Scope and Prosody in Multiple Wh-Questions

Abstract

The prosodic marking of the wh-scope has been a good testing ground to shed light on syntax-prosody mapping. Many accounts have been proposed based on various theoretical models, including the E-feature agreement system, the Multiple Spell-Out Model, Contiguity Theory, and the Wrap-XP Model. However, most previous studies focused on the constructions with a single wh-phrase, and few studies paid attention to multiple wh-questions. This paper presents novel data from production experiments to show the prosodic patterns of multiple wh-questions in Korean, for which none of the previous accounts makes correct predictions. This study proposes a new alignment constraint considering the scope relations between wh-words. The necessity of such a constraint suggests that the prosodic structures for wh-scope interpretations are not the direct outcome of syntax and phonology but the aggregation of syntax, phonology, and semantics.

1. Introduction

This paper investigates the relationship between semantic scope and prosody, putting special attention on the wh-scope of multiple wh-questions. In wh-in situ languages such as Korean and Japanese, wh-scope is decided by the association with a question marker such as -nunci(Q) or -ni(Q) in the Korean example in (1).

(1)		Minho-nun	[Yumi-ka	nwukwu-lul	mannass-nunci]	kwungkumhayha-ni?
		Minho-top	Yumi-nom	who-acc	met-q	wonder-q
	a.	‘Does Minho wonder who Yumi met?’
	b.	‘For which x, x a person, does Minho wonder whether Yumi met x’

Since the question marker -ni(Q) in (1) is scope-neutral, the wh-phrase nwukwu can be associated with either -nunci(Q) or -ni(Q). Consequently, the sentence is scopally ambiguous. When the wh-phrase nwukwu ‘who’ is associated with the embedded complementizer -nunci, it has embedded scope, as shown in (1a). Alternatively, if the wh-phrase is associated with the matrix complementizer -ni, it is interpreted with a matrix scope, as demonstrated in (1b). This ambiguity between the indirect question interpretation (i.e., embedded scope) and the direct question interpretation (i.e., matrix scope) can be resolved by prosody. For example, in South Kyeongsang Korean (henceforth SKK), the domain of wh-scope is found to correlate with the span of so-called wh-intonation (i.e., F0 compression or high plateau) (Hwang 2011a, 2011b, 2015). The examples (2) illustrate embedded and matrix scope wh-questions in SKK, which are also morphologically marked by the yes-no question ending -na and the wh-question ending -no, respectively. As shown in the diagrams in (2), the semantic scope of wh-phrases can be realized phonetically with the span of the high flat F0 contour, which starts from the wh-phrase and continues to its associated complementizer (Q). It ends at the embedded Q in the case of embedded-scope wh-phrases (2a) in Figure 1, whereas it ends at the matrix Q in the case of the matrix-scope wh-phrases (2b) in Figure 2. The domain of this high plateau is called wh-intonation. The same wh-intonation pattern is observed even when the scope-neutral question marker is used, in which case prosody is the only clue to disambiguation.

(2)		Minho-nun	[Yumi-ka	nwukwu-lul	mannass-nunci]	kwungkumhayha-na/-no?
		Minho-top	Yumi-nom	who-acc	met-q	wonder-ynq/- whq
	a.	‘Does Minho wonder who Yumi met?’
	b.	‘For which x, x a person, Minho wonders whether Yumi met x?’

In order to explain the wh-intonation pattern, many accounts have been proposed based on various theoretical models, including the E-feature agreement system (Deguchi and Kitagawa 2002; Kitagawa 2005), the Multiple Spell-Out Model (Ishihara 2007), Contiguity Theory (Richards 2010), and the Wrap-XP Model (Smith 2011). However, most previous studies focus on the constructions with a single wh-phrase, and very few studies have paid attention to multiple wh-questions. Therefore, this paper aims to identify the major prosodic phenomena that are generally observed in multiple wh-questions and to illustrate how the semantic scope relationship of wh-phrases can influence the formation of wh-intonation. The study of the prosodic scope marking, including a wider range of data, helps us reveal the essential nature of wh-intonation, which has implications for the interfaces of phonology, syntax, and semantics.

The rest of this paper is structured as follows. In the following sections, I examine the prosodic pattern of multiple wh-phrases in production. Section 2 offers a background on the syntactic characteristics relevant to wh-questions as well as an overview of the prosodic properties of South Kyeongsang Korean. Section 3 reviews the previous studies on the correlation between prosody and semantic scope of wh-phrases, and makes a prediction about the prosodic formations of multiple wh-phrases based on the previous studies. Then, in the following sections, I provide empirical data on the phonetic implementation of wh-intonation in multiple wh-questions in SKK based on a production test. I propose new constraints to account for the surface realization of wh-intonation of multiple wh-questions, based on the observations. Finally, I conclude the paper in Section 7.

2. Background

2.1. Syntactic Properties of wh-Questions

2.1.1. Syntactic Structure and Complementizers

The schematic syntactic structure of (1) and its scope interpretations are illustrated in (3).

The diagram in (3a) shows how the embedded wh-phrase associates with the embedded complementizer (Q), yielding a yes/no question format. Korean speakers allow the wh-phrase to relate to the matrix interrogative ending (Q), as shown in (3b). Given that Korean is a wh-in situ language, it does not require wh-movement to determine scope.

Instead, sentence types and associated wh-scopes can be morphologically indicated through morphological markers at the end of the sentence. Generally, the choice of sentence endings (matrix clause complementizers) varies according to the sentence type, tense specification, and utterance style. The details of interrogative complementizers in two Korean varieties, Seoul Korean and South Kyeongsang Korean, are given below in

Table 1. Interrogative sentence-final particles.

Style		Property			Seoul K	SKK
Formality	Politeness	Q	wh	Verb Tenses
formal	high	+	±	past/present/future	-pnikka	-pnikka
informal	high	±	±	past/present/future	-eyo	-eyo
informal	low	±	±	past/present/future	-e
informal	low	+	±	past/present/future	-ni, -nya
informal	low	+	–	past/present/future		-na
informal	low	+	–	present (be)/future		-ka
informal	low	+	+	past/present/future		-no
informal	low	+	+	present (be)/future		-ko

. The formal endings are shown to be question-specific yet scope-neutral, as indicated by [±] on the wh column. In contrast, the informal endings in South Kyeongsang Korean (SKK) exclusively indicate wh-scope (-na: embedded scope (an indirect wh-question interpretation) vs. -no: matrix scope (a direct wh-question interpretation)).1

The embedded complementizer -nunci in SKK, analogous to ‘whether’ in English, is crucial for forming both yes/no and wh-questions, as demonstrated in (4).

(4)	a.	Minho-nun	[Yumi-ka	Suci-lul	mannass-tako]	malhayss-ta
		Minho-top	Yumi-nom	Suci-acc	met-dec	wonder-dec
		‘Minho said that Yumi met Suci’.
	b.	Minho-nun	[Yumi-ka	Suci-lul	mannass-nunci]	kwungkumhayss-ta
		Minho-top	Yumi-nom	Suci-acc	met-q	wonder-dec
		‘Minho wondered whether Yumi met Suci’.
	c.	Minho-nun	[Yumi-ka	nwukwu-lul	mannass-nunci]	kwungkumhayss-ta
		Minho-top	Yumi-nom	who-acc	met-q	wonder-dec
		‘Minho wondered who Yumi met’.

This complementizer is not optional for interrogative embedded constructions and plays a role in forming what is known as a wh-island, thereby potentially restricting the embedded wh-phrase from taking a matrix scope.

2.1.2. Wh-Island Effect

As in (3), Korean does not require overt wh-movement, in contrast to English (5), where dislocating a wh-phrase from its original position to its scope position is obligatory.

Syntactically, overt wh-movements are subject to the wh-island constraint (Chomsky 1973). In other words, the wh-island constraint prohibits the embedded wh-phrase ‘whom’ from extending its scope beyond the boundaries of an island, formed by an interrogative element such as ‘whether’. Hence, the matrix scope of ‘whom’ in (6) is unavailable due to a violation of wh-island constraint.

(6)	a.	[Does Minho ask [whom Yumi met ___ ]]?
	b.	* [Whom does Minho ask whether Yumi met ___]]?

