2. Theoretical background

Jackendoff (Reference Jackendoff1972: 352) said of the sentence in (6) that it as a ‘a contrast in meaning … produced by a difference in the choice of pitch accent [read, intonation]’.

1. (6) All the men didn't go.

When the reading in (7) is intended, negation is associated with the presupposition, giving rise to the representation in (8). That is, it was expected that some quantity of men did not go, and what is asserted is that that quantity is all of the men.

1. (7) ∀ > ¬ (‘none’)

1. (8) A accent (fall): negation is part of the presupposition, not the focus

   Presupposition: λQ [Q of the men didn't go] is well-formed/under discussion

   Assertion: all ∈ λQ [Q of the men didn't go]

When the reading in (9) is intended, however, negation is associated with the focused quantifier (all), giving rise to the presupposition and assertion in (10). That is, there is some expectation that some men went, and what is asserted is that that number is not all.

1. (9) ¬ > ∀  (‘not all’)

1. (10) B accent (rise): negation is associated with the focus (i.e., the assertion), not presupposition

   Presupposition: λQ [Q of the men went] is well-formed/under discussion

   Assertion: all ∉ λQ [Q of the men went]

Jackendoff did not restrict this account to sentences with subject all and negation. Indeed, he offered a similar approach for other negation sentences with all or many in object position (seen in (11)), because (see the unfortunate example in (12)), and focused constituents elsewhere.

1. (11) I didn't see ALL of the men.

   (Jackendoff (Reference Jackendoff1972: exx. (8.181)–(8.182))

1. (12) Max doesn't beat his wife because he LOVES her.

   (Jackendoff (Reference Jackendoff1972: exx. (8.185)–(8.156))

For example, Jackendoff claimed that in both versions of the ambiguous sentence in (13) (his (6.137)–(6.139) collapsed),

1. (13)

   1. (a)

   2. (b)

Jackendoff (Reference Jackendoff1972: 260) further argued that ‘the difference between them always appears at the end of the sentence’.

Liberman & Sag (Reference Liberman, Sag, LaGaly, Fox and Bruck1974) were among the first to attempt a revision of Jackendoff's account, arguing instead that with sentences where negation takes wide scope, the speaker is questioning the addressee's assumptions. The result is that there is a contradiction contour, and a distinct terminal rise. Their key example, presented in (14), accompanied by a fall-rise contour, may be thought of as a rejection addressed to someone who has claimed that elephantiasis is incurable.

1. (14) Elephantiasis isn't incurable.

While Ladd (Reference Ladd1980) agreed with the general direction of Liberman & Sag's (Reference Liberman, Sag, LaGaly, Fox and Bruck1974) account, he pointed out that their contradiction contour and the fall-rise contour are not the same phenomenon, and that while both result in a rising sentence contour, the beginning of the sentence is noticeably different. For example, in response to someone who thinks that elephantiasis is fatal, a speaker might reply with (14), with a high pitch accent toward the beginning of EL (the contradiction contour). However, in response to someone concerned about his imminent death as a result of contracting elephantiasis or rabies, a speaker might reply with (14) with a high pitch accent on TI (the fall-rise/contrastive contour). Ladd also pointed out that while the contradiction contour cannot be embedded (as Liberman & Sag had observed), a fall-rise contour can be. The implication, then, is that sentences such as All the men didn't go could result in a rising pattern for different reasons.

Ladd (Reference Ladd1980) thus shared the observation that there are intonational differences correlated with interpretation, but saw the effect as pragmatic, rather than semantic. According to Ladd's account, the primary message of the fall-rise contour is ‘focus within a given set’. An example dialogue from Ladd (Reference Ladd1980: 153 ex. (16)), presented here as (15), illustrates this point: there is the presupposition that B fed something, and B's assertion combined with the use of the fall-rise contour (indicated with Ladd's ̌ notation) indicates that s/he fed the cat, and cats are members of the set evoked by A.

1. (15) A: Did you feed the animals?

   B: I fed the ̌ cat.

That the sets are contextually relevant and can be ad-hoc is illustrated by other examples, such as (16) (Ladd's (Reference Ladd1980) example (24)).

1. (16) A: What would you think of getting [ – ] a dog?

   B: A ̌ stove, maybe.

As Ladd (Reference Ladd1980: 154) pointed out, dogs and stoves may share membership in a set of things like ‘things we can afford’ or ‘material possessions we can allow ourselves’ – sets that the speaker and hearer could perhaps agree upon. (See his example (25), page 154.) Ladd's reasoning behind the all example is as follows: Because fall-rise focus on all cannot place it as a member of a superset (because there is no greater quantity than all), then the interpretation must be ‘not all’. Thus, a scopal relation where negation takes scope over the universal quantifier will be associated with a fall-rise (and not a falling) contour.

Ward & Hirschberg (Reference Ward and Hirschberg1985) presented what is easily the most extensive pragmatic account of fall-rise contour to date. They first argued (pages 754–755) that Ladd's (Reference Ladd1980) conception of ‘focus within a given set’ cannot be correct because of examples illustrating that B does not have to evoke membership within the set mentioned by A. For example, in (17) (their (19)), A is asking about the route that B took traveling in Philadelphia. The dialogue is felicitous, they argued, if B does not know that Walnut ends at 34th. What is not at issue is membership in the set of streets. (The \ / notation was used by Ward & Hirschberg to indicate the syllable hosting the fall-rise accent pattern.)

