2.1 Direct Reference Approaches

In early work in phonetics and psycholinguistics, phonological and phonetic phenomena were taken to directly reflect syntactic structure (Lehiste, 1973; Cooper & Sorensen, 1977; Cooper & Paccia-Cooper, 1980). For example, intonational contours, or lengthening, were analyzed as signaling syntactic boundaries. In models of categorical grammar (Steedman, 1991), there is a matching between syntactic structure and prosodic structure where the former corresponds to the latter. In these cases, either there is no level of phonological representation or this level is isomorphic to syntactic representation.

However, the main arguments in favor of direct reference to syntax are phonological phenomena, identified in several languages, that are described as referring to specific syntactic information (e.g., various syntactic categories or syntactic configurations). Crucially, the syntactic properties needed to account for such phenomena are too varied and too particular to be captured by generalizations over syntactic categories and configurations, like lexical head or lexical phrase. An illustrative case is provided by a set of phenomena in the phonology of Kimatuumbi, described by Odden (1987, 1990). The phonological processes of shortening, phrasal tone insertion, initial tone insertion, lengthening, and glide formation are conditioned by an array of different kinds of syntactic properties (e.g., heads, phrasal nodes in particular configurations, initial or final edges of particular types), as well as lexical properties (e.g., class, lexical category of the elements affected).

Kaisse (1985) is one of the main proponents of DRT. Kaisse’s proposal is based on external sandhi phenomena that show domains of application conditioned by syntactic information, in particular by syntactic organization, like the c-command relation, and by particular syntactic category labels. Lenition in Gilyak illustrates the former case. In this language, word-initial obstruents are voiced after nasals and spirantized after vowels, but only if the target of lenition is in a word that c-commands the preceding word. Vowel deletion in Modern Greek could illustrate the latter case, in the account provided in Kaisse (1977), as different rules are described to apply in V-PP and V-Adv combinations. However, as noted in Kaisse (1990), a prosodic hierarchy account of the Greek sandhi phenomena is possible through a generalization in broad syntactic and prosodic terms where heads and phrases play a crucial role in establishing the prosodic domains for vowel sandhi (Condoravdi, 1990).

At least some of the phenomena supporting DRT can be (re)analyzed within Hayes’s (1990) theory of Precompiled Phrasal Phonology. The behavior of the phenomena described in the paragraphs above resembles that of phenomena that apply at the word level and below to specific morphemes or words. An example of variation in the form of morphemes that do not strictly follow from phonological factors is provided by the alternation in the Spanish feminine definite article, which appears as [el] or [la]. The first allomorph is conditioned by certain nouns that start with a stressed [a], whereas the second appears elsewhere, including other syntactic categories beginning in stressed [a]. Allomorphy involving verb and clitic clusters in several Romance languages has similar properties, being sensitive to a preceding verb, in some cases to the type of verbal inflection, as well as to phonological properties (Peperkamp, 1997, for Italian; Vigário, 2003, for European Portuguese). These morphophonological alternations can be viewed as different allomorphs listed in the lexicon. Hayes’s proposal is that phrasal phonological phenomena that refer to specific syntactic information are treated in a parallel way, as phrasal allomorphs. The lexicon includes phrasal allomorphs and the phonological instantiation frames where they are inserted. These frames specify the conditioning environment of the allomorph, which is partly syntactic. For example, the phonological process of shortening in Kimatuumbi, which was described above as supporting the DRT, is instead a case of precompiled phrasal phonology according to Hayes.

