1.1. Properties of Connected Clauses and clefted Relative Clauses

This section provides an overview of the syntactic, semantic, and prosodic properties of Connected Clauses and clefted Relative Clauses (clefted Relatives from now on), as well as a summary of recent results showing that clefted Relatives induce garden-path effects in the absence of prosody and additional context supporting the more complex presupposition structure of Relative Clauses (e.g., the Question sentence in [2-b]). Examples (2-a) and (2-b) are repeated here as (3) and (4) respectively, with their syntactic and Focus structures marked in brackets.

Example (3) shows one prototypical it-cleft structure in English which is cued by a question. Declerck (1988), Higgins (1973) and Percus (1997) categorize such it-cleft sentences as specificational, because they specify a value (the clefted element, i.e., the gardener) for a variable (the non-clefted element, i.e., the person who painted the door). The non-clefted parts are often presupposed (e.g., Prince, 1978 and Delin, 1992), yet here we explicitly show it in the contextual question to avoid any ambiguity. Thus, in (3), the Determiner Phrase the gardener provides new information to the discourse and is in focus.² More specifically, it carries Identificational Focus that exhaustively identifies a target (Kiss, 1998; Gussenhoven, 2007). The Complementizer Phrase that painted the door, is given instead, serving as the background (e.g., Percus, 1997; Collins, 1991, 2006).

When preceded by a different question as in (4), the same string expresses a different meaning. Here, the whole Determiner Phrase (including the Complementizer Phrase) the gardener that painted the door specifies the identity of the person that called, thus providing new information and carrying identificational focus.

As indicated through bracketing and in the syntactic analyses in (5), these interpretive differences map onto well-established structural distinctions: The Complementizer Phrase that painted the door in (3) is a Connected Clause, but a restrictive Relative Clause in (4). While several analyses of it-cleft structures have been proposed (Akmajian, 1970; Higgins, 1973; Percus, 1997; Hedberg, 2000; Reeve, 2010), they all agree that the Complementizer Phrase in a typical it-cleft (i.e., Connected Clause) occupies a higher syntactic position than that in restrictive Relative Clauses, either because of extraposition or base-generation in a higher projection. A detailed review of the different syntactic analyses of clefts (which the interested reader can find in Reeve, 2010) is beyond the scope of this paper, as it would not affect our arguments and predictions. In contrast, in (4) the Relative Clause is nested within the Determiner Phrase it modifies (i.e., the gardener). The actual Connected Clause that called in this sentence occupies the same position as the Connected Clause shown in (5-a) and carries the relevant presupposition, but is reduced (Declerck, 1983b) or truncated (Hedberg, 2000) to better illustrate the ambiguity. For illustrative purposes, here we adopt the structure proposed by Hedberg (2000) to exemplify the attachment contrast.³

However, despite clefts’ prominence in the theoretical syntax and semantics/pragmatics literature (Akmajian, 1970; Higgins, 1973; Prince, 1978; Atlas & Levinson, 1981; Declerck, 1983a, 1984; Hedberg, 1990; Delin, 1992, 1995; Percus, 1997; Davidse, 2000; Hedberg, 2000; Lambrecht, 2001; Collins, 1991; Sharvit, 2003; Reeve, 2010; Den Dikken, 2013; Hedberg, 2013), the experimental work on the prosodic and processing properties of clefts is much more limited, let alone the comparison between the two structures of interest here.

In terms of the prosody of prototypical it-clefts, the shared intuition in the theoretical literature is that the clefted element that carries new information is obligatorily accented, while the non-clefted part that contains old information is typically unaccented (Prince, 1978; Delin, 1992, 1995; Gussenhoven, 2007). Few corpus-based studies analyzed the prosody of prototypical it-clefts, but they either included too few items to draw reliable generalizations (Hedberg & Fadden, 2007, American English) or focused solely on the prosody of prototypical it-clefts like (3) and/or the pragmatic function of prosody (Collins, 1991, 2006; Herment & Leonarduzzi, 2012; Bourgoin et al., 2021, British English). As a result, not only is systematic experimental work on prototypical it-clefts lacking, but the prosodic patterns associated with structurally complex clefted phrases (e.g., clefted Relatives) remain almost completely unexamined.

Arnhold (2021) highlighted such a lack of experimental work on the prosody of cleft structures and performed carefully controlled production and perception experiments, testing whether syntactically marked focus (it-clefts) differs prosodically from unmarked syntax (simple subject-verb-object structures) under different focus conditions (narrow vs. broad). While this work did not investigate the prosody of clefted Relatives, it is relevant here for providing a detailed analysis of prosodic correlates of standard it-clefts (i.e., our Connected Clauses).

It-clefts in Arnhold’s production study always had a narrow focus on the subject (- Who was buying an island? - It was the royal who was buying an island), considering that “clefts are not well-formed in broad focus” (Arnhold, 2021). Results reveal a similar prosodic pattern for narrow-focused elements in both marked and unmarked syntax (- Who was buying an island? - The royal was buying an island), in contrast to the broad-focus condition (- What happened? - The royal was buying an island).

Because the main goal of Arnhold’s (2021) study was to examine the interaction of syntactic and prosodic marking of focus, the analyses focused more on comparing the prosody of narrow vs. broad focus conditions, rather than providing detailed acoustic analyses for clefts with a narrow focus. Nevertheless, the perception study included clefts with a broad focus as auditory stimuli (- What happened? - It was the royal who was buying an island), which was briefly summarized to have “nearly identical” intonation pattern with sentences having unmarked syntax and broad focus.⁴

In sum, while the prosody literature has examined the properties of standard it-clefts, including Arnhold’s detailed comparison of the prosody of it-clefts and focus marking in unmarked syntax, no work to date has carefully examined whether and how prosody disambiguates Connected Clauses and clefted Relatives.

Similarly, to our knowledge, this ambiguity has also been ignored in the psycholinguistics literature. Guo et al. (2024b, a) is the first study investigating the processing of this type of sentence. In two silent reading studies, we showed that disambiguating towards a clefted Relative reading led to lower acceptability ratings and longer reaction and reading times than Connected Clauses. The results suggest that, similar to other instances of ambiguity involving restrictive modifiers, clefted Relatives elicit garden-path effects in both online and offline processing in the absence of explicit prosody and a supporting context.⁵

In other words, clefted Relatives are less preferred, consistent with the parser’s general tendency to avoid Relative Clauses. This parsing preference is expected because while clefted Relatives are adjuncts, Connected Clauses are “quasi-arguments” of clefts. Even when reduced, Connected Clauses are always part of clefts’ semantics, and possibly also their syntax. Connected Clauses also carry fewer unsupported presuppositions than clefted Relatives, which also explains why they are easier to parse in the absence of a supporting context. Analyzing the prosodic disambiguation of these structures, therefore, adds to the broader issue of whether and how nested garden paths are prosodically disambiguated and to what extent prosodic disambiguation reduces the processing load of restrictive modifiers and allows listeners to sidestep the garden-path effect observed in the absence of (explicit) prosody.

1.2. Clefted Relatives and Phrasing: Nested garden paths at the syntax-prosody interface

In the previous section, we showed that Connected Clauses and clefted Relative Clauses, though string-identical, exhibit clear structural and interpretive differences. We also saw that, in the absence of explicit prosodic cues or contextual information, clefted Relatives give rise to garden-path effects (Guo et al., 2024a, 2025).⁶ A central aim of this paper is to determine whether this ambiguity can be resolved prosodically, and if so, how. Before turning to empirical investigation in Section 2, we situate this question within the broader literature on the interface between prosody and syntax (this section) and between prosody and semantics (next section). While no prior work has directly examined the prosody of clefted Relatives or how they are distinguished from Connected Clauses, existing research will allow us to formulate predictions about both prosodic phrasing and prominence, grounded in their structural and interpretive differences.