1. (17) A: Did you go straight up Walnut?

   B: To Thirty-\fourth/.

Fall-rise can also accompany a superset, as in (18) (also their (18)).

1. (18) A: Are you sending me mail?

   B: I'm sending \peo/ple mail.

And fall-rise can be inappropriate even given membership in a set, as in (19) (their (17)).

1. (19) A: Did they have a boy or a girl?

   B: #They had a \boy/.

Moreover, fall-rise intonation need not accompany set membership: the dialogue in (15) above is still interpretable with falling intonation on cat.

Ward & Hirschberg (Reference Ward and Hirschberg1985) argued that the purpose of the fall-rise contour is to convey ‘speaker uncertainty’ about the appropriateness of an utterance in a given context, and specifically about a salient relationship between discourse entities. This may involve set membership, but not necessarily. What matters is that the speaker who employs the fall-rise contour perceives there to be some possible scale (see Hirschberg Reference Hirschberg1985), and uses this contour, because s/he is uncertain about (i) whether it is appropriate to evoke a scale at all, (ii) if some scale is appropriate, which scale should be chosen, or (iii) given a scale, which value (or scalar alternative) should be chosen.

Ward & Hirschberg (Reference Ward and Hirschberg1985) noted that because in their account the contour is indicative of a speaker's knowledge of contextually-relevant scales and scalar alternatives, a fall-rise contour should not be tied to any meaning (or scopal relation between quantification and negation) in particular, and should not force negation to take wide scope over the quantifier. Two predictions thus follow, which are relevant to the quantificational sentences in question. First, it should be possible to observe a fall-rise contour without negation taking wide scope over the quantifier (that is, with all taking wide scope over negation). Ward & Hirschberg provided the example in (20) below, in which B responds to A's utterance using a fall-rise contour. Here, B's use of fall-rise is an indication of speaker uncertainty. B does not know whether A wants to know which meeting at least some of the men missed, or whether a quantifier scale should be evoked.

1. (20) A: The foreman wants to know which meeting some of the men missed.

   B: \All/ the men didn't go to the last one.

Second, we predict that we should be able to observe negation taking wide scope over the quantifier without a fall-rise contour (i.e., with a falling contour). As an example, Ward & Hirschberg pointed out that the embedded clause in (21) (following I know that) can be uttered with a falling contour while still favoring negation taking wide scope.

1. (21) George said that everyone had left for the game by five, but I know that all the men didn't go that early.

This rejection of a one-to-one correspondence between prosody and interpreation and an appeal to a pragmatic account is echoed in Kadmon & Roberts (Reference Kadmon, Roberts, Farley, Farley and McCullough1986), whose main example is presented in (22), with the prosody–interpretation correlations and scopal relations in (23).

1. (22) He doesn't hate most of the songs.

1. (23)

   1. (a) falling contour in response to ‘negative’ question

      scope: most > negation

   2. (b) fall-rise contour in response to ‘positive’ question

      scope: negation > most

We note in passing that it is possible to find similar attested examples in an online search:

1. (24) Although they didn't win many awards during their performing years,

   Led Zeppelin's lasting influence has garnered them several Grammy Hall of Fame Awards, as well as an induction into the Rock & Roll Hall of Fame.

   (‘Led Zeppelin and the Recording Connection Audio School’,

   http://www.recordingconnection.com/artists/led-zeppelin)

1. (25) AMC's superb “Mad Men” didn't win many awards on the night, but the two they did counted …

   (‘Sepinwall on TV: Recapping the Emmys’,

   http://www.nj.com/entertainment/tv/index.ssf/2008/09/sepinwall_on_tv_recapping_the.html)

Kadmon & Roberts (Reference Kadmon, Roberts, Farley, Farley and McCullough1986: 16) acknowledge that the prosody of a given sentence does indeed appear to favor one interpretation over another, but that these different prosodic patterns also differ with respect to the contexts in which they occur, and the explicit or implicit topical question they address – the Question Under Discussion (QUD) (using terminology developed by Roberts Reference Roberts, Yoon and Kathol1996). In a nutshell, their claim is that prosody does not directly determine scopal relations. It conveys partial information about the structure of the discourse, and this structure determines scope. The hearer is then left to reconstruct the best fitting (and simplest) interpretation, and does this in large part by retrieving the relevant implicit question, which is part of the information structure, from the preceding utterance context. This question is connected to a presupposition skeleton along the lines of Jackendoff (Reference Jackendoff1972) in that the question is either ‘negative’ or ‘positive’, and the presupposition either encodes negation or not (in which case negation is focused).

What follows from this discussion is that while it is certainly possible that a fall-rise (or more generally, non-falling) contour may be more likely to favor negation taking wide scope over the subject quantifier, such a contour is neither necessary nor sufficient for this interpretation. A certain degree of variability should also be expected, if prosody is tied to speaker uncertainty in a discourse context and to implicit questions that are part of the information structure. Thus, while there may be a correlation between sound and meaning with these ambiguous sentences, there is not a one-to-one mapping, especially between sentence-final contour and scopal relation. One wonders whether – in the face of this variability – hearers can still associate certain prosodic information with certain interpretations (qua scopal relations). In the next section, we review relevant experimental work in this area, which suggests that hearers are attuned to auditory correlates to sentence interpretation and can use prosody to reconcile ambiguity in other cases. These findings suggest that a similar pattern of results could be identified for our target sentences involving quantification and negation.