More recently, other models have been proposed that are close to DRT in the sense that they argue for a direct influence of syntax in the application of phonological processes (Seidl, 2001; Dobashy, 2003; Ishihara, 2003, 2007; Kahnemuyipour, 2004; Kratzer & Selkirk, 2007; Pak, 2008; Samuels, 2009, 2010). In the model of Seidl (2001), there are two levels of representation mapped from (morpho)syntactic structure: one that is defined by syntactic phases (Chomsky, 2001), and another that is defined by theta-domains, or domains where theta-roles are assigned. There are also two types of phonological processes, depending on the level of representation where they apply. The first type, called early rules, operates at the first level of representation, and thus has access to all syntactic information. The second type, called late rules, operates at the second level of representation, a level that is not purely syntactic and where prosodic domains are projected from theta-domains. This dichotomy is reminiscent of Kaisse’s (1985) segregation of P1-rules, which are defined as those primarily sensitive to syntactic information. However, the phonological process of shortening in Kimatuumbi, for example, that is seen as a P1-rule in Kaisse’s terms (and in the DRT view), and a precompiled phrasal rule in Hayes (1990), is analyzed by Seidl as a late rule, operating at the level of representation projected from theta-domains.

Other proposals (Dobashy, 2003; Ishihara, 2003, 2007; Kahnemuyipour, 2004; Kratzer & Selkirk, 2007; Pak, 2008; Samuels, 2009, 2010) have developed the idea that phonological constituents are mapped from syntactic phases, through a process known as Multiple Spell-out (cf. Uriagereka, 1999; Chomsky, 2001). In Chomsky’s minimalistic theory of syntax, CP and vP correspond to two phase domains, and thus to phonological phrases. Spell-out domains, namely VP and IP/TP, have also been argued to be mapped as phonological phrases (Dobashy, 2003; Ishihara, 2007). The assumption that phases and spell-out domains are essentially the same has led to the theory of Phonological Derivation by Phase (Samuels, 2009, 2010), according to which spell-out domains are the only domains required by phonology. Other spell-out domains may also correspond to prosodic constituents (Kratzer & Selkirk, 2007; Pak, 2008): for example, the Intonational Phrase may correspond to the spell-out of Comma Phrase, and the Prosodic Word can be understood as corresponding to the spell-out domain of lexical heads.

2.2 Indirect Reference Approaches

2.3 Current Research

Some research comparing different approaches to the syntax–phonology interface has been produced, but detailed empirical comparisons and discussions within and across languages are still missing. Studies that have compared relation-based and end-based approaches often arrive at opposite conclusions (e.g., Bickmore, 1990 on Kinyambo; Cho, 1990 on Korean). It has been suggested that some languages are more amenable to one or the other approach, and that languages may also combine both, with some prosodic constituents being more end-based and others more relation-based (Chen, 1990; Inkelas & Zec, 1995; Ghini, 1993). Such a scenario could be a reflection of demarcation and grouping as driving forces of the organization of speech into constituents. Studies comparing the edge-based approach and the Match Theory have also reached contradictory conclusions (e.g., Cheng & Downing, 2012; Truckenbrodt & Féry, 2015). As to the DRT and the PHT, it has been proposed that it might be the case that there are two strategies available for mapping phonological structure from syntactic structure (Elordieta, 2008). Why different possibilities may be at play, across and/or within languages, is yet to be understood. Most past and current research has focused on a restricted set of phonological processes in a limited set of languages, and on a particular level of representation/constituent type for a given language. Few studies have examined the full range of phonological representations in a given language. Even rarer is the attempt to simultaneously consider different types of phonological phenomena to test proposals of syntax–phonology mapping (see section 4). Continued research on these issues is needed to find principled ways of explaining the syntax–phonology interface at all relevant levels within and across languages.