As for a wh-in situ language such as Korean, it is traditionally understood, following Huang (1982), that an in situ wh-phrase undergoes covert movement at LF even though there is no overt movement. However, it is controversial whether the covert wh-movements are also susceptible to the wh-island constraints. On the one hand, it is argued that (1) cannot yield a matrix scope interpretation for a wh-phrase, positing such embedded clauses as islands (Hong 2004). This perspective implies that covert movements are restricted by the wh-island constraint. On the other hand, it is claimed that the matrix scope interpretation is feasible (Suh 1987; Ishihara 2002; Y.-S. Choi 2006; Hwang 2011a, 2011b, 2015), thereby questioning the application of wh-island constraints to covert movements and challenging the classification of these clauses as islands. Specifically, many studies, including Hwang (2011a, 2011b, 2015), highlighted the significance of wh-intonation. They suggest that when appropriate prosody indicating matrix scope is provided, the wh-island effect is nullified. Further discussion about wh-intonation can be found in Section 2.2.2.

This becomes even more complicated when multiple wh-phrases appear in a single sentence, as shown in (7).

(7)		Minho-nun	[nwu-ka	nwukwu-lul	mannass-nunci]	kwungkumhayha-ni?
		Minho-top	who-nom	who-acc	met-q	wonder-q
	a.	‘Does Minho wonder who met whom?’
	b.	‘For which x, x a person, does Minho wonder who met x’.
	c.	‘For which y, y a person, does Minho wonder whom y met’.
	d.	‘For which x, y, x a person, y a person, does Minho wonder x met y’.

The sentence in (7) has multiple interpretations. However, judgments on those scope interpretations vary across previous studies. Shimoyama (2001) reported strong wh-island effects, indicating that only example (7a) is valid. Nishigauchi (1986, 1990); Saito (1994); and Richard (1997) argued that acceptable interpretations are those where both wh-phrases are assigned the same scope, making (7a) and (7d) acceptable, but not (7b) and (7c). Kurata (1991) further differentiated between (7b) and (7c), finding (7c) acceptable but not (7b). Conversely, Ishihara (2003) suggested that all scope interpretations can be considered viable given an appropriate context.

Adopting Ishihara’s perspective that all potential scope combinations are plausible, this study aims to explore the correlation between each scope interpretation and its prosodic realization in SKK.

2.2. Prosodic Properties

In this section, we turn our attention to the prosodic properties of SKK.

2.2.1. Prosodic Phrasing

The prosodic categories of a commonly posited prosodic hierarchy are shown in Figure 3.

In some studies on prosodic categories, including from Pierrehumbert and Beckman (1988), a Phonological Phrase is further divided into two sub-levels: Major Phrase/Intermediate Phrase (a higher level) and Minor Phrase (a lower level).

According to Jun (1993), the Korean prosodic structure is hierarchically organized, as shown in Figure 4. An Intonational Phrase can contain more than one Accentual Phrase, which corresponds to the Phonological Phrase in Figure 3, and an Accentual Phrase contains one or more phonological words.

With regard to South Kyeongsang Korean, there is little research investigating its prosodic constituents. To the best of my knowledge, there is only one study for South Kyeongsang Korean (SKK) (Kim and Jun 2009). Kim and Jun (2009) separated a Phonological Phrase into two levels by additionally positing an Intermediate Phrase above an Accentual Phrase, following the hierarchical structure proposed by Pierrehumbert and Beckman (1988).2 However, recent studies on this matter, i.e., Ito and Mester (2007, 2009) and Selkirk (2009), have also argued that just one category of phonological phrase should be posited because the further distinction of the Phonological Phrases (Intermediate Phrase/Major Phrase and Minor Phrase) is a syntagmatic notion; the distinction of Minor and Major phrases are not categorical but relational such as minimal and maximal projections in tree structures. Following this, this paper assumes that only one Phonological Phrase is posited in South Kyeongsang Korean.

Phonetics and phonology studies on Korean wh-questions, including Cho (1990) and Jun and Oh (1996), have reported that wh-phrases introduce changes in phonological phrasing in the sentence. For example, a wh-phrase creates a single prosodic unit with the following unaccented words. The prosodic unit relevant to wh-scope in SKK is the Phonological Phrase, which corresponds to Minor Phrase in Richards (2000, 2010, 2016) and Major Phrase in Hwang (2011b).

2.2.2. Wh-Intonation: Pitch Compression and High Plateau

According to Hwang (2011a), wh-intonation is phonetically realized as either F0 pitch compression or high plateau in SKK. It is argued that this phonetic variation comes from the alternating accent patterns of the wh-phrases.

(8)	Minho-nun	[Yumi-ka	nwukwu-lul	mannass-nunci]	kwungkumhayha-no?
	Minho-top	Yumi-nom	who-acc	met-q	wonder-whq
	‘For which x, x a person, does Minho wonder whether Yumi met x?’

The wh-phrase nwuku-lul in (8) bears alternating accent patterns LH(H) ~ HH(L). On the one hand, when the wh-phrase is produced with a rising tone (LHH), a high plateau is formed, as shown in Figure 5. On the other hand, when the wh-phrase is produced with a falling tone (HHL), a pitch compression is formed, as shown in Figure 6. In addition, Hwang shows that both patterns can be used even in the same situation by a single speaker. Considering that either a pitch compression or a high plateau truly relies on the alternative accent patterns of a wh-phrase, this study focuses on the spans of a high plateau or a pitch compression in multiple wh-phrases, not the specific accent patterns (either a high plateau or a pitch compression).

3. Models and Their Predictions

The syntax/semantics and prosody interface phenomenon on wh-scope has been analyzed by many different models, including the E-feature agreement system (Deguchi and Kitagawa 2002; Kitagawa 2005), Chomsky’s Multiple Spell-Out model (Ishihara 2007), Contiguity Theory (Richards 2010), and the Wrap-XP Model (Smith 2011). However, most of the models in previous studies were developed based on single wh-questions. Hence, this section explores the main ideas of these models and their predictions on the prosodic formation of multiple wh-questions.

The examination of model predictions encompasses six specified conditions that relate to the potential associations between wh-phrases and complementizers, as outlined in example (9).

(9)		Two types of multiple wh-questions
	a.	Both wh-phrases in the embedded clause: [   …   [ WH1  WH2 … V-Q] V-Q? Possible scope combinations:
		i.	Embedded	embedded
		ii.	Embedded	Matrix
		iii.	Matrix	embedded
		iv.	Matrix	Matrix
	b.	One in the matrix clause, the other in the embedded clause: [  WH1 [ …  WH2 … V-Q] V-Q? Possible scope combinations:
		i. Matrix		embedded
		ii. Matrix		Matrix

Note that the models were developed based on Tokyo Japanese and Fukuoka Japanese. Depending on the language used in the model, the relevant prosodic constituents to wh-intonations were different with either Major phrases or Minor phrases. The further distinction of phonological phrases relevant to the wh-intonation in SKK has not yet been discussed. Due to the lack of research on prosodic phrasing in SKK, Hwang (2011a, 2011b) also simply followed the phrasing structures shown in North Kyeongsang Korean. Since the discussion on the details of prosodic phrasing is out of the scope of this research, I will use “Phonological phrase” in the remainder of this paper. As such, the terms Minor phrase and Major phrase in each model are replaced with Phonological phrases in the prediction and the analysis.

3.1. The E-Feature Agreement System

Deguchi and Kitagawa (2002) introduced a syntactic analysis based on the operation Agreement (Chomsky 1998; Chomsky 2001) in order to explain the correlation between wh-scope and prosody. They proposed the following Emphatic features and assumed that those features undergo Agreement.

(10)	E(mphatic)-features
	a.	Uninterpretable E-feature: optionally assigned to INFL (or to T).
	b.	Interpretable E-feature: on wh-phrases.

According to Deguchi and Kitagawa, the uninterpretable E-feature on INFL acts as a probe, and the interpretable E-feature on a wh-phrase acts as a goal.3 These two features undergo agreement, which is called “E-agreement”. The example (11) illustrates how prosody and wh-scope correlations are established in this model. At LF, the E-feature on INFL induces covert movement of a wh-phrase to get the correct scope interpretation, as shown in (11a). However, at PF, as shown in (11b), a pair of E-features undergoing agreement is linearly scanned and comes to be phonetically implemented as wh-intonation. The emphatic accent falls on a wh-phrase carrying the goal, and pitch-eradication follows it and continues to an INFL containing the probe.