3. Experimental background

A number of studies over the years (covering a range of phenomena and employing a variety of paradigms) have investigated the role of prosody in sentence disambiguation. As we will show, the general picture that emerges is that experimental participants do assign some weight to prosody in determining the interpretation of a potentially ambiguous string, but how much weight they assign varies among speakers, hearers, and contexts. And importantly, a conspicuous gap exists when it comes to systematically probing the quantificational sentences that originally interested Jackendoff and others.

In a set of three studies, Speer et al. (Reference Speer, Crowder and Thomas1993) showed that participants use prosody to disambiguate structurally ambiguous sentences and to determine the focus or presuppositional structure. In one study, for example, participants heard a prerecorded, potentially ambiguous sentence, then selected from two choices the appropriate paraphrase of it. Among the stimuli were sentences with ambiguous pronominal reference, phrase boundaries (e.g. The dog may attack (#) Gwen), conjunction patterns among NPs (e.g. Either Sam # or Susan and Lara # vs. Either Sam or Susan # and Lara #) and differences in syntactic structure (e.g. They are cooking APPLES vs. They are COOKING apples). Participants were sensitive to the prosodic pattern when choosing their paraphrase, but more so for syntactic ambiguities than for placement of focus, as indicated by a calculation of difference scores using hits and false alarms based on participants’ choice of intended meaning.

Price et al. (Reference Price, Ostendorf, Shattuck-Hufnagel and Fong1991) recorded professional radio announcers reading a range of structurally ambiguous sentences that were embedded in discourse contexts. They then presented the excised sentences in isolation to naïve participants, who were given a choice between two contexts in which the sentence could have appeared and asked to select the best match. They found considerable variation among the speakers, as well as variability in participants’ success rate based on the speaker and the sentence types, with the overall percentage correct reported ranging from 54% to 96%. Listeners were largely successful in distinguishing between apposition and NP attachment (The neighbors who usually read [the dailies / , the Daleys,] were amused), with a 91–92% correct rate reported. However, their performance with attachment ambiguities (e.g. Raoul murdered the man [with a gun]) was less consistent, with the percentage correct ranging between 63% and 78%, and clausal ambiguity (e.g. Mary was amazed and Dewey was angry/Mary was amazed Ann Dewey was angry), with the percentage correct ranging from 54% to 88%. These rates are important to keep in mind when evaluating the results of our Perception experiments later in the paper.

Other studies have focused more specifically on scopal ambiguity with negation. For example, Cooper & Paccia-Cooper (Reference Cooper and Paccia-Cooper1980) presented participants with ambiguous target sentences such as the ‘negation–because’ sentence in (26), followed by a disambiguating follow-up sentence or with the target sentence embedded in a disambiguating paragraph.

1. (26) Dick didn't fly the kite because it was a beautiful day.

Participants significantly lengthened the vowel of the word preceding the because clause (here, kite) when producing the interpretation where because took wide scope over negation (i.e., ‘It was because of the day that Dick didn't fly the kite’). The length of the vowel was approximately 30 ms longer relative to the length of the vowel when negation took wide scope over the because clause. This effect, however, only surfaced when participants were made aware of the ambiguity with isolated target sentences.

Hirschberg & Avesani (Reference Hirschberg, Avesani, Botinis, Kouroupetroglou and Carayiannis1997, Reference Hirschberg, Avesani and Botinis2000) presented English and Italian speakers with a range of ambiguous sentences, including negation interacting with a because clause (e.g. William isn't drinking because he's unhappy), attachment ambiguities, focus-sensitive operators (only, even), and the scope of none. Speakers were shown two paragraphs, each favoring a different interpretation, side by side, and asked to read them both aloud. Afterwards, they answered a comprehension question. Speakers produced no consistent results for the none sentences or for attachment ambiguities. They did, however, produce differences for the ‘negation–because’ sentences, as captured in Table 1.

Table 1

Pattern of results for ‘negation–because’ sentences in Hirschberg & Avesani (Reference Hirschberg, Avesani, Botinis, Kouroupetroglou and Carayiannis1997, Reference Hirschberg, Avesani and Botinis2000).

The contour data in the far right column can also be re-cast by saying that of the 13 sentences in which there was no falling contour (adding the differences of 3 and 10), 10 were observed with negation taking wide scope over because. Thus, a non-falling contour was more likely when negation took wide scope. For only/even sentences, participants regularly placed the nuclear stress in the focused phrase in the sentence.

Further evidence for the role of prosody in disambiguating ‘negation–because’ sentences in English comes from Koizumi's (Reference Koizumi2009) self-paced silent reading studies. In one experiment, participants were shown sentences that favored one of two possible interpretations, as in (27), then answered a question about the sentence.

1. (27)

   1. (a) Jane didn't purchase the white blouse because it had a stain.

      (because > negation)

   2. (b) Jane didn't purchase the white blouse because it suited her.

