Participants

Twenty-nine five-year-olds (range=4;11–5;11, mean=5;05, 11 boys, 18 girls) and twenty-six seven-year-old children (range=6;11–7;08, mean=7;03, 9 boys, 17 girls) participated in the experiment. The children were recruited from kindergartens and primary schools in the Niederrhein area of Germany (northwest Germany). Informed written consent was obtained from caregivers/parents. Twenty-eight adult speakers (mean age 23 years, 5 male, 23 female) acted as controls. Most of them were undergraduate students of psychology or linguistics at the University of Potsdam. All participants were native speakers of German and had no reported history of speech/language or other developmental deficits. The children received stickers for their participation. The adult participants were given course credit or a financial reimbursement.

Materials and design

Coloured picture stories were constructed to elicit the production of four different referents with three distinct information statuses each: new, given, and accessible. Picture stories were used rather than animated films, as it has been observed that children tend to make more referential errors with movies than with static pictures (Pratt & MacKenzie-Keating, Reference Pratt and MacKenzie-Keating1985). The stories were presented using a touch screen (for details, see below). In the new condition, the target referent occurred for the first time. In the given condition, a referent was first introduced, and occurred again as target referent in the immediately following picture. In the accessible condition, a referent was introduced at the beginning of the story, and re-occurred only after four (in one story) or five (in three stories) other intervening pictures, respectively. Between the two occurrences of the target referent, two other referents, one animate and one inanimate, played a part in the plot of the story to ensure that the activation of the target referent's representation was reduced.

In the studies on German discussed in the ‘Introduction’ (Baumann, Reference Baumann2006; Baumann & Grice, Reference Baumann and Grice2006), the target referents were always in phrase-final position. In order to make the results of the present study comparable to previous research, the stimuli were therefore also created in such a way as to elicit referents in phrase-final position. The target pictures showed simple transitive events (e.g. hugging, drawing), in which the target referents had non-agentive semantic roles (e.g. recipient, beneficiary). All four target referents were animals, which were assumed to be known to five-year-old children on the basis of results from a pilot study conducted with eleven children (age five) before. The words (the names of the target referents) were disyllabic and contained mostly sonorant segments to facilitate pitch tracking: Kamel (/kaˈmeːl/ ‘camel’), Möwe (/ˈmøˈvə/ ‘seagull’), Biber (/ˈbiˈbɐ/ ‘beaver’), and Biene (/ˈbiˈnə]/ ‘bee’). Note that Kamel – unlike the other three words – has an iambic stress pattern.

Since the intonational phenomena of interest can be best observed in full noun phrases, several measures were taken to reduce the likelihood of pronoun use for the target referents. It has been observed that in narration, young children pursue a ‘thematic subject strategy’: they seek to identify the main protagonist of a story, which they then tend to encode with a pronominal expression; this happens less often with more peripheral characters (e.g. Karmiloff-Smith, Reference Karmiloff-Smith1979). The stories were therefore constructed in such a way that the target referents played only a secondary role in the plot. In addition, the target referent and the agent shared the same grammatical gender so that pronoun use would lead to ambiguity and is therefore largely avoided (e.g. Arnold & Griffin, Reference Arnold and Griffin2007). Finally, the semantic role of the target referent switched from agentive to non-agentive in the given condition to discourage a ‘parallel role strategy’, whereby children use pronouns when co-referring expressions occur in the same role in consecutive events (Hickmann, Reference Hickmann2003). An example of a story containing a ‘new’ and an ‘accessible’ referent is provided in Figure 3, a ‘given’ story is provided in Figure 4.

Fig. 3.

Example of a picture story (order from top left to bottom right) for eliciting a referent with new and with accessible status. The target referent is the seagull (encircled) introduced in picture 2 (new condition), which occurs again in picture 8 (accessible condition).

Fig. 4.

Example of a picture story (order from top left to bottom right) for eliciting a referent with given status. The target referent is the seagull introduced in picture 4, which occurs again in picture 5 (the target picture).

Each referent appeared once in each condition. The new and the accessible condition occurred within the same story, that is, the introduction of the referent served for the new condition and its re-introduction as accessible. Referents in the given condition were presented in a separate story. Four additional stories (three pictures each) from another experiment served as fillers. In total, each participant told twelve stories. Four different lists were constructed by varying the order in which the stories appeared. Stories containing the same target referent (e.g. the ‘new/accessible’ story containing Kamel and the ‘given’ story containing Kamel) were always separated from each other by three other stories. Each time the story with the new and the accessible referent preceded the story with the given referent, but new/accessible stories and given stories were mixed (i.e. it is not the case that participants first told all new/accessible stories and then all given stories). Participants were randomly assigned to one of the four lists.

