The syntax-prosody interface: current theoretical approaches and outstanding questions

2.1 Architecture of the grammar

Much work on the syntax-prosody interface deals with questions of constituency: how are words in an utterance related to one another hierarchically? In syntax, constituency is defined on the basis of syntactic constituency tests, whose goal is to manipulate syntactic constituents defined by syntactic structure. These tests provide evidence that sentences are hierarchically structured: even though words are pronounced in a linear fashion, they are related to one another in terms of membership in syntactic phrases. At the syntax-prosody interface, prosodic constituency is diagnosed in terms of prosodic boundaries and prosodic domains, which are demarcated by such phenomena as the distribution of pitch accents and boundary tones, the domains of application of segmental (“sandhi”) processes, and phonetic cues such as pauses, duration, and the scaling of tonal accents (Pierrehumbert 1980; Nespor and Vogel 1986; Selkirk 1980, Selkirk 1986; for a review, see Ladd 2008 [1996]). A central research question in work at the syntax-prosody interface concerns whether—and to what extent—prosodic constituents are isomorphic with syntactic constituents.

All linguistic interfaces are ultimately concerned with questions relating to the architecture of the grammar. At the syntax-prosody interface, in addition to the question of how syntactic structure and prosodic constituents correspond to one another, the direction of interaction is also important to consider. Within generative (and more specifically, minimalist) theories of syntactic structure, the “Y-model” of the grammar (Figure 1) is commonly assumed to operate uni-directionally, in which syntactic structure influences prosodic structure, and not vice-versa (Chomsky 1995).

Figure 1:

The Y-model of the grammar.

Specifically, this approach predicts that phonological well-formedness constraints on the prosodic organization of utterances, such as eurhythmic constraints, will not (and cannot) affect how syntactic constituent structure is derived, and therefore will have no effect on the interpretation of meaning. Alternatives to this approach include models in which syntactic structure has access to phonological information, and thus may influence syntactic structure formation, such as whether or not a movement operation takes place (Richards 2010), as well as models in which mismatches between the prosodic and syntactic structure of sentences are attributed to variation in the underlying syntactic structure, derived from the application of optional syntactic processes such as extraposition (Wagner 2005, Wagner 2010, Wagner 2015).

2.2 Prosodic structure theory

Much theoretical work on prosody assumes prosodic structure theory (Selkirk 1981 [1978]; Beckman and Pierrehumbert 1986; Nespor and Vogel 1986; Pierrehumbert and Beckman 1988), which posits that the prosodic organization of sentences can be represented using a hierarchically ordered structure that is distinct from syntactic structure. Prosodic structure is part of the phonological grammar, and mediates between syntactic constituent structure and the phonetic output of the speech stream; phonological and phonetic phenomena thus target constituents and domains relating to prosodic structure rather than syntactic structure. The domains present in prosodic structure are formalized via the universal prosodic hierarchy, a system of hierarchically ordered domains of increasing sizes (Selkirk 1981 [1978]).

Prosodic structure theory is part of a rich tradition used to describe the prosodic properties of languages from a typological perspective, under the assumption that all languages use the prosodic hierarchy as a universal system of prosodic organization. The prosodic hierarchy illustrated above in Figure 2 represents a minimalist version of the hypothesis, as argued for in recent work (Selkirk 2011; Elfner 2012, Elfner 2015; Ito and Mester 2013); however, research on specific languages has included proposals that additional prosodic categories be included as part of the universal hierarchy. For example, languages with lexical pitch accent systems like Japanese and Basque arguably require the inclusion of additional intermediate categories such as the Minor/Accentual Phrase and the Major Phrase, which together take the place of the Phonological Phrase (φ) in the hierarchy in Figure 2 (McCawley 1968; Pierrehumbert and Beckman 1988; Selkirk and Tateishi 1988, Selkirk and Tateishi 1991; Jun and Elordieta 1997).^[1] Conversely, it has also been proposed that the prosodic systems of some languages require reference to fewer domains than are represented in the hierarchy in Figure 2. For example, the Inuit languages have been claimed not to require reference to the phonological phrase level, showing evidence only of word-level and intonational phrase-level domains (Arnhold 2014; Arnhold et al. to appear). For further discussion of prosodic typology, see the papers collected in Jun (2005) and Jun (2014), and references cited therein.

Figure 2:

The prosodic hierarchy.

2.3 Direct reference and phase-based approaches

Prosodic structure theory asserts that the representation of prosodic constituent structure is distinct from syntactic constituent structure; it is derived (at least in part) through reference to syntactic domains, but subject to its own set of prosodic well-formedness constraints. This type of approach is often referred to in the literature as an indirect referenceapproach to the syntax-prosody interface because domain-sensitive phonological processes refer to syntactic structure only as mediated by prosodic structure (Selkirk 1981 [1978], Selkirk 1986; Nespor and Vogel 1986).