      (negation > because)

Koizumi found that sentences favoring a because > negation scopal relation were read more quickly than negation > because sentences (a pattern consistent with Frazier & Clifton Reference Frazier and Clifton1996). However, this difference in reading time went away when the sentences were embedded in an if clause, which carries with it its own continuation rise, which neutralized the intonational difference between the two sentences. When a line break was inserted before the because clause (which had the effect of inserting a prosodic break before because), participants displayed a preference for the interpretation where because took wide scope.

With the above studies as background to how experimental participants interpret the role of prosody in sentence ambiguation, we now turn to ambiguous sentences involving quantification and negation, our target sentences in this study. Baltazani's (Reference Baltazani, Andronis, Debenport, Pycha and Yoshimura2002, Reference Baltazani2003) complementary production and perception studies in Greek offer the first strong suggestion that speakers and hearers use prosody to disambiguate such sentences. Baltazani (Reference Baltazani, Andronis, Debenport, Pycha and Yoshimura2002) focused on three types of items, captured in English in (28).

1. (28)

   1. (a) He's not watching TV because he's bored. (negation–because)

   2. (b) They did not eat many apples. (negation and object quantifier)

   3. (c) Three nurses helped every doctor. (subject and object quantifiers)

The tokens for each item type varied SVO/SOV word order, since Greek allows for variable word order. In the production study, participants viewed a question that either did or did not contain negation, thereby allowing the positive/negative status of the QUD to be manipulated. They then saw the target sentence offered as an answer to the question, and were asked to read the question–answer pair aloud (see (29)).

1. (29) A: How many problems [did/didn't] the students solve?

   B: The problems they solved are not many.

These productions were then presented to participants in a forced choice task. Participants heard the excised target sentence in isolation and were asked to choose a suitable answer.

Baltazani (Reference Baltazani, Andronis, Debenport, Pycha and Yoshimura2002) found that nearly every one of her Greek speakers reliably produced ambiguous sentences with negation with two distinct contours related to the two interpretations. Moreover, hearers were by and large able to correctly identify their corresponding interpretations. When presented with sentences with an OV word order in which the quantifier was focused, hearers were more likely to interpret the quantifier taking wide scope over negation (51% quantifier > negation vs. 31% negation > quantifier). By contrast, when presented with VO sentences in which negation was focused, hearers were more likely to interpret negation taking wide scope over the quantifier (62% negation > quantifier vs. 29% quantifier > negation). Interestingly, however, while sentences involving subject–object quantifier interaction (like (28c) above) were produced in a manner similar to those like (28b), hearers were not able to unambiguously select the corresponding interpretation based on production alone. The combined production–perception experimental findings indicate that Greek speakers do reliably disambiguate sentences using prosody. However, hearers’ ability to retrieve the correct interpretation on the basis of prosody alone depends on the lexical items that interact and give rise to the ambiguity.

The observation of depressed percentages in the perception studies also underscores the fact that while there are clear and significant trends in hearers’ ability to recruit prosody to assign meaning, what may matter most is not their overall success rate, but rather the difference in responses between conditions in which one or the other interpretation is favored. Note, however, that although these findings are promising, they were found in Greek – a language that may require such prosodic manipulation, or at least be more highly sensitive to it than English. Thus, it still remains an open question whether such a pattern could be observed in English for the quantificational sentences in question here. Recent studies aimed at addressing precisely this question have come up relatively empty handed.

McMahon et al. (Reference McMahon, Lidz and Pierrehumbert2004) designed four short children's stories, each of which included a sentence with ambiguous pronominal reference and an ambiguous sentence involving the interaction of a universal quantifier with negation. The story context and a continuation sentence favored one of the two scopal relations, as in (30).

1. (30) Every bunny didn't jump over the fence …

   1. (a) not a single one jumped over. (∀ > negation)

   2. (b) only some did. (negation > ∀)

Eleven parents were recorded reading these stories to their children, and sentences were later coded holistically for falling/rising contour. The authors found no discernible pattern with respect to sentence-final contour, reporting a consistent 58% falling contour across all sentences.

The excised target sentences were then included in a perception study, during which hearers were told explicitly that prosody could be used to disambiguate sentences. During the test session, participants were asked to choose between two images representing what had immediately happened before a target sentence was uttered. They were given an unlimited amount of time to inspect the images before listening to the target sentence and making their selection. While participants were successful with the pronominal reference items, they were largely at chance with the quantificational sentences, selecting the interpretation with negation taking wide scope 59% of the time, regardless of the interpretation intended by the speaker. This preference persisted despite the intonational contour. See Table 2.

Table 2

Distribution of percentage of perceived interpretation of sentence-final contour for every–negation sentences in McMahon et al. (Reference McMahon, Lidz and Pierrehumbert2004).

These results, then, could be taken as demonstrating that there is in fact no connection between prosody and sentence interpretation. However, the experimental design may have minimized the role of prosody. For example, the pictures clearly disambiguated the sentences, leaving little room for prosody to play a central role in production. The continuations following the target sentence may also have played a significant role in that they may have been able to host the informative pitch accenting, minimizing the need for such prosodic information in the target sentence itself. Moreover, the construction of the dialogue may have favored a continuation rise in both cases. It is not possible to conclude from this study, then, that speakers do not offer the hearer such prosodic cues, and if speakers did not consistently provide prosodic cues to interpretation in this study, it is not possible to conclude that hearers cannot recruit cues that are not consistently there.

