Rooth (1985; 1992) introduced a theory of linguistic focus which has proven to be particularly insightful in understanding notions like focus and givenness, and in accounting for focus-related prosodic phenomena such as contrastive stress or prosodic question-answer congruence. The basic idea is that every expression in language comes with two associated meanings: Its regular denotation and a set of alternatives. Minimally, this set of alternatives comprises the expression itself. But if the expression or one of its sub-constituents is marked as focused (noted syntactically with an F-feature on that node), then contextually relevant alternatives to that constituent are introduced. Depending on which elements within a constituent are marked with an F-feature, the content of the corresponding alternative set changes. Consider the alternatives to a proposition, which vary depending on whether the subject (5a) or object (5b) is marked as focused via F:

One fact that any account of focus must explain is why (5a) is felicitous in this context while (5b) is not. The crucial insight in Rooth’s theory is that prosodic focus is anaphoric, and that a pronunciation that fails to find a proper antecedent in the context will be infelicitous, much like using a pronoun without a proper antecedent. This anaphoric requirement is attributed to the focus operator ~. The ~ operator introduces the presupposition that some member of the alternative set to the constituent it attaches to is salient in the discourse.6 The examples in (5) have the representations in (6). In this and the following examples, we indicate F-marked material by SMALLCAPS and non-F-marked material in the scope of ~ by boldface. The antecedent that ~ is anaphoric to is not specially marked, but will be indicated in the text.

The two structures differ in the anaphoric requirement that ~ introduces because the alternative sets differ: (6a) involves a set of alternatives of the form x read Moby Dick, with varying substitutions for John, while (6b) involves a set of alternatives of the form John read x, with varying substitutions for Moby Dick. The context provides a question that consists of propositions of the form x read Moby Dick, and therefore only the anaphoric requirement in (6a) is satisfied in this context.

The use of prosodic focus is usually obligatory when possible; for example, in (5a), it would be infelicitous to respond to the question with the default pronunciation with an accent on both the subject and object. Current focus theories differ in how they explain the obligatoriness of prosodic focus. Williams (1997) proposes a principle Don’t overlook anaphoric possibilities, and Schwarzschild (1999) a violable OT constraint AVOIDF, which minimizes the use of F-markers. Wagner (2005) and Sauerland (2005), on the other hand, explain the obligatoriness of focus marking with the principle Maximize Presupposition (Heim 1991).

The ~ operator has no direct phonological exponent.7 Rather, the presence of ~ is only detectable through its effect on prosodic prominence within its sister constituent. According to Rooth (1992), there is a straightforward relationship between ~ and prosodic prominence in its attachment site, which he refers to as its scope. Truckenbrodt (1995) makes this condition explicit as follows:8

This correlation between scope and prosodic prominence makes interesting predictions about the relationship between syntax and prosody, since the scope of ~ in English is by no means fixed. For example, there could be prosodic focus marking within a subject, leaving the prominence of the VP intact (8a), or main prominence could fall within the subject (8b), with the VP being deaccented. The difference, in Rooth’s theory, is one of scope:

The sentence in (8a), where only man is reduced and the pitch accent on the VP (on Dick) remains, requires an antecedent of the form the x man. In contrast, the sentence in (8b), where the entire string man was reading Moby Dick is reduced, requires an antecedent of the form the x man was reading Moby Dick. This difference arises through the different scopes of ~, and is reflected in the prosodic rendition of the utterance.

The notion scope of ~ (i.e., the attachment site of ~) will play a central role in this paper. Importantly, it has to be distinguished from what is referred to as the “breadth” of focus (Ladd 1980: 75). For example, a sentence with focus on the VP is often said to show broad focus (Eady et al. 1986) compared to a sentence with narrow focus on the object, while focus on the entire sentence is sometimes called neutral (e.g., Eady et al. 1986) or wide (e.g., Büring 2016). The width of focus corresponds in Rooth’s theory to the attachment site of F within the scope of ~:

Wide focus is licensed by a contextually relevant proposition of any shape (e.g., Mary was sleeping in the context What was everyone doing?). VP-Focus requires alternatives of the form John was x-ing (e.g., John was swimming). Narrow focus on the object, as in (9c), requires alternatives that only differ in the choice of object (e.g., John was reading Leaves of Grass).