3.1 Phonological Constituents: How They Are Constructed

There is general agreement that the structure of phonological representation at the word level and above is a hierarchy of phonological constituents that results from the interaction of a limited set of morphosyntactic information with phonological principles or constraints (e.g., Nespor & Vogel, 1986; Ghini, 1993; Selkirk, 2000, 2011). The syntactic basis of prosodic constituents consists mostly of generalizations over morphosyntactic categories and structures, so that a prosodic constituent may not be built with reference to specific categories of words or phrases. However, the precise syntactic information that is relevant to phonology depends on the particular theory of the syntax–phonology interface and (the particular version of) the theory of syntax assumed to account for syntax–phonology mapping (see section 2.2). Phonological principles or constraints form the phonological basis of prosodic constituents. Across PHT, eurythmic conditions have been acknowledged, including constituent size/length/weight, balance/symmetry/uniformity and particular directions of asymmetries, as well as speech rate. Many of these factors have been shown to contribute to prosodic constituency both at the word and phrasal levels. For example, size restrictions determine that a given prosodic constituent contains a minimum, a maximum, or an exact number of constituents of a lower level (Nespor & Vogel, 1986; Ghini, 1993; Selkirk, 2000, 2011; Ito & Mester, 2003). A common size restriction is prosodic binarity, shown to be relevant at the foot, prosodic word, and prosodic phrasal levels in several languages. Prosodic binarity requires that a constituent is structurally binary. Size restrictions expressed in number of syllables, which may not be accounted for by prosodic binarity, have also been described in various languages (Delais-Roussarie, 1996; Jun, 2003; Elordieta, Frota, & Vigário, 2005; Frota, 2014). Preferences in terms of relative size/weight determine that prosodic constituents of similar size are formed (Nespor & Vogel, 1986; Ghini, 1993; Frota, 2000; Sândalo & Truckenbrodt, 2002; D’Imperio et al., 2005). An alternative to similar size constituents are phonologically constrained asymmetries. The most common one determines that constituents in prominent positions, i.e., heads, are structurally complex or longer/heavier than non-heads (Dresher & Van der Hulst, 1998; Ghini, 1993; Frota, 2000; Prieto, 2005). Besides head-dependent asymmetries, edge-dependent asymmetries have recently been proposed, as the requirement that a prosodic constituent starts with a strong constituent, that is one which is not lower in prosodic structure than the constituent that immediately follows (Selkirk, 2011). Speech rate also impacts on prosodic constituency: increasing speech rate promotes larger constituents, so that size and balance restrictions may depend on tempo (Nespor & Vogel, 1986; Ghini, 1993; Prieto, 2005).

The phonological basis of prosodic constituents has been implemented within Optimality Theory as phonological markedness constraints that establish the well-formedness of a given prosodic constituent (Selkirk, 2000, 2011). These constraints are proposed to interact with the morphosyntax–phonology mapping constraints, which express the syntactic basis of prosodic constituents, to produce phonological constituency. Importantly, phonological constraints play a crucial role in determining higher (e.g., intonational phrase, phonological phrase) and lower (e.g., prosodic word) phonological constituents.

3.2 Phonological Constituents: How Many Are There?

A major research question has been determining how many prosodic category types there are in the prosodic hierarchy. Proposals have differed according to theories of the syntax–phonology mapping and theories of the structural relations between constituents of prosodic structure (see section 3.3). However, three basic prosodic categories relevant to the syntax–phonology interface are commonly defined: the prosodic word (PW), the phonological phrase (PhP), and the intonational phrase (IP). Each of these prosodic categories bears a relation to a syntactic constituent type, respectively a word-like (lexical) morphosyntactic unit, a phrase-like syntactic unit, and a clause-like syntactic unit (Nespor & Vogel, 1986; Truckenbrodt, 1995; Selkirk, 2011). Beyond these, the repertoire of possible prosodic constituent types has varied considerably across theories, languages, and studies. By and large, the variation results from the addition of word-like or phrase-like categories that respectively fill up the space between PW and PhP and between PhP and IP. A simplified picture of this variation is provided in Table 1.

The definition of possible phonological constituents goes hand in hand with the definition of the properties of prosodic structure. If recursive structures are generally assumed (Ito & Mester, 2009, 2012, 2013; Elfner, 2012), different layers of recursion of the same basic prosodic category (e.g., PhP) may form different prosodic constituents (e.g., maximal PhP, minimal PhP, non-minimal PhP). What is the set of prosodic category types within and across languages remains an open empirical question.