They argued that even in multiple wh-questions, there is just one extended emphatic domain, as shown in (12). They also show that when only the second wh-phrase is emphasized with an accent (12b), the sentence becomes uninterpretable. According to them, this occurs because, although the emphatic domain is formed at PF through the E-feature agreement, the presence of identical wh-features in both wh-phrases leads to intervention effects (wh-island effect) at the LF. Specifically, the non-emphasized ‘who’ phrase obstructs the LF movement of the emphasized ‘what’ phrase. Moreover, given the constraints of the E-feature system where only a single wh-phrase may be marked with an emphatic feature, it becomes impossible to produce the correct prosodic forms (i.e., emphatic accents on both wh-phrases in (12d)). To address this issue, they have adopted the “wh-cluster hypothesis”, as proposed by Saito (1994). According to this hypothesis, the second wh-phrase moves up to merge with the preceding wh-phrase, creating a cluster. This entire wh-cluster then undergoes movement at LF.

Following this idea, we can predict the prosodic formation in

Table 2. Prosodic phrasing of multiple wh-sentences predicted by the E-agreement system.

Prosodic Phrasing	Position of wh		Wh-Scope
	WH1	WH2	WH1	WH2
a. NP-NOM [Emb WH1-NOM WH2-ACC V-Q12] V-YNQ?    (     )(       )	S-Emb	O-Emb	Emb *	Emb
b. NP-NOM [Emb WH1-NOM WH2-ACC V-Q1] V-WHQ2?              #(           )	S-Emb	O-Emb	Emb	Mat **
c. NP-NOM [Emb WH1-NOM WH2-ACC V-Q2] V-WHQ1?        (                )	S-Emb	O-Emb	Mat	Emb
d. NP-NOM [Emb WH1-NOM WH2-ACC V-Q] V-WHQ12?        (     )(           )	S-Emb	O-Emb	Mat	Mat
e. WH1-NOM [Emb NP-NOM WH2-ACC V-Q2] V-WHQ1? (                       )	S-Mat	O-Emb	Mat	Emb
f. WH1-NOM [Emb NP -NOM WH2-ACC V-Q] V-WHQ12? ?(                      )	S-Mat	O-Emb	Mat	Mat

* Emb—Embedded, ** Mat—Matrix.

. The scope relation is marked with numbers on wh-phrases and complementizers. Under the E-feature agreement framework, when the anticipated prosodic pattern leads to unacceptable interpretations, such instances are denoted with “#”.

Except for the conditions (a), (b), and (d) in Table 1, it is expected that a Phonological Phrase starts at the first wh-phrase and continues to the associated complementizer. The second wh-phrase is encompassed in the Phonological Phrase initiated by the first wh-phase, so its pitch will be deleted. Particularly, in (f), since two wh-phrases are not adjacent, even if both wh-phrases initially receive the emphatic accents, the following focus reduction process will delete the pitch of the second wh-phrase. Regarding (b) in Table 1, as shown in (6b), the first wh-phrase lacks an emphatic accent and does not constitute a single Phonological Phrase by itself.

3.2. The Multiple Spell-Out Model

Ishihara (2003, 2004, 2005, 2007) applied Chomsky’s (2001) and Fox and Pesetsky’s (2005) Multiple Spell-Out Model to a cyclic derivation of prosody for proper wh-scope marking. The details regarding phase and multiple spell-out systems are as follows.

(13)	Phase and Multiple Spell-Out (Chomsky 2001)

(14)	Phase and Multiple Spell-Out (Fox and Pesetsky 2005)

As seen in (13) and (14), Chomsky (2001) and Fox and Pesetsky (2005) identified the Spell-Out domain differently. While Ishihara (2007) follows the basic concepts of Fox and Pesetsky (2005), he additionally assumed that phrases that are adjoined to a phase, such as adjuncts and A’-moved material, are excluded from the spell-out domain.

Ishihara also assumed that focus features participate in an agreement process; however, this agreement involves items containing focus features, specifically one associated with a wh-phrase and another with a question particle (Q). According to Ishihara (2007), after the agreement between two focus features takes place, the complement of the phase head (TP) is transferred to the interface level. Then, at PF, it creates a prosodic domain; a focus feature on a wh-phrase initiates the generation of wh-intonation from a wh-phrase to the right edge of a spell-out domain. Since a C head is not included in the Spell-Out domain (TP), he suggests that the C head is phonologically cliticized to the preceding phrase in order to contain C in the wh-intonation domain.

According to Ishihara, when a prosodic domain is created, all phonological materials at the interface level are included, even for the materials that were transferred at earlier spell-out cycles. This means that the prosodic domain incrementally gets bigger and bigger as spell-out takes place cyclically, as illustrated in (15).

At each spell-out domain, wh-intonation is created by the phonological rules in (16).

(16)	a.	P-focalization rule:
		If αFOC bears FOCUS, add x’s to αFOC at a metrical line until a new line is formed.
	b.	Post-FOCUS Reduction (PFR) Rule:
		If αFOC bears FOCUS and precedes β, and αFOC’s peak (after P-focalization) is at Line n, then delete an x of β on Line n-1.
		(Ishihara 2003)

Unlike the E-agreement system, the Multiple Spell-Out Model assumes that phonological wh-scope marking is an outcome of a phase-by-phase syntactic derivation, not a result of the direct phonology-syntax/semantics interaction.

In Ishihara’s Multiple Spell-Out system, wh-intonation is created within the Spell-Out domain containing both a wh-phrase and an associated complementizer. When a prosodic domain is created, all phonological materials at the interface level are included, even the materials that are transferred at earlier spell-out cycles. For multiple wh-questions, if the first wh-phrase has a wider scope than the second wh-phrase as shown in (c) and (e) in

Table 3. Prosodic phrasing of multiple wh-sentences predicted by the Multiple Spell-Out system.

Prosodic Phrasing	Position of wh		Wh-Scope
	WH1	WH2	WH1	WH2
a. NP-NOM [Emb WH1-NOM WH2-ACC V-Q12] V-YNQ?    i.(     )(      )    ii.(               )	S-Emb	O-Emb	Emb *	Emb
b. NP-NOM [Emb WH1-NOM WH2-ACC V-Q1] V-WHQ2?        (     )(           )	S-Emb	O-Emb	Emb	Mat **
c. NP-NOM [Emb WH1-NOM WH2-ACC V-Q2] V-WHQ1?        (                )	S-Emb	O-Emb	Mat	Emb
d. NP-NOM [Emb WH1-NOM WH2-ACC V-Q] V-WHQ12?        i.(     )(           )       ii.(                   )	S-Emb	O-Emb	Mat	Mat
e. WH1-NOM [Emb NP-NOM WH2-ACC V-Q2] V-WHQ1? (                       )	S-Mat	O-Emb	Mat	Emb
f. WH1-NOM [Emb NP -NOM WH2-ACC V-Q] V-WHQ12?  i.(           )(             )  ii.(                       )	S-Mat	O-Emb	Mat	Mat

* Emb—Embedded, ** Mat—Matrix.

, both wh-phrases will be phased together in one single prosodic domain. After the agreement between the second wh-phrase and the embedded complementizer happens first, the wh-intonation between them will be created at the embedded CP spell-out domain. Then, after the agreement between the first wh-phrase and the matrix complementizer takes place, the wh-intonation from the first wh-phrase to the matrix complementizer will be created, as boosting the focalized wh-phrase first and then reducing all the elements up until the end of the matrix complementizer. As a consequence, the pitch accent on the second wh-phrase will be deleted. In South Kyeongsang Korean, the pitch accent deletion will result in either a high plateau or a pitch compression.

For the cases where both wh-phrases have the same scope as shown in (a), (d), and (f) in Table 3, after the agreements between two wh-phrases and their associated complementizer happen, both wh-phrases go to the same spell-out domain. In the spell-out domain, both wh-phrases are boosted at first, and then the pitch accents of all the following elements are reduced up until the end of the complementizer. As long as there is no additional rule that forces preservation of the boosted pitch accents of wh-phrases, the pitch accent of the second wh-phrase should be reduced as shown in (ii) in (a), (d), and (f) because it falls into the pitch accent reduction domain. However, Ishihara’s (2003) experimental data shows the prosodic phrasing in (i), and he assumed that the pitch reduction happened only after the second wh-phrase. As for (b) in Table 3 where the first wh-phrase has narrower scope than the second wh-phrase, the first wh-phrase will be phased by itself because the first wh-phrase is outside of the domain for the wh-intonation of the second wh-phrase, so the boosted pitch on the first wh-phrase will be preserved. Even though he mentioned that the interpretation (b) is available in a certain context, as shown in (17), he did not include this scope relation in his experiments by simply following the claims that (b) is an illegitimate reading (Kurata 1991, Saito 1994, Shimoyama 2001).