Rooth’s theory of focus also allows us to make precise what we mean by the terms focused and given: A constituent that is in the scope of ~ and F-marked is marked as focused; a constituent that is in the scope of ~ (or contained in an F-marked constituent) and not F-marked is marked as given; and a constituent that is not in the scope of ~ is neither marked as focused nor given.9 This last category is similar to what others have called discourse new, but the notion of discourse new plays no direct role in Rooth’s theory.10

Returning to Lieberman’s (1963) example, the alternatives theory of focus can explain why deaccenting nine in a stitch in time saves nine is infelicitous. A shift in prominence to stitch suggests that the whole idiom is in the scope of ~, and that stitch is F-marked and therefore marked as focused, while the rest of the idiom is not and therefore marked as given. This prosody thus requires an antecedent of the form an x in time saves nine. But such an alternative is highly unlikely to be already contextually salient, nor can it easily be accommodated: Since this is an idiomatic expression, substituting an alternative for stitch will lead to an expression that only has a literal meaning.11

We can use Rooth’s formalism to make explicit possible points of cross-linguistic variation in focus marking. A commonly held view is that the difference between English and French with respect to focus and givenness must be related to the considerably different prosodic systems of the two languages (e.g., Jun & Fougeron 2000; Féry 2001). Two potential types of phonological sources of variation, or P-differences, are listed in (10):12

Our experiment tests two concrete claims about P-differences, one for each of (10a) and (10b).

A different potential source of cross-linguistic variation stems from the syntax and semantics of ~, which we refer to as S-differences. One possibility is that ~ might vary between languages in its precise presupposition. The idea that (some) Romance languages only mark corrective focus is of this type, and could be captured in Rooth’s approach by building “correctiveness” into ~. In this paper, we propose a second possibility, that languages could instead vary in the syntactic scope possibilities of ~. We summarize these two potential S-differences in (11a) and (11b) respectively:

We test for S-differences by varying the context such that depending on the content and scope of ~, focus/givenness marking should or should not be possible.

In the following subsections, we review various concrete hypotheses about P-differences and S-differences for how English and French may differ in how they employ prominence shifts to encode focus.

One explicit hypothesis about why French differs from English (and other languages like German) is proposed in Féry (2001; 2014). Féry (2001) argues that stress-languages like English encode focus via the placement of pitch accents on particular words, while phrasing-languages like French encode focus via prosodic phrasing. This fits with the idea that in French, accentuation is more generally a property of phrases (Delattre 1938; among others), while in English, it is a property of words. According to Féry (2014), this has the consequence that in English, any syllable can be prosodically reduced, while in French, only entire phonological phrases can be.

The two focus marking strategies can result in superficially similar patterns. In French, various studies found a post-focal reduction in pitch (Di Cristo 1998; Jun & Fougeron 2000; Féry 2001; Dohen & Loevenbruck 2004), similar to what is often reported in English. In responses to subject wh-questions, for example, both English and French show reduced pitch on the VP (Breen et al. 2010 and Xu & Xu 2005 for English; Féry 2001 for French). This is as expected under what we call the Phrasing Hypothesis: Since the VP is mapped to a separate phonological phrase, post-focal reduction is possible in French in this case.

The differences between the two types of languages emerge when looking at cases in which the post-focal material does not form a separate phonological phrase. Hamlaoui et al. (2012) found that in noun-adjective sequences, adjectives are not prosodically reduced when focus is placed on the noun. The explanation proposed in Féry (2014) is that the adjective is not assigned to a separate phonological phrase, and hence there can be no post-focal compression of pitch and more generally, there is no post-focal prominence reduction. Féry (2014) and Destruel & Féry (2015) present evidence that post-verbal adjuncts show pitch reduction when the verb is focused while post-verbal arguments do not. The proposed explanation is that internal arguments are phrased with the verb, while adjuncts are mapped to their own separate phonological phrase, and hence can be reduced.

The Phrasing Hypothesis not only makes predictions about when focus can be marked by prosodic prominence, but also about how it should be marked. One prediction is that the phonetic cues used to mark focus in French should be the same ones that correspond to the encoding of phrasing, and might differ from those used in English which has been argued to encode focus via the level of prominence on particular constituents. Most prior studies did not check for intensity cues for focus in French, which are typically associated with focus prominence in English, and moreover did not directly compare English with French. Any observations about differences in the phonetic realization of focus between the languages have been gleaned from different studies that used various methods, which provides another motivation for the present study.

We summarize the predictions of the Phrasing Hypothesis as follows:13

The Phrasing Hypothesis assumes that French and English differ due to a P-difference along the lines of (10a). We now turn to a P-difference along the lines of (10b), where the scope of ~ is marked by something other than a shift in prominence.

Another way in which French has been argued to differ from English is that it has been reported to show an additional prosodic strategy to mark focus, namely a high or rising tone at the beginning of the focused constituent. The initial high, which is often referred to as the accent d’insistance, is optional and can occur in an accentual group in addition to the high tone present on the final syllable. It typically occurs on one of the first two syllables (e.g., Grammont 1914; Fouché 1935; Delattre 1938; Jun & Fougeron 2000), although function words are skipped (Jun & Fougeron 2000; Post 2000a; Welby 2006). It is often analyzed as a boundary tone rather than as a pitch accent (Jun & Fougeron 2000; Féry 2001; Welby 2006).