At the level of prosodic phrasing, although Relative Clauses form a single syntactic constituent with the nominals they modify, they are well known to tolerate prosodic separation, resulting in syntax-prosody mismatches (Shattuck-Hufnagel & Turk, 1996). In this section, we aim to show that clefted Relatives constitute an exceptional case due to independent grammatical factors, and we predict the prosodic phrasing of Connected Clauses and clefted Relatives should align with their respective syntactic structures.

One strong argument for the existence of a separate prosodic level of representation is that syntactic structures and prosodic phrasing are not always aligned (e.g., Shattuck-Hufnagel & Turk, 1996 and Turk & Shattuck-Hufnagel, 2014). Syntax-prosody mismatches suggest that prosodic phrasing is determined by the interaction of what we will refer to as faithfulness and balance constraints which operate over and shape hierarchically organized prosodic representations. Faithfulness constraints ensure that structural information is faithfully encoded at the syntax-prosody interface; balance constraints govern the optimal size of prosodic phrases. Syntax-prosody mismatches can arise from a conflict between these two types of constraints.⁷

One example of faithfulness constraints is the Edge-Alignment constraint (Selkirk, 1986, 1996), which requires alignment of the left/right edge of a syntactic phrase with the left/right edge of their corresponding prosodic constituent. This constraint further interacts with the Wrap-XP constraint (Truckenbrodt, 1995, 1999) which demands each syntactic phrase be contained in a phonological phrase, thereby preventing the division of syntactic phrases into multiple phonological phrases. An alternative approach in the Match Theory (Selkirk, 2009, 2011) involves a set of Match constraints that do not specifically refer to left or right edge alignment and, instead, emphasize the correspondence between syntactic and prosodic constituency at clause, phrase, and word levels.

Regardless of the specific account of this mapping relation, following faithfulness constraints, a single syntactic phrase should typically align with a single prosodic phrase, while distinct constituents are mapped to separate prosodic phrases, as evidenced by prosodic boundaries. Example (6) shows one such instance described in Büring (2013), with syntactic structures indicated by square brackets and prosodic phrasing by round brackets. In (6-a), fancy books forms a single complex Determiner Phrase. There is a smaller prosodic boundary between the two words compared to (6-b), where Nancy and books are two separate Determiner Phrases with a clear prosodic break in between.

Phrasing, however, is not determined solely by faithfulness principles. It is also shaped by constraints which require prosodic phrases to be neither too long nor too short. In phonological theory, constraints on the size of prosodic phrases have been formalized as well-formedness requirements. For instance, Selkirk (2000) proposed the Binary Minimum constraint, which requires a minimally binary prosodic length of a major (phonological) phrase, and the Binary Maximum constraint, which restricts a major phrase to have “at most two minor/accentual phrases” (Selkirk, 2000, p. 244). The principle of Uniformity, proposed by Ghini (1993), similarly suggests rhythmic balance across prosodic constituents, which can be achieved by modulating the number of phonological words and/or by modulating speech tempo (we will go back to Ghini’s insight in our discussion section). Constraints on the size of prosodic phrases have also been invoked to explain phrasing patterns observed in both comprehension and production in the psycholinguistics literature (e.g., Gee & Grosjean, 1983; Fodor, 1998; Watson & Gibson, 2004; Fodor, 2002; Augurzky, 2005 and Cooper & Paccia-Cooper, 1980). For example, in the comprehension literature, Frazier and Fodor (1978) proposed that sentences are parsed into units of similar length; similarly, in the literature on production, Ghini (1993) and Gee and Grosjean (1983) observed that speakers tend to produce balanced prosodic phrases and, like Frazier and Fodor, suggested that “a string is ideally parsed into same length units” (Ghini, 1993, p. 56) in speech comprehension. These so-called eurhythmic or balance constraints are motivated by insights from both phonological theories and cognitive/physiological considerations. A different take on size effects in phrasing is put forward by Watson (2002) and Watson and Gibson (2004), who suggest that the placement of intonational boundaries is partly governed by planning and recovery mechanisms in production. In this model, while phrasing is primarily determined by the syntactic and semantic properties of a sentence, speakers are more likely to produce an intonational boundary before and after large constituents (see also Breen et al., 2011 for a systematic comparison of this proposal with balance accounts).⁸

Independent of the source of size constraints, the effects of faithfulness and balance constraints do not always align. One classic example of such tension, first described in Chomsky (1965), is illustrated in (7-a). The string the cat that caught the rat that ate the cheese contains multiple nested clauses that form a single syntactic constituent, as in (7-b). However, this sentence is typically produced with a flatter prosodic structure, shown in (7-c), i.e., with phrases of similar sizes and prosodic breaks between each head noun (cat/rat) and the corresponding Relative Clauses.⁹ In what follows, we use squared bracketing ([]) to represent syntactic structure, and double vertical bars (||) to indicate prosodic phrasing.

For one family of approaches (e.g., Nespor & Vogel, 1986; Selkirk, 2000; Truckenbrodt, 1995 and Turk & Shattuck-Hufnagel, 2014), these asymmetries show that sentences have a prosodic structure that is derived from their syntactic structure but essentially can be non-isomorphic to it, and this prosodic structure directly determines speech prosody. The properties of the prosodic structure itself result from the complex interaction of faithfulness and balance principles. Separate phrasing for Relative Clauses, for instance, illustrates a case where balance is ranked above faithfulness.

Wagner (2005, 2010) proposed an alternative account of such syntax-prosody mismatches involving Relative Clauses, which essentially amounts to questioning whether they constitute a genuine case of mismatch. Wagner points out that the separate phrasing in (7-c) is compatible with an alternative syntactic analysis in which the Relative Clause is not nested within the phrase it modifies but is instead extraposed to a higher syntactic position, as shown in (8). Extraposition involves movement of the Relative Clause out of the DP it modifies and to a higher position in the syntactic structure, creating two separate syntactic constituents which match the separate phrasing in (7-c). That extraposition is a grammatical option for these sentences is demonstrated by examples like (9), which shows that temporal adverbs (e.g., yesterday) can intervene between a head noun (cat) and the Relative Clause (that caught the rat). The interpolation of temporal adverbs (which belong to the main clause and not to the Relative Clause) forces extraposition of the Relative Clause (Hulsey & Sauerland, 2006) and serves to demonstrate the availability of this alternative structure.

One important consequence of Wagner’s account is that, since extraposition is not always a grammatical option, separate phrasing for Relative Clauses should be contextually restricted, i.e., such syntax-prosody mismatch should only be observed when extraposition is licensed. Support for this account comes from Relative Clauses whose head is an idiom chunk (Wagner, 2005), for instance (10-a). It is generally assumed that extraposition is not a grammatical option for this kind of Relative Clause, illustrated in (10-b): Inserting an adverb (last year) between the head noun and the Relative forces extraposition (as shown in example [9]) and leads to degraded acceptability (Hulsey & Sauerland, 2006). Under Wagner’s (2005) account, the ungrammaticality of extraposition in these environments explains why syntax-prosody mismatches such as (10-c) are not observed.

In sum, Wagner’s account of syntax-prosody mismatches involving Relative Clauses makes a specific, testable prediction which singles it out from other accounts: If separate phrasing is the reflection of an alternative, extraposed structure, it should only be observed in environments which license extraposition.

In the remainder of this section, we argue that clefted Relatives constitute another environment where extraposition is banned and that studying the prosody of clefted Relatives and Connected Clauses provides a novel testing ground for Wagner’s account. To begin with Connected Clauses: Not only do these structures allow extraposition, but several syntactic analyses argue that they involve obligatory extraposition (e.g., Reeve, 2010). That Connected Clauses allow extraposition is visible from examples such as (11): Inserting a temporal modifier last night between the clefted noun and the Connected Clause forces extraposition and yields a perfectly grammatical structure.

Applying the same test to Clefted Relatives in (12), however, gives very different results, which provides a first indication that clefted Relatives resist extraposition.