In opposition to indirect reference approaches are direct reference approaches. Instead of assuming that domain-sensitive phonological and phonetic processes reference prosodic structure, direct reference approaches assume that these domains can be derived directly from syntactic structure and the mechanisms of syntactic spell-out. These approaches assert that the domains created by prosodic structure are actually just syntactic domains, thus rendering prosodic structure superfluous to the grammatical derivation. Domain-sensitive phonological processes thus directly reference syntactic structure. Early examples of work proposing direct reference approaches to the syntax-phonology interface include Cooper and Paccia-Cooper (1980), Gee and Grosjean (1983), and Kaisse (1985).

Direct reference approaches initially received critical attention because of the seemingly many clear counter examples where prosodic domains do not match syntactic domains (Selkirk 1981 [1978], Selkirk 1986; Nespor and Vogel 1986). However, relatively recently, new proposals assuming direct reference have emerged, primarily due to the emergence of theories of cyclic, phase-based, and multiple spell-out, independently proposed to be an integral component of syntactic computation (Uriagereka 1999; Chomsky 2000, Chomsky 2001). If sentences are spelled-out incrementally, then perhaps these spelled-out chunks correspond to prosodic domains, and may be able to account for apparent mismatches between syntactic and prosodic domains. If we can thus derive prosodic domains without reference to prosodic structure as an intermediate step, using only independently-motivated domains from syntactic theory, perhaps prosodic structure is therefore not necessary to account for domain-sensitive phonetic and phonological patterns.

A topic of debate within phase-based approaches to prosodic domains regards the extent to which these domains are prosodic or syntactic in nature. The strong hypothesis is that phasal domains are prosodic domains, and that there is no need to posit a separate prosodic representation as part of the phonological grammar. Work assuming this strong hypothesis do not (necessarily) espouse the idea that prosodic well-formedness constraints (including eurhythmic constraints) play a role in sentential prosody; rather, such factors are considered to occur late in the grammatical derivation, after the creation of prosodic domains. For work exploring versions of this “strong” hypothesis relating to cyclic or phase-based spell-out, see, for example, Seidl (2001), Wagner (2005), Wagner (2010), Wagner (2015), Pak (2008), Newell (2008), Dobashi (2013), and Newell and Piggott (2014).

Conversely, the “weak” version of this hypothesis maintains that while phases (or other types of domains created by cyclic/multiple spell-out) play a role in defining prosodic domains, phase theory and prosodic structure theory are not incompatible with one another. Rather, prosodic domains are created on the basis of phasal domains rather than syntactic constituent structure. Work incorporating both phase theory and prosodic structure theory include Dobashi (2003), Ishihara (2007), Kratzer and Selkirk (2007), and Selkirk (2009). A critical study comparing phase-based approaches with non-cyclic (edge-based/syntactic) approaches can be found in Cheng and Downing (2012) and Cheng and Downing (2016).

2.4 Prosody and word order

In addition to its role in providing information about syntactic constituent structure, another relevant area of investigation is the relationship between word order and prosodic structure. One main area of inquiry the degree to which prosodic structure interacts with linearization (Kayne 1994; Fox and Pesetsky 2005), resulting in a postsyntactic reordering of terminal nodes or phrases, as has been argued to occur in domains such as second-position clitic phenomena (Halpern 1992; Werle 2009; Huijsmans 2015), pronoun postposing in Irish (Bennett et al. 2015, Bennett et al. 2016), clitic right dislocation in Romance languages (López 2009), the fronting of prosodic (but not syntactic) constituents in Latin, Classical Greek and Russian (Agbayani and Golston 2010, Agbayani and Golston 2016; Agbayani et al. 2011), scrambling in Japanese (Agbayani et al. 2015), and alternations between VSO/VOS word order in Austronesian and Mayan languages (Clemens 2014; Clemens and Coon to appear).

Provided that linearization is one of a number of operations that occur during syntactic spell-out (whether the operations at spell-out occur serially or in parallel), it is perhaps not surprising that prosody may play some role in determining how terminal elements are linearized. However, if it is indeed the case that such manipulations occur only at spell-out, they should not have any effect on the semantic, pragmatic, or information structural interpretation of the manipulated structures. This has been argued to be the case for the phenomena cited above. Under a conservative approach regarding the ability of prosody to manipulate word order, any manipulation that also has an interpretational effect (such as, for example, topicalization) must be analysed in such a way as to ensure such movement occurs in the syntactic component preceding spell-out.