Procedure

Data treatment

Statistical analysis

All statistical analyses were performed using the statistical software R (R Development Core Team, 2008). The two research questions stated above were (1) whether five- and seven-year-old children adhere to the suggested information status–pitch accent mapping, and (2) if not, how the children differ from adults. We are thus interested in whether children show a clear mapping between information status and pitch accent type, and if this mapping is the one put forth in the literature. We also want to know if the age of a speaker has any influence on the choice of pitch accent. The statistical method used to address these questions was logistic regression. In other words, it was tested whether the outcome of a binary dependent variable was influenced by several predictor variables. The different pitch accent types were recoded as binary variables (e.g. ‘H* yes/no’). It was then tested whether the probability of a given accent (e.g. H*) occurring was influenced by the two (categorical) variables information STATUS (with the three levels new, given, and accessible) and AGE (with the three levels five years, seven years, and adults). The differences in the distribution of pitch accents across information statuses and age groups were analyzed using logit mixed-effects models (LME; Bates & Sarkar, Reference Bates and Sarkar2007). Mixed-effects models include both fixed effects (independent variables) that are repeatable (such as the experimental condition in an experiment, e.g. information status), and random effects (variables) that are randomly sampled from a much larger population (such as certain words or the particular speakers who took part in the experiment). In all subsequent analyses, the random effects WORD (i.e. the various animal terms used by the participants) and SUBJECT (i.e. the individual participants) were initially included in the regression model. They were removed from the model if the variance explained by them was effectively zero. If the explained variance was not zero, likelihood ratio tests were used to test whether the random effect parameters were justified in the model. Random slopes were initially also included in the models, but as in most cases the models did not converge, they were excluded in all models to achieve consistency across models. All reported p-values were obtained by Laplace approximation. (For reasons why Bonferroni correction and other p-value adjustments are inappropriate for this type of analysis, see Gelman & Hill, Reference Gelman and Hill2006.) However, mixed-effect models could not be used where there were too few cases in one of the categories: a rule of thumb for binomial models regarding the absolute upper limit for the number of parameters in the model is min(outcome A, outcome B)/10. Thus when there are only about twenty cases in one category, a model with three factors (two fixed, one random) would already be overfit. In these cases, Generalized Linear Models (GLM; the glm function in R) were employed, with only STATUS and AGE as factors instead. In the case of the accent H+L*, zero occurrences in some groups cause problems for both LME and GLM in R. To circumvent these problems, a z-test for two proportions was used instead, comparing the proportion of the accent type in each condition with its proportion in each of the other two. For this test, proportions are transformed into z-values (Fisher z-transformation) and then compared to each other (see Hinkle, Wiersma & Jurs, Reference Hinkle, Wiersma and Jurs2003).

What is the target model?

The present results call into question such a clear mapping, and more particularly the existence of a special accent for accessible referents. Speakers always accented new referents, which is consistent with claims in the literature (e.g. Chafe, Reference Chafe1974; Lambrecht, Reference Lambrecht1994) and previous production studies (e.g. Brown, Reference Brown, Cutler and Ladd1983; Terken & Hirschberg, Reference Terken and Hirschberg1994). However, no particular accent type emerged as the prototypical ‘newness accent’. The accent H*, which is assumed to be a marker for newness (Pierrehumbert & Hirschberg, Reference Pierrehumbert, Hirschberg, Cohen, Morgan and Pollack1990), and the related L+H* were used by adult speakers in about 40% of the cases, but about 30% of new referents were realized with L*+H accents. This finding is at odds with the suggestion that L* accents signal givenness (Chen et al., Reference Chen, Den Os and De Ruiter2007; Pierrehumbert & Hirschberg, Reference Pierrehumbert, Hirschberg, Cohen, Morgan and Pollack1990). In contrast to new referents, referents in the given condition were significantly more often deaccented, in accordance with previous reports (e.g. Brown, Reference Brown, Cutler and Ladd1983). On the other hand, almost 70% of the given referents still received an accent, in line with previous findings that mere repetition of a referent does not necessarily lead to deaccentuation (e.g. Terken & Hirschberg, Reference Terken and Hirschberg1994). Interestingly, the proportion of L* accents was even lower in the given condition than in the new condition, which further questions the status of L* as conveying givenness. There was also no evidence for the existence of a special accessibility accent H+L*, as suggested by Baumann and Grice (Reference Baumann and Grice2006). This accent type was used only twice in the accessible condition, and was rare in the data overall. In this study, accessible referents patterned more with new referents than with given ones – in their preference to be accented, and partly in the type of pitch accent used (e.g. L*+H), as claimed by Chafe (Reference Chafe1994: 75). Although adults used L+H* more often in the accessible condition than in the new condition, this accent was not used exclusively for accessible referents. The fact that accessible referents were not realized with a distinct accent and were sometimes also deaccented like many given referents supports Lambrecht's (Reference Lambrecht1994) view that there is no clear phonological correlate for accessible referents, and that speakers decide whether to treat accessible referents like given or like new referents.