Di Cristo (1998) reports that an early high peak is the predominant realization of at least some types of foci (see also the discussion in Welby 2003). Beyssade et al. (2009) present evidence in favor of a phrase initial high as a general strategy to mark focus, and argue that French differs from English in using “Intonational Highlighting” rather than the distribution of pitch accents to mark focus. We refer to this as the Initial High Hypothesis. The Initial High Hypothesis is compatible with the Phrasing Hypothesis, since the high tone serves as a cue to phonological phrasing. The placement of an initial high is compatible with post-focal compression, so it is in principle orthogonal to the main question we explore here, concerning the conditions on prominence shifts. However, it is conceivable that the initial high could be an alternative strategy to mark focus when other tools such as prominence shifts are not available.

The status of the initial high tone as a focus marker is controversial. Jun & Fougeron (2000; and references therein) identify rhythm, style, and speaker as the main factors influencing its occurrence. Welby (2006) argues that the best predictor for the presence of the initial high is phrase length. While German & D’Imperio (2016) observed the initial high to be more likely to be found at the left edge of focused constituents, they also found a higher occurrence on long constituents that were not focused. Rhythmic reasons were also found to play a role in Astésano et al. (2007). Féry (2001) states that the initial high is often used purely for expressive purposes or in news speech, and is not specifically tied to focus realization.

While our study mainly focuses on the different predictions of the Phrasing Hypothesis and Scope Hypothesis (discussed in Section 2.6) for post-focal reduction, our data lends itself to look for initial highs as a marker of focus. The predictions of the Initial High Hypothesis can be summarized as follows:

We now turn to two hypotheses concerning possible S-differences between French and English.

Perhaps the most widespread idea about how French (and other Romance languages) differs from English (and other Germanic languages) is that the former only marks focus in cases of (metalinguistic) correction (Lambrecht 1994: 343; Ladd 2008: 236). What it means to be corrective has only been discussed informally. Gussenhoven (2007: 91) characterizes it as follows: “When the focus marks a constituent that is a direct rejection of an alternative, either spoken by the speaker himself (‘Not A, but B’) or by the hearer, the focus is ‘corrective’ […]”. Note, however, that a constituent cannot be rejected by another constituent; rather, a correction is a discourse relationship between alternative assertions.14 One way to make more precise what it means for a speech act to be corrective is that it suggests an amendment to a previous speech act. A corrective assertion proposes to update the common ground (the agreed beliefs shared by speaker and hearer) with a different proposition than the one asserted by the previous assertion, which it aims to reject. With respect to the distribution of prosodic focus marking, the Corrective Focus Hypothesis predicts the following:

Some previous experimental results appear to support the Corrective Focus Hypothesis. Jun & Fougeron (2000) and Dohen & Loevenbruck (2004) present evidence showing that French marks focus prosodically with post-focal reduction in corrective contexts, but they did not include a comparison case with non-corrective focus. Our experimental design, described in Section 3, includes both corrective and non-corrective utterances, and varies phonological and syntactic size in the post-focal domain in ways that will allow us to put both the Corrective Focus and Phrasing Hypotheses to the test.

One way to make the Corrective Hypothesis explicit is to try to build it into the semantics of ~. Suppose that French uses an operator ~_Fr, which encodes that the complement of ~ has to be a correction of the antecedent. Its English counterpart ~_En would not require correctiveness. But note that the contribution of ~_Fr only makes sense if it attaches to a constituent that denotes an assertion, since is not clear what it means for a constituent to be corrective. For example, if ~ attaches to a DP, how would the requirement that the DP be corrective be interpreted? The only sense we can give to this is that the DP must form part of a bigger assertion which stands in a corrective relation to a previous one. It seems then that the Corrective Focus Hypothesis as outlined here requires non-compositional reference to the larger structure that a focused constituent is part of. There is another way, however, to think about the difference between French and English that avoids this non-compositionality, to which we turn now.

Suppose that French differs from English in that the focus-sensitive operator ~ involved in prosodic focus marking has a much narrower set of syntactic scope possibilities in French compared to English, which results in a narrower set of pragmatic contexts in which focus can be marked. We will call this the Scope Hypothesis.

While syntactic scope seems like a natural point of cross-linguistic variation given Rooth’s (1992) formalism, we are not aware of any work where a hypothesis of this type has been proposed to account for cross-linguistic differences in prosodic focus marking. While the literature on focus often refers to different types of focus (e.g., Gussenhoven 2007), Rooth’s theory of alternatives does not actually distinguish different focus types, and any apparent differences in focus type must be explained as interactions with other factors. We argue that the scope of ~ may be helpful in explaining apparent effects of focus types.