Jackson (Reference Jackson2006) presented four speakers with nearly 200 sentences involving the scopal interaction of a small set of lexical items in different syntactic positions. Items included negation, the universal quantifier every, and the indefinites a and a few. Each sentence was accompanied by two still images, illustrating the two interpretations (e.g. a group of circles, each of which hit a different square vs. a group of circles, all hitting the same square, for the sentence Every circle hit a square). Participants were asked to produce the sentence in a way that favored one of the two interpretations. Jackson's findings hinted at durational differences in the lexical items being correlated with the interpretation favored by the speaker. However, the conditional ranking among these durations was quite complex and the number of speakers extremely small, making generalization beyond his results rather difficult. Productions were also not presented to hearers.

Given the previous experimental studies, the following picture emerges. First, participants of studies conducted in English and other languages can recruit prosodic information when disambiguating a range of sentence types. Their ability to do so, however, is dependent upon a range of factors, including the lexical items, type of ambiguity, speaker, and experimental design. Second, for the sparse evidence we have for ambiguous sentences with quantification and negation in English – precisely those sentences that have driven the debate over the years – there is still an open question regarding whether or not hearers can recruit prosodic information in the service of arriving at the correct interpretation. The current research was designed to fill this gap.

4. Stimuli from Production experiment

The auditory stimuli for the two Perception experiments reported in this paper were taken from a larger Production experiment, reported in more detail in Syrett et al. (to appear). The aim of that study was to elicit productions of ambiguous sentences in a discourse context for the purpose of finding prosodic correlates to sentential interpretation (and, therefore, scopal relation). We therefore present key aspects of the Production experiment here as a backdrop to the Perception experiments, which made use of a subset of speaker sound files.

5. Perception Experiment 1

The joint purpose of the two Perception experiments was to determine whether hearers can use surface level auditory cues provided by speakers to arrive at the intended interpretation of scopally ambiguous sentences. To this end, each of the experiments served a specific purpose. In Perception Experiment 1, we sought to determine whether hearers could use the prosodic delivery of a target sentence to properly situate it in a discourse context indicated by two possible continuation sentences. In Perception Experiment 2, we sought to determine whether hearers could use the thread of a discourse context to choose the prosodic version of a target sentence that was most compatible with the interpretation favored by that context.

Tokens for both perception studies were contributed by a subset of naïve speakers from the Production experiment and an experimenter. This experimenter was the second author: a female with extensive vocal musical training who was acquainted with the theoretical and experimental literature, and who was therefore well acquainted with the prosodic patterns claimed to be correlated with the various interpretations. The two experiments were run separately. Each experiment began with a brief training session, in which participants were introduced to the task structure as well as the possibility of disambiguating sentences with prosody, using items with ambiguous pronominal reference. Each experiment was completed in under 30 minutes.

5.1 Experimental method

5.1.1 Participants

Forty-nine undergraduates participated in Perception Experiment 1. Data from four additional participants were excluded from the analysis, because the participants indicated that English was not their native language. One other participant was excluded due to consistently low response times (many below 150 ms).

5.1.2 Stimuli

Items for Perception Experiment 1 were contributed by four speakers from the Production experiment (one male and two females) and the experimenter. The naïve speakers were selected on the basis of their high comprehension scores in the Production experiment (above 75%) and their consistent and clear production of distinct versions of the two interpretations of each sentence in their respective contexts, which largely reflected the manner of production discussed in the literature.

There were 48 experimental items. These included 24 minimal pairs of sentences in which the same sentence was produced in two distinct manners, each favoring an interpretation supported by a previous discourse context. Each speaker contributed six minimal pairs. For the test items, there were two with all and negation, and one with many/most and negation. For the control items, there were two with negation and because, and one with one of the focus-sensitive operators (only or even). These items are provided in Appendix A. The items were blocked by speaker, and then further divided within the block to separate the minimal pair members. During the experiment, the presentation of the items within each block was randomized by the stimuli-presentation software. The order of the blocks was predetermined: the female experimenter, a naïve male, and the two naïve females.

5.1.3 Procedure

Items for both of the Perception experiments were presented in a quiet laboratory setting at either a Dell or an iMac computer using the SuperLab stimulus presentation software (Cedrus Corporation 2012) and a SuperLab response pad. Participants wore noise-cancelling headphones.

Each item in Perception Experiment 1 had the same structure. First, participants viewed the target sentence in the middle of the screen (e.g. All the moms didn't allow eyeliner). This visual stimulus was accompanied by a speaker's production of the sentence, which was intended as part of a cohesive discourse context that favored one possible interpretation. Pitch tracks for two renditions of one target by the experimenter (Speaker 1) are presented in Figure 1. Additional examples for both Perception experiments are presented in Appendix B.

Figure 1

Pitch tracks for two versions of the sentence All the moms didn't allow eyeliner by Speaker 1 from two different contexts, each favoring a different scopal relation.

