Introduction

Focus denotes that part of an utterance that is important to a speaker or that the speaker considers to be most informative for the listener (e.g., Lambrecht, 1994). Both cross-linguistically and within a given language, there are different ways to signal focus: via prosody, morphology, or syntax. Common to all of these different types of marking is that they increase the prominence of a focused element within an utterance. This increased prominence, in turn, makes the focused element more salient to the listener. Increased salience has an effect on online language processing: Targets (phonemes, artificially added clicks, or entire words) are detected faster in focused than unfocused spoken language material (Cutler, 1976; Cutler & Fodor, 1979), focused phrases undergo deeper semantic processing (Wang et al., 2009; Wang et al., 2011), and focused phrases are retained better in memory (Birch & Garnsey, 1995; see review by Cutler et al., 1997, for more studies showing how prosodic prominence affects the integration of words in an existing discourse-model). However, it seems to be the case that there are notable individual differences in the perception and interpretation of prosodic prominence. Most notably, differences between the sexes have been reported (e.g., Wang et al., 2011).

The purpose of the present study was to further investigate the role of focus on memory, in particular on the recall of alternatives to the focused element that have been mentioned in the discourse context. A second research question concerns possible differences between men and women. In order to prepare the ground, we present the theoretical framework of Alternative Semantics (Rooth, 1985, 1992), discuss the difference between pragmatic and semantic uses of focus, review sex differences in pragmatic and prosodic processing, and summarize previous findings on focus and memory.

Focus alternatives

Approaches to focus that make use of the notion of alternatives (most notably Alternative Semantics by Rooth, 1985, 1992, 2016, but also Roberts, 2012) assume that focus refers to “alternative phrasal meanings” (cf. Rooth, 2016: 19). We can distinguish between so-called pragmatic and semantic uses of focus (cf. Krifka, 2008; Velleman & Beaver, 2016). In pragmatic use, focus gives rise to a pragmatic inference. For instance, for “Mary introduced [BILL]F to Sue” (example from Rooth, 1992), where the focused element is spoken with nuclear pitch accent (indicated here with capital letters), a listener might draw the inference that Mary introduced Bill to Sue but not to anybody else, negating alternative propositions like “Mary introduced [TOM]F to Sue” or “Mary introduced [ANN]F to Sue.” However, the inference can be cancelled, as in “Mary introduced [BILL]F to Sue. And she introduced [ANN]F to Sue.” By contrast, semantic use of focus affects the truth conditions of a sentence. This is most easily demonstrated for cases where a focus-sensitive operator like the particle “only” associates with the focus. The sentence “Mary only introduced [BILL]F to Sue” asserts that of all the alternative individuals Mary could have introduced to Sue, she introduced Bill and nobody else to Sue. This assertion cannot be cancelled. If it turns out that Mary introduced a second person to Sue, then the first sentence states an untruth.

When we talk about alternative sets in this paper, we refer to the set of referents that can replace the focused referent in the given context. Following Rooth (1992), the focused element is also a member of the set. Thus, in the examples discussed above, the alternative set consists of {Bill, Tom, Ann}.

Focus marking

In different languages, linguistic focus can be marked in different ways: Syntactically by employing non-canonical word orders like fronting or clefts, morphologically by adding focus markers, and prosodically by using particular intonation patterns. In German, as in English, prosodic focus marking is the most common. In German speech-production studies, Baumann, Becker, Grice, and Mücke (Baumann et al., 2007, see also Baumann et al., 2006) observed increases in prominence-lending parameters such as increased duration, higher pitch peak, an increased pitch excursion and hyperarticulation on the accented syllables in focused words. Different pitch accents can be identified based on pitch height and pitch excursion. Pierrehumbert and Hirschberg (1990) catalogued different accent types according to their function. L+H accents – that is, accents with a low starting point, are identified as conveying the salience of some scale. Out of this class of accents, Pierrehumbert and Hirschberg describe the L+H* accent as conveying “that the accented item – and not some alternative related item – should be mutually believed” (p. 296), that is, L+H* is used to mark contrastive focus. However, as Roessig, Mücke, and Grice (Roessig et al., 2019) point out, the mapping between focus type and pitch accent type is not quite as clear-cut as that: Grice et al. (2017) collected data from five participants producing the same words in broad, narrow, and contrastive focus conditions. The authors analyzed the realizations of accents both qualitatively and quantitatively. They found that not all speakers showed the expected categorical shift from H+!H* to H* to L+H* for broad, narrow, and contrastive focus, respectively. All speakers did show quantitative shifts of the three measured dimensions in the expected direction, though (see also Kügler & Gollrad, 2015, for similar results). As far as interpretation is concerned, Watson et al. (2008) showed (for English) that the interpretative domains of pitch accent types partly overlap: Participants listened to spoken instructions and had to manipulate objects presented on a display. Eye movements revealed participants’ anticipations of referents during on-line comprehension: While stimuli spoken with an L+H* accent strongly biased listeners’ anticipations to a contrastive referent, H* accents were compatible both with contrastive and new referents. Thus, there is a certain amount of variability both in the production and in the interpretation of pitch accents for focus.

In contrast to prosodic focus, which gives rise to a cancellable pragmatic inference and, as we have just seen, varies both in production and comprehension, so-called focus-sensitive particles like only or also are structure-sensitive operators that associate with a focused element and relate the value of the focused expression to a set of alternatives (Koenig, 1991: 33). Thus, reference to a set of focus alternatives is part of the meaning of a focus particle. A sentence containing an inclusive/additive focus particle presupposes that it is also true for at least one alternative proposition. A sentence containing an exclusive focus particle asserts that it is only true if it holds for no alternative proposition. Independent of the exact operation performed by the particle, it requires alternatives for its interpretation. This has been called “conventionalized sensitivity to focus” by Beaver and Clark (2008). We believe that the activation of alternatives in the presence of a focus particle might be “stronger” than after pitch accents because the particle serves as an additional, non-negligible cue for focus, whereas “mere” prosodic focus might be overheard or interpreted differently. As discussed in the previous paragraph, there is no one-to-one mapping between a given pitch accent type and its interpretative domain. Due to their clearly defined lexical meaning, focus particles resolve this ambiguity in favor of an interpretation requiring alternatives. In addition, in a sentence containing a focus particle, reference to alternatives is an integral part of the interpretation process which cannot be skipped.