3.3 On Recursion

An answer to the question about how many prosodic category types are possible crucially depends on another central research theme in phrasal phonology, namely to what extent phonological representations have formal properties distinct from syntactic representations. At one extreme, proposals have been developed that posit that phonology and syntax are inherently distinct in that phonological representations consist of strings whereas syntactic representations consist of trees (Chomsky & Halle, 1968; Selkirk, 1981, 1974; Idsardi, 1992; Neeleman & Koot, 2006). Generative phonology, unlike generative syntax, started as a theory of strings structured through boundary symbols (Chomsky & Halle, 1968). Recursion naturally follows as a property of tree-based representations, but not of strings. Thus, syntactic representations are recursive, whereas phonology lacks generalized recursion in the syntactic sense (Neeleman & Koot, 2006). Boundaries have also been appealing for accounts of speech planning, which favor a view of prosodic phrasing based on processing requirements instead of grammatical mappings (Watson & Gibson, 2004). Most phonological theories, however, have departed from string representations to propose hierarchical structures: “we speak trees, not strings” (Pierrehumbert & Beckman, 1988 p. 160). Prosodic Hierarchy Theories (PHT) are tree-based, but whether recursion is allowed or seen as a natural property of phonology is a matter of contention.

The idea that syntax and phonology are related but phonological representations have formal properties that distinguish them from syntactic representations in fundamental ways has been developed both by relation-based and edge-based PHT (see section 2.2). These properties were first known as the Strict Layer Hypothesis, which enforces strict layered prosodic trees (Selkirk, 1984; Nespor & Vogel, 1986). Thus, recursion is ruled out, and in particular recursive representations that would mirror syntax, like (i) domination relations involving a constituent B that is dominated by and contained within A, and that also dominates a constituent A, as illustrated in (1a), or (ii) sisterhood relations involving nodes of different categories, as in (1b–c). A reformulation of the Strict Layer Hypothesis replaced an all-in-one restriction on prosodic representations by a set of constraints (Selkirk, 1995). These domination constraints included Layeredness (which rules out the configuration in 1a), Exhaustivity and NoLevelSkipping (which ban the configurations in 1b–c), NonRecursitivity (which penalizes a node dominating another of the same category, as in 1c–d), and Headedness (which captures the hierarchical organization whereby a constituent must dominate a constituent of the immediately lower level, as in 1e).

(1)

In this view, recursion is highly marked and only arises due to pressures on phonological representations to fulfill other requirements. Even models that make extensive use of recursive constituents, as in Ito and Mester (2009, 2012), generate these structures by violations of constraints against recursion. Two types of recursion may be distinguished: unbalanced and balanced recursion (Van der Hulst, 2010). Forms of bounded recursion, usually restricted to a limited set of structures, have been acknowledged and usually illustrate the first type, depicted in (1c). This is the case for the prosodic phrasing of affixes and clitics with words in many languages, which involves adjunct-like representations and thus level skipping (e.g., Booij, 1996; Peperkamp, 1997; Vigário, 2003). Such phrasing may be understood as a sort of last-resort configuration arguably due to phonological markedness constraints related to well-formed PW, which require stress. A different case is that of balanced recursion (Ladd, 1996; Frota, 2000), a configuration where a prosodic node dominates nodes of the same category type, illustrated in (1d). This has been recently captured by the prosodic sisterhood constraint EqualSisters (Myrberg, 2013). Unlike unbalanced recursion, balanced recursion has no analogue in syntactic structure.

Match Theory is at the other extreme of the spectrum by proposing not only that phonology is tree-based, but that phonological representations are recursive by default like syntactic representations, given the inherent relation between prosodic categories and syntactic categories (Selkirk, 2011; Selkirk & Lee, 2015a). Departures from syntactic-like recursion need to be explained by phonological constraints on domination and sisterhood relations between constituents of the prosodic hierarchy.