(17)	a. Naoya:	[dáre-ga	bíiru-o	nónda ka ]	obóeteru?
		who-nom	beer-acc	drank Q	remember
	‘Do you remember who drank beer?’
	b: Mari:	[dáre-ga	BÍIRU-o	nónda ka ] -wa	obóete-nai.
		who-nom	beer-acc	drank Q—top	remember-neg
	‘I don’t remember who drank BEER’.
	c. Naoya:	[dáre-ga	NÁNI-o	nónda ka ]-wa	obóeteru no?
		who-nom	what-acc	drank Q- top	remember
	‘Whati do you remember who drank ti?’
	d: Mari:	[dáre-ga	WÁIN-o	nónda ka ] -wa	obóeteru yo.
		who-nom	wine-acc	drank Q—top	remember
	‘I remember who drank WINE.’

To ascertain the prosodic patterns for the scope relationship, where WH1 is embedded within WH2 in the matrix, an experimental investigation is necessitated.

3.3. The Contiguity Theory

Regarding the correlation between prosody and wh-scope, Richards (2006, 2010) has an opposite view of the E-agreement system and the Multiple Spell-Out Model. Contrary to the E-agreement system and the Multiple Spell-Out Model, which propose that a certain syntactic operation on relevant features may affect its phonological realization, Richards suggested that syntactic operations such as wh-movement or wh-in situ happen when prosody requires them.4 He proposed a universal condition on wh-prosody that accounts for the motivation for wh-movement: when a wh-phrase in situ cannot form a proper prosodic domain for wh-scope in the following fashion (18) and (19), wh-movement takes place in order to create the proper prosodic domain.

(18)	Universal condition on wh-prosody (Richards 2006, 2010):
	Given a wh phrase α and a complementizer C where α takes scope, α and C must be separated by as few minor phrase boundaries as possible for some level of minor phrasing.

(19)		Algorithm for wh-domains:
	a.	For one end of the larger Minor Phrase, use a Minor Phrase boundary that was introduced by a wh-phrase.
	b.	For the other end of the larger Minor Phrase, use any existing Minor Phrase boundary.

Two important factors for constructing the prosodic domain are the position of the complementizer and the placement of Minor Phrase boundaries (i.e., the prominence of either the Left or Right boundary). In Korean, a left edge is more prominent (Jun 1998), and a complementizer appears at the sentential-final position. According to the Contiguity Theory regarding a wh-question, the procedure to form the prosodic domain for matrix wh-scope in (20), hence, is like (21).

(20)	Yengwu-nun	[Mila nwuna-ka	mwusun kok-ul	yencwuhayss-nunci]	mwuless-no?
	Yengwu-top	Mila sister-nom	which song-acc	played-q	asked-whq
	‘For which x, x a song, Yengwu asked whether Mila played x’

(21)		NP	[EmbeddedNP	wh	Verb-C(Q)]	Verb-C(WHQ)
	a.	(     )	(      )	(	)	(      )
	b.	(     )	(      )	(		)

Following the algorithm in (19), a larger Minor Phrase containing the wh-phrase and the associated complementizer (a matrix Q) is created by keeping the Minor Phrase boundary associated with the Left edge (prominent edge) of the wh-phrase, skipping the immediately following one, and using the Minor Phrase boundary associated with the right edge of the complementizer (a matrix Q) as shown in (21b). Likewise, since the proper prosodic domain for wh-scope can be created successfully, Koreans can leave wh-phrases in situ. In other words, in Richards’ view, the wh-intonation in Korean is the result of a universal condition on wh-prosody, not because of the effect of the syntactic operation.

Hence, under the algorithm of Richards’ Contiguity Theory, the left edge of each wh-phrase and the right edge of its associated complementizer is used for creating a large Phonological phrase. This large Phonological phrase is referring to the domain of a high plateau or a pitch compression. When a large Phonological phrase, including a wh-phrase and its associated complementizer, cannot be formed, such as in English, wh-movement takes place in order to create the proper prosodic domain. As a result, in multiple wh-questions the first wh-phrase will not be in the same prosodic domain with its associated complementizer because the second wh-phrase in the stimuli always appears before complementizers and the new prosodic domain starts at the second wh-phrase as shown in

Table 4. Prosodic phrasing of multiple wh-sentences predicted by the Contiguity Theory.

Prosodic Phrasing	Position of wh		Wh-Scope
	WH1	WH2	WH1	WH2
a. NP-NOM [Emb WH1-NOM WH2-ACC V-Q12] V-YNQ?    *(        )(       )	S-Emb	O-Emb	Emb *	Emb
b. NP-NOM [Emb WH1-NOM WH2-ACC V-Q1] V-WHQ2?        *(      )(           )	S-Emb	O-Emb	Emb	Mat **
c. NP-NOM [Emb WH1-NOM WH2-ACC V-Q2] V-WHQ1?       *(       )(           )	S-Emb	O-Emb	Mat	Emb
d. NP-NOM [Emb WH1-NOM WH2-ACC V-Q] V-WHQ12?       *(      )(            )	S-Emb	O-Emb	Mat	Mat
e. WH1-NOM [Emb NP-NOM WH2-ACC V-Q2] V-WHQ1?  *(           )(           )	S-Mat	O-Emb	Mat	Emb
f. WH1-NOM [Emb NP -NOM WH2-ACC V-Q12] V-WHQ12?  *(           )(            )	S-Mat	O-Emb	Mat	Mat

* Emb—Embedded, ** Mat—Matrix.

. In the Contiguity Theory, it is illicit for a wh-phrase and its associated complementizer to be phased separately.

In order to resolve this problem, the first wh-phrase is supposed to move next to its associated complementizer to make them be in one prosodic domain together as a last resort. This syntactic movement, however, does not happen in Korean. In other words, Richards’ Contiguity Theory cannot be applied to the prosodic domain formation for multiple wh-questions in Korean.

3.4. The Wrap-XP Model

Smith (2011) proposed an alternative model based on Optimality Theory (Prince and Smolensky 2002). To account for the wh-intonation, Smith introduced three constraints related to the phonological phrasing for wh-scope. They are shown in (22).

(22)	Constraints on the wh-intonation:
	a.	Wrap-C	Every C[+wh] must be in the same phrase as some associated wh element.
	b.	Wrap-WH	Every wh element must be in the same phrase as some associated C[+wh].
	c.	Align-L(wh, MiP)	The left edge of every wh element is aligned with the left edge of some MiP.

(23)	The ranking of constraints:
	Wrap-C >> Align-L (wh, MiP) >> Wrap-WH

The ranking in (23) is derived from multiple wh-questions as well as single wh-questions. First, let us take a look at the example of a single wh-question (2), repeated in (24).

(24)	Minho-nun	[Yumi-ka	nwukwu-lul	mannass-nunci]	kwungkumhayha-no?
	Minho-top	Yumi-nom	who-acc	met-q	wonder-whq
	‘For which x, x a person, does Minho wonder whether Yumi met x’

The ranking (Wrap-C >> Align-L (wh, MiP) >> Wrap-WH) rules out the candidate (25a), which cannot create a single prosodic domain including a wh-phrase (WH1) and an associated complementizer (Q1).

When this ranking is applied to the sentence including multiple wh-phrases, as shown in (26), the prosodic domain, which starts from the first wh-phrase and ends at the end of the sentence, is predicted to be the winner.

(26)	enu nwuna-ka	[Mila nwuna-ka	mwusun kok-ul	yencwuhayss-nunci]	mwuless-no?
	which sister-nom	Mila-nom	which song-acc	played-q	asked-whq
	‘For which sister, x a sister, x asked which song Mila played.’

The constraint ranking Wrap-C >> Align-L (wh, MiP (henceforth, PoP)) >> Wrap-WH from Smith’s (2011) Wrap-XP Model predicts the following.

As seen in

Table 5. Prosodic phrasing of multiple wh-sentences predicted by the Wrap-XP Model.

Prosodic Phrasing	Position of wh		Wh-Scope
	WH1	WH2	WH1	WH2
a. NP-NOM [Emb WH1-NOM WH2-ACC V-Q12] V-YNQ?   (       )(       )	S-Emb	O-Emb	Emb *	Emb
b. NP-NOM [Emb WH1-NOM WH2-ACC V-Q1] V-WHQ2?        (                 )	S-Emb	O-Emb	Emb	Mat **
c. NP-NOM [Emb WH1-NOM WH2-ACC V-Q2] V-WHQ1?       (                  )	S-Emb	O-Emb	Mat	Emb
d. NP-NOM [Emb WH1-NOM WH2-ACC V-Q] V-WHQ12?       (      )(            )	S-Emb	O-Emb	Mat	Mat
e. WH1-NOM [Emb NP-NOM WH2-ACC V-Q2] V-WHQ1? (                       )	S-Mat	O-Emb	Mat	Emb
f. WH1-NOM [Emb NP -NOM WH2-ACC V-Q12] V-WHQ12?   (            )(           )	S-Mat	O-Emb	Mat	Mat

* Emb—Embedded, ** Mat—Matrix.