If ~ in French is indeed more restricted in its scope possibilities than in English, there are a number of possibilities for how restricted it might be. One possibility is that in French, ~ is a sentential operator (similar to always), which can attach to any constituent that denotes a proposition, whereas in English, it can attach to many different constituents, including DPs (similar to only). Under this version, the scope possibilities in French are more constrained than in English, where it is assumed that ~ can attach to constituents of any conjoinable semantic type (e.g., Rooth 1992). This idea is able to account for some of the known differences between the two languages. Consider again the examples from Cruttenden (2006) (cf. (1)–(4)):

In (15a), two focus operators are involved, each of which takes scope over a sub-constituent of the utterance. The two ~ operators attach to DPs and the antecedents for focus marking are within the same utterance. In (15b), the contrast that B evokes is one between at least two potential alternative utterances, You mean you did all the work vs. You mean your sister did all the work, and the focus operator can attach to the propositional node.15 If in French, ~ can only attach to clausal nodes denoting propositions, embedded uses of ~ should be impossible. Thus, the focus structure in (16) with embedded ~ should not be possible, and hence focus should not be marked prosodically. However, prosodic focus marking should be possible in examples equivalent to (15b), illustrated in (17), since ~ can attach to the root in this case to introduce a focus presupposition that is licensed by the context.

Thus, the Scope Hypothesis can rationalize some of the differences found in Cruttenden (2006). Note that corrective examples are always compatible with ~ taking wide scope and attaching to the clause node that denotes a proposition, precisely because entire utterances are juxtaposed. What might intuitively appear like a “type of focus” effect might instead simply involve differing scopes of ~. We summarize the predictions of the Scope Hypothesis in (18):

The Scope Hypothesis has some similarity with the Corrective Focus Hypothesis: As we noted, requiring focus to be corrective imposes a condition on the entire utterance, not just the focused constituent. According to the Scope Hypothesis, attaching to the entire utterance—and not to smaller nodes—is the real difference, which gives rise to the appearance that French only marks one type of focus using prominence.

The Scope Hypothesis is different from the Corrective Focus Hypothesis in that it attributes cross-linguistic variation in prosodic effects of focus to differences in the syntactic scope possibilities of the ~ operator rather than its semantic content. We presented the idea that ~ in French only attaches to constituents that denote propositions, but there are actually variants of this proposal. For example, perhaps ~ in French can only attach to root nodes, that is, to syntactic objects that denote speech acts. Either way, it often makes similar predictions to the Corrective Focus Hypothesis. The crucial difference is that the Scope Hypothesis predicts that a prominence shift should be possible in French whenever the relevant scope is possible, irrespective of whether the antecedent is an assertion that stands in a corrective relation to the present assertion. Our experiments were designed to tease some of the divergent predictions apart, but the results stop short of resolving the issue, as we will see. If the Scope Hypothesis is correct, we should find independent syntactic evidence for scope differences. The observation that French resorts to marking focus with clefts more often than English (e.g., Lambrecht 1994) could be seen as a reflex of the difference in scope options for ~. Clefts are bi-clausal, and clefting might be a strategy to single out a constituent by putting it into a separate clause and adjoining ~ to it. A more thorough discussion of such syntactic differences is beyond the scope of this paper.16

We created 16 item sets which varied across the five conditions (corrective focus, contrastive focus, embedded parallelism, cleft, control). Each condition within an item set had lexically identical target constituents. As is conventional in the field, we will henceforth refer to each set of dialogues with similar linguistic material differing only in condition as an item (see Supplementary file 1 for a complete list of our experiment stimuli).

Below, we illustrate each condition and discuss the predictions for English versus French based on the Scope Hypothesis, Corrective Focus Hypothesis, Phrasing Hypothesis, and Initial High Hypothesis. In so-called corrective focus contexts, the entire prior assertion serves as the antecedent for focus marking. We can understand a corrective assertion in terms of how it updates the common ground: A corrective assertion proposes to update the common ground with a new proposition that aims to amend a previous context update. The focus in this case is the sub-constituent in which the two assertions differ. In Roothian terms, the representation of corrective focus involves ~ with wide scope over the clausal node:17

The focus alternatives involved in corrective focus are alternatives to the entire clausal node, as in (19)–(20), or possibly alternative speech acts, if operators like assert are represented in the syntax:

A corrective statement contrasts entire propositions (or perhaps speech acts) with each other. We call any contrast between constituents of a smaller size contrastive focus. In such cases, a sub-constituent is contrasted with linguistic material in an antecedent statement. For example, a red bike is contrasted with a blue bike in (22), and une soup froide with une salade froide in (23). Note that our contrastive focus differs from corrective focus in that it does not reject the proposition asserted in the antecedent proposition, but proposes to add an additional proposition to the common ground; one that contains a similar part, such as bought an x bike:

In the contrastive condition, the focus operator must take narrower scope than in the corrective condition. In examples (22) and (23), attaching ~ either to the VP or the DP would lead to a focus presupposition that is fulfilled by an antecedent in the context, as shown in (24).18 Based on the Scope Hypothesis, either representation should be possible in English only, since in French, ~ should not be able to attach to a DP or a VP.