Even more strikingly, extraposition of a clefted Relative over an intervening Connected Clause is completely ungrammatical, as shown in (13):¹⁰

These examples suggest that, unlike Connected Clauses, clefted Relatives resist syntactic extraposition and thus, syntactically, they can only be nested within the head nouns they modify. Under Wagner’s account, the ban on extraposition leads us to expect no syntax-prosody mismatches in this environment: Clefted Relative should be prosodified as a single phrase with its host, as shown in (14-a), and the separate phrasing in (14-b) should be unattested.

Empirically validating the predicted prosodic phrasing in (14-a) can thus provide a novel argument for Wagner’s extraposition account, supporting the claim that the alignment between prosody and syntax may be more systematic than previously thought, at least in the domain of Relative Clauses.

While the source of the ban on extraposition in clefted relatives is not immediately relevant here (but see Poschmann & Wagner, 2016 for an overview of factors affecting the availability of extraposition), it is tempting to speculate that it may also be related to the different attachment sites of Connected Clauses and clefted Relatives. Extraposing a clefted Relative would produce a structure indistinguishable from a Connected Clause. A rational speaker, as defined by Clifton et al. (2002, 2006) and Frazier et al. (2006), would therefore avoid such ambiguity by employing prosody and syntax in an internally consistent and rational way. This is, in essence, the account proposed by Grillo et al. (2025) to explain the prosody of nested garden-path sentences, a class of constructions to which our contrast of interest belongs. A number of recent findings on the prosodic disambiguation of nested garden paths (Grillo & Turco, 2016; Grillo et al., 2018, 2023, 2025) consistently show that nested structures tend to be prosodically grouped with their syntactic hosts in such environments. Our predictions for clefted Relatives, therefore, are in line with observations from this broader literature. For clarity and continuity, we postpone a thorough discussion of this point until the Discussion.

In sum, this section has argued that clefted Relatives offer a novel testing ground for theories of syntax-prosody mapping, particularly Wagner’s extraposition account of phrasing mismatches. Before empirically evaluating these predictions, the next section revisits the differences in information structure between clefted Relatives and Connected Clauses, demonstrating how this contrast can further inform ongoing debates concerning the interface between prosody and semantics.

1.3. Clefted Relatives and Prominence: The semantics-prosody mapping

As discussed in Section 1.1, Connected Clauses and clefted Relatives differ in their syntactic and information structures. In ‘standard’ clefts (Example [3]), only the clefted DP is in focus, while the Connected Clause introduces backgrounded material and can be optionally omitted. In contrast, because clefted Relatives are syntactically nested within the DP they modify, the entire DP (including Relative Clause) receives focus, and any background information is introduced by an optionally realized (but always presupposed) Connected Clause (Example [4]). This contrast makes clefted Relatives a compelling test case for theories of the prosody-semantics interface. While different approaches make similar predictions for Connected Clauses, their predictions diverge on prominence assignment within clefted Relatives, providing a unique opportunity to evaluate competing accounts. Before addressing this question, we first review the broader literature on the prosodic realization of focus to situate clefted Relatives within the relevant theoretical context. We then examine different families of accounts of the semantics-prosody mapping and consider their respective predictions for the prominence patterns of our target structures.

It is commonly observed that the meaning of a sentence varies with the context in which it occurs and that these differences in meaning are often expressed acoustically. For example, (15) illustrates that the same string of words (John ate pizza) may be realized with prosodies corresponding to the different information that each answer provides. In (15-a), the word pizza provides new information, which answers the question and most naturally receives prosodic prominence. Conversely, John carries new information when answering the question in (15-b) and would typically be acoustically emphasized in that context.

Semantically, this difference in acoustic realization can be captured with Rooth’s (1985) theory of alternative semantics: Focus marks the presence of alternatives relevant to the interpretation of a sentence. A focused constituent introduces a set of contextual alternatives, which are evaluated against the current question under discussion or contrastive expectations. This helps explain why focused elements are typically accented—they signal new, contrastive, or informative content—whereas backgrounded or given material remains unaccented and prosodically reduced (see also Hoeks et al., 2023; Rooth, 1985, 1992; Gussenhoven, 2007; Féry & Krifka, 2008; Breen et al., 2010).

While the meanings and pronunciations of (15-b) and (15-a) differ, the prosodic prominences associated with the focused information in each retain some similarities. For example, such prominence is typically characterized by longer duration, higher F0, and higher intensity on focused elements relative to given elements, at least for some languages (including American and British English), and despite some across-experiment differences in the consistency of these features (Cooper et al., 1985; Eady & Cooper, 1986; Xu & Xu, 2005; Breen et al., 2010; Arnhold, 2021; Lee et al., 2015). Xu (2019) further summarized the most consistent correlates of focus in English as i. post-focal compression of F0, and ii. longer duration and higher F0 on the focused word. Taking (15) as an example, we should expect to find a longer duration and higher F0 for John in (15-b) than (15-a) (and vice versa for pizza), and a lower F0 for ate in (15-b) than (15-a).

However, on the post-lexical level, whether this semantics-prosody mapping is direct (Cooper et al., 1985; Eady & Cooper, 1986; Lieberman, 1963; Pell, 2001; Xu & Xu, 2005; Xu et al., 2015) or indirect (e.g., Ladd, 2008; Pierrehumbert, 1980; Gussenhoven, 2004 and Wagner, 2005) is a contentious matter. From the perspective of direct approaches, aspects of meaning in a sentence, such as emotion, focus, or grouping, are directly linked to a set of acoustic features. In contrast, for indirect approaches, meaning is first mapped onto linguistic representations (syntactic and/or prosodic), over which different phonological rules are applied and which ultimately determine the acoustic properties of utterances. Whether that intermediate level is syntactic or metrical (as the prosodic hierarchy of intonational phonology models) is beyond the scope of this paper; here we are primarily concerned with testing the predictions of direct and indirect accounts.

One issue with this literature is that the same term is sometimes used to refer to different concepts, making it difficult to compare and validate the theories’ predictions. The use of “focus” is one such example. While both accounts consider focus an important factor that affects sentence prosody, indirect approaches generally use the term as we have defined it above, i.e., as marking the presence of alternatives relative to the interpretation of the sentence. Direct approaches, on the other hand, are somewhat internally inconsistent. Early direct-approach studies such as Cooper et al. (1985) and Eady & Cooper (1986) (and implicitly in Pell, 2001) use the term (main) focus to indicate “primary stress” or prominence in a sentence, which relates to the prosodic aspect. Xu (2005, 2019) and Xu et al. (2015) instead see the word focus as a communicative function. Particularly, Xu defines focus as the part of the information that speakers intentionally emphasize to convey its importance to the listener, while newness “is associated only with memory retrieval” (Xu, 2019, p. 330). However, this definition does not clarify how speakers determine which part of the information is important or, crucially, how to predict which elements are focused in a sentence. To address this issue, in the following discussion, we will reserve the word Focus to refer to the semantic aspect associated with a word/phrase’s contribution to information structure, and prominence when referring to the acoustic properties of focused elements.