A consequence of the lack of agreement on the formal properties of possible prosodic structures is the difficulty in defining phonological constituents. Theories of constituency that posit generalized recursion tend to equate different levels of prosodic phrasing with different constituents that function as prosodic domains for particular phonological processes in a given language (Ito & Mester, 2009, 2012; Selkirk, 2011; Elfner, 2012; Elordieta, 2015). For example, Ito and Mester (2012) suggest that in Japanese the minimal PhP is the domain of accentual culminativity, as there can be at most one accent per minimal PhP, unlike in the other levels of the same constituent, that is, non-minimal PhP or maximal PhP. Differently, theories of constituency that make a restricted use of recursion tend to differentiate levels of constituency from phrasing levels, as only the former are domains for categorical phonological processes (Downing, 1999; Frota, 2000, 2012; Vigário, 2010; Downing & Kadenge, 2015). If two phonological processes refer to two different domains, then two prosodic category types are involved (for example, PW and a higher-level constituent), and not recursion-based constituents defined as the higher and lower levels of the same prosodic type (e.g., minimal and maximal PW).

4.1 Types of Prosodic Structure-Sensitive Phenomena

There is a wide range of phenomena that have been shown to appeal to prosodic constituent structure (e.g., Nespor & Vogel, 1986; Inkelas & Zec, 1990; Hayes & Lahiri, 1991; Frota, 2000, 2012; Horne & Oostendorp, 2005; Frota & Prieto, 2007; Árnason, 2009; Vigário, 2010; Borowsky et al., 2012; Selkirk & Lee, 2015b). They are characterized as sensitive to size/length/weight and eurythmic factors as well as speech rate, and thus reflect the phonological basis of prosodic constituency. Several phenomena may converge to refer to a particular prosodic constituent in a given language, and different constituents are distinguished as the domain of different (sets of) phenomena.

Prosodic structure-sensitive phenomena include sandhi processes (assimilations, lenitions, fortitions, deletions, insertions, and so on) involving segments and tones. For example, a sandhi rule of Bengali phrasal phonology is /r/ assimilation, where /r/ can optionally assimilate to a following coronal consonant (Hayes & Lahiri, 1991). This process is sensitive to PhP domains, as assimilation is only obtained within the PhP. In European Portuguese, vowel sandhi processes such as vowel deletion or semivocalization apply within the IP (Frota, 2000). An illustration of tonal rules sensitive to phonological phrasing is provided by many Bantu languages, like Chicheŵa (Kanerva, 1990; Truckenbrodt, 1995). For example, a rule of tone retraction shifts the tone of the final syllable to the second mora of the penultimate syllable when the final syllable ends a PhP. The phonology and phonetic realization of segments and tones have also been shown to be sensitive to prosodic constituent-initial strengthening at the PW and phrasal levels (Keating et al., 2003; Vigário, 2003; Pan, 2007). Similarly, prosodic constituent-final lengthening or shortening has been described in many languages (Nespor & Vogel, 1986; Turk & Shattuck-Hufnagel, 2007).

Rhythmic and intonation phenomena are also sensitive to prosodic structure. Stress clash resolution strategies, for example, apply within a given prosodic domain and are constrained by prosodic conditions on the organization of PW and phrase-level prominences (Nespor & Vogel, 1989; Frota, 2000; Prieto, 2005). The distribution of intonation events and the realization of their tonal targets also reflect prosodic structure. The relation between pitch accents and prosodic constituency may vary across languages, so that a pitch accent is required on each PW, as in Cairene Arabic (Hellmuth, 2007); on each Prosodic Word Group, as in Brazilian Portuguese (Vigário & Fernandes-Svartman, 2010); on each PhP, as in Bengali (Hayes & Lahiri, 1991) and Northern European Portuguese (Vigário & Frota, 2003); or on each IP, as in Standard European Portuguese (Frota, 2000, 2014). Languages with a lower prosodic domain for accentuation thus show a denser distribution of pitch accents, whereas languages with a higher prosodic domain for accentuation show a sparse distribution of pitch accents. Pitch scaling, pitch reset, and final lowering are phenomena that also refer to prosodic constituency (Ladd, 1996; Truckenbrodt, 2007b). This is shown, for example, by pitch scaling in German, where prosodic constituents define reference lines for tone height and partial reset is found at the left edge of IP (Truckenbrodt, 2002; Truckenbrodt & Féry, 2015).