Memory representation of focused elements and alternatives

Fraundorf et al. (2010) contrasted two different accountsFootnote 1 for the effect of prosodic focus on memory. The first account, the granularity account, has been proposed by Sturt, Sanford, Stewart, and Dawydiak (Sturt et al., 2004). These authors tested short-term memory effects of focus with a change-detection paradigm. Participants saw a set-up sentence and a critical sentence. Depending on the structure of the set-up sentence, the critical sentence was either in broad focus or had narrow focus on one of its constituents. The critical sentence was presented a second time, and participants had to indicate whether or not a word in the sentence had changed compared to the first presentation. Participants were better in detecting coarse violations as in replacing “the man with the hat” with “the man with the dog” compared to replacing it with “the man with the cap.” However, if “the man” was focused, participants detected more of these semantically close replacements. Sturt and colleagues therefore suggested that focus increases the granularity (the fineness) of the semantic analysis on a word, such that a word not in focus will be represented at a cruder granularity level than a word in focus (see also Sanford et al., 2006). More recent neurophysiological studies support the view that a focused element is subjected to a deeper, more fine-grained semantic analysis (Wang et al., 2009, 2011). A different account, dubbed “contrast account” by Fraundorf et al. (2010), assumes that the use of specific pitch accents not only modulates the representation of the accented word but also that of a contrast item or a set of contrast items, respectively. Fraundorf et al. (2010) found that contrastive pitch accents increased not only memory for a focused element but also memory for its alternatives. In particular, in their Experiment 3, they showed that a contrastive L+H* pitch accent very specifically improved participants’ ability to discriminate between the focused element for which a statement was true and a mentioned alternative (the “contrast item” in the authors’ terminology), for which the statement was false. In the following paragraphs, we present their study and two others (Fraundorf et al., 2013; Spalek et al., 2014) in some depth because these three studies form the basis of our own experiment.

Fraundorf et al. (2010) were the first to assess memory for focus alternatives. They presented participants with spoken little stories like example (1):

  1. (1)

    Both the British and the French biologists had been searching Malaysia and Indonesia for the endangered monkeys.

    Finally, the {British/ French} spotted one of the monkeys in {Malaysia/ Indonesia} and planted a radio tag on it.

The critical words ({British/ French}; {Malaysia/ Indonesia}) were either realized with an H* pitch accent or a (contrastive) L+H* pitch accent. Experiment 3 of Fraundorf and colleagues is most relevant for us: Participants had listened to the stories in the lab and returned a day later to answer recognition questions, for example:

  1. (2)

    The British scientists spotted the endangered monkey and tagged it – true or false?

Participants’ accuracy in accepting a correct target was increased if it had previously been marked by an L+H* pitch accent compared to an H* pitch accent. That is, participants were more likely to respond “true” to (2) if they had heard “Finally, the BRITISHL+H* spotted one of the monkeys …” the day before than if they had heard “Finally, the BritishH* spotted one of the monkeys …”. Critically, contrastive prosodic focus also helped participants to reject a false claim. That is, they were more likely to respond “false” to (2) if they had heard “Finally, the FRENCHL+H* …” previously, compared to “Finally, the FrenchH* …”. Focus marking did not affect performance for unrelated lures that had not been present in the alternative set presented on day one, like, for example, “Portuguese scientists.” That is, participants’ memory for mentioned alternatives increased if the focused item had been presented with an L+H* pitch accent, and they demonstrated this by remembering that the alternative had not been the element for which the proposition was true. In a later study, Fraundorf et al. (2013) repeated their findings with written stimuli where the focused element was realized with capital letters. In addition, they reported that the memory benefit only extends to contextually plausible alternatives.

Spalek et al. (2014) investigated the contribution of focus particles to the memory for focus alternatives: Participants listened to dialogues (Experiment 1) or narratives (Experiment 2), in which a set of three elements was introduced, see (3) and (4).

  1. (3)

    Matthias receives a parcel with shirts, trousers, and jackets. He considered what he liked.

  2. (4)

    He kept (a) only/ (b) even/ (c) the shirts.

After having listened to a block of ten of these little stories, participants had to answer questions about the dialogues. For critical items, the question was a recall question about the alternative set, for example: “What was in the parcel?” Spalek and colleagues looked at correctly recalled focus items (here: shirts) and alternatives (here: {trousers, jackets}) separately. While recall for the focus items was always better than recall for the alternatives, recall for the alternatives was improved in the presence of a focus particle compared to the control condition without a particle. There was no difference between only and even, that is, it did not matter whether the particle was inclusive or exclusive. Spalek and colleagues concluded that alternatives were made salient by a focus particle because reference to alternatives is part of the lexical meaning of a focus particle. This increased salience led to better memory encoding and hence, improved recall.

There are important differences between the studies by Fraundorf et al. (2010, 2013) and Spalek et al. (2014). Apart from the obvious language difference (Fraundorf et al.’s experiments were carried out in English and those of Spalek et al. in German), the alternative set introduced by Spalek and colleagues was larger (three elements) than the one introduced by Fraundorf and colleagues (two elements). Also, Spalek and colleagues used a recall task, whereas Fraundorf and colleagues investigated recognition memory. While recognition memory can rely at least to some extent on item familiarity, recall requires complete recollection of the studied episode (see Yonelinas, 2002, for a review on familiarity and recollection in memory). Most importantly, though, the manner of focus marking differed. Fraundorf and colleagues used prosodic focus marking, whereas Spalek and colleagues used focus-sensitive particles. As discussed above, the inference caused by prosodic focus marking is cancellable, that is, it can be taken back without causing any linguistics violations, and pitch accents like H* and L+H* do not map strictly onto information-structural categories like “new information” and “contrastive information.” Therefore, it seems likely that prosodic focus marking leaves more room for individual differences in processing. By contrast, reference to focus alternatives is a necessary part of the meaning of focus particles. That is, a focus particle “needs” alternatives in order to be interpreted, and hence, activating an alternative set is not optional. Seen like this, focus particles might be regarded as a stronger cue to the presence of alternatives than mere prosodic markings. One of the aims of the present experiment was to directly compare the study by Spalek et al. – including particles – with a replication where focus is marked with prosody only. Thus, the question is whether accent by itself can also make alternatives more salient and therefore improve their recall. Observing a memory benefit in recall for the focused element alone would support the granularity account, whereas observing a memory benefit either for recall of the focused element and its alternatives or for recall of the alternatives alone would join Fraundorf et al. (2010, 2013) and Spalek et al. (2014) as support for the contrast account.