, since Wrap-C is the highest ranked, when both wh-phrases have different semantic scopes, the prosodic domain is predicted to start at the first wh-phrase and end at the matrix complementizer. When both wh-phrases have the same semantic scope, as the associated complementizer is phased with the second wh-phrase, the Wrap-C constraint is not violated. Thus, the separate phrasing of the first wh-phrase is predicted as an outcome because one single Phonological phrase from the first wh-phrase to the associated complementizer violates the second-ranked constraint Align-L (wh, PoP).

In sum, the models’ predictions on the prosodic domain of multiple wh-questions differ according to two key factors: the treatment of each wh-phrase’s prominence, leading to its segmentation into a distinct prosodic domain, and the onset and offset of the pitch reduction process, which is associated with the size of the pitch reduction domain. The application of these factors is influenced by each model’s theoretical perspectives, such as whether syntax precedes phonology, phonology precedes syntax, or both are considered simultaneously. Consequently, these factors lead to two potential prosodic structures for each scope relation among multiple wh-phrases, as detailed in

Table 6. The summary of predictions by each model.

The Conditions for Multiple wh-Sentences	EA: E-Agreement MS: Multiple Spell-Out Model CT: Contiguity Theory WM: Wrap-XP Model
	EA	MS	CT	WM
a. NP-NOM [Emb WH1-NOM WH2-ACC V-Q12] V-YNQ?    i.(      )(         )      ii.(                 )	✓ ✗	✗ ✓	✗ ✗	✓ ✗
b. NP-NOM [Emb WH1-NOM WH2-ACC V-Q1] V-WHQ2?         i.(       )(             )        ii.(                  )	✗ ✗	✓ ✗	✗ ✗	✗ ✓
c. NP-NOM [Emb WH1-NOM WH2-ACC V-Q2] V-WHQ1?       i.(      )(           )       ii.(                 )	✗ ✓	✗ ✓	✗ ✗	✗ ✓
d. NP-NOM [Emb WH1-NOM WH2-ACC V-Q] V-WHQ12?       i.(     )(            )        ii.(                    )	✓ ✗	✗ ✓	✗ ✗	✓ ✗
e. WH1-NOM [Emb NP-NOM WH2-ACC V-Q2] V-WHQ1? i.(           )(           ) ii.(                       )	✗ ✓	✗ ✓	✗ ✗	✗ ✓
f. WH1-NOM [Emb NP -NOM WH2-ACC V-Q] V-WHQ12? i.(           )(           ) ii.(                       )	✗ ?	✗ ✓	✗ ✗	✓ ✗

✓: The model’s prediction.

. The predictions from each theory or model are summarized in Table 6 (✓: prediction success, ✗: prediction failure). As shown in Table 6, the predictions of each theory or model vary. This calls for an experiment to investigate what is the phonetic realization of multiple wh-scope relationships.

Most empirical data regarding prosody and wh-scope in Korean are restricted to less complex sentences that include only one wh-question, as shown in (2). As a result, it remains unknown what the phonological realization of the semantic scope of multiple wh-phrases is and whether the correlation between wh-intonation and wh-scope is maintained in the constructions of multiple wh-questions. Moreover, it is difficult to examine whether the suggested models succeed in accounting for the relationship of wh-scope and wh-intonation in Korean multiple wh-questions. In the later sections of this paper, I will describe an empirical investigation of wh-intonation in multiple wh-question constructions in Korean, which shows that none of the suggested models in the previous literature explains the data.

4. Experiment

4.1. Materials and Methods

In order to examine the domain of wh-intonation when a sentence includes multiple wh-phrases, the stimuli in this experiment consisted of three different types of sentences: baseline sentences without wh-items, single wh-sentences, and multiple wh-sentences.

In the stimuli of single wh-sentences, two different positions of wh-phrases in an embedded clause were included, as shown in (28).

(28)	The conditions for single wh-sentences	position of wh	wh-scope
	a. NP-NOM [Emb WH-NOM NP-ACC V-Q] V-YNQ?	S	Emb
	b. NP-NOM [Emb WH-NOM NP-ACC V-Q] V-WHQ?	S	Mat
	c. NP-NOM [Emb NP-NOM WH-ACC V-Q] V-YNQ?	O	Emb
	d. NP-NOM [Emb NP-NOM WH-ACC V-Q] V-WHQ?	O	Mat

In (28a) and (28b), wh-phrases are in the subject position in the embedded clause, and the wh-phrase in (22a) has embedded scope, but the one in (28b) has matrix scope. In (28c) and (28d), wh-phrases are in the object position in the embedded clause, and their semantic scope is embedded clause in (22c) but the matrix clause in (22d). In order to lead to the intended wh-scope interpretation, the distinct question complementizers -na (ynq) or -no (whq) were used; -na (ynq) indicates the embedded scope of wh-phrases and -no (whq) indicates the matrix scope of wh-phrases.5 Sixteen sentences (= 4 sets x 4 conditions) were created. They are in Appendix A.

To investigate the prosodic domain of multiple wh-questions, sentences were constructed with two wh-phrases, adhering to the two different types of multiple wh-questions outlined in (29). For instance, wh-phrases were located either in the same clause (both in the embedded clause) or in the different clauses (one in the matrix clause and the other in the embedded clause). The specific conditions for multiple wh-questions are as follows.

(29)	The conditions for multiple wh-sentences	Position of wh		wh-scope
		WH1	WH2	WH1	WH2
	a. NP-NOM [Emb WH1-NOM WH2-ACC V-Q1,2] V-YNQ?	S-Emb	O-Emb	Emb	Emb
	b. NP-NOM [Emb WH1-NOM WH2-ACC V-Q1] V-WHQ2?	S-Emb	O-Emb	Emb	Mat
	c. NP-NOM [Emb WH1-NOM WH2-ACC V-Q2] V-WHQ1?	S-Emb	O-Emb	Mat	Emb
	d. NP-NOM [Emb WH1-NOM WH2-ACC V-Q] V-WHQ1,2?	S-Emb	O-Emb	Mat	Mat
	e. WH1-NOM [Emb NP-NOM WH2-ACC V-Q2] V-WHQ1?	S-Mat	O-Emb	Mat	Emb
	f. WH1-NOM [Emb NP-NOM WH2-ACC V-Q] V-WHQ1,2?	S-Mat	O-Emb	Mat	Mat

Similar to single wh-questions, the positions of wh-phrases and the scope of wh-phrases were controlled in the stimuli of multiple wh-questions. In (23a–d), both wh-phrases are located in the same clause. Each wh-phrase can have either embedded or matrix scope, so the four different scope relations between the two wh-phrases (2 × 2) were tested. In (23e–f), two wh-phrases are in the different clauses. The first wh-phrase is in the subject position in the matrix clause, and the second wh-phrase is in the object position in an embedded clause. In Korean, wh-phrases cannot be associated with a lower-level clause complementizer, so the first wh-phrase can have matrix scope only. On the other hand, the second wh-phrase can take either an embedded or matrix scope. Consequently, the study tested the two distinct scope relations (1×2) between the wh-phrases. Although this creates unequal stimulus conditions across two types of wh-phrases, including both variations in the positions of wh-phrases allows us to investigate the presence of consistent prosodic patterns across them. All the associations between wh-phrases and complementizers (Q) are marked with numbers in (29). Four different sets were tested, varying the embedded verbs: yencwuhata ‘play’, mekta ‘eat’, mantulta ‘make’, and masita ‘drink’. These embedded verbs were selected in consideration of segmental context; they start with sonorant consonants.

One set of stimuli where both wh-phrases are embedded is listed in (30). The remainders are in Appendix B.