A more radical approach to the syntax-phonology interface asserts that prosody can also have a direct effect on syntactic structure, affecting domains as syntactic movement and thus word order. Such an approach can be found in Contiguity Theory, as in (Richards 2010, Richards 2016, Richards 2017), where it is proposed that syntactic agreement and selectional relations are sensitive to the prosodic characteristics of specific languages, and that syntactic movement is sometimes triggered to help resolve prosodic constraints on contiguity.

3.1 Evidence for recursion

A topic of increasing importance in the study of the syntax-prosody interface is the role of recursion in the creation of prosodic domains. Under the Strict Layer Hypothesis (Selkirk 1981 [1978], Selkirk 1984, Selkirk 1986; Beckman and Pierrehumbert 1986; Nespor and Vogel 1986; Pierrehumbert and Beckman 1988), prosodic structure was hierarchically organized but adhered to a strict constraint banning the recursion of prosodic categories. In this way, prosodic structure was thought to differ fundamentally from syntactic structure.

Recursion in the prosodic domain does not necessarily equate to recursion in the syntactic domain, which has a narrower definition. Rather, because prosodic structure deals with the basic units of syntax, the recursion of prosodic categories may simply reflect compound structures of two prosodic words (Ito and Mester 2007), hierarchically ordered phrasal (XP) structure (Elfner 2012, Elfner 2015; Ito and Mester 2012, Ito and Mester 2013), or the coordination of two or more sentences (Selkirk 2009; Myrberg 2013), without reference to syntactic details such as category labels. Evidence of recursivity in the creation of prosodic domains, as in Match Theory, has been argued to provide support for a more transparent version of the syntax-phonology interface than was assumed under the Strict Layer Hypothesis (Selkirk 2011; Elfner 2012, Elfner 2015; Ito and Mester 2013).

Phonetic evidence has been used to identify differences in the relative boundary strength of structures that show recursivity in the syntactic structure. The relative strength of prosodic boundaries can be measured by comparing the acoustic measures of prosodic cues such as F0, duration, and intensity, and relative cues such as F0 scaling and downstep. Work arguing in favour of recursion in prosodic structure on the basis of prosodic boundary strength include Lehiste (1973), Ladd (1986), Ladd (1988), Kubozono (1989), Kubozono (1992), Wagner (2005), Wagner (2010), Féry and Truckenbrodt (2005), Myrberg (2013), and Elordieta (2015). Phonological arguments in favour of recursion can be found in Selkirk (2011) on Xitsonga, and Elfner (2012, 2015).

3.2 Match Theory

The presence of recursion in prosodic structure plays a crucial role in Match Theory (Selkirk 2009, Selkirk 2011), a recently-proposed indirect reference approach to the syntax-prosody interface, and in work assuming this framework (e.g. Elfner 2012, Elfner 2015; Ito and Mester 2013; Myrberg 2013; Clemens 2014; Bennett et al. 2015, Bennett et al. 2016). In Match Theory, prosodic structure is derived directly from a family of syntax-prosody Match constraints, which call for correspondence between syntactic constituents (word, phrase, clause) and prosodic domains (ω, φ, ι). Schematically, the three mapping constraints can be represented as follows:

Take, for example, the phrasal level, governed by the constraint Match-Phrase. This constraint can be conceptualized in terms of correspondence between sets of terminal nodes, as proposed in Elfner (2012, 2015):

This constraint requires that a correspondence relation (in the sense of McCarthy and Prince 1995) hold between every phrasal constituent in the syntactic component and every φ-level domain in the phonological component, defined such that for every set of terminal nodes dominated by a given syntactic constituent, the phonological exponents of this same set is dominated by a φ constituent. Because syntactic structure is hierarchical, therefore, prosodic structure, under this theory, will also be hierarchical. However, because Match-Phrase does not preserve category distinctions (such as VP, DP, or NP), the resulting prosodic structure will consist of a recursive hierarchy of phonological phrase (φ)-level domains.

One area which requires further investigation is the proposal that the grounding of prosodic structure depends on the three proposed building blocks of syntactic structure—words, phrases, and clauses. The syntactic definitions of these terms can be defined in a number of ways, and may not even have a universal interpretation.