In the present paper, we ask the following research questions: Do we observe a memory benefit for alternatives to prosodically focused elements in a recall task? That is, will we replicate the findings of Fraundorf et al. (2010) with a different task and with a larger alternative set (three items instead of two)? An additional question concerns the comparison of focus marked by prosody (arguably the weaker cue and the type of marking that allows for more individual variance) compared to focus marked by the additional presence of a focus particle. We use the same paradigm and the same stimuli as Spalek et al. (2014) (only slightly altered to replace focus marking with particles by focus marking through accent), which allows for a direct comparison of the results. Finally, we include participants’ sex as a factor in our design to test if females and males show equal memory benefit. In the following sections, we outline why we might expect differences between men and women.

Sex differences in memory

It is well documented that women outperform men in episodic memory tasks if the material to be memorized is verbal, whereas men outperform women for episodic memory on visuo-spatial materials (e.g., Herlitz et al., 1997; Herlitz & Rehnman, 2008; Lewin et al., 2001).

In a meta-analysis on the brain areas underlying working memory, Hill et al. (2014) observed that women and men activate gender-specific networks during working memory tasks (see also Goldstein et al., 2005, for an fMRI study on auditory verbal working memory). While the neural differences were not accompanied by corresponding performance differences (in fact, groups were matched on working memory scores), they support the assumption that memories are encoded and retrieved somewhat differently between the sexes.

While the previously reported studies argue for a general memory benefit for women, women might also process prosodic cues differently from men and this might affect their memory for focus alternatives. In the next section we summarize findings supporting this hypothesis before turning to our study.

Sex differences in semantic and intonation processing

It has been suggested that sex differences exist in the functional organization of the brain for language (e.g., Shaywitz et al., 1995). This finding has been hotly contested, with a number of reviews and meta-analyses unable to replicate these sex differences (e.g., Ihnen et al., 2009; Wallentin, 2009). Kaiser et al. (2009) find that sex effects in brain organization are very variable and depend on many factors, including choice of analysis and choice of threshold. This idea is seconded by Harrington and Farias (2008), who observed that, if the analysis method is held constant, sex effects do not obtain in all language tasks. Bearing this cautionary advice in mind, there is evidence that sex differences might underlie some areas of language processing that are relevant for the task used in the present paper:

Kansaku et al. (2000) observed sex differences in the distribution of neural activity in a story-listening task. They argue that sex differences only become apparent when subjects are required to process the global structure of sentences. Kaiser et al. (2007) found sex differences in lateralization (with stronger left lateralization for women) in a silent free narration task. Sex differences in the underlying brain activation were also obtained in a task testing coherent story comprehension and reasoning about false beliefs (Frank et al., 2015). These findings suggest that sex differences in language processing are particularly strong when participants have to process longer sequences of coherent text, which is exactly what participants had to do in the recall task used here. However, sex differences have also been obtained for processing single words. Wirth et al. (2007) report an earlier and longer-lasting N400 component in the EEG of women compared to men when participants had to passively read related or unrelated word pairs. The authors argue that women conducted a deeper semantic analysis on these word pairs than men (see also Wang et al., 2011).

Sex differences in the interpretation of intonation have most often been investigated in the context of emotional speech processing. Wildgruber et al. (2002) observed sex differences in the brain areas that were involved in a task where participants listened to two renderings of the same sentence, spoken by the same woman in a happy, neutral, sad, or angry voice, and they had to decide which of the two renderings was more “explicit.”

Findings from an EEG-study by Schirmer et al. (2002) on emotional speech processing suggest that female listeners integrate prosodic and semantic information much faster than male listeners. Schirmer et al. (2005) built on this finding by looking at whether men and women also differ in their pre-attentive reaction to neutral and emotional stimuli. They investigated the mismatch negativity in the EEG of participants who listened to bi-syllabic stimuli (da-da) spoken in neutral or emotional (angry/happy) prosody. While both sexes reacted to deviants by presenting the classic MMN component, only women varied in the size of the MMN depending on whether it was in response to an emotional or to a neutral deviant (see also Hung & Cheng, 2014, for very similar findings).

Furthermore, the results of a neuroimaging study by Schirmer et al. (2004) on emotional speech processing suggest that male listeners are less susceptible to intonation manipulations as evidenced by reduced IFG (inferior frontal gyrus) contrast activations for incongruent versus congruent speech presentation.

Given these findings in the literature, we a priori decided to test a gender-balanced sample in the present study to further explore the possible influence of participant sex on how prosodic focus is exploited when creating a representation of a just heard story.

Study design and method

We set up a delayed recall experiment with auditory stimulus presentation and orthographic prompts. A combined within-subjects plus within-items design was employed with each participant and each experimental item tested in both experimental conditions: broad focus and contrastive focus.

Participants

107 young adults were recruited from the participant database LingEx sustained by the Leibniz-Zentrum Allgemeine Sprachwissenschaft. All participants were native speakers of German, over 18 years of age, and 57 were female). Due to a clerical error, 13 subjects were tested although they had been tested shortly before using a similar stimulus set in a related fMRI experiment (Spalek & Oganian, 2019). These 13 participants (ten female) were excluded from the test sample. The remaining 94 participants (47 female) ranged in age from 18 to 39 years with an overall mean of 25.4 years (SD = 4.9; Mfemale=25.4; Mmale=25.2).