The auditory stimulus was presented three times. The participant then saw a second screen, on which were two possible continuations to the target sentence, one above the other (e.g. A: They were all in agreement; B: Only the moms of the older girls let their daughters wear it). Participants were asked to choose the most natural continuation. In this way, we placed the target sentences into a mini discourse. The correct choice was counterbalanced between A and B. We coded a correct response as one that corresponded to the interpretation intended by the speaker, based on the fact that the speaker had responded correctly to the comprehension question when uttering these sentences as part of the contexts in the Production experiment. Perception participants had a maximum of 10 seconds to make their decision. If no response was made during that time, the trial would end, and the next one would begin. Most speakers rendered a response well before this time. There were only eight instances where a participant made no response during the time allotted. Responses registered in less than 200 ms (n = 4) were not included in the analysis.

5.2 Results

We first looked at the percentage of correct responses for each test item type, overall and by scopal relation. These results are presented in Table 3. The percentages observed here may seem somewhat low overall, given much higher success rates observed in other psycholinguistic studies, such as lexical decision or discrimination tasks.Footnote ² To explain this rate, we first refer to the results of previous studies reviewed in the experimental background section, which report comparable rates. Indeed, these low percentages are not specific to the ambiguous quantificational sentences. Second, as a point of comparison, we note that the percentage of correct responses for the sentences with negation and a because clause were similar, with an overall percentage correct of 69.3%, and 62.4% when the because clause was intended to take wide scope, and 76.0% when negation took wide scope over the because clause.

Table 3

Average overall percentage correct in per item type and per speaker in Perception Experiment 1.

To determine whether the average scores for each item type and each corresponding scopal relation were above chance, we ran a binomial probability test on the overall average (the far left column of Table 3) (p = ·5, 99% Confidence Interval (CI)). Indeed, despite the lower-than-anticipated percentages, the averages were significantly above chance level: all > negation: 2.7, p < .01; negation > all: 3.1, p = .002; many/most > negation: 2.5, p = .01; negation > many/most: 5.1, p < .0001. This is the first indication that hearers were able to successfully use the speakers’ prosodic delivery of the target sentences to arrive at the correct interpretation. This analysis was complemented by a χ² analysis on the distribution of responses, based on the number of participants responding at each percentage level. As would be expected, the findings were significantly above chance: all > negation: χ²(7) = 20, p < .01; negation > all: χ ²(7) = 32, p < .01; many/most > negation: χ²(4) = 21.45, p < .01; negation > many/most: χ²(4) = 31.45, p < .01.

Admitting the possibility that there could be a response bias contributing to these scores, we ran an additional d′ analysis using signal detection theory (Macmillan & Creelman Reference Macmillan and Creelman2005), which is a measure of hearer sensitivity. Hit rates and false alarm rates were used to calculate a d′ score for each participant for the test items (d′ = (z(Hits)–z(False Alarms)). Scores can range from 0 (no detectability) to roughly 4. The average d′ score was 1.02, which was significantly above chance (0) (t(43) = 8.82, p < .0001). The hit and false alarm rates are plotted in Figure 2. Importantly, hit rates are higher than false alarm rates.

Figure 2

Hit and false alarm rates for participants’ responses to test items in Perception Experiment 1.

Following the experimental session, participants were asked to comment on their strategy for responding to the items, and whether any aspect of the items themselves made making a decision either easier or harder (following an approach used successfully by Syrett & Kawahara, published online 14 November 2013). We took these open-ended responses and categorized them based on key words in each participant's response. Responses were coded as mentioning ‘word-level stress’ if they included any of the following words: inflection (with regard to specific words), tone(s) (on words), emphasis, stress, loud, and soft. Responses were coded as mentioning ‘phrase-level intonation’ if they included words such as up, down, end, rais(ed), low(er), drone, monotone, pitch, and inflection (with regards to entire sentence). Responses were coded as mentioning ‘prosodic break’ if they contained words such as breath(e), pause(s), and break(s). This coding method allowed for participants’ responses to be coded as belonging to one or more categories. Only three of the participants’ responses could not be categorized, because their responses were too vague. Two participants failed to complete a survey. We present the distribution of participant responses in Table 4.

Table 4

Participants’ categorized responses to the post-experiment survey regarding their response strategies in Perception Experiment 1.

The responses indicate that when hearers were listening to the sentences to select a plausible continuation sentence, they were attending to those aspects of the sentences that speakers were prosodically manipulating to favor one interpretation over another. It is also interesting that most hearers did not indicate that they attended to whether the sentence was rising or falling, suggesting that there were cues beyond sentence-final contour enabling them to make a decision.

6. Perception Experiment 2

Our aim in Perception Experiment 2 was to determine whether participants could use the information structure provided by a discourse context to identify the prosodic version of the sentence that was a best fit, given the interpretation that had been favored.

6.2 Results

As in Perception Experiment 1, we analyzed the percentage of correct responses for each item type, both overall and by scopal relation. The results are presented in Table 5.

Table 5

Average overall percentage correct per item type and per speaker in Perception Experiment 2.

In this experiment, hearers were quite successful at identifying the prosodic match, given the preceding discourse context. Responses were well over 50% in all cases, with the exception of the ‘negation > all’ items, which we return to momentarily. As before, we point out that responses to control sentences in which a because clause and negation scopally interacted exhibited comparable percentages: 82.4% correct overall, with 79.3% correct when the because clause took wide scope and 85.4% when negation took wide scope.