At the word level, we may gain some insight from examining how prosodic words are formed in morphologically-complex polysynthetic languages. For example, Windsor (2017) proposes that Match-Word in Blackfoot references syntactic heads (X⁰), while Guekguezian (2017) proposes that Match-Word in Chukchansi references complex X⁰ as derived via head movement. An additional complication, however, may arise from the Match constraints themselves, which may not be universally interpreted. For example, in polysynthetic languages such as Inuit, phonological “words” may in fact be syntactically phrasal, but correspond prosodically to prosodic words (Compton and Pittman 2010; Arnhold et al. to appear).^[2]

At the phrase level, the particulars of how syntactic trees are built will have some effect on how Match-Phrase is evaluated. For instance, if bare phrase structure (Chomsky 1995) is assumed rather than X-bar structure, syntactic phrases and words may be more aptly characterized as maximal and minimal projections, with fewer intermediate (X′) projections present in the syntactic representation. As discussed in Bennett et al. (2016), one complication that arises when assuming bare phrase structure is that certain elements must then be analyzed as being simultaneously dominated by maximal and minimal projections, predicting that such elements will be simultaneously mapped to two prosodic categories through the application of Match constraints. Bennett et al. (2016: 189) suggest as a solution to this problem, that certain Match constraints be prioritized (or perhaps ranked) over others. For example, they argue that pronouns in Irish, being simultaneously maximal (as DP) and minimal (as D), are preferentially mapped as ω rather than φ; similarly, CPs are mapped to ι (as clauses) rather than φ (as maximal projections). It remains a matter of debate whether such prioritization is universal or is the result of language-specific constraints/rankings, or whether some languages allow constituents to be mapped simultaneously to multiple prosodic categories.

A different type of complication arises from the presence of intermediate (X′ level) structure in syntactic phrases containing an overt specifier, head and complement, as in the following hypothetical structure (where Y, X, and Z are assumed to be phonologically overt):

(3)

X-bar structure

In such structures, strict application of the Match-Phrase constraint as defined in (2) would result in a non-binary φ-phrase (Y X Z) rather than binary (Y(XZ)),^[3] as illustrated in (4) and (5), respectively.

(4)

Ternary structure, derived from (3)

(5)

Binary (recursive) structure, preserving binary branchingness

Bennett et al. (2016) derive the preference for the structure in (5) in Irish to a high-ranking prosodic Binarity constraint requiring φ to dominate no more than two ω, rather than requiring sensitivity to bar-level categories and syntactic branchingness on the part of Match-Phrase. It remains an empirical question whether ternary prosodic structures occur in languages, or whether binary (or binary-branching) structures are universally preferred.^[4] For further discussion of the relationship between syntactic branchingness and prosodic eurhythmic preferences for binary structures, see, for example, Ghini (1993).

At the clausal level, there are a number of ways in which Match-Clause may be interpreted in the prosodic representation. As discussed in Selkirk (2011), clauses can be defined syntactically (e.g. as CP) or pragmatically (e.g. as the ‘illocutionary’ clause). As noted in Selkirk (2011), there is empirical support behind both possibilities as being the basis for intonational phrases in Match Theory (as well as its predecessors in prosodic structure theory), and it is proposed in that work that both Match(illocutionary clause, ι) and Match(clause, ι) may exist in linguistic grammars. On a different note, Hamlaoui and Szendrői (2015, 2017) propose that the mapping of intonational phrases is syntactically defined, and may vary from language to language depending on language-specific surface syntactic patterns: specifically, they argue that intonational phrase edges are defined by the highest extended projection of the verb. For further discussion of these issues, as well as empirical discussion, see Selkirk (2011) and Hamlaoui and Szendrői (2015, 2017) and the references cited within.

Finally, it is worth discussing how the Lexical Category Condition (Selkirk and Shen 1990; Selkirk 1995; Truckenbrodt 1995, Truckenbrodt 1999) relates to both Match-Word and Match-Phrase. This condition embodies the observation that syntactic functional elements and their projections behave differently in terms of syntax-prosody mapping as compared to syntactic lexical elements. More specifically, function words and their projections are thought not to be visible to syntax-prosody mapping constraints. It remains a matter of debate whether this condition, originally proposed as part of an end-based approach to syntax-prosody mapping, should be incorporated into Match Theory, which makes different base assumptions regarding how syntactic constituents are mapped onto prosodic constituents. This topic, specifically as relates to Match Theory, is further discussed in Selkirk (2011), Elfner (2012, 2015), and Selkirk and Lee (2015).

Match Theory thus provides a theory of syntax-prosody mapping in which prosodic structure maps directly from syntactic structure. As conceived in Selkirk (2011), this mapping is governed by violable correspondence constraints that may be outranked by conflicting constraints, such as those governing prosodic well-formedness, as in an Optimality Theoretic framework. As will be discussed in the next section, any mismatches between syntactic and prosodic structure are modeled using constraint interaction.