In both corrective focus and contrastive focus, the antecedent for focus marking is in a separate utterance, introduced in the first part of the dialogue in our examples. There are also cases of prosodic focus marking in which a constituent within the same sentence serves as an antecedent. These sentences therefore involve an embedded parallelism, since two constituents with parallel structure appear in the same utterance:19

In Roothian terms, such examples would be analyzed as involving two ~ operators, with narrow scope over the contrasting DPs (Rooth 1992), which serve as the antecedent of the focus marking for each other:

Based on the Scope Hypothesis, a representation like (27), which in the following we will simply call parallelism, should be impossible in French.

The predictions of each focus condition for adjectival modification under the four accounts we consider are summarized in (28)–(30). Note that the Phrasing and Initial High Hypotheses both differ from the Scope and Corrective Focus Hypotheses, but make the same predictions across all types of focus examined.

French and English might differ not just in their use of prosodic focus, but also in the trade-off between syntactic and phonological means of encoding focus, as discussed in the literature on Romance vs. Germanic (Vallduví 1992; Lambrecht 1994; Cruttenden 1997; Ladd 2008; among others). One salient difference between the two languages is that clefts are used more frequently in French, in particular for subject focus (Clech-Darbon & Rialland 1999; Hamlaoui 2008; 2009). We therefore included a cleft condition, with an it-cleft in English, (31), and a c’est-cleft in French, (32). The type of focus we used with clefts is corrective.

The constituent of interest occurs early in the sentence in the cleft condition, as opposed to in all other conditions. This makes direct comparison of the prosodic results impossible, and hence we will exclude this condition from most of our prosodic analyses. The reason for including clefts was to establish their naturalness compared to non-cleft realizations of focus. The concern was that not using a cleft might be anomalous in French. These results are discussed in Section 6.

In order to evaluate whether focus is marked prosodically at all, we also included a control condition in which there is no potential antecedent for focus marking, and both target words are contextually new:

By design of the experiment, the participant is asked to respond with a particular scripted sentence. Hence, in the conditions in which a cleft is not used, participants must either mark focus prosodically (i.e., by shifting prominence in an informative manner or perhaps by using an initial high tone), or not mark it at all.

We varied two factors between item sets, to determine if they might play a role in whether focus is marked prosodically. The first is phonological size. Ladd (1996) was the first to suggest that the size of the constituent involved in a contrast may be important for determining whether a prominence shift can occur in Romance, specifically in Italian. An effect of phonological size in French could potentially provide evidence for the Phrasing Hypothesis, if longer constituents are more likely to be phrased separately. Ladd’s original observation is compatible with phonological size playing a role. As a first test of this, we varied the phonological size of the given material between items within our 16 item sets involving adjectival modification: 8 involved given constituents of 2 syllables and 8 of 1 syllable. If phonological size matters, this manipulation could have an effect on the possibility of prosodic focus marking in French. Examples of adjectival modification in the corrective focus condition showing the given word as 1 vs. 2 syllables are provided in (35) and (36) for French with bleu ‘blue’ and mignonne ‘cute’. (See Supplementary file 1 for all examples in English and French.)

It could be that the difference in word order between nouns and adjectives in French interacts with prosodic focus marking. Pre-nominal adjectives in French usually undergo liaison with the following noun, while liaison between the noun and a post-nominal adjective is more variable (Durand & Lyche 2008; Kilbourn-Ceron 2017). This can be seen as an indication that the two word orders differ in their prosodic phrasing: An adjective-noun sequence may be obligatorily phrased together as one domain, while a noun-adjective sequence may be phrased as two domains, even if the correlation between liaison and phrasing has been noted to be far from perfect (Post 2000b; Pak & Friesner 2006; Côté 2011).20 Noun-adjective sequences are also less likely than adjective-noun sequences to show clash resolution on the first word (Post 2000a; 2000b). If post-focal reduction correlates with phrasing, we would expect it to be more likely in noun-adjective sequences where adjectives are more likely to be phrased separately. In order to test for such an effect of word order, we included 4 item sets with pre-nominal adjectives and 12 item sets with post-nominal adjectives. Post-nominal adjectives have been shown in the previous examples, such as (35). An example of pre-nominal adjectival modification in French is in (37):21

In order to maintain comparability across word orders, the focused word always came first in linear order (focused word > given word). Beyond phonological size, focus marking might also interact with the syntactic size of the constituents involved. We tested for such effects in Experiments 2 and 3, discussed in Sections 4 and 5. The three experiments were part of a single study, but they are presented separately following the logic of the design of the sub-experiments.