(30)		Yengwu-nun	[enu nwuna-ka	mwusun kok-ul	yencwuhayss-nunci]
		Yengwu-top	which sister-nom	which song-acc	played-q
		mwuless-no?	(-na for (a))
		asked-whq	(-ynq)
	a.	which sister (embedded scope)  =  which song (embedded scope)
		‘Did Yengwu ask which sister played which song?’
	b.	which sister (embedded scope)  <  which song (matrix scope)
		‘For which y, y a song, Yengwu asked which sister played y?’
	c.	which sister (matrix scope)    >  which song (embedded scope)
		‘For which x, x a sister, Yengwu asked which song x played?’
	d.	which sister (matrix scope)    =  which song (matrix scope)
		‘For which x, y, x a sister, y a song, Yengwu asked whether x played y?’
		Enu nwuna-ka	[Mila nwuna-ka	mwusun kok-ul	yencwuhayss-nunci]
		Which sister-nom	Mila (sister)-nom	which song-acc	played-q
		yencwuhayss-nunci]	mwuless-no?
		played-q	asked-whq
	e.	which sister (matrix scope)      >  which song (embedded scope)
		‘Which sister asked which song Mila played?’
	f.	which sister (matrix scope)      =  which song (matrix scope)
		‘For which x, y, x a sister, y a song, x asked whether Mila played y?’

In the stimuli, D-linked wh-phrases such as enu nwuna ‘which sister’ were used in order to reduce the lexical ambiguity between wh-indefinite and wh-interrogative. For each target interrogative sentence, the short context and the proper answer were provided to lead the specific scope interpretation. One example set of contexts is shown in Appendix C.

4.2. Participants and Procedure

Eight native speakers of South Kyeongsang Korean (five females and three males) participated in the recording. All participants were born and grew up in South Kyeongsang province (3: Ulsan, 2: Busan, 1: Changwon, and 2: Grew up in Ulsan and moved to Busan to attend colleges) and had no history of speech or hearing impairment. Participants were recruited on Korean social network websites (facebook.com, accessed on 12 March 2024) and through word of mouth.

The experiments were separated into three sessions: (i) non-wh, (ii) single-wh, and (iii) multiple whs. For each session, I briefly explained the procedure for the recording to the speakers. Since the target sentences were provided with the contexts and answers in the written scripts, the participants were asked to read them silently first in order to get the intended scope interpretations. Participants were instructed to give natural renditions at a comfortable speed. For example, they were asked to read the sentences naturally as if they were talking to their family members or friends. Practice time was given in order to induce natural-sounding speech. While the recording was being made, when the participants misread the sentences or produced unnatural long pauses, they were asked to repeat the sentences.6 When participants themselves wanted to re-record the sentences, those sentences were re-recorded. Note that it was a difficult task for participants because multiple wh-questions are not very frequently used in daily conversation. In order to alleviate the difficulty, particularly for multiple wh-questions, I provided additional information to make it clear what the intended interpretation was. For example, I verbally mentioned the expected answers. They were told that they were going to ask a question to get those answers. In order to encourage the participants to produce natural speech, I produced the answer parts to the target question sentences that the participants produced, as shown in (31).

(31)	a.	which sister (embedded scope)  =  which song (embedded scope)
		nye.
		‘yes’
	b.	which sister (embedded scope)  <  which song (matrix scope)
		Moccaluthu-yo.
		Mozart-dec.
		‘It was Mozart’
	c.	which sister (matrix scope)    >  which song (embedded scope)
		‘For which x, x a sister, Yengwu asked which song x played?’
		Mila nwuna-ka-yo.
		Mila sister-nom-dec
		‘It was Mila’.
	d.	which sister (matrix scope)    =  which song (matrix scope)
		‘For which x, y, x a sister, y a song, Yengwu asked whether x played y?’
		Mila nwuna-ka Moccaluthu-lul yencwuhayss-nunci mwuless-eyo.7
		Mila sister-nom Mozart-acc  played-c      asked- dec.
		‘(He) asked whether Mila (sister) played Mozart’.
	e.	which sister (matrix scope)    >  which song (embedded scope)
		‘Which sister asked which song Mila played?’
		Mila nwuna-ka-yo.
		Mila sister-nom-dec
		‘It was Mila’.
	f.	which sister (matrix scope)    =  which song (matrix scope)
		‘For which x, y, x a sister, y a song, x asked whether Mila played y?’
		Cia nwuna-ka Mila nwuna-ka Moccaluthu-lul yencwuhayss-nunci mwuless-eyo.8
		Cia nwuna-ka Mila sister-nom Mozart-acc played-q asked- dec.
		‘Cia (sister) asked whether Mila (sister) played Mozart’.

In addition, while the two specific morphological markers -na (ynq) or -no (whq) were used in the stimuli, if the participants indicated that other markers such as -nuntey (whq) were the most frequently used in their daily conversation, using them was allowed as a substitute for the given morphological markers in order to help participants produce the sentences easily. Dropping case markers such as the -ul accusative marker was also allowed.

Recording was conducted both in a lab and over Zoom. The recording of three participants was conducted in the phonetics lab in the Department of Linguistics at Stony Brook University. A portable Zoom H6 Handy recorder and a Shure SM 10A-CN microphone were used for the recordings. The remaining five participants were done over Zoom rather than in the lab due to COVID-19 safety concerns. The procedure of conducting the experiment via Zoom paralleled that of the in-lab setup, with participants instructed to be in a quiet room. The speech was recorded using both a separate recorder and the built-in recording function on Zoom as a backup measure. While the recording did capture sound through the speaker, incidental background noise, such as that from a computer’s cooling fan, was also present. However, given that the focus of this study is on the overall pitch contour rather than segment-level analysis, these noises did not compromise the quality of the data. 16 target sentences for single wh-questions and 24 multiple wh-questions were recorded with fillers. The filler sentences vary in terms of length and 1–3 adverbial phrases were included in each sentence. Of the total 180 sentences per participant, 44 sentences (non-wh) were recorded in session 1, 36 sentences (single wh) were recorded in session 2, and 90 sentences (multiple whs: one scope relation per sentence) were recorded in session 3. In each session, the sentences were pseudorandomized to ensure that the sentences with identical lexical elements or those implying similar scope relationships did not appear consecutively. The recording sessions for each speaker took approximately 40 to 60 min, including practice time and the breaks. All participants were paid USD 20 for their participation at the end of the recording session.

5. Results

5.1. Measurements

The data from eight participants were digitized at a 44.1 kHz sampling rate and 16-bit quantization. Analysis was carried out in Praat, version 6.0.33. For each utterance, syntactic phrase boundaries were labeled manually, as illustrated in Figure 7. The example shown in Figure 7 is a multiple wh-phrase question in which two wh-phrases appear in the embedded clause, and their scope relationship is WH1 > WH2 (WH1: matrix scope, WH2: embedded scope). While there are some microvariations in the pitch contour, the high flat pitch pattern (or the pitch compression) is observed from the end of the first wh-phrase to the syllable preceding the matrix clause complementizer. In this way, the pitch patterns of each sentence were examined by focusing on the prosodic domain of wh-intonation.

(32)	Yengwu-nun	[enu nwuna-ka	mwusun kok-ul	yencwuhayss-nunci]
	Yengwu-top	which sister-nom	which song-acc	played-c(q)
	mwuless-no?
	asked-c(whq)
	‘For which x, x a sister, Yengwu asked which song x played?’

Note that the prosodic patterns were largely consistent across both the sets and participants. Therefore, the pitch contours described in Section 5 illustrate the dominant patterns observed among participants based on a single set of stimuli. For details on variations in pitch patterns, please see Appendix D.

5.2. Phonetic Description of the Intonation of Non-Multiple wh-Questions

Before taking a look at the results of multiple wh-questions, let us briefly take a look at the results of baseline recording in non-wh question constructions and single wh-question constructions. Figure 8 shows the F0 contour of a complex sentence that has no wh-phrase.

(33)	Yengwu-nun	[Mila nwuna-ka	Moccaluthu-lul	yencwuhayss-nunci]
	Yengwu-top	Mila sister-nom	Mozart -acc	played-c(q)
	mwuless-na?
	asked-c(ynq)
	‘Did Yengwu ask whether Mila played Mozart?’

Figure 9 and Figure 10 below present the F0 contour of the same construction as in (33), but with a wh-phrase in an object position within the embedded clause as in (34), corresponding to the embedded scope and the matrix scope, respectively.

(34)	Yengwu-nun	[Mila nwuna-ka	mwusun kok-ul	yencwuhayss-nunci]
	Yengwu-top	Mila sister-nom	which song-acc	played-c(q)
	mwuless-na?
	asked-c(ynq)/-c(whq)
	a. ‘Did Yengwu ask which song Mila played?’
	b. ‘For which y, y a song, Yengwu asked Mila played y?’

Figure 11 and Figure 12 below present the F0 contour of a construction with a wh-phrase in the subject position at the embedded clause, corresponding to the embedded scope and the matrix scope, respectively.