None of the participants reported hearing loss, language disorders, or had a history of a neurological disease. The procedure was approved by the ethics committee of the Deutsche Gesellschaft für Sprachwissenschaft (German Linguistic Society, https://dgfs.de/de/inhalt/ueber/ethikkomission.html). All participants provided written informed consent and were instructed that they could withdraw from the study at any time.

Materials

All materials are based on Experiment 2 in Spalek et al. (2014). Participants listened to short texts containing two context sentences and a critical sentence (cf. example (5)). The first context sentence introduced a set of three elements (nouns) from taxonomic categories (e.g., furniture, tools, fruits) and connected a person with the setting. The three elements (target words) from the first sentence ranged in frequency between 0.01 and 196 occurrences per million (cf. DLEX database; Heister et al., 2011). The norms for the taxonomic categories to which the target words belong can be found in Schröder et al. (2012). The second context sentence continued the story and, in most cases, indicated a choice or selection to be made by the person. The critical third sentence complemented the context sentences and re-mentioned one of the three elements from the first sentence. Half of the experimental items contained the definite determiner before the re-mentioned noun and half of them did not, depending on what sounded natural in the given contexts. Care was taken that, across items, the focused element in the critical sentence was equally often the first, second, or third element from the first context sentence. Two versions were constructed for each critical sentence: (a) a control condition with broad focus (H+!H* pitch accent) and (b) a version with a contrastive focus (L+H*pitch accent) on the aforementioned element.

figure a

We used a set of 80 experimental items (45 targets and 35 fillers) plus six practice items. The fillers had the same structure as the targets and presented lists of three elements from various categories, either taxonomic or non-taxonomic (cf. Table 2 for examples). Half of the fillers were presented with broad focus and half with contrastive focus in the critical third sentence (see Appendix A for a transcription of all target items. All sound files are provided in the Online Supplementary Materials, https://osf.io/txq5r).

A professional female speaker was recorded in a sound-shielded booth using a directional Sennheiser (ME64) microphone and an Edirol E09 solid state recorder (44.1 kHz sampling frequency, 16-bit resolution). The speaker had a middle German accent close to the standard variety of German. Two versions of each critical practice, filler and target sentence were recorded – a broad focus version (H+!H* pitch accent on the critical word) and a contrastive focus version (L+H* pitch accent on the critical word). Note that, in contrast to the stimuli used in Spalek et al. (Spalek et al., 2014, Experiment 2), no focus-sensitive particles (e.g., “only”) were used in the present study. The difference in the information structure for the critical sentences resulted from prosodic focus and was thus signaled by means of intonation.

The critical sentences for the broad focus condition were elicited with the pre-recorded wh-question “Was war los?” (“What happened?”) produced with a typical declining intonation contour (H* L-%; GToBI notation, cf. Grice & Baumann, 2002). The resulting broad focus answers of our speaker showed the German default declarative accentuation pattern, which assigns an H+!H* tone accent (GToBI notation) to the re-mentioned noun (L-% sentence final edge tone, GToBI). Contrastive focus productions of the critical sentences were responses to pre-recorded yes/no questions like “Sie hat die TISCHE ausgesucht?” (“She chose the TABLES?”). These yes-no questions were realized with a low nuclear tone accent (L*; GToBI) associated with the target noun (TABLES in the above example) followed by a high-rising interrogative contour (H-^H% edge tone, GToBI). Our speaker’s contrastive focus answers were characterized by a low-rising nuclear tone accent (L+H*, GToBI) on the focused constituent (i.e., the re-mentioned noun) in the critical third sentences and showed a default low sentence final edge tone (L-%, GToBI).

Acoustic analyses conducted with Praat (Boersma & Weenink, 2018, Version 6.0.16) confirmed that the critical elements were longer, had an increased sound pressure level and showed a steeper F0-excursion for the contrastive condition (L+H* pitch accent) compared to the broad focus condition (H+!H* pitch accent). Table 1 presents means and standard errors for the acoustic parameters duration, mean intensity, maximum and minimum pitch as well as pitch range for both focus conditions. Figure 1 displays averaged intonation contours of the re-mentioned elements (focused words) and the respective lexically accented syllable for the two focus conditions. Figure 2 illustrates the pitch contours for the duration of the critical sentences for both focus conditions.

Table 1 Mean acoustic parameters of the re-mentioned elements in the critical sentences (e.g., Regale in (5a, 5b), target items only)
Full size table
Fig. 1
figure 1

Mean pitch contour of focused element in the critical sentences for contrastive and broad focus condition (left: F0 contour across critical word; right: F0 contour for critical syllable; target items only). Error bars represent the standard error of the mean

Full size image
Fig. 2
figure 2

Pitch contours of the critical sentences (n=90) for contrastive vs. broad focus condition (gray: broad focus pitch contours; red: contrastive focus pitch contours); dashed window roughly indicates the position of the critical word

Full size image

The 45 target items were presented on a pseudo-randomized list, interspersed with the 35 filler items. Two versions of this randomization list were employed (A and B). These two versions complemented each other in terms of focus condition for the target items. That is, a target item with broad focus on version A of the list was presented with contrastive focus on version B. The sequence of the items (targets and fillers) was identical for both versions of the presentation list. All in all, the experiment contained eight experimental blocks (ten items each, six or seven targets per block) plus a unique initial training block with six items (four filler-like items, two target-like items). Stimuli were randomized within blocks such that neither the same stimulus type (target, filler) nor the same focus condition (broad, contrastive focus) was presented more than three times in a row. In addition, taxonomic categories did not reoccur within blocks. Moreover, stimuli with identical taxonomic categories were arranged in experimental blocks with maximal distance from each other. Participants were quasi-randomly assigned to one of the two versions of the presentation list (participants’ sex was balanced across lists). That is, each participant heard a target item only once in a particular focus condition. Twenty-three out of 45 targets were presented with contrastive focus on version A of the list (22 targets realized with broad focus) and vice versa on version B. Seventeen out of 35 fillers were presented with contrastive focus and 18 with an intonation contour for broad focus on both versions of the list.