Under this definition, we expect the two accounts to make aligned predictions for sentences like (15-a) or (15-b), based on the prosodic pattern described above: The word carrying focus has more prominent acoustic features than words that are given or carry a lower informational load. This is because, for cases like (15-a) and (15-b) where focus falls on a single word, space is too limited to decide whether information structure directly determines the prominence of that word or whether this is mediated by intermediate phonological representations (e.g., a +accent feature associated with focused elements in a phonological representation). This explains why the two approaches are often hard to differentiate, as many production studies used such simple structures as items, where their predictions align. Even more sophisticated experimental work, such as Breen et al. (2010), fails to clearly distinguish between the two accounts, despite offering strong evidence on the prosodic correlates of focus.¹¹

Distinguishing the two accounts, therefore, boils down to not a matter of precision of measurement but to constructing linguistic contrasts for which the two accounts indeed make different predictions. We suggest that this can be achieved by leveraging structural nesting, as in (16). Here, the DP pizza with anchovies contains nested materials, and in this context, each part of it carries new information and is in focus. Therefore, direct accounts should predict overall higher prominence for all pitch-accentable words across the focused phrase (i.e., pizza and anchovies). Indirect accounts, in contrast, make specific and localized predictions about accent assignment, which, at least in this “all new” context, are relatively independent of the informational content of a word and are determined based on rules that apply to syntactic and phonological representations (e.g., the Nuclear Stress Rule, Chomsky & Halle, 1968; Zubizarreta, 2016). Specifically, the Nuclear Stress Rule would predict higher prominence to fall on the most deeply nested word anchovies compared to pizza.¹²

Similar reasoning can also be applied to our target structures, repeated in (17-a) and (17-b). Here, both families of accounts predict greater prominence on the clefted element in (17-a), which is in focus and carries new information, than on the Connected Clause, which is not in focus and carries given/old information. However, their predictions diverge for clefted Relatives in (17-b). Since the entire DP carries focus, we expect direct accounts to predict higher prominence for all pitch-accentable words (gardener, painted, and door). As they contribute equally to conveying new information, there should be no difference in the prominence levels among elements in the Relative Clause. In contrast, indirect accounts more specifically predict the location of the nuclear stress to fall on the most deeply nested word in the structure, i.e., door in (17-b), resulting in a prominence asymmetry even when all elements in the Relative Clause are in focus.

One key reason why we chose to test the two accounts using Connected Clauses and clefted Relatives is that their focus structures are immediately accessible. An additional reason is that in (16), the DP is globally unambiguous.¹³ Lack of ambiguity prevents us from asking a number of questions we can afford when testing clefted Relatives, including the contribution of prosody to the disambiguation of this understudied contrast, in addition to the questions about phrasing and the syntax-prosody interface discussed in Section 1.2.

1.4. The present study

In this study, we will address the broader questions of whether and how the contrast of clefted Relatives and Connected Clauses is prosodically disambiguated in production (Experiment 1), and whether listeners are sensitive to these prosodic differences in comprehension (Experiment 2). Based on our impressionistic judgments, previous results on similar structural contrasts (e.g., Grillo & Turco, 2016; Grillo et al., 2018; Grillo et al., 2025 and Poschmann & Wagner, 2016), and the assumption that different syntactic and semantic properties of sentences are reflected in their prosody, we expect Connected Clauses and clefted Relatives to be produced with different prosodic patterns. Similarly, we expect prosody to strongly ameliorate comprehension of clefted Relatives, a structure which we have independently shown to pose severe problems for the parser in silent reading (Guo et al., 2024a, 2025). These questions, however, can only be reasonably addressed through careful experimental work.

In addition, as clarified in the previous sections, the structural and interpretive contrast between the two structures allow us to address the following, more specific theoretical questions about the interface between prosody and syntax/semantics:

Q1. Are the respective phrasing properties of clefted Relatives and Connected Clauses consistent with the predictions of Wagner’s (2005, 2010) extraposition account and Grillo et al.’s (2025) Rational Speaker account? These accounts predict that clefted Relatives and their hosts will be produced as a single prosodic phrase and that Connected Clauses, on the other hand, should be prosodified separately. If our production experiment showed that speakers consistently produce separate phrasing for clefted Relatives and their head nouns, this would provide strong evidence against these accounts.

Q2. Do the prominence properties of these structures align better with the predictions of direct or indirect accounts of the semantics-prosody mapping? The two families of accounts make similar predictions for Connected Clauses (i.e., higher prominence on the clefted Noun and deaccenting on the Connected Clause itself). Their predictions differ for clefted Relatives, i.e., only indirect accounts predict localized higher prominence on the final, most embedded Noun of the clefted Relatives.

2.1. Participants

Eight native British English speakers originating from different regions of the UK participated in the experiment in a soundproof booth at the University of York. One participant was excluded due to producing too many errors and hesitations, leaving us seven participants for analysis (age range = 28–36, mean = 32.3, SD = 3.5, women = 4). Participants gave their informed consent and were paid for their participation. Each subject participated in two sessions that were separated by at least one week, and each session focused on a single critical structure. This study was approved by the Ethics Committee of the Department of Language and Linguistic Science, University of York.

2.2. Materials

Each condition included 24 Question-Answer pairs as in (18-a) and (18-b). Each target sentence was structured as It was + the NP1 + that + was + Verb + the NP2 (NP stands for Noun Phrase). The questions in the Connected Clause condition directly asked about the agent of the Verb Phrase in the answer, while the questions in the Clefted Relative condition were kept constant as in (18-b). This design ensured grammatical disambiguation of the two readings: Matching/mismatching the content of the question and the Complementizer Phrase forced either a Connected Clause reading (match) or a clefted Relative reading (mismatch) of the target sentence.¹⁴ Target sentences are prosodically controlled across items where we kept the number of syllables and the position of lexical stress constant within each region: NP1 is always trisyllabic with lexical stress on the first syllable, denoting a profession or property of people; the Verb is transitive and disyllabic with lexical stress on the first syllable, and it forms a phrase with NP2, which is monosyllabic. All sentences were in the past tense.

Experimental items were interspersed with 48 filler sentences that varied in their syntactic structures, but approximately matched experimental items in length. Twelve fillers were also preceded by context questions to make half of all items form Question-Answer pairs.

2.3. Procedure

Before the experiment, participants were asked to fill in a questionnaire about their language background and other demographic information. They were then instructed by a researcher to silently scan the entire (question and) sentence before reading aloud. They should then produce the question (if any) and sentence naturally and fluently at normal speed. When presented with a Question-Answer pair, participants were instructed to imagine themselves in a conversation: asking the Question, as if seeking information, and responding with the Answer, as if informing someone unaware of the answer. Items were automatically presented on a computer screen, and participants were recorded with a headset microphone connected to the PC using the software ProRec 2.4 (Mark Huckvale, University College London), at a sampling rate of 44.1 kHz and 32-bit resolution.

Experimental stimuli were initially divided into two lists, each containing only one structure type (i.e., either Connected Clause or clefted Relative) to ensure that participants saw only one condition per session. This was a within-subjects design: All participants completed both conditions across two sessions, with the order of conditions counterbalanced across participants. All items were pseudo-randomized within each list. To control for potential sequence effects, two additional lists were created by reversing the item order of the originals, leading to a total of four lists. Experimental items were separated by at least one filler item in between.

Every session started with four practice items, followed by 24 experimental items interspersed with 48 fillers, leading to a total of 76 items for each participant in each session. The entire experiment lasted approximately 40 minutes.

2.4. Data processing and analysis

Before data analysis, segmentation was performed automatically using the Montreal Forced Aligner (McAuliffe et al., 2017). Duration, F0, and intensity were automatically detected by means of scripts run in Praat software (Boersma, 2001). The results of the automatic procedure were checked and manually corrected (blinded to the condition the sentence belonged to) in case of errors.¹⁵

To address the broader question of whether and how speakers prosodically disambiguate Connected Clauses and clefted Relatives, we analyzed the acoustic properties of productions, following Eady & Cooper (1986), Breen et al. (2010) and Arnhold (2021). Firstly, to examine cues for early disambiguation, we calculated the temporal alignment of F0 maximum within the accented syllable of NP1 (i.e., the first syllable), which hereinafter is referred to as peak alignment. This measurement is expressed as a percentage of the syllable’s duration and is calculated as follows. Values above 100 indicate that F0 peak on NP1 occurred after the end of the first syllable, whereas values between 0 and 100 reflect peaks that fall within the syllable.