A binomial probability test (p = .5, 99% CI) on the overall average for each item type and each corresponding scopal relation revealed that nearly all averages were significantly above chance level: all > negation: 5.0, p < .00001; many/most > negation: 27.4, p < .00001; negation > many/most: 6.5, p < .00001. The one chance-level pattern was from negation > all: –1.0, p = .31. Thus, in the majority of cases, hearers were overwhelmingly successful at pairing the information from the discourse context with the prosodic version of the target sentence that best matched the favored interpretation and scopal relation. A χ² analysis on the distribution of responses by participant also revealed that it was significantly different from chance: all > negation: χ²(6) = 44.5, p < .0001, negation > all: χ ²(6) = 23.33, p = .0007; many/most > negation: χ²(3) = 56.51, p < .0001; negation > many/most: χ²(3) = 43.54, p < .0001.

As in Perception Experiment 1, we calculated a d′ score for each participant for the test items. The average d′ score was 1.26, which was significantly above chance (0) (t(36) = 8.09, p < .0001). The hit and false alarm rates are plotted in Figure 3. As before, hit rates are higher than false alarm rates.

Figure 3

Hit and false alarm rates for participants’ responses to test items in Perception Experiment 2.

The results demonstrate that participants were not choosing at random; nor were they biased towards one prosodic contour over another. Instead, their choice of speaker rendition was based on the information provided in the previous context, which set up a particular information structure and a particular scopal relation.

7. General discussion

This work is, to our knowledge, the first to systematically test seminal claims by Jackendoff (Reference Jackendoff1972) that differences in scopal relations in sentences with quantification and negation in English are tied to differences in prosodic delivery. We first identified a set of target sentences involving the interaction of quantification and negation from a controlled Production experiment in which the sentences were embedded in a disambiguating discourse context and speakers prosodically differentiated between members of a minimal pair in a manner largely consistent with claims from the previous literature. We then asked whether – given these clear differences correlated with sentence-level meaning – hearers could use such information in the speech signal to arrive upon the correct, intended interpretation. In two Perception experiments, we provided an affirmative answer to this question.

In Perception Experiment 1, we asked whether hearers could use prosody to situate a target sentence in a discourse context that was indicated by two possible continuation sentences. We found that when speakers use prosody to favor an interpretation of a scopally ambiguous sentence that was supported by a discourse context, hearers can successfully recruit this information to assign the correct interpretation. In Perception Experiment 2, we asked whether hearers could use the thread of a brief discourse context to choose the prosodic version of a target sentence that was most compatible with the interpretation favored by that context. Here, we found that participants recruited the information structure to assign an interpretation to a sentence, and were able to select the prosodic rendition of the sentence that best corresponded to that interpretation. The only notable exception to participants' successful performance in this task was with universal quantification–negation sentences where negation took wide scope. This outlier underscores the importance of testing a range of quantifiers and syntactic positions, as generalizations cannot be made just from the universal quantifier all in subject position alone.

Thus, we conclude that when speakers provide them, hearers can recruit auditory cues to disambiguate scopally ambiguous sentences in precisely those cases where these cues are informative about the information structure in the discourse context, which is in turn informative about scopal relations. Given these findings, we must also conclude that psycholinguistic studies focused on participants' ability to disambiguate scopally ambiguous sentences (perhaps even more specifically, with negation, or perhaps more generally involving the interaction between two logical operators) should take auditory cues in the speech signal into account.

As we mentioned in the introduction to this paper, one area in which this conclusion is directly relevant is in acquisition research aimed at identifying whether language learners are able to access the full range of interpretations of quantifier–negation sentences, such as our test sentences and the prototypical sentence from this line of research in (38).

1. (38) Every horse didn't jump over the fence.

Musolino's (Reference Musolino1998) initial observation was that children for whatever reason seem to be locked into the reading where negation takes narrow scope (i.e., the ‘surface scope’ reading) and have great difficulty accessing the ‘inverse scope’ reading where negation takes wide scope (his ‘observation of isomorphism’). This pattern has been replicated across a variety of languages by a number of different researchers, using lexical items beyond the universal quantifier in subject position (see Musolino Reference Musolino, de Villiers and Roeper2011, for a review). However, subsequent research manipulating a variety of contextual factors has revealed that children are in fact able to access the inverse scope reading, and that it is indeed within their grammatical capacity (see Musolino & Lidz Reference Musolino and Lidz2006; Gualmini et al. Reference Gualmini, Hulsey, Hacquard and Fox2008; Viau, Lidz & Musolino Reference Viau, Lidz and Musolino2010). Thus, they simply display an overwhelming preference for the surface scope reading, but are in principle able to access both readings. One might ask, given the current results, what role prosody has to play in children's performance. Musolino & Lidz (Reference Musolino and Lidz2006: fn. 13) remark that ‘we are not aware of a single study on the acquisition of universal quantification in which prosodic cues are manipulated’. To our knowledge, at the time of this paper going to press, this appears to still be the case.