The structure of the English and French experiments was as parallel as possible given independent syntactic differences between the languages. The experiments were run in a latin-square design, where each participant saw one condition from each item set, but an equal number of trials from each condition across the experiment. Each trial consisted of a pseudo-dialogue with an auditorily presented context and a scripted response to be read aloud by the participant, followed by a response rating provided by the participant. A sample dialogue for English is in (38):

In order to elicit the most natural conversation possible, the auditorily presented contexts were pre-recorded. There were English, Québec French, and European French versions of the pre-recorded stimuli, each recorded by a monolingual female native speaker of the respective language in her twenties.

After filling out a consent form and language questionnaire, participants read the instructions and the experimenter also verbally went over them. To familiarize participants with the task, there were four practice dialogues at the beginning of the experiment.

The actual experiment then proceeded as follows. Participants saw the complete dialogue on the screen as in (38), and read the dialogue silently. When ready, they pressed a key to hear the pre-recorded first part of the dialogue in their headphones, and were immediately able to utter their response. The reason the participants were given their scripted response before hearing the context was that without preparation, it would have taken too long to absorb the content to be able to respond within a reasonable period of time. Further, the entire dialogue was presented because otherwise, participants might have already pre-planned a response with a particular prosody, and then have become confused if their own response did not fit the given context.

After recording their response, participants saw a second screen requesting a naturalness rating of their response given the context. The precise wording was: “Please rate how natural your response was given what the other person said. [1 = completely unnatural, 5 = completely natural]”. We included unrelated filler dialogues that were interspersed among test trials, so each participant completed 60 trials in total (30 test trials; 30 fillers). The overall experiment lasted 15–25 minutes; no time limit was imposed.

Participants were recruited from our participant pool as well as through online ads. A total of 35 English and 33 French speakers participated. Among the French groups, we recruited 16 European French speakers and 17 Québec French speakers. Regarding possible differences between speaker groups, we were particularly interested in the potential influence of English on the French speakers.22 This is because, as mentioned above, there is evidence from language contact, between Chinese and neighboring languages, that prosodic features of focus marking can spread, perhaps because it is a useful tool to encode information status that can be imported without dramatic changes to other parts of the grammar (Xu et al. 2012). We checked for influence of English on French as follows. Following Akita (2005), we categorized the French speakers into two groups based on global accent ratings conducted by three English-speaking judges on English production data collected from each participant. We excluded 4 European French speakers and 1 Québec French speaker who were judged to have high influence of English in order to focus on how French speakers without significant English influence use prosody. The final number of participants whose data were analyzed is thus: 35 English speakers, 12 European French speakers, and 16 Québec French speakers.23

Our study deliberately focuses on prominence shifts as a means to encode focus.24 Our approach is to look for quantitative differences in focus effects across conditions that would reflect prominence shifts, and to compare across languages to see whether any differences can be observed. The reason for this is that we wanted to minimize the theoretical assumptions that go into the interpretation of our data. The different phonological systems of the two languages are such that superficially very similar pitch contours, for example, might have very different phonological interpretations, which makes a direct comparison of a phonological transcription more difficult. Importantly, our main predictions are about focus being encoded or not. Any quantitative differences that we find in particular focus contexts can serve to falsify claims that focus is not marked prosodically. The downside of examining quantitative cues is that we might overlook phonological differences where the languages seem similar given the phonetic realization. Thus, even if we find that focus is marked phonetically in similar ways in a particular context, this does not necessarily mean that the phonological representation of focus in that condition is the same. As a consequence, our results may remain compatible with multiple phonological interpretations, but this does not diminish the main findings about the interaction of focus type with whether or not focus is marked prosodically at all.

In looking at cues to focus, we begin by examining quantitative measures of relative prominence between words of interest, rather than the absolute prominence of particular words. Although the latter has been shown to be most relevant in encoding focus (see Ladd 1980; Wagner 2005; Ladd 2008; among others), absolute measures of prominence are harder to interpret. For example, a word placed earlier in a phrase will tend to have higher intensity than a word later in a phrase (Poschmann & Wagner 2016), as the latter might have lower pitch due to declination and downstep effects (cf. Ladd 2008). The prominence of a word will also be influenced by its segmental make-up. Since each two-word sequence appears in all 5 conditions, we can control for such by-item effects.

In all of our conditions, the focused constituent always preceded the given constituent. In English, material preceding foci is often optionally accented. That is, when the last accented word within an intonational phrase (carrying nuclear stress) is particularly prominent, it can easily be preceded by accented words even if it carries narrow focus. On the other hand, post-focal material must be deaccented or at least heavily reduced since, arguably, focus fixes the relative prominence between words, but does not necessarily proscribe the phonetic realization. This difference in the degree of perceived prominence between prenuclear and nuclear accents has been observed for both English and French (Hualde et al. 2016). We therefore decided to investigate the difference between early focus followed by a post-focal word and all-new controls. For example, in the dialogue for corrective focus in (39), we are interested in quantitative measures correlating with prominence on the target constituents steel and knives.