• (Time_{F0 maximum} – Time_{1st syllable onset}) / (Time_{1st syllable offset} – Time_{1st syllable onset}) ∗ 100

On the specific issue of phrasing differences between the two structures raised in Question 1, if clefted Relatives are prosodified with their hosts, as predicted by Wagner (2005) and Grillo et al. (2025), then we expect to observe a stronger prosodic boundary between NP1 editor and the Complementizer that in Connected Clauses than in clefted Relatives. Comparing pre-boundary lengthening between NP1 and the Complementizer for the two structures should allow us to diagnose the strength of this boundary. Thus, the following measurement was extracted:

• Raw duration of the three syllables of NP1, in milliseconds (ms)

To further examine the general prosodic differences between the two structures, we extracted and analyzed the following three measurements for every pitch-accentable word (i.e., NP1 editor, Verb working, and NP2 shift). Also, to address Question 2, we focused on the comparison between NP2 vs. Verb in the two structures, where we expect the two families of accounts to make different predictions.

• Raw duration of words, in ms

• F0 range (F0_maximum – F0_minimum), in semitones (st)

• Mean intensity, in decibels (dB)

Additionally, we conducted a post hoc analysis comparing the F0 peak at NP2 (shift) and NP1 (editor) across the two conditions. This measurement (hereinafter referred to as F0 scaling) was intended to indicate, among other acoustic cues, whether the focus on NP2 was contrastive or non-contrastive, and was calculated as follows (in semitones):

• 12 × log₂ (F0 Max_NP2/F0 Max_NP1)

If NP2 carries contrastive focus, we expect a higher F0 peak on NP2 than on NP1, resulting in a positive F0 scaling. If the focus on NP2 is non-contrastive, the F0 peak on NP2 should be lower than on NP1, leading to a negative F0 scaling. Values at zero would indicate a same level of F0 peak on the two NPs.

We analyzed acoustic properties in R (v. 4.3.1) using RStudio (v. 2023.6.2.561). Linear mixed-effects regression models in the lme4 package (Bates et al., 2015) were used for these measurements with the maximal random effects structure while allowing for model convergence (Barr et al., 2013). ANOVA comparisons were used to examine the contribution of fixed effects to models, and emmeans (Lenth, 2023) for post hoc analyses on simple structural effects at each region where interactions were observed. Similar analyses were applied to the other analyses discussed later in this paper.

For models relating to the broader question and Question 1, in which Structure is the only independent variable, sum coding was applied for factor contrasts of Structure. Connected Clauses and Relative Clauses were set to –0.5 and 0.5 respectively. For models testing Question 2 in which both Structure and Region are independent factors, we set the fixed effects to Regions (NP2 vs. NP1 vs. Verb, the element in italics indicated the reference level), Structures (CC vs. RC), and the interaction between them.

2.5. Results

To illustrate the general pattern of the recorded utterances, one pair of examples from Experiment 1 is shown in Figures 1 and 2, corresponding to examples of pitch contours of the Connected Clause and Relative Clause recordings respectively. Time-normalized contours of the two structures are provided in Figure 1 in Appendix A.

Figure 1

Example of waveform and pitch contour of a Connected Clause production, segmented word-by-word.

Figure 2

Example of waveform and pitch contour of a clefted Relative Clause production, segmented word-by-word.

2.5.1. F0 peak alignment at NP1

The temporal alignment of the F0 peak to the length of the first syllable in NP1 across the two structures is illustrated in Figure 3. Analysis showed a strong tendency for delayed peak alignment at NP1 in Connected Clauses compared to clefted Relatives (β = –11.42, SE = 5.69, t(6.01) = –2.01, p = .09). Post hoc analysis found that one participant showed the opposite pattern in the delayed peak, while the remaining six speakers illustrated a significantly delayed peak at NP1 in Connected Clauses compared to clefted Relatives (β = –16.63, SE = 2.98, t(13.27) = –5.58, p < .001).

Figure 3

Box plot of temporal alignment of F0 peak relevant to the first syllable of NP1 (editor) in Connected Clauses and clefted Relatives. The box spans the interquartile range (25th to 75th percentile), and the median is in between. Whiskers represent the most extreme values within 1.5 times the interquartile range. Jittered points represent individual data observations.

2.5.2. Raw duration of syllables in NP1

Differences in raw durations of the three syllables in NP1 (e.g., editor) are shown in Figure 4 and Table 1. We observed significant structural effects in the same direction at the second (β = –8.18, SE = 2.42, t(301.07) = –3.38, p < .001) and the third (β = –38.89, SE = 13.49, t(5.94) = –2.88, p = .028) syllables which were shorter in Relative Clauses than Connected Clauses. No significant differences between structures were found at the first syllable in NP1 (p = .40). The overall pattern suggests a similar duration for the first syllable in the two structures, but shorter second and third syllables in Relative Clauses.

Table 1

Average duration in ms of syllables in NP1 (editor) by structure (with Standard Errors in parentheses).

	Syllable 1	Syllable 2	Syllable 3
CC	197(6.95)	91.8(3.38)	200(19.3)
RC	195(7.93)	83.5(4.17)	161(9.38)

Figure 4

Duration in ms of syllables in NP1 (editor) by structure. Error bars show Standard Errors of the mean. The x-axis represents the syllable position within NP1, where “1st”refers to the first syllable, “2nd” to the second, and “3rd” to the third.

2.5.3. Raw Duration of words

Localized effects of Structure were found for word duration as presented in Table 2 and Figure 5. The model comparison showed a significant interaction between Region and Structure (AIC = 11196, χ2(2) = 86.23, p < .001). Crucially, the difference in structural effects at the Verb was significantly smaller than at NP2 (β = –36.89, SE = 8.90, t(952.09) = –4.15, p < .001), indicating localized effects as predicted by the indirect account. This is further supported by post hoc analyses which revealed that NP2 was produced significantly longer in Relative Clauses than in Connected Clauses (β = 51.7, p < .001), while the pattern was opposite for NP1, which was longer in Connected Clauses (β = –36.9, p = .002). No differences in the duration of the Verb were observed (p = .11).

Table 2

Average duration in ms of NP1 (editor), Verb (working), and NP2 (shift) by structure (with Standard Errors in parentheses).

	NP1	Verb	NP2
CC	474(19.4)	305(6.8)	354(12.8)
RC	438(18.3)	320(10.8)	405(21.9)

Figure 5

Duration in ms of NP1 (editor), Verb (working), and NP2 (shift) by structure. Error bars show Standard Errors of the mean.

2.5.4. F0 Range

For F0 range, we observed a similar localized pattern to duration, illustrated in Table 3 and Figure 6. Models revealed a significant interaction effect between Structure and Region (AIC = 4344.7, χ2(2) = 42.93, p < .001). Importantly, the difference in the F0 range between the two structures was significantly smaller at Verb than at NP2 (β = –2.26, SE = 0.43, t(869.95) = –5.29, p < .001). This is supported by post hoc analysis where F0 range at NP2 was significantly larger for Relative Clauses than Connected Clauses (β = 2.14, p < .001). No significant differences were found between the two structures at NP1 (p = .36) or the Verb (p = .77).

Table 3

Average F0 range in st of NP1 (editor), Verb (working), and NP2 (shift) by structure (with Standard Errors in parentheses).

	NP1	Verb	NP2
CC	5.68(0.99)	2.74(0.33)	1.89(0.28)
RC	5.11(0.75)	2.64(0.16)	4.12(0.74)

Figure 6

F0 range in st at NP1 (editor), Verb (working), and NP2 (shift) by structure. Error bars show Standard Errors of the mean.

2.5.5. Mean intensity

Similarly, localized effects in intensity were observed as shown by Table 4 and Figure 7. Models showed significant interactions between Structure and Region (AIC = 5239.7, χ2(2) = 10.51, p = .005). However, we did not find significantly different structural effects at the Verb compared to at NP2 (p = .13). The post hoc analysis showed a significant effect of Structure at all three regions in the same direction: Relative Clauses were produced with higher intensity than Connected ones at NP1 (β = 3.07, p = .017), Verb (β = 3.92, p = .011) and NP2 (β = 4.60, p = .007).

Table 4

Average intensity in dB of NP1 (editor), Verb (working), and NP2 (shift) by structure (with Standard Errors in parentheses).