Previous authors have indicated in passing that they were sensitive to the possible role of prosody, but never systematically controlled for this factor as an independent variable. For example, Lidz & Musolino (Reference Lidz and Musolino2002: 130) say that, ‘When making these statements, the experimenter holding the puppet was instructed to say the sentences in a way that is the most naturally compatible with the appropriate reading on which the sentence was a true description of what had happened in the story. This step was taken to ensure that if there are any prosodic cues associated with the different scope readings, they would be provided to the child subjects by the experimenter holding the puppet’. A similar statement is echoed in Lidz & Musolino (Reference Lidz and Musolino2005: 87–88). It is far from clear, however, what the ‘most naturally compatible’ prosody was in those studies – or, given the results of our Production experiment, what this would be, even from puppeteer to puppeteer. On a related note, Gualmini et al. (2008: fn. 14) say of a previous study of Gualmini's that the children's lower acceptance rate in one condition ‘might be due to the particular intonation used by the speaker who recorded the stimuli for that experiment’ but say nothing about what this intonational difference might have been, or how it could have produced a difference.

Gualmini et al. (2008: 219) do state explicitly that their test sentences, such as (39), in a context supporting negation taking wide scope, were ‘uttered with the intonation that is required by the inverse scope interpretation in adult English: stress on every, de-stressed wasn't, and rising intonation on delivered’.

1. (39) Every letter wasn't delivered.

Why they describe the ‘required’ intonation this way is unclear, given (i) the claims in the theoretical literature, which do not describe the required or favored prosody in this way, and (ii) the lack of experimental evidence to this effect at the time that their paper was written. This point aside, children in their experiment did access the interpretation where negation took wide scope over the quantifier 80% of the time. However, the authors do not attribute the increase in accessing this reading to the prosody, and it is not even possible to identify the contribution of prosody in these experiments, given that this factor was not controlled for by introducing a similar condition in which this prosody was not used, or by ensuring that a consistently falling contour was used for the every > negation reading.

Lidz & Musolino (Reference Lidz and Musolino2005) and Musolino & Lidz (Reference Musolino and Lidz2006) refer to the McMahon et al. (Reference McMahon, Lidz and Pierrehumbert2004) study in a footnote, as ‘evidence that adult speakers do not normally use prosody or intonation to indicate the scope of a quantificational subject with respect to negation’ (Lidz & Musolino Reference Lidz and Musolino2005: fn. 10). However, in Section 3, we reviewed reasons why we think that it is not possible to make this generalization based on that study. Even if it were, though, note that this finding says nothing about whether the participants in their study may have been sensitive to any prosodic cues that may have been present in the delivery of the sentences. It therefore remains a wide open question whether children are sensitive to the auditory cues associated with quantifier–negation sentences, as the adult participants in our perception studies were.

Building upon a discussion in Musolino & Lidz (Reference Musolino and Lidz2006: 832), we would like to suggest that the contextual manipulations in the previous studies may have been successful not only in and of themselves but because they also carried with them prosodic cues that may have made the ‘inverse scope’ reading more salient (or easier to access in processing). For example, the contrast sentences in that paper, such as (40), may have provided children with the necessary contrast in scalar alternatives, coupled with fall-rise prosody needed to access the reading where negation took wide scope.

1. (40) Every horse jumped over the log [and/but] every horse didn't jump over the fence.

Given that ‘log’ and ‘fence’ form members of a scale (i.e., ‘things that the horses could have jumped over’), both should be instantiated with fall-rise prosody – especially given the continuation rise on log (see Ward & Hirschberg Reference Hirschberg1985). Moreover, given the contrast between the conjoined positive and negative statements, negation is surely focused in the second statement, meaning that negation is not part of the presupposition and the QUD is positive (see Jackendoff Reference Jackendoff1972, Kadmon & Roberts Reference Kadmon, Roberts, Farley, Farley and McCullough1986) – precisely the scenario that favors fall-rise and negation taking wide scope. Of course, the findings of Viau et al. (Reference Viau, Lidz and Musolino2010) reveal that children can access the inverse scope reading at a comparable rate without the explicit contrast, given other salient contextual manipulations and experiment structuring, so it cannot be the case that the prosodic correlate is necessary – an observation consistent with claims by Ward & Hirschberg (Reference Ward and Hirschberg1985) reviewed earlier.

There is also promising independent evidence that children can, in fact, recruit prosodic information to calculate the partitive scalar implicature associated with some (Miller et al. Reference Miller, Schmitt, Chang, Munn, Brugos, Clark-Cotton and Ha2005), to retrieve information about pronominal reference (Maratsos Reference Maratsos1973), to resolve attachment ambiguities (Snedeker & Yuan Reference Snedeker and Yuan2008), and to assign syntactic structure (Beach, Katz & Skowronski Reference Beach, Katz and Skowronski1996; Syrett Reference Syrett, Franich, Iserman and Keil2010). A fruitful area for future research, then, would be to try to paint a clearer picture of what the precise role of prosody is when children are able to access the ‘inverse scope’ readings that can be so elusive for them. Such research would allow us to see at what stage in development this sensitivity to the auditory correlates of sentence interpretation first manifests itself, and determine how grammatical principles and processing capacity interact in language development. The current research demonstrates that part of becoming an adult language user involves acquiring the ability to recruit prosodic cues to access the interpretations of scopally ambiguous sentences in a discourse context. The findings of the complementary Perception experiments reported in this paper underscore the importance of attending to the manner in which scopally ambiguous sentences are presented in the course of psycholinguistic experiments, as hearers do demonstrate sensitivity to the relation between prosodic rendition of these sentences, the information structure of the discourse context, and the scopal relation.