(35)	Yengwu-nun	[enu nwuna-ka	Moccaluthu-lul	yencwuhayss-nunci]
	Yengwu-top	which sister-nom	Mozart -acc	played-c(q)
	mwuless-na?
	asked-c(ynq)/-c(whq)
	a. ‘Did Yengwu ask which sister played Mozart?’
	b. ‘For which x, x a sister, Yengwu asked x played Mozart?’

As shown in Figure 8, when there is no wh-phrase in a sentence (33), neither pitch compression nor high plateau are observed. As for single wh-questions in (34) and (35), a high plateau extends from the wh-phrase to the complementizer, contingent upon its scope. Regardless of the syntactic position of the wh-phrase—whether as a subject or an object—within the embedded clause, this acoustic feature remains consistent. For example, in the pitch contours depicted in Figure 9 and Figure 11 for the embedded scope, a high plateau appears from the wh-phrase to the embedded complementizer. Conversely, in the contours for the matrix scope, as shown in Figure 10 and Figure 12, the high plateau spans from the wh-phrase to the matrix complementizer. The pitch contour patterns shown in Figure 9, Figure 10, Figure 11 and Figure 12 conform to Hwang’s (2011b) observation of wh-intonation in South Kyeongsang Korean.

5.3. Phonetic Description of the Intonation of Multiple wh-Questions

In the analysis of multiple wh-questions, the data are categorized into two types depending on the positions of the wh-phrases: (i) both wh-phrases in the same clause (i.e., the embedded clause) or (ii) one wh-phrase in the matrix clause and the other in the embedded clause.

The pitch contours of the sentences, including two wh-phrases in the same clauses as in (30a–d), repeated in (36), are shown in Figure 13, Figure 14, Figure 15 and Figure 16. Observe that the prosodic phrasing and the span of high-plateau wh-intonation vary depending on the scope relationship between the two wh-phrases.

(36)		Yengwu-nun	[enu nwuna-ka	mwusun kok-ul	yencwuhayss-nunci]
		Yengwu-top	which sister-nom	which song-acc	played-c(q)
		mwuless-no?	(-na for (a))
		asked-c(whq)	(-c(ynq))
	a.	which sister (embedded scope)  =  which song (embedded scope)
		‘Did Yengwu ask which sister played which song?’
	b.	which sister (embedded scope)  <  which song (matrix scope)
		‘For which y, y a song, Yengwu asked which sister played y’
	c.	which sister (matrix scope)    >  which song (embedded scope)
		‘For which x, x a sister, Yengwu asked which song x played’
	d.	which sister (matrix scope)    =  which song (matrix scope)
		‘For which x, y, x a sister, y a song, Yengwu asked whether x played y’.

When both wh-phrases take the same scope, each wh-phrase lies in a different prosodic phrase. As schematically indicated in (37), a wh-specific Phonological phrase begins at each wh-phrase, and the Phonological phrase, including the second wh-phrase, ends at the complementizer that both wh-phrases are associated with: the embedded complementizer for (37a) corresponding to Figure 13 and the matrix complementizer for (37b) corresponding to Figure 16.

(37)	a.	NP [Emb WH1 WH2 V-Q12] V-Q	WH1 = WH2 (embedded)
		( )  (  )(     )(  )
	b.	NP [Emb WH1 WH2 V-Q ] V-Q12	WH1 = WH2 (matrix)
		( )  (  )(        )

When two wh-phrases have different scopes, the formation of the prosodic domain is different depending on which wh-phrase has a wider scope (matrix scope). When the first wh-phrase has a narrower scope than the second wh-phrase (WH1(embedded scope) < WH2 (matrix scope)) as in (38a) (corresponding to Figure 14), each wh-phrase lies in a different prosodic phrase. Each Phonological phrase is initiated at each wh-phrase. The prosodic domain, including the second wh-phrase, ends with the matrix clause complementizer. However, when the first wh-phrase has a wider scope than the second wh-phrase (WH1 (matrix scope) > WH2 (embedded scope)) as in (38b) (corresponding to Figure 15), both wh-phrases lie in the same prosodic phrase, which begins at the first wh-phrase and ends at the matrix clause complementizer.

(38)	a.	NP [Emb WH1 WH2 V-Q1] V-Q2	WH1(embedded) < WH2 (matrix)
		( )  (  )(          )
	b.	NP [Emb WH1 WH2 V-Q2] V-Q1	WH1 (matrix) > WH2 (embedded)
		( )  (           )

The prosodic configurations for wh-intonation of the scope relations in (37b) (WH1 (matrix) = WH2 (matrix)) and (38a) (WH1 (embedded) < WH2 (matrix)) are the same. In order to further examine whether the prominence of the wh-phrases can be different depending on the scope relations shown in (37b) and (38a), the maximum pitch of the wh-phrases was measured. Their averages are in

Table 7. The average maximum pitch of the wh-phrases.

	WH1	WH2
a. WH1 (matrix) = WH2 (matrix)	259.25 Hz	256.83 Hz	p = 0.88
b. WH1 (embedded) < WH2 (matrix)	244.78 Hz	268.79 Hz	p < 0.05

.

When the second wh-phrase had a larger scope than the first wh-phrase, the phonetic prominence was increased on the second wh-phrase. This phenomenon was consistently observed across participants.

The pitch contours of the sentences, including two wh-phrases in different clauses in (24e–f) and repeated in (39), are in Figure 17 and Figure 18. Recall that the first wh-phrase in the matrix clause always has matrix scope, but the second wh-phrase in the embedded clause can take either matrix scope or embedded scope.

(39)		Enu nwuna-ka	[Mila nwuna-ka	mwusun kok-ul	yencwuhayss-nunci]
		Which sister-nom	Mila (sister)-nom	which song-acc	played-c(q)
		yencwuhayss-nunci]	mwuless-no?
		played-c(q)	asked-c(whq)
	a.	which sister (matrix scope)    >  which song (embedded scope)
		‘Which sister asked which song Mila played?’
	b.	which sister (matrix scope)    =  which song (matrix scope)
		‘For which x, y, x a sister, y a song, x asked whether Mila played y?’

Schematic representations of prosodic phrasing are given in (40). When the second wh-phrase has narrower scope (embedded scope) than the first wh-phrase (matrix scope), both wh-phrases are in one prosodic domain that starts from the first wh-phrase and continues to the matrix complementizer as shown in (40a) corresponding to Figure 17. When the second wh-phrase has the same scope as the first wh-phrase (matrix scope), the two wh-phrases are in separate prosodic domains, as shown in (40b), corresponding to Figure 18.

(40)	a.	WH1 [Emb NP WH2 V-Q2] V-Q1	WH1(matrix) > WH2 (embedded)
		(             )
	b.	WH1 [Emb NP WH2 V-Q] V-Q12	WH1 (matrix) > WH2 (matrix)
		(      )(       )

These results show that the semantic scope relationship between the two wh-phrases plays a more important role in deciding their prosodic domains than the syntactic positions of wh-phrases (i.e., whether in the matrix clause or in an embedded clause).

In sum, regardless of the positions of wh-phrases, when the first wh-phrase has a wider scope than the second wh-phrase, two wh-phrases are phrased together in the same prosodic domain. Otherwise, two wh-phrases are in separate prosodic phrases. In the next section, these results will be compared to the predictions by the models introduced in the previous literature.

6. Discussion

The summary of the predictions of each theory and model in Table 5 is repeated in

Table 8. The comparison of model predictions and empirical findings.

The Conditions for Multiple wh-Sentences	EA: E-Agreement MS: Multiple Spell-Out Model CT: Contiguity Theory WM: Wrap-XP Model
	EA	MS	CT	WM
a. NP-NOM [Emb WH1-NOM WH2-ACC V-Q12] V-YNQ?    i.(       )(         )   ii.(             )	✓ ✗	✗ ✓	✗ ✗	✓ ✗
b. NP-NOM [Emb WH1-NOM WH2-ACC V-Q1] V-WHQ2?       i.(     )(            )      ii.(                 )	✗ ✗	✓ ✗	✗ ✗	✗ ✓
c. NP-NOM [Emb WH1-NOM WH2-ACC V-Q2] V-WHQ1?       i.(     )(           )       ii.(                 )	✗ ✓	✗ ✓	✗ ✗	✗ ✓
d. NP-NOM [Emb WH1-NOM WH2-ACC V-Q] V-WHQ12?       i.(     )(           )        ii.(                   )	✓ ✗	✗ ✓	✗ ✗	✓ ✗
e. WH1-NOM [Emb NP-NOM WH2-ACC V-Q2] V-WHQ1? i.(            )(            ) ii.(                       )	✗ ✓	✗ ✓	✗ ✗	✗ ✓
f. WH1-NOM [Emb NP -NOM WH2-ACC V-Q] V-WHQ12? i.(            )(            ) ii.(                      )	✗ ?	✗ ✓	✗ ✗	✓ ✗
The number of correct predictions	4	3	0	5

, incorporating a comparison with the empirical results. It marks the results of the experiment as well as the theories that are consistent with the results by shading.