	NP1	Verb	NP2
CC	61.7(1.43)	57.8(1.13)	52.8(0.97)
RC	64.8(1.51)	61.8(1.40)	57.4(1.37)

Figure 7

Intensity in dB at NP1 (editor), Verb (working), and NP2 (shift) by structure. Error bars show Standard Errors of the mean.

2.5.6. F0 scaling

Figure 8 illustrates the F0 scaling between NP2 (e.g., shift) and NP1 (e.g., editor) for both Connected Clauses and clefted Relatives. We observed a significant effect of Structure: Relative Clauses had significantly higher F0 scaling than Connected Clauses (β = 3.82, SE = 1.56, t(5.92) = 2.47, p = .049). This suggests that the difference in F0 peak between NP2 and NP1 was larger in Connected Clauses than in clefted Relatives. Crucially, both conditions showed an average negative F0 scaling, meaning that NP2 had a lower F0 peak than NP1 even in Relative Clauses, suggesting a non-contrastive focus on NP2 in clefted Relatives.¹⁶

Figure 8

F0 scaling at NP2 (shift) vs. NP1 (editor) by structure in semitones. Error bars show Standard Errors of the mean. According to the formula used above, positive values indicate a higher F0 peak on NP2 than NP1; negative values indicate a lower peak on NP2 than NP1; zero indicates equal peak values for the two regions.

3.1. Participants

Sixty-four native speakers of English (30 women) located in the US (age range = 20-to-50, mean = 33.8, SD = 8.1) were recruited via the online recruitment platform Prolific (www.prolific.com). We used recruitment filters to ensure that all participants had no language, vision, or hearing-related disorders. All participants gave their informed consent and were compensated for their participation.

3.2. Materials

This experiment contained 24 experimental stimuli, each comprising two parts: a short written context ending with a Wh-question (similar to the contextual questions in Experiment 1) and an audio sentence acting as an answer. One stimulus is shown in example (19). We employed a 2 × 2 within-subjects design crossing Context and Prosody: The written context and question were designed to elicit either a Connected Clause or Relative Clause interpretation of the target sentence. The target stimuli were sentences from Experiment 1, produced by a trained phonetician with either Connected Clause or clefted Relative prosody.

Specifically, the Connected Clause contexts (including questions), for instance (19-a), were manipulated such that the first noun phrase in the answer (e.g., the editor) was in focus, providing new information. This is consistent with Experiment 1, as shown by the example audio item with pitch contour in Figure 9. In contrast, to ensure a natural flow of context and target questions and sentences, the clefted Relatives context (19-b) introduced two alternative referents for the head of the clefted DP, each performing different actions (in all items we took care to avoid repetition of lexical content in the context and target sentences). While the whole DP was given (entailed by the context), it is still in focus. In the target sentence, the noun introduces a set of alternatives (e.g., editor) which are given in the context, and the Relative Clause provides a restriction over that set. This is a notable difference with the design of Experiment 1, in which the head noun (and the Relative Clause) was only introduced in the target sentence (recall that the context question was: Which one of them was identified?). Due to this difference in information structure, the prosody of the Relative Clause stimuli in Experiment 2 differed slightly from that observed in Experiment 1. Although modeled on the general prosodic patterns of Experiment 1, the speaker in Experiment 2 was instructed to produce the target sentences, keeping in mind the contextual cues provided to participants in the comprehension task. This resulted in a less pronounced tonal excursion at NP1 in Experiment 2 compared to Experiment 1, visible in the pitch contour example in Figure 10.

Figure 9

Example of waveform and pitch contour of a Connected Clause prosody in Experiment 2, segmented word-by-word.

Figure 10

Example of waveform and pitch contour of a clefted Relative Clause prosody in Experiment 2, segmented word-by-word.

Experimental items were balanced for condition using the Latin Square design and were interspersed with 36 fillers, preceded by 3 practice items. In total, each participant completed 63 trials. To ensure participants’ attentiveness, half of each of the experimental and filler items contained comprehension questions that targeted different parts of the context to avoid strategic reading of the context. The proportion of Yes and No answers to the comprehension questions was balanced.

3.3. Procedure

This experiment followed a paradigm in Arnhold (2021) and was performed on the Gorilla Experiment Builder (www.gorilla.sc). In each trial, after reading the context and question, participants listened to a recording with either matched or mismatched prosody to the given context as shown in example (19). Next, they were asked to judge whether they thought the audio sentence was an acceptable answer for the context and question by choosing Yes or No. Every judgment was followed by a graded confidence rating that asked about their certainty in that judgment, namely Not confident, Somewhat confident, or Very confident. Finally, participants needed to answer a comprehension question, if any. The experiment took approximately 30 minutes to complete.

3.4. Data processing and analysis

To remove artifacts in the results, trials with a Reaction Time of less than 200 ms (too short to be a realistic response time) or longer than 10000 ms (unreasonably long response time for this task, potentially indicative of distraction) in the acceptability judgment task were excluded from data analysis. This led to a total of 52 items being removed, accounting for 3.38% of the data. Statistical analysis focused on two measurements.

Binary Acceptability. We analyzed the binary acceptability responses (Yes vs. No) using generalized logistic mixed-effects models (GLMER) in the lme4 package (Bates et al., 2015) with a binomial distribution. Fixed effects included Context (Connected Clause-leading vs. clefted Relative-leading), Prosody (Cooperative vs. Conflicting), and their interactions, with a maximum random effects structure while allowing for model convergence. The factor contrasts were set to be (0.5, –0.5) for both factors, where the Connected Clause-leading Context and the Cooperative Prosody weighted –0.5.

Combined 6-pt responses. Apart from binary acceptability responses, we also modeled the responses by combining the binary acceptability data with graded confidence ratings. Such a combination led to a 6-point scale, ranging from 1-Very confident unacceptable to 6-Very confident acceptable, which closely mimics a Likert scale. For statistical analysis, we employed the cumulative link mixed-effects models (CLMM) in the ordinal package (Christensen, 2023), considering the ordinal nature of the data. We used the same fixed and random effects structure as in the GLMER model.

3.5. Results

Binary acceptability. As illustrated in Figure 11, the acceptability rating was significantly lower for the Conflicting than Cooperative prosody (β = –1.74, SE = 0.19, z = –9.03, p < .001) and also lower for clefted Relative-leading contexts than Connected Clause-leading ones (β = –1.00, SE = 0.33, z = –3.01, p = .003). Interestingly, we found a significant interaction between Prosody and Context on the acceptability score: The difference in acceptability between the two Prosody conditions was greater under the clefted Relative-leading Context than the Connected Clause one (β = –1.38, SE = 0.34, z = –4.09, p < .001). The interaction effect is also supported by post hoc analyses that showed no significant difference in acceptability between the two contexts when the Prosody was Cooperative (p = .44); with Conflicting prosody, on the other hand, acceptability was significantly lower for clefted Relative-leading contexts than Connected Clause-leading ones (β = –1.69, SE = 0.34, z = –4.92, p < .001).

Figure 11

Proportion of Yes answers in acceptability judgment by condition. Error bars show Standard Error of the means.

Combined 6-point responses. Figure 12 presents the distribution of the 6-point rating across conditions. Overall, the pattern is consistent with the binary data. Conflicting Prosody to the Context received significantly lower ratings than the Cooperative one (β = –0.76, SE = 0.21, z = –3.62, p < .001), while no significant difference between Contexts was found (p = .77). Similar to binary results, 6-point responses also illustrated a significant interaction between Prosody and Context on the ratings (β = –1.70, SE = 0.33, z = –5.20, p < .001): The difference in acceptability between Conflicting and Cooperative Prosody was greater for a Relative Clause-leading Context, compared with Connected Clause-leading context.

Figure 12

Distribution of the combined 6-point ratings in acceptability judgment by condition, ranging from No-Very confident to Yes-Very confident. Error bars show the Standard Error of the means.