To obtain measurements for the target constituents, we automatically annotated the data using the Prosodylab-Aligner (Gorman et al. 2011). Using Praat scripts (Boersma & Weenink 2011), we extracted duration, intensity, and pitch measurements on the focused constituent and the following given constituent; in (39), steel and knives respectively. We then computed measures of relative prominence by looking at the difference between the focused and given constituents. We used measures sensitive to the ratio rather than the difference between the physical measure (difference in semitones of maximal pitch; in log duration; in intensity measured in db). The use of logarithmic difference measures is motivated by much psychoacoustic work which shows that perceptual differences in pitch, loudness, and duration are better modeled as ratios rather than as differences in absolute values (e.g., Breen et al. 2010).

In this section, we present the results of Experiment 1, which directly compares adjectival modification items across focus conditions in English and French. We first observe whether focus is prosodically marked, and if so, whether the type of focus conditions this marking. Specifically for French, we examine whether word order or the syllable size of the given item influences focus marking. Secondly, we investigate precisely how focus marking is achieved when focus marking does occur: Does the prominence shift occur because of a boost on the focused constituent or a reduction on the given constituent, or a combination of both? Finally, we take a closer look at French in light of the predictions of the Initial High Hypothesis.

To recap our overall predictions, if S-differences are the main reason why French and English differ in their use of prosody, we expect the languages to share similar relative measurements in at least some focus contexts—in particular, in corrections. If P-differences in terms of phrasing are the main reason why French and English differ, we expect items that do not phonologically allow for focus marking in French to be different from their English counterparts across all focus types.

Our results confirm previous findings that French can shift prominence at least in certain discourse contexts (e.g., Di Cristo 1998; Féry 2001; Cruttenden 2006; Ladd 2008), and more specifically, that European French marks corrective focus by post-focal pitch reduction (Jun & Fougeron 2000; Dohen & Loevenbruck 2004). They also confirm findings in earlier literature that English shifts prominence for focus purposes (e.g., Eady et al. 1986; Eady & Cooper 1986; Cruttenden 2006; Ladd 2008; Breen et al. 2010). Importantly, our results show for the first time experimentally that there is indeed an interaction between focus type and language when it comes to the marking of prosodic focus. While such interactions are implicitly assumed in earlier descriptions of the languages (e.g., in Ladd 2008), no study has actually shown evidence for this interaction in a direct comparison.

Our results show that French and English both prosodically encode corrective focus in similar ways, but differ with respect to other types of focus, especially parallelism: English uses prosody to mark focus in these contexts, but French does not. The fact that French and English pattern the same for corrective focus but not in other focus contexts is in line with the Scope Hypothesis, where ~ can only mark wide scope corresponding to a propositional-sized constituent in French, as well as with the Corrective Focus Hypothesis, where the locus of variation between English and French is in distinct semantics/pragmatics for the ~ operator.

Our results provide direct evidence against the Phrasing Hypothesis, which predicts that a prominence shift will not occur within a DP, independent of focus type. It is noteworthy that word order (adjective-noun vs. noun-adjective) did not have an effect on whether there was post-focal reduction. Noun-adjective sequences are more likely to be phrased separately, at least if we take the presence or absence of liaison to correlate with phrasing (cf. Post 2000b; Durand & Lyche 2008; Kilbourn-Ceron 2017). If this correlation between liaison and phrasing is correct, the Phrasing Hypothesis would predict a greater likelihood of reduction depending on word order (see also Féry 2004). There was also no effect of the size of the given constituent as measured in syllables, which again might have provided evidence for an effect of phrasing. Additionally, we found no evidence for the Initial High Hypothesis, in that we did not find that an initial high tone was used across focus contexts, instead of or in addition to shifting prominence.

It is interesting that French uses intensity as a cue to encode focus prominence, given that early studies of French found that intensity does not play a role in encoding accentuation (Parmenter & Blanc 1933; Delattre 1939), and other, more recent studies have not investigated intensity as a cue to focus (Jun & Fougeron 2000; Féry 2001, among others). Our results are in line with those of Dahan & Bernard (1996) and Hamlaoui et al. (2012), who did observe that intensity plays a role in prosodic focus-marking in French.

Earlier, we discussed whether the size of the constituents involved matters in whether prosodic focus marking is possible based on Ladd’s (1996) observation that larger constituents allow for deaccentuation more readily in Italian. From Ladd’s original comment, it is not evident exactly what is meant by “larger constituent”: It could relate to phonological size or to syntactic size. We have observed that manipulating phonological size in terms of syllable count (1 vs. 2 syllables) does not appear to impact prosodic focus marking. We turn now to the results of two sub-experiments that manipulate syntactic size, in ways that interact with phonological phrasing that would lead to very different predictions based on the Phrasing Hypothesis.

