degrees of freedom, was used to calculate p-values. In each of the subject and object datasets, a model with the dependent variable Mean F0, fixed factor Focus Type, and random factors Speaker, Item, and Clause Subtype (such as, e.g. SVO, SV, XSV for subjects) was run (with a random intercept for each predictor but no random slopes). The broad-focus condition acted as the intercept. The results in Table 9 show that the S_–1 F0 values in subject focus contexts, both non-contrastive and contrastive, are significantly lower than those of preverbal subjects in broad-focus utterances. None of the other contexts significantly differ from the intercept (broad focus).

Table 9

Statistical results for F0 (st) of subject-final syllables (S_–1). Asterisks mark levels of significance: p < 0.001(***).

Focus type	p-value	β	SE	t
SF	<0.001***	–2.052	0.368	–5.576
SCF	<0.001***	–2.996	0.479	–6.252
broad (intercept)	<0.001***	2.489	0.39	6.376
VPF	0.76	–0.121	0.396	–0.306
OF	0.695	–0.286	0.73	–0.39
OCF	0.281	1.127	1.043	1.081

Turning to the object dataset, Table 10 and Figure 10 summarize the facts of F0 realization of object-final syllables, O_–1. Note that contrastively focused objects were attested exclusively postverbally and are not considered here. Similarly, subject focus is not included, since all focused subjects in the dataset were preverbal – that is, no subject-focus OV strings (e.g., OVS_F) were attested.

Table 10

Summary of the F0 values of the object-final syllable (O_–1) in the object dataset, by focus type.

Focus type	n	Mean (O–1), st	SE (O–1), st
OF	20	0.77	2.501
broad	20	1.985	2.029
VPF	30	1.58	2.149

Figure 10

F0 values of object-final syllables (O_–1) in the object dataset, by focus type.

Figure 11 shows average F0 values over OV strings in the object dataset. Here, all three conditions considered closely align in their prosodic realization, including the height of the final rise on O_–1.

Figure 11

Averaged F0 contours in OV strings in the object dataset, smoothed at 0.2. On the x-axis, ticks mark syllable onsets.

The same set-up of a linear mixed-effects model as described for the subject dataset was used on the object dataset. Additionally, a model with VP-focus as the intercept was run, given that focus projection accounts also make predictions for VP-focus. The results in Table 11 show that neither model detected a significant difference between the F0 values on O_–1 in the different focus conditions.

Table 11

Statistical results for F0 (st) of object-final syllables (O_–1). Asterisks mark levels of significance: p < 0.05 (*).

Focus type	p-value	β	SE	t
Intercept: broad
OF	0.219	–1.243	0.979	–1.27
broad	0.049*	1.535	0.722	2.127
VPF	0.076	–1.747	0.933	–1.873
Intercept: VPF
OF	0.62	0.503	1.002	0.503
broad	0.076	1.747	0.932	1.873
VPF	0.777	–0.211	0.738	–0.287

To recap, narrowly focused subjects, contrastive and non-contrastive, are realized markedly differently from subjects in broad-focus contexts. They carry a falling F0 contour, as opposed to a rising contour typical of constituents in broad-focus utterances. In contrast, narrowly focused objects and those in broad and VP-focus contexts carry the same default rising contour.

5.2 Duration of the stressed (initial) syllable

Table 12 and Figure 12 summarize the durational properties of initial syllables in subjects in the subject dataset. The results show that the stressed syllable receives extra duration in narrowly focused subjects (contrastive and non-contrastive), as compared to subjects in broad focus.

Figure 12

Duration of the stressed (initial) syllable in subjects in the subject dataset, by focus type.

Statistical analysis of the duration data in the subject dataset was performed with a linear mixed-effects model, using the lmer function in the lme4 package (Bates et al. 2015) and the lmerTest package (Kuznetsova et al. 2017) for R (R Core Team 2020). Duration of the initial syllable acted as the dependent variable, Focus Type as the fixed factor, and Speaker, Item, Clause Subtype and Syllable Count as random factors. Syllable Count was included to account for the fact that nouns in the subject and object datasets were of variable syllable counts, and the duration of individual syllables decreases as the syllable count increases (polysyllabic shortening; Lehiste 1972). In the subject dataset, the broad-focus condition acted as the intercept. The results in Table 13 show that the duration of the stressed syllable was significantly greater in narrowly (non-contrastively and contrastively) focused subjects than in subjects in broad-focus utterances. No significant difference between subjects in broad focus and in other contexts was detected.

Turning to objects, Table 14 and Figure 13 provide a summary of the durational values of the stressed syllables in objects in the object dataset. The results show that the initial syllable in narrowly focused objects does not receive extra lengthening, as compared to objects in broad or VP-focus.

Figure 13

Duration of the stressed (initial) syllable in objects; one outlier, with the value > 700 ms (VP-focus) is not shown.

The same linear mixed-effects model as described for the subject dataset was used on the object dataset. A model with VP-focus as the intercept was run as well. The results in Table 15 show that neither model detected a significant difference between the durations of stressed syllables in the three conditions.

To recap, in narrowly focused subjects, greater duration of the stressed syllable was found, as compared to broad-focus contexts. This was not the case with narrowly focused objects: here, the duration of the initial syllable did not exceed its counterpart in broad or VP-focus.

5.3 Summary

The analysis of the acoustic parameters of narrowly focused subjects and objects and those of subjects and objects in broad focus yielded the following results. Most focused subjects received a falling F0 contour, identified as (L)H* La, while all subjects in broad focus carried a rising contour, L* Ha. In contrast, most objects, narrowly focused and found in broad and VP-focus contexts, received the rising F0 contour, L* Ha. These findings were corroborated by a quantitative analysis of the boundary tone height. Narrowly focused subjects (contrastive and non-contrastive) had significantly lower boundary tone values than subjects in broad focus. The F0 values were not different for objects in any of the contexts. The duration of the stressed syllable in subjects was found to be greater under narrow focus, as compared to the broad-focus condition. This was not the case for objects, where the duration of the stressed syllable did not vary significantly across focus conditions. The results are schematized in Table 16.

Table 16

Summary of the results.