To summarize, Experiment 2 asked whether listeners are sensitive to the prosodic differences between Connected Clauses and clefted Relatives identified in Experiment 1. A main effect of Prosody showed that listeners use prosodic cues to disambiguate the two structures. Additionally, the interaction effect shows that listeners are more likely to judge the clefted Relative prosody as acceptable in the context of Connected Clauses than they are to accept the prosody of Connected Clauses in a context that licenses clefted Relatives. We comment on these results in the General Discussion.

A. Additional measurements for Experiment 1

Figure 1 shows time-normalized intonation contours of clefted Relative Clauses and Connected Clauses productions from Experiment 1, expressed in semitones relative to 100 Hz. F0 measurements were extracted using a Praat script that identified labeled intervals in the TextGrid and divided each interval into 20 equal-length sub-intervals, resulting in 21 pitch values per word. The extracted F0 values were then converted to semitones and visualized in R (v. 4.3.1). The contours reflect mean values across participants and items. The divergence between the two contours, especially at NP2 (e.g. scene), illustrates distinct prosodic patterns associated with the two structures.

Figure 1

Time-normalized intonation contours of Connected Clause (CC) and clefted Relative Clause (RC) productions in semitones. X-axis labels represent an example target sentence, shown word by word.

Figure 2 illustrates the by-participant F0 Scaling patterns in the two structures.

Figure 2

By-participant mean F0 scaling (in st) for Connected Clause (CC) and clefted Relative Clause (RC) productions in Experiment 1. Positive values indicate a higher F0 peak on NP2 than NP1; negative values indicate a lower peak on NP2 than NP1; zero indicates equal peak values for the two regions.

B. List of experimental items from Experiment 1

Response sentences from experimental items used in Experiment 1. Every sentence was preceded by a context question that either elicits a Connected Clause or clefted Relative Clause reading of the sentence. For example, for item 1, the question of the Connected Clause condition is Who was leading the talk?, while the clefted Relative Clause condition is Which one of them was identified?.

1. It was the manager that was leading the talk.

2. It was the monarchist that was rowing the boat.

3. It was the murderer that was locking the door.

4. It was the maniac that was waiting the trial.

5. It was the auditor that was dealing the drug.

6. It was the humorist that was leaving the scene.

7. It was the governor that was running the gang.

8. It was the moralist that was guarding the truth.

9. It was the messenger that was walking the dog.

10. It was the modeler that was humming the tune.

11. It was the editor that was working the shift.

12. It was the royalist that was hunting the deer.

13. It was the bureaucrat that was reading the form.

14. It was the mortgager that was viewing the file.

15. It was the novelist that was riding the mare.

16. It was the officer that was writing the case.

17. It was the lecturer that was moving the desk.

18. It was the journalist that was launching the book.

19. It was the analyst that was heating the room.

20. It was the medalist that was hiding the score.

21. It was the carpenter that was making the tool.

22. It was the gardener that was loading the car.

23. It was the loyalist that was holding the sign.

24. It was the lumberer that was learning the trade.

C. List of experimental items from Experiment 2

Context questions and audio stimuli in written version from experimental items used in Experiment 2. Participants either read a context eliciting a Connected Clause answer (CC Context) or a context targeting a clefted Relative Clause answer (RC Context), and then listened to an audio stimulus producing the sentence with either a Connected Clause or clefted Relative Clause prosody.

1. CC Context: You just started working in a big company and you were confused as to who was in charge of your first meeting, so you asked an experienced colleague: Who was leading the talk?

   RC Context: You just joined the company where your friend works, which has two managers. One is very vocal and deals with operations and the other very silent and tends to take a less prominent role. After a meeting, you want to know which one your friend had trouble with, so you ask: Which one of them did you have issues with?

   Sentence: It was the manager that was leading the talk.

2. CC Context: The Reality TV show you were watching with your friend was chaotic, there was a variety of famous people doing weird things. You got distracted and lost track of who was doing what, so you asked your friend: Who was rowing the boat?

   RC Context: You went down to the river bank with your friend and there were two famous monarchists, one was clapping non-stop while the other worked really hard. Back at home, your friend mentioned that one of them was in a TV show he happened to work on and you wanted to know who, so you asked: Which one of them was on the TV show?

   Sentence: It was the monarchist that was rowing the boat.

3. CC Context: You and your friend were watching a horror film that he had already watched several times. There was a scene where the front door of the house was suddenly shut from outside accompanied by the sounds of keys. After a while you got really curious and wanted to know who did that, so you asked your friend: Who was locking the door?

   RC Context: You and your friend were watching a horror film in which two murderers acted as a killing team. At one point one of them is holding the victim while the other is securing the house. You lost track of things and can’t remember which one of them was married to the policewoman, so you asked: Which one of them was married to the policewoman?

   Sentence: It was the murderer that was locking the door.

4. CC Context: In the past few weeks, there had been several arrests of hardened criminals in your town. The local newspaper you read this morning said that a major criminal trial was starting today, but no details were given. Later on at the local courthouse, you forgot who was related to that particular case, so you asked your friend who works there: Who was awaiting the trial?

   RC Context: There were two notorious maniacs in your town, one of them was already captured and another was still at large. You heard that one of them had died suddenly but you didn’t catch the details, so you asked your friend: Which one of them died?

   Sentence: It was the maniac that was awaiting the trial.

5. CC Context: You were playing a card game at a company event and you didn’t realize until after the last round that the person who shuffled is not from your office group, so you ask an experienced colleague: Who was dealing the cards?

   RC Context: You and your colleague were playing a card game at a company event and met two auditors from another group. One of them joined the game, the other one said he’d rather just watch. After the game, your friend told you that one of them used to live with him, so you asked: Which one of them did you use to live with?

   Sentence: It was the auditor that was dealing the cards.

6. CC Context: You were watching a musical with your friend. There was a commotion in the seats around you that distracted you as one of the performers was escorted off stage, so you asked your friend: Who was leaving the scene?

   RC Context: You were taking a walk with your friend when you saw two popular humorists from your town when all of a sudden there was a car accident. One the humorists, stayed put and called the police, the other one was rushing away. Later on you were wondering which one of them performed in the most recent musical, so you asked: Which one of them was in the musical we saw last week?

   Sentence: It was the humorist that was leaving the scene.

7. CC Context: You and your friend have been watching a TV series about politics. Yesterday the trailer indicated that the secret leader of a criminal group would be revealed in today’s episode. However, you missed the show, so later that night you asked your friend: Who was running the gang?

   RC Context: You and your friend have been watching a TV series about politics. In yesterday’s episode there were two governors having conflicts with each other. One of them was secretly leading a criminal group while the other focused on reforming criminal law. You missed the new episode today and when your friend mentioned a scene with FBI questioning, you asked your friend: Which one of them was interrogated?

   Sentence: It was the governor that was running the gang.

8. CC Context: Your friend was covering a beauty pageant for the local newspaper and told you that there was an important guest counting the vote. You are curious, so you ask: Who was counting the vote?

   RC Context: Today you went to a local election event with your friend. It was crowded and there were also two congressmen present. One of them was talking to the volunteers while another kept tallying the number of ballots. On the way back, you wanted to know which one of them had worked with your friend, so you asked: Which one of them did you work with?

   Sentence: It was the congressman that was counting the vote.

9. CC Context: As a newcomer to an advertising company and you don’t know many people there yet. One day, while you are hanging out with one of your colleagues who has worked there for a long time, you saw somebody from work passing by with a Golden Retriever, and out of curiosity, you asked your friend: Who was walking the dog?

   RC Context: You were hanging out with your colleague from your big advertising company when two of the company’s messengers walked by. One of them was carrying a heavy bag and the other had a Golden Retriever. After they left, you remembered that years ago your friend said he dated one of them but you can’t remember clearly, so you asked: Which one them did you use to date?

   Sentence: It was the messenger that was walking the dog.

10. CC Context: You are attending a workshop with your friend. You heard someone produce the melody of your favourite song. However, without glasses you can’t see clearly, so you asked your friend: Who was humming the tune?