We analyzed the complex sentence data in the same way as the adjectival modification data, comparing measurements on the focused constituent (the matrix subject) with measurements over the given constituent (the matrix verb to the end of the clause). The data in Figure 6 look similar to the adjectival modification item results (Experiment 1; Figure 1) in that French consistently shows a difference between corrective focus and control like English, but parallelism is not distinguished for the most part in French, while it is in English, at least in intensity. One notable difference is that pitch seems to be irrelevant as a cue to focus in complex sentences in English, in contrast to what was observed for the adjectival modification data.

For the complex sentence items, we tested for significant effects of Focus and interactions with Language using a mixed model linear regression, like for the adjectival modification items. Note that given the low number of items, we could not fit by-item random slopes for this data set. Table 4 shows the results.

The results for complex sentences replicate those we observed for adjectival modification in that there is a significant interaction between Focus type and Language when looking at intensity, but only when we consider parallelism and contrastive focus. In other words, English encodes focus in parallelism and contrastive focus contexts using intensity, but French does not. Corrective focus seems to be marked similarly in French as in English.

The results thus appear to support the Corrective Focus Hypothesis, which predicts that focus is only encoded in corrective contexts in French, and speak against the Scope Hypothesis, at least in the version where ~ can attach to any proposition-denoting node. If ~ could attach to any proposition-denoting node in French, we would expect focus marking for all types of focus in French—even in parallelism, as outlined in (45)—and, thus, for French to be similar to English. This prediction is not borne out since French only marks focus in the corrective focus condition.

We can interpret these results, however, under an alternative formulation of the Scope Hypothesis; one that is more restrictive, where ~ can only attach to root nodes in French. This scope restriction in French would rule out the possibility of focus marking in parallelism even for complex sentences. In other words, (46b) would be impossible in French because ~ is attaching below the root node (cf. (45)):

This version of the Scope Hypothesis and the Corrective Focus Hypothesis make the same predictions for our experiment, and we can therefore not arbitrate between them. The predictions of the two hypotheses are summarized as follows:

Finally, the results speak against the Phrasing Hypothesis, which would predict focus marking in all conditions in the complex sentence cases since the given constituent is a VP and maps to a phonological phrase, and hence can be deaccented. In contrast to the complex sentence items, the Phrasing Hypothesis predicts, as outlined above, no focus marking in any condition in the adjectival modification cases. In other words, the fact that the French data look qualitatively similar in both complex sentences and adjectival modification is unexpected under the Phrasing Hypothesis. Note that this is the same prediction as the proposition-based Scope Hypothesis, which cannot be upheld given these results:

We now turn to a third type of syntactic configuration, relative clause modification, which was aimed at further distinguishing the Phrasing Hypothesis from the other hypotheses, and further testing the influence of phonological and syntactic size.

We submitted the relative clause data set to the same analysis as the other items, comparing the focused constituent (the relative clause noun head) to the given constituent (the relative complementizer and the rest of the clause). The means for relative intensity, pitch, and duration are illustrated in Figure 7.

Again, we tested for significance using mixed effects models, reported in Table 5. There is no statistical evidence that English and French differ in how they mark the various focus types in relative clause items. Most crucially, there is no difference between corrective focus and the other two types of focus. These results are unexpected under both the Corrective Focus Hypothesis and the Scope Hypothesis. In part, this may be a power issue, as we are considering less data compared to the case of adjectival modification (8 vs. 16 items).31 However, given that the complex clauses (6 items) replicated our basic finding, we expect that this is not the full story.32

French seems more similar to English when it comes to how focus marking works in the case of relative clauses compared to the other two types of constructions we investigated. While the interpretation of these results remains unclear, the findings could point to an interaction with phrasing as expected under the Phrasing Hypothesis. It seems plausible that relative clauses, but not head nouns, are mapped to their own intonational phrase. Note, however, that the Phrasing Hypothesis still faces three challenges in light of our overall results: First, why does focus type matter in the other two types of syntactic constructions (adjectival modification and complex sentence items)? Second, why do relative clauses and complex clauses not pattern together, based on the idea that the given constituent should map to a separate phonological phrase in both cases?

Clearly, the results we have found for relative clause modification pose a challenge for our account in terms of the Scope Hypothesis. It would be important to obtain more data on the prosody of relative clauses in a follow-up study.

Overall, the hypothesis inspired by Ladd (1996) that a larger domain for the given item would facilitate prosodic focus realization in French is not borne out in our data. The results for the complex sentence items show an interaction with type of focus, replicating the results from the adjectival modification data. This interaction is not predicted by the Phrasing Hypothesis, nor by the Scope Hypothesis in which ~ can attach to any node that denotes a proposition: The Phrasing Hypothesis would predict prominence shifts for all foci in both complex sentences and relative clause modification. The proposition-based Scope Hypothesis would make the same prediction for the complex sentences (see (43) and (46)). The data from relative clause modification remain puzzling for all hypotheses, and a thorough look at a broader range of syntactic configurations seems necessary before this result can be interpreted.