Apparatus

Participants were seated in front of a Belinea TFT monitor and wore a Sennheiser PC131 headset with an integrated microphone. Stimulus presentation and recording of the answers was controlled by Neurobehavioral Systems Presentation software (Version 16.5).

Procedure

The experiment started with an on-screen instruction informing participants about the structure of the experiment and the task they would have to perform. The experiment consisted of two paired phases: (a) an encoding phase (ENC) during which the participants listened to the short texts and (b) a cued recall phase (REC). Thus, phases of auditory presentation alternated with test phases in which participants were cued for overt recall. Subjects were informed that the recall phase was time-limited and that they were supposed to respond aloud. After the instructions, subjects performed six practice trials (a block of six three-sentence texts (ENC) followed by six questions (REC)) and were allowed to adjust the sound volume of the headphones. Table 2 presents a schematic of the experiments’ procedure.

Table 2 Experimental procedure
Full size table

In the encoding phase, each trial began with a central fixation cross displayed for 500 ms. Then, the three-sentence text was presented auditorily, sentence by sentence with a silent pause of 500 ms duration between sentences. After a silent interval of 4 s the next trial started automatically. The average trial length in the encoding phase was about 12 s. Once ten texts (i.e., one ENC block) had been presented, the recall phase was initiated automatically. First, a screen informed the participants that the phase with the queries was about to begin (display duration 2 s.). Then, after a central fixation cross was presented for 500 ms, a test cue (question) appeared on the screen for 3 s followed by a ‘#’ symbol that was displayed for the remaining trial duration. Subjects were supposed to respond aloud as soon as the ‘#’ symbol appeared on the screen. Two types of questions were used as recall cues depending on whether the respective trial was a filler or a target. For the target items the questions aimed at the three elements presented in the first sentence of the text (e.g., “Which pieces of furniture were there in the furniture shop?” for example (5), above). For filler trials we asked 25 information questions and 10 yes/no questions on the texts presented in the respective encoding phase. These questions aimed at the action performed (e.g., “What did Susanne do with the bedspreads?”), the name or sex of the person involved (e.g., “What was the name of the person who ate ice cream?”), the location of the action (e.g., “Where was Sophie?”) or the element that was re-mentioned (e.g., “What was it that Stefan lost during the rugby game?”). The set of fillers was included to discourage participants from memorizing the list of three elements only. Participants had 20 s to respond aloud and their answers were recorded. In order to not lose information if a participant responded too early, that is, while the question was still on the screen, recordings actually started from the onset of the test cue. Participants could terminate trials with a button press as soon as they thought their answer was sufficient. We instructed the participants to indicate aloud if they did not know the answer to a question and then to press the button to terminate the respective trial. Immediately after the time out or after the participants’ button press, the next trial was initiated. All items were tested for recall in the same order as presented during the encoding phase. Thus, the number of trials and approximately the amount of time delay between encoding and recall was kept constant and subjects could easily keep track of the sequences. Between blocks, participants were asked to perform a ten-step n-backward counting task using decrements of two to nine and to take a small self-paced break. This procedure should reduce interference effects between blocks, because some taxonomic categories (e.g., fruit, vegetables, animals, furniture, plants) were used more than once in the experiment (but only once within a given block). After the experiment, a questionnaire was administered asking the participants for basic demographic information, what they thought the experiment was about, and whether they employed any specific strategies. An entire testing session lasted about 50 min. The participants were paid €8 in compensation.

Results

The recorded answers were transcribed and recall accuracy per word was coded independently by an assistant annotator and the first author. Recall for each word was coded binarily, that is, a word was coded as “recalled“ (1), “not recalled“ (0), or it was excluded (NA). If participants mentioned a synonym (e.g., “shovel” for “spade”; cf. Duden, 2007, 2014, 2018) or a hyponym instead of a target word (e.g., “handbag” instead of “bag”) we coded this as a correct response. Cases of coding mismatch between the two raters were discussed with the second author and a binding coding scheme was established. Appendix B presents all synonym and hyponym decisions used in our coding scheme. If a participant indicated that he/she did not know an answer for a whole trial this was coded as not recalled (0) for all three words to be recalled for the respective item. One experimental item (item 44) was excluded from the analyses, because it resulted in uncertainty about the correct answer in the recall phase for a majority of participants. Further, 34 trials (out of 4,136 trials, i.e., approx. 0.8% of the data) could not be analyzed because the respective audio recordings were empty. We decided to exclude these trials completely, because we could not decide afterwards whether the data loss was due to technical failure or because participants forgot to indicate that they did not know an answer to a question. The raw data are available in the Online Supplementary Materials (https://osf.io/txq5r).