    RC Context: You and your friend were attending a 3D printing workshop where there were two instructing modelers. During the practice session, one kept producing the same melody while the other was very quiet. After the workshop, you remembered that your friend said that one of them was his roommate. You wanted to know who, so you asked your friend: Which one did you use to live with?

    Sentence: It was the modeler that was humming the tune.

11. CC Context: There was a severe incident last night at your printing company. As the site manager, you need to know all the details, but you can’t remember who was in at the time, so you asked your colleague: Who was working the shift?

    RC Context: The section of the newspaper you recently joined has two editors, one of them has been on leave for the past two months and the other you met just as they were finishing-up for the day. As you enter the office kitchen, you hear that one of them got a prize. You want to know who so you ask the colleague: Who got a prize?

    Sentence: It was the editor that was working the shift.

12. CC Context: You and your friend are watching a political debate on animal welfare on TV and you remember that your friend’s newspaper recently published compromising photographs of a number of the debaters. You remember one of the debaters was involved in hunting, but you can’t remember who, so you asked your friend: Who was hunting the deer?

    RC Context: Last week you and your journalist friend watched a political debate with guests from the whole political spectrum, including two royalists. Both of them claimed to be animal activists but one of them was photographed hunting in the forest. Your friend mentioned today that their newspaper interviewed one of them and you wanted to know which one, so you asked: Who did your company interview?

    Sentence: It was the royalist that was hunting the deer.

13. CC Context: Your friend submitted an application to the town hall two months ago. This type of application has to be read and discussed publicly, so it can take a long time before they are approved. Yesterday, your friend told you that the application was finally accepted. You are happy it’s finally sorted and curious about the process, so you ask: Who was reading the form?

    RC Context: Today you accompanied your friend to a government office where you saw two bureaucrats behind the desk. One of them was chatting with an intern, while another was checking some documents. After you left, you want to know which one was dealing with your friend’s application, so you asked: Which one of them was in charge of your acceptance?

    Sentence: It was the bureaucrat that was reading the form.

14. CC Context: As a lawyer, you are responsible for cases related to private lending issues. One day, coming back from your lunch break, you realize that a sensitive document that was on your desk had been taken by somebody for review. To ensure the confidentiality of the case, you asked one of your colleagues: Who was viewing the file?

    RC Context: You went to your friend’s office today to have coffee with him and you saw two mortgagers having an argument: one of them was keeping busy going through a document while the other kept asking for attention and criticizing them. At the cafe, you asked your friend: Which one of them started the fight?

    Sentence: It was the mortgager that was viewing the file.

15. CC Context: Your friend invited you and some of their colleagues to an afternoon at a country farm. After the introductions, someone excitedly rushed off to try out horseback riding, but soon after fell off. You ask your friend: Who was riding the mare?

    RC Context: Your friend invited you and some colleagues to a social event at a horseback riding center. Amongst them were two famous novelists, one that spent the day sitting in the sun chatting with people and the other spent the day riding. During the dinner, you wanted to know which one your friend once admired, so you asked: Which one did you admire when you were at university?

    Sentence: It was the novelist that was riding the mare.

16. CC Context: You are assigned to review an old lawsuit. You want to know as much details as possible, so you ask the managing partner: Who was writing the case?

    RC Context: You have been robbed while visiting a friend’s town and he takes you to the police station. There are only two officers there. One of them helped filling the paperwork while the other was making phone calls. After you left, your friend said that one of them was once in a reality TV show. So you ask: Which one of them was on TV?

    Sentence: It was the officer that was writing the case.

17. CC Context: You fell asleep during a class. However, your dream was interrupted by some grating noise. You are really confused and sleepy, so you ask your classmate nearby: Who was moving the desk?

    RC Context: You and your friend volunteered to organise an event in your department this afternoon. Two lecturers also came to help with the venue setup, one of them focused on wall decoration and the other helped rearranging the furniture. You wanted to know which one they used to have problems with, so you asked: Which one of them did you have issues with?

    Sentence: It was the lecturer that was moving the desk.

18. CC Context: Your office recently had a job cut and you have to take over some advertising work in the publishing section. To your surprise, you found that the new edition you were assigned to work on has already been thoroughly advertised. You wanted to know who did such an excellent job, so you asked your colleague: Who was launching the book?

    RC Context: Two journalists in your office were recently laid off. On their last day of work, you and your colleague saw one of them conducting online interviews, while the other one was finalising the advertising campaign for the new edition of a famous novel. On your way home, your colleague mentioned that a renowned company just offered a position to one of them. Out of curiosity, you ask: Who was offered the job?

    Sentence: It was the journalist that was launching the book.

19. CC Context: It was freezing today so you were worried the office would be cold as well. However, after entering the office you found it was actually very warm, although the heater was off. You wonder who was so kind to warm up the office, so you ask your colleague who alway arrives early: Who was heating the room?

    RC Context: You just joined a research group on a new project. One day, one of the two analysts in the group kept adjusting the thermostat to ensure all were warm, while the other one was really hungry and focused on eating. After lunch, you want to know which one of them was in charge of the previous project, so you ask a colleague nearby: Who was the former project leader?

    Sentence: It was the analyst that was heating the room.

20. CC Context: In a high school sports competition, you noticed that there was a sudden uproar near the scoreboard because it was blocked by someone you couldn’t see very well. After the crowd went away, you asked your friend next to you: Who was hiding the score?

    RC Context: Your friend’s sports group proudly has two state medalists in basketball this year. When attending their informal competition today, you saw one of them surprisingly block the scoreboard for nearly one minute on purpose, while the other continued to dribble. After the game, you wanted to know which one of them won the competition last year, so you asked your friend: Who was the champion last year?

    Sentence: It was the medalist that was hiding the score.

21. CC Context: You and your friend are shopping in a big furniture store. You went to the garden section and your friend to the woodworking section. When you reunited, your friend said that there was a special demonstration and give-away of hand-made instruments built on site which can be used for assembling furniture without too much effort. You didn’t see the process, so you ask him: Who was making the tool?

    RC Context: Today you and your friend participated in a workshop on making carpentry instruments. The workshop was delivered by two carpenters. One of them gave verbal instructions, while the other demonstrated the physical process. When walking back home, you wanted to know which one had advised your friend on how to fix their chair, so you asked: Who advised you on how to repair your chair?

    Sentence: It was the carpenter that was making the tool.

22. CC Context: You friend is going on a road trip tonight so you came to their house to help them pack. When you arrived, you were surprised to see someone else carrying luggage to the already half-filled car. You wanted to know who was helping him, so you asked: Who was loading the car?

    RC Context: While walking with your friend, you saw two gardeners busy in front of your neighbour’s house, one of which was taking boxes out of the house while the other was packing them into your neighbour’s jeep. Afterwards, your friend said that one of them previously happened to work in their home. You wanted to know which one so you asked: Which one worked for you?

    Sentence: It was the gardener that was loading the car.

23. CC Context: Your friend is an activist and he convinced you to join him at a demonstration. There were lots of very different people marching with you and you were a bit confused by the different affiliations. Back home, you remembered a strange guy standing alone in the middle of the square with a big board and cryptic words on it, so you asked your friend: Who was holding the sign?

    RC Context: While jogging this morning, you and your friend saw two loyalists on the street trying to publicise a new activity. One of them was carrying a big board with words on it, while the other was talking with a speaker. After getting back home, your friend mentioned that he once met one of them in a restaurant and had a chat. You want to know which one so you ask: Which one did you meet at a restaurant?

    Sentence: It was the loyalist that was holding the sign.

24. CC Context: You are a new member of a landscaping company. This morning your colleague is giving you a tour of the company and you observed an old introductory manual on the table. You want to know who that belonged to, so you ask: Who was learning the trade?

    RC Context: Today you and your colleague are visiting a landscaping company where you saw two lumberers near the field. One of them looked very experienced, while the other was still in the training phase. After you left the field, your colleague told you that he once lived next to one of them. You want to know more, so you ask: Which one used to be your neighbour?

    Sentence: It was the lumberer that was learning the trade.