Stress assignment has also been described as evidence for prosodic structure. For example, in Yidin the presence of stress shows that two consecutive suffixes (like [daga-ɲu] in [gumári-dagá-ɲu] ‘to have become red’) behave as a PW (Nespor & Vogel, 1986). In Chimwiini, vowel length only surfaces in words located at the right edge of a PhP, as this is the position of phrasal stress, and thus phrasal stress identifies the right edge of this prosodic constituent (Selkirk, 2011). More generally, and depending on the language, the location of prosodic prominence is related to the right or left edge of specific prosodic constituents (like the PW, PhP, or IP; Nespor & Vogel, 1986; Selkirk, 2011).

This wide range of phenomena provides potential cues to prosodic structure. Such cues show great variation across languages, given that they may be present or absent in a given language, and vary with respect to the types of prosodic categories they signal across languages. Zerbian’s (2007) study of cues to prosodic phrasing across Bantu languages illustrates how different cues can signal the same level of constituency in different languages (e.g., the IP is cued by blocking of high tone at the right edge in Northern Sotho, and by deletion of high tones within the IP domain in Kinyambo), and how the same phenomena may cue boundaries of different constituent types (e.g., penultimate lengthening signals the PhP in Chicheŵa, but the IP in Northern Soho). Final lengthening at the IP level, although widespread, is apparently absent in some languages, like Chimwiini, Estonian, and Finnish. In Finnish, the preservation of the language-specific quantity system seems to constrain final lengthening (Nakai et al., 2009). Prosodic structure-sensitive phenomena thus appear to be mostly driven by language-particular phonologies.

4.2 Properties of Prosodic Structure-Sensitive Phenomena

Prosodic structure-sensitive phenomena have been identified as belonging to one of three kinds of prosodic rules or constraints (Selkirk, 1980; Nespor & Vogel, 1986): domain span, domain limit or domain edge, and domain juncture. The structural properties of these three kinds are described in Table 2, following the notation in Selkirk and Lee (2015a), where Ψ‎ and χ‎ stand for the phonological properties involved in the phenomenon and “(. . . )π‎” refers to the relevant prosodic structure conditions. Ψ‎ and χ‎ denote the different types of phonological phenomena that appeal to prosodic constituent structure, and include segmental features, tonal features, intonational features, representations of tonal associations, and representations of rhythmic structures (see section 4.1). π‎ is a variable over the set of prosodic category types in prosodic structure (see section 3.2).

/r/ assimilation in Bengali (Hayes & Lahiri, 1991) and some of the vowel sandhi processes in European Portuguese (Frota, 2000) are examples of domain span rules respectively sensitive to the PhP and IP domains. In European Portuguese, semivocalization of stressless high vowels (i/u) followed by another vowel applies both within and across words and PhPs, but not across IPs, as shown by the [w] surface form of the underlined vowels in (o boato alastrou)_IP ‘the rumor spread’ versus (alastrou o boato)_IP (infelizmente)_IP ‘the rumor spread, unfortunately.’

Tone retraction in Chicheŵa, which occurs at the right edge of PhP (Kanerva, 1990), and partial pitch reset in German, found at the left edge of IP (Truckenbrodt & Féry, 2015), illustrate domain edge rules. Prosodic constituent-initial strengthening and prosodic constituent-final lengthening or shortening phenomena (Hayes, 1989; Beckman & Edwards, 1990; Turk, 2012), which may be sensitive to different prosodic category types, are also instances of domain edge rules. In French, for example, different degrees of initial strengthening have been found for different prosodic constituents (PW, Accentual Phrase, IP; Keating et al., 2003).