Since we were primarily interested in the recall of the alternative set, we carried out two analyses, splitting the data into recall of the alternatives (e.g., “tables” and “beds” in example (5)) and into recall of the focused element (e.g., “shelves” in (5)). First, we analyzed the effect of contrastive focus on the probability to recall contextual alternatives (from here onwards: recall probability) to the focused element. Descriptive statistics pointed towards higher recall probability in the contrastive focus condition (M = 57.6% contextual alternatives correctly recalled) compared to the broad focus baseline condition (M = 55.1% contextual alternatives correctly recalled). Moreover, recall accuracy (across both focus conditions) was considerably lower for male than for female participants. We applied logistic mixed models in the statistical computing environment R (version 4.0.0) with the lme4 package (version 1.1-23) (cf. Bates et al., 2015) to assess the effects of focus and participants’ sex as well as their interaction on recall for contextual alternatives. Focus condition (broad focus vs. contrastive focus) and sex (male vs. female) were sum-coded. We started out with the maximal random effects structure given the experimental design, as advocated in Barr, Levy, Scheepers, and Tily (2013) (formula: glmer(recall~focus*sex+(1+sex|item)+(1+focus|participant)+(1|word),family=binomial, control=glmerControl(optimizer = "bobyqa",optCtrl=list(maxfun=2e5)))). “Item” refers to a story (e.g., example 5), “word” to a single word (e.g., “tables” in example 5). Random slopes were added to allow for the possibility that the effect of intonation differed across participants and that there were sex-dependent random effects of item. Likelihood-ratio tests (with the “anova()”-function) showed that the random slope for intonation improved model fit only marginally (χ2(2) = 5.15, p = .076), and it was therefore left out. Four control variables were tested via likelihood-ratio tests for their potential effects on recall for contextual alternatives: the version of the randomization list, position of the focused word in the list of elements (scaled and centered), target word frequency,Footnote 2 and (scaled and centered) trial number (χ2(1) = 6.39, p < .05). Allowing for the possibility that the effect of trial number differed across participants by adding a random slope for trial number on the participant intercept drastically improved model fit (χ2(2) = 63.7, p < .001). Finally, we tested (based on the suggestion of an anonymous reviewer) whether adding the quadratic component of trial number by using a polynomic predictor further improved model fit, which it did (χ2(4) = 14, p < .01). The resulting parsimonious model (cf. Table 3) showed facilitatory effects of contrastive focus (B = 0.15, |z| = 2.83, p = .005) and trial number on recall probability for contextual alternatives (B = 25.83, |z| = 3.09, p = .005, linear component). Contrastive intonation on the focused element improved recall probability for contextual alternatives by 3.5% compared to the broad focus condition (cf. Fig. 3, left bar plot). Recall probability was higher for trials later in the experiment compared to earlier trials. In accordance with the descriptive findings above, the model revealed that male participants showed a generally lower recall probability for alternatives (-18.6%) as evidenced by a sex effect on contextual alternative recall (B = -0.79, |z| = 3.24, p = .001). Furthermore, we found a focus × sex interaction (B = -0.20, |z| = 1.97, p = .049), that is, a reduced focus effect on recall probability for contextual alternatives in male compared to female participants.

Table 3 Fixed-effect estimates (top) and variance estimates (bottom) for GLMER of recall for contextual alternatives (recall probability ~ focus * sex + poly(trial number,2) + (1+sex | item) + (1| word) + (1+poly(trial number,2) | participant), n = 8271, log-likelihood: -4705), coding scheme: sum coding
Full size table
Fig. 3
figure 3

Model predictions for recall probability of contextual alternatives. Bar charts split by focus. Left: Pooled results. Middle: Female data. Right: Male data

Full size image

Given that the interaction between focus and sex was significant, we carried out separate analyses for females and males. The models were identical to the full models (apart from the fact that the main effect of sex and the random slope of sex on the item intercept were no longer included). For females, the effect of focus was significant (B = 0.24, |z| = 3.24, p = .001), as was the linear component of (centered) trial number (B = 17.82, |z| = 2.34, p = .020). For males, the only significant effect was an effect for the linear component of (centered) trial number (B = 18.31, |z| = 2.78, p = .005). Critically, the effect of intonation was not significant (B = 0.04, |z| = 0.62, p = .537)

Second, we analyzed the effect of prosodic focus on the recall probability of the focused element. Descriptive statistics showed that the focused element was correctly recalled in 68% of the trials in the broad focus condition and in 69% of the trials in the contrastive focus condition. We started out in the same way as for alternative recall, with the maximal model justified given the design. Since there was only one focused word per item, we did not need separate random effects for word and item but used only one. We eliminated random slopes if they did not improve model fit. Again, control predictors were added to the most parsimonious model one by one and kept if they improved model fit. The final model contained fixed effects of focus condition and sex and their interaction and a fixed effect for experimental list. We modelled random intercepts for participants and items and the random slope for focus condition on the participant intercept. This model revealed that the recall of the focused element was not affected by focus (B = 0.06, |z| = 0.65, p = 0.51), but again by participants’ sex (B = -0.84, |z| = 4.07, p < .001). A summary of this model and a figure of the model predictions is presented in Appendix C.

In order to explore whether the focus × sex interaction for the recall of contextual alternatives described above may be associated with distributional differences between male and female participants (beyond a shift in mean), we analyzed the variability of the contrastive intonation effects on recall probability (for alternatives). We summarized the effect of intonation on recall probability in a single variable per participant: gain. Gain corresponds to the percent contextual alternatives recalled if the focused element is presented with contrastive focus minus percent contextual alternatives recalled if the focused element is realized as broad focus. Influential subgroups with differential susceptibility regarding the prosodic manipulation may result in a non-normal (e.g., bimodal) distribution for contrastive focus gain. Such a finding would call for additional studies testing a battery of covariates that may account for such susceptibility differences (e.g., pitch discrimination abilities, ability to spot contrastive intonation). Moreover, participants may exist, who, while they are able to discriminate contrastive from non-contrastive intonation, do not benefit from contrastive intonation. In this case we should also observe deviations from the normal distribution across participant data or across data subsets split by participants’ sex. Consequently, we calculated the contrastive focus gain for all observational units separately in order to be able to visualize the order of magnitude of the contrastive focus gain and its dispersion. Figure 4 shows the distributions of contrastive focus gain aggregated over target words, items, and participants. Target-, item-, and participant-related distributions more or less resemble each other: the distributions did not violate the assumptions of a standard normal distribution (Shapiro-Wilk Test, p > .05 for the three datasets), and they were unimodal with a mean of approximately 2–3% contrastive focus gain. Figure 5 illustrates that female participants show more contrastive focus gain compared to male. However, both male and female histograms look similar except for the fact that the female distribution is skewed towards higher values. Neither of the distributions violates the assumptions of the standard normal distribution (Shapiro-Wilk Test, p > .05 for both datasets).