The experimental results can be summarized as follows. In most cases, multiple wh-phrases induce separate prosodic domains for their own (a, b, d, f). The exception is when the linearly preceding wh-phrase takes a wider scope (matrix scope) than the next wh-phrase. In this case, the second wh-phrase does not induce a separate prosodic domain for its wh-intonation, whether the two wh-phrases are in the same clause (c) or in different clauses (e). The predictions summarized in Table 8 show that none of the theories and models in the previous literature introduced in this paper successfully predict the prosodic domain formation of all multiple wh-questions. This calls for the modification of the model.

I propose a new constraint and a new ranking of constraints that can deal with all the results of multiple wh-questions in South Kyeongsang Korean, based on the Wrap-XP Model, which seems the most promising among the previous models because it predicts the most cases correctly. The empirical findings reveal distinctive prosodic patterns that align with scope relations, suggesting a more direct interconnection among prosody, syntax, and semantics rather than a sequential relationship where syntax precedes prosody or vice versa. Consequently, an Optimality Theory (OT) approach may be more plausible, as it considers all constraints simultaneously rather than sequentially.

The main observation from the results is that (i) when the linearly preceding wh-phrase takes wider scope than the following wh-phrase, the prosodic phrase for the wh-intonation starts at the first wh-phrase and the pitch accents of the following elements, including the second wh-phrase are deleted, and (ii) in all other cases, a wh-specific phonological phrase is initiated at each wh-phrase. Based on this observation, I suggest a new Align-L constraint in (41), which can initiate a new phonological phrase at each wh-phrase depending on the scope relationship between wh-phrases. Align-L-Scope (wh, PoP) takes the linear order of two wh-phrases and their scope relationship into account. Specifically, this constraint considers only two adjacent wh-phrases at once.

(41)	A new constraint:
	Align-L-Scope (wh, PoP)	Align the left edge of some PoP when it takes a wider scope than a linearly preceding wh-phrase.

I also propose that Align-R constraints should be posited in the grammar. In the original Wrap-XP Model (Smith 2011), Align-R constraints are not included because the model assumes that the right edge of a Phonological phrase for wh-intonation is aligned with the associated complementizer by the constraint Wrap-C. However, according to the definition of Wrap-C, it does not require the right edge to align with a complementizer and a Phonological phrase. It is satisfied as long as a wh-phrase and an associated complementizer are in the same phrase.

(42)	The definition of Wrap-C constraint:
	Wrap-C	Every C[+wh] must be in the same phrase as some associated wh element.

Since Smith (2011) considered only two candidates (43a) and (43b) in her study, the right edge alignment did not stand out. However, as seen in (43), when another possible candidate (43c) is added, the constraints suggested by Smith (2011) are not sufficient to choose the right result (43a).

To rule out the wrong candidate (43c), Align-R(C, PoP), as shown in (44), should be included but low-ranked.

(44)	The definition of Aligh-R constraint:
	Align-R(C, PoP)	The right edge of every C element is aligned with the right edge of some PoP.

As seen in (45), the revised ranking Wrap-C >> Align-L (wh, PoP) >> Wrap-WH, Align-R (C, PoP) draws the right result.

Let us now take a look at whether the newly revised constraint ranking, including the new suggested constraints Align-L-Scope (wh, PoP) and Align-R (C, PoP,) extracts the correct results.

(46)	The new ranking of constraints: Aligh-L-Scope (wh, PoP) >> Wrap-C >> Align-L (wh, PoP) >> Wrap-WH, Align-R (C, PoP)

In the following tableaux, a simplified configuration is given as an input. The semantic scope relationship is indicated with numbers. The observed phonetic realization of wh-intonation is shaded.

Let us first consider the cases where two wh-phrases take the same scope. In (47), since both wh-phrases have embedded scope (WH1: Embedded = Wh2: Embedded), none of the candidates violate the highest ranked constraint Align-L-Scope (wh, PoP). The more general Align-L (wh, PoP) constraint induces the presence of a phrase break at each wh-phrase. As a result, (47b) and (47d) are ruled out. The comparison between (47a) and (47c) shows that Align-R (C, PoP) brings about the presence of a phrase break at the associated complementizer, and (47c) is ruled out.

When two wh-phrases have matrix scope (WH1: Matrix = Wh2: Matrix), as shown in (48), Wrap-WH, a constraint ranked higher than Align-R (C, PoP), militates against the presence of a phrase break at the embedded complementizer which is not associated with any of the two wh-phrases.

Let us now take a look at the cases where two wh-phrases take different scopes. In (49), the second wh-phrase has a wider scope than the first wh-phrase (WH1: Embedded < Wh2: Matrix). The highest-ranked constraint Align-L-Scope (wh, PoP) induces the presence of a phrase break at each wh-phrase. In evaluating candidates (a) and (c), (a) is ruled out due to the severe violation of the second-highest-ranked Wrap-C. While candidate (c) violates Wrap-C once due to the phasing of C(Q1) in a separate domain from WH1, candidate (a) doubly violates the Wrap-C as neither C(Q1) nor C(Q2) resides in the same phrase with their corresponding whs. Thus, with the second-highest-ranked Wrap-C, closing a Phonological phrase at the embedded complementizer is avoided.

In (50), the first wh-phrase has a wider scope than the second wh-phrase (WH1: Matrix > Wh2: Embedded). Align-L-Scope (wh, PoP) is not applied here. The second highly ranked Wrap-C, thus, forces two wh-phrases and two complementizers to be phrased together.

To summarize, the ranking for wh-intonation in Korean should be Align-L-Scope (wh, PoP) >> Wrap-C >> Align-L (wh, PoP) >> Wrap-WH, Align-R (C, PoP). This conclusion shows that more fine-grained constraints in terms of the scope relationships of multiple wh-phrases play a crucial role in determining the initiation of new prosodic boundaries of wh-phrases. In addition, even though Align-R (C, PoP) is low-ranked, it is necessary to determine the end of the prosodic domain of wh-intonation.

As we have seen, the competing approach to the correlation between prosody and syntactic structure, particularly for wh-scope, ends up working well only with the limited data. The newly added data in this paper show that the semantic scope relations of multiple wh-phrases influence prosodic phrasing of wh-intonation. The insertion of the left edge of wh-intonation is decided by the scope relations of wh-phrases. The proposed constraint Align-L-Scope (wh, PoP) reflects the semantic scope relations to phonological phrasing. This informs that prosodic structure has to do with semantics as well as syntax. Future studies are required to see if this left-edge alignment constraint can be stretched to new types of structures other than wh-questions, such as the old-new information structure or the focus structure that involves changes in prosodic phrasing.

7. Conclusions

This study shows that South Kyeongsang Korean stands as evidence that the syntax-phonology interface is sensitive to the scope relationship between wh-phrases. The evidence from multiple wh-questions has shown that the syntax-semantics-prosody interface constraint that determines the initiation of wh-intonation in South Kyeongsang Korean is Align-L-Scope (wh, PoP), repeated in (51a). Align-R (C, PoP) as well as Wrap-C contribute to determining the end of wh-intonation.

(51)	Newly added constraints:
	a. Align-L-Scope (wh, PoP)	Align the left edge of some PoP when it takes a wider scope than a linearly preceding wh-phrase.
	b. Align-R(C, PoP)	The right edge of every C element is aligned with the right edge of some PoP.

In fact, Smith (2011) pointed out that regarding wh-prosodic constructions, the Fukuoka Japanese data do not serve as unambiguous evidence for the existence of the constraint Wrap-WH. South Kyeongsang Korean data also provide no evidence that the constraint Wrap-WH is required to construct the proper prosodic phrases for wh-intonation. In Korean, the proper prosodic phrasing for wh-intonation is imposed by the relationship between wh-phrases and the complementizer Rather than Wrap-WH.

Most previous studies on wh-scope have paid attention only to the syntax-phonology interface, mainly investigating, for example, which aspects of syntactic structure have phonological consequences or what kind of phonological information is responsible for the syntactic operation. However, the current study finds that the prosodic structures for wh-scope interpretations are not the direct outcome of syntax and phonology but the aggregation of syntax, phonology, and semantics.