Stress clash resolution strategies, at least in some accounts, require reference to more than one prosodic category type, as in the case of stress shift or deletion/weakening of the first stress in the clash between PW within the same PhP, or in the case of stress-strengthening when the adjacent stresses belong to different PhP within the same IP (Nespor & Vogel, 1989; Prieto, 2005). These illustrate domain juncture rules. Other examples of domain juncture rules, involving different prosodic category types from the syllable-foot levels to phrasal levels, are provided in Nespor and Vogel (1986) and Kula and Bickmore (2015).

Differing from earlier work, a more restrictive theory of sensitivity to prosodic structure has been proposed in Selkirk and Lee (2015a). In this approach, phonological phenomena may only refer to a single prosodic constituent. This theory, by excluding reference to multiple prosodic categories, excludes the possibility of domain juncture rules. It remains an open research question whether all prosodic structure-sensitive phenomena can be expressed with the structural descriptions in Table 2, and whether these may be minimally reduced to domain span and domain edge constraints.

A further research issue is whether a unified theory of the hierarchical organization of speech may provide a common understanding of the various manifestations of prosodic structure, that is, of all attested prosodic structure-sensitive phenomena, including prominence and intonation phenomena, segmental and tonal sandhi, speech timing, domain-initial strengthening, and so on. Proposals that differentiate accentual phenomena from other prosodic-sensitive phenomena (Gussenhoven & Rietveld, 1992; Himmelmann & Ladd, 2008) or rhythmic and speech timing phenomena from others (van der Hulst, 2009) are not infrequent. Work on the prosodic hierarchy has traditionally been rule-based (Selkirk, 1984, 1986; Nespor & Vogel, 1986; Hayes, 1989; Truckenbrodt, 1999), prominence-based (Hayes, 1984; Nespor, 1990; Beckman & Edwards, 1990, 1994), or intonation-based (Beckman & Pierrehumbert, 1986; Jun, 2005). There are still not many languages studied where several types of prosodic structure-sensitive phenomena are examined in relation to the possible set of prosodic category types that characterize the language. In studies where this is done, it is often the case that a single hierarchy of constituents accounts for the phenomena observed (Hayes & Lahiri, 1991; Jun, 1996; Frota, 2000, 2014; Árnason, 2009). An integrated theory of prosodic structure remains a goal to be pursued.

4.3 Other Evidence

Evidence for prosodic structure also stems from work in language processing and language development. Investigations of the processing of prosodic structure have shown that adult listeners are sensitive to different prosodic category types (e.g., Li & Yang, 2008, on Mandarin Chinese), and that prosodic structure plays a role in word segmentation, lexical processing, and the computation of syntactic structure (Cutler, Dahan, & Donselaar, 1997; Christophe et al., 2004; Millotte, Wales, & Christophe, 2007). Boundaries of prosodic phrases, in particular, are instrumental to lexical access and to establish perceptual groupings into constituent structure. Besides local cues to prosodic boundaries, other cues that signal hierarchical prosodic structures, such as intonation and speech timing, have been shown to be used in artificial grammar learning and to outrank segmental transitional probabilities (Langus et al., 2012). On the production side, prosodic boundaries have been shown to be related to speech planning and recovery in language processing (Watson & Gibson, 2004). Prosodic constituency, for example the number of PW within a phrase, also plays a role in language production planning (Wheeldon & Lahiri, 1997; Lahiri & Wheeldon, 2011). Infant listeners and young children are also sensitive to prosodic structure, including both higher and lower prosodic phrases (Jusczyk, Hohne, & Mandel, 1995; Gout, Christophe, & Morgan, 2004; Homae et al., 2007; Shukla, White, & Aslin, 2011). They are able to use prosodic boundaries to constrain lexical access and syntactic processing. They are also especially sensitive to units at the edges of prosodic phrases. Building on this information, models of language learning have shown that several aspects of prosodic structure may play a key role in language development (Gutman et al., 2015).

The findings from language processing and language development, as well as the continued appeal to behavioral and non-behavioral experimental methods and paradigms for the analysis of prosodic representations, promise to significantly contribute to the understanding of prosodic structure.