Fig. 4
figure 4

Histograms for contrastive focus gain (100*(contrastive recall – broad recall)) across three aggregated datasets. Upper left: aggregated across target words. Upper right: aggregated across items. Lower: aggregated across participants

Full size image
Fig. 5
figure 5

Aggregated across participants per sex; means displayed as solid lines in histograms. Upper: male participants. Lower: female participants

Full size image

Based on the suggestion of an anonymous reviewer, we also did a post hoc investigation of the item variability: While all words were introduced as hyponyms to a hyperonym that was named in the third critical sentence, some of these hyperonyms were conventional, like, for example, FURNITURE or FRUIT, whereas others were more ad hoc like CLEANING UTENSILS or WRITING MATERIALS. We classified the items as “conventional” (n = 28) and “ad hoc” (n = 9). A further seven items were classified as “unclear.” These were items with a clear hyperonym where the individual words were less typical for this hyperonym, for example “paint brush” and “file/ rasp” for TOOLS or cases where, in retrospect, the hyperonym struck us as too generic, like PERSONS, where a more fitting hyperonym would have been PROFESSIONS. Figure 6 visualizes how the gain is distributed across these different types of items. While words from items with conventional hyperonyms are evenly distributed across negative and positive gain, more words from the ad hoc hyperonyms contribute to the overall recall benefit for contrastive intonation (i.e., positive gain). Even more strikingly, all words from the unclear categories contribute to the overall recall benefit. These observations suggest that the beneficial value of focus intonation for forming an alternative set is particularly high for items that do not normally co-occur. Focus indicates the presence of alternatives, which causes a listener to scan the linguistic environment for possible alternatives. If the resulting set is one that already co-occurs often, like “apples, pears, plums,” the additional effect of focus for creating a salient set is not very strong. However, for items that have no or no strong previous relationship, focusing one of them causes the listener to view them as members of a set (the alternative set), thereby increasing their salience. The view that contrastive focus marking does not (or only minimally) increase recall for sets of co-hyponyms received is further supported by the overall recall performance (averaged across the two levels of the focus condition), which was highest for hyponyms of conventional hyperonyms (mean = 59.4%, SD = 14.7), followed by recall for elements from ad hoc hyperonyms (mean = 53.5%, SD = 8.7) and unclear cases (mean = 47.7%, SD = 15.5). Note that our experiment was not designed to test these questions; further research investigating the effect of focus on memory recall for different types of alternative sets will be needed to systematically address this question.

Fig. 6
figure 6

Contrastive focus gain for each word. Positive values indicate improved alternative recall if the focused element was produced with a contrastive accent

Full size image

Discussion

In a delayed-recall experiment, we presented equal numbers of male and female participants with auditory discourses in two different versions and assessed their memory for contextual alternatives as a function of the focus structure (broad focus, contrastive focus), which was conveyed prosodically (H+!H* vs. L+H* pitch accents realized on the focused item). The analyses show that recall for contextual alternatives was facilitated by contrastive focus. Interestingly, this effect was mainly driven by female participants. In what follows, we first discuss our findings with regard to the memory representation of discourse and different means of focus marking. Second, we discuss the finding that women showed a greater effect of prosodic focus marking than men in our task.

Memory representations and means of focus marking

Previous studies (Fraundorf et al., 2010; Spalek et al., 2014) had already revealed that explicit focus marking improves memory for focus alternatives. Fraundorf et al. (2010) had shown that contrastive focus improved recognition memory for alternatives. Specifically, their participants were better able to reject claims that were true for the focused item but not for the alternative if the focused element had been produced with a contrastive pitch accent. By contrast, Spalek et al. (2014) had demonstrated that recall for contextual alternatives is facilitated in the presence of focus particles (e.g., “only the apples”), that is, in cases of “conventionalized” focus sensitivity (Beaver & Clark, 2008). In both of these studies, as well as in the present study, the focused element in the critical sentence was already contrastive for contextual reasons, independent of its prosodic realisationFootnote 3: In the set-up sentence, a list of three elements is mentioned (e.g., tables, beds, shelves; see example 5). In the critical sentence, one of these three elements is repeated. Thus, a listener is bound to ask herself or himself: “And what about the other two things?” Therefore, what the present study (and those by Fraundorf et al., 2010, and Spalek et al., 2014) investigated was not an effect of focus per se but whether additional explicit marking of the focused element increased the salience of its alternatives. Arguably, prosodic focus cues signal focus less reliably since pragmatic inferences caused by a pitch accent can be cancelled – and might not be drawn by all individuals in the first place. The focus particle is an additional, non-negligible marker and as its lexical meaning requires access to focus alternatives, it makes the presence of alternatives more salient, as argued by Spalek et al. (2014). The present study’s result that prosodic focus facilitates recall for contextual alternatives but not the focused element complements the previous findings in two aspects. First, the results of Fraundorf et al. (2010) were replicated, but with a recall task that requires full recollection of the studied episode. In addition, the alternative set consisted of two elements in the study by Fraundorf and colleagues and of three elements in the present study, which poses even stronger demands on memory. Second, our results deliver empirical evidence in favor of the contrast account (e.g., Braun & Tagliapietra, 2010; Fraundorf et al., 2010) and to the disadvantage of the granularity account (cf. Sanford et al., 2006) as recall for contextual alternatives was boosted whereas memory for the focused element was not improved by focus marking in our study. This means that although a focused element may be encoded more specifically (see the review by Cutler et al., 1997; Sanford et al., 2006; Wang et al., 2011), related words (i.e., contextual alternatives) are not suppressed. In fact, focus seems to trigger a deeper semantic processing and/or enhance attention processes spreading benefit especially to the members of a contrast set. Note that we are specifically referring to Experiment 3 in Fraundorf and colleagues. In their first two experiments, where the authors probed recognition of the focused element, L+H* accent did improve memory for the focused element. We can only speculate why we did not obtain a memory benefit for the focused element itself in our recall study. While Fraundorf and colleagues quizzed participants on what happened in their stories (and what did not, the title of Fraundorf et al., 2013), we quizzed them on the content of the entire alternative set. Put differently, if we partition the stimuli in set-up sentence and critical sentence, we probed for the content of the set-up sentence, whereas Fraundorf and colleagues probed for the content of the critical sentence. Thus, a strategic component might have entered into processing such that participants in Fraundorf et al. oriented their attention more to the focused element whereas our participants concentrated more on the three list items.

Attentive readers may have noticed that we tested approximately three times the number of participants as in the precursor study (i.e., Spalek et al., 2014, Experiment 2). The reason for this was that we expected prosodic focus to yield smaller effect sizes compared to focus using focus particles. In addition, we tested two experimental groups, male and female listeners, which also made it necessary to increase the sample size. The effect size of prosodic focus on contextual alternative recall for our gender-balanced sample was approximately 3.5% compared to 4.5% effect size for the focus particle manipulation in Spalek et al. (2014, Experiment 2: 33 participants, 12 male). However, in a recent corrigendum to the original article, Spalek et al. report that, after correcting several coding errors, the effect in Experiment 2 (that is, the one we had based our study on) was no longer significant. Descriptively, the advantage had shrunk to 2% (while a 6% advantage was still present after corrections in their Experiment 1). Thus, the question whether focus signaled by focus particles triggers enhanced recall for discourse representations compared to focus cued by focus accenting remains undecided. Future studies may investigate this issue by using both focus types, contrasting the one manipulation against the other with the same test population.

Are there alternative explanations for the memory benefit caused by contrastive intonation? As an anonymous reviewer has pointed out to us, rather than claiming that focus improved recall for alternatives, we can only say that acoustic salience improved memory for focus alternatives, since the critical element in the third sentence was already “inherently” contrastive due to the way the stimuli were set up. This is in line with Calhoun (2009), who argues that there are constructions that are inherently contrastive (she uses the term “kontrastive”) but that the salience of the members of the alternative sets is affected by prominence. In particular, she claims that “[t]he more prominent a word than expected (on the basis of its syntactic/discourse properties), the more salient the alternative set, and therefore the more likely a contrastive reading” (p. 74), and also with the “effort code” described by Gussenhoven (2004), who argues that increased effort in production such as wider excursion of the pitch movement is often interpreted to express the “significance” of what is being said. This, in turn, is most often grammaticalized in the function of (contrastive) focus. The anonymous reviewer suggested the presence of contrastive intonation might not actually improve recall of focus alternatives. Rather, since the listener expects to hear an L+H* accent, its absence might hamper processing. Thus, rather than observing a memory benefit for alternatives with contrastive focus intonation, we might observe a memory detriment for alternatives with neutral intonation. We agree that it is notoriously difficult to determine the direction of an effect if only two conditions are compared. However, since it was the recall for alternatives (and not the recall for the focused element) that was affected by the experimental manipulation, the conclusion that intonation affects the salience of the mental representation of an alternative set and, therefore, its later recall, is still justified.

Individual differences in memory for focus alternatives

As a secondary research question, we had set out to compare the performance of males and females in recall of focus alternatives. We have found both a main effect of sex such that women remembered significantly more alternatives than men and an intonation × sex interaction effect such that women benefited more from the prosodic cue than men did. The main effect is in line with reports of female superiority in episodic memory tasks with verbal materials outlined in the introduction (e.g., Herlitz et al., 1997; Herlitz & Rehnman, 2008; Lewin et al., 2001).

Resolving the sex × intonation interaction effect revealed that only the female group showed a significant memory gain if the focused element has been contrastively accented. In a recent replication study of the present finding in Vietnamese (Tjuka et al., 2020), the authors also found an interaction of sex × intonation, and observe that, even descriptively, the memory benefit for contrastive focus on alternative recall is only present for the female sample. However, as Fig. 5 illustrates, there are individuals among both males and females who show a large gain, individuals in both groups who do not show any gain, and individuals who show a reverse gain. Thus, it has to be kept in mind that the following discussion applies to the groups, not to an individual male or female listener.

Our result that male listeners do not make use of prosodic focus for the recall of heard discourse as much as female participants do is in line with the finding that sex differences in language processing are more marked during the later stages of processing related to semantic processing and interpretation (cf. Wang et al., 2011; Wirth et al., 2007). Given the nature of our task where participants had to listen to three-sentence stories and later answer recall questions about these stories, it is not surprising that this should be another task domain where sex differences obtain. Two reasons why males might make less use of the prosodic information can be assumed: Either they do not perceive it as clearly as females do or perceiving the prosodic information does not lead them to the same interpretation.

There is some evidence that men and women differ both in the intonational patterns they use themselves in speaking and in their processing of intonation: Haan and van Heuven (1999) describe the intonation patterns of Dutch speakers in different types of utterances. The female range on the majority of acoustic measures was wider than the male range. Similar findings are reported by Daly and Warren (2001) for New Zealand English. These authors report a greater pitch range and greater pitch dynamics for women than for men, and this effect is more pronounced in story telling than in reading aloud sentence lists. These differences in production might in part be able to explain differences in sensitivity for intonation, supporting the first assumption that the male group did not perceive prosodic focus as clearly as the female group did.

As discussed in the Introduction, Schirmer et al. (2002) reported that male listeners do not integrate prosodic and semantic information as quickly as female listeners do. If male listeners cannot exploit intonation to the same extent as female listeners, their discourse representations may also differ: Delayed or incomplete integration of the prosodic information might have precluded increases in salience in the mental representation of the alternatives for the male group. Therefore, the two conditions would not have seemed different to them, which explains the null effect. The assumption that the male group processed both intonation conditions in the same way pragmatically is even more likely because, logically, the prosodic realization of the focused element does not add any new information about its information structural status – contextual information alone is sufficient to interpret the focused element as contrastive (i.e., contrasting with the other two list items).

There is, however, an alternative explanation for the reduced effect on alternative recall in the male group: Recall accuracy generally reduced in male compared to female participants for both contextual alternatives and focused items. Thus, reduced overall recall in male participants might have resulted in less statistical power to observe a focus effect in this group (floor effect). The attempt to identify influential subgroups underlying the focus × sex interaction with an additional distribution analysis was not successful. This analysis rather indicated that continuous traits may play a role in contrastive focus processing across participant sex. Auditory working memory and sustained attention may be candidate measures to be included in future studies elaborating on our findings.

Conclusion

Prosodic focus conveyed by intonation improved subsequent recall memory for alternatives to the focused element. This finding supports accounts like the contrast account that posit that focus signals the relevance of alternatives for the interpretation of an utterance, thereby increasing the salience of these alternatives in a listener’s mind. The memory improvement was only observed for women, not for men, in accordance with sex differences reported for memory, discourse processing, and the integration of intonational information with language interpretation.