In the light of these results, the hypothesis that particular accent types are used to signal distinct information states seems difficult to maintain, at least with respect to natural speech production. Rather, the findings back the basic assumption that accentuation of new referents is obligatory, whereas the realization of given (and accessible) referents is less constrained. This view has been expressed by several researchers, based on perception data: the absence of accentuation always signals givenness, whereas the presence of an accent does not constrain the interpretation in such a way (e.g. Lambrecht, Reference Lambrecht1994). This study completes the picture from the production side.

Why then did other studies find that pitch accent type plays such an important role (Baumann, Reference Baumann2006; Baumann & Grice, Reference Baumann and Grice2006; Chen et al., Reference Chen, Den Os and De Ruiter2007)? What seems crucial here is to make two distinctions: first, the distinction between perception and production, and second that between natural speech and read speech. As noted by Trueswell and Tanenhaus (Reference Trueswell, Tanenhaus, Trueswell and Tanenhaus2005), it is uncertain whether the visual world paradigm (as used by Chen et al., Reference Chen, Den Os and De Ruiter2007), with its restricted visual and linguistic environment, produces results that can be generalized to the processing of natural speech in a discourse context. A particular accent may lead listeners in an eye-tracking experiment to come up with an interpretation quickly, because the task requires them to figure out where to move the computer mouse next. It is not certain that they would do the same in a natural discourse situation, which provides a wealth of other cues that facilitate the identification of a referent (e.g. discourse context, syntactic construction). From a production perspective, this conversely means that speakers do not have to leave the burden of providing crucial referent information to the intonation (more specifically: the type of pitch accent); they offer their listeners other additional sources, and may use pitch accent type differentiation for other purposes, as long as they comply with the convention to mark new referents with an accent.

In the case of off-line judgment tasks (as employed by, e.g. Baumann & Grice, Reference Baumann and Grice2006), the reason for the emergent pitch accent type preferences seems of a different nature. Just as word choice is determined by the linguistic register, it has been established that there is also register-dependent prosodic variation (Yaeger-Dror, Hall-Lew & Deckert, Reference Yaeger-Dror, Hall-Lew, Deckert, Leistyna and Meyer2003). It has furthermore been observed that listeners are sensitive to ‘formal intonation’ as a cue to identify speech as read-out rather than spontaneous (e.g. Laan, Reference Laan1997). It seems plausible that people may hold (more or less) strong (prescriptive) views about what characterizes ‘good intonation’ when they listen to read-out stimuli (as in the Baumann & Grice, Reference Baumann and Grice2006, study), but that does not necessarily mean that they adhere to these ‘standards’ themselves in natural discourse. If speaking mode (i.e. read vs. spontaneous) were to have an impact on the intonational marking of information status, this could also explain the discrepancies between the present study and Baumann's (Reference Baumann2006) analysis of read-out newspaper texts. H+L* may be an appropriate marker for accessible referents in read speech, but may be used much less, or used in a different function (as discussed below) in spontaneous speech. Thus whether we find strong correlations between information status and pitch accent type may depend not only on modality (perception vs. production), but also on speaking mode (read vs. spontaneous).

The adults' narrations also differed from scripted speech in that they featured many high boundary tones, often in combination with L*+H accents (see Table 4). This rising pitch at the end of a clause typically indicates that there is ‘something more to come’, and is commonly referred to as ‘continuation intonation’ (e.g. Cruttenden, Reference Cruttenden1997). By using continuation intonation, the speakers signalled that the utterance was part of a larger structure, namely the story they were telling. This fact illustrates again the multi-functionality of intonation; it can be used to signal information status, but it is among others also a marker of discourse structure. Continuation intonation is not typically found in read-out speech, unless the text contains enumerations, for example. Written text, and in particular punctuation, superimposes strong restrictions on speakers' choice of boundary tones. This is different in spontaneous speaking, where speakers decide for themselves how to connect phrases with each other. Again, it becomes clear that speaking mode influences intonation. This point is particularly important for the study of child language acquisition, as the target model for children will be what their interlocutors produce spontaneously. Although being read to by an adult is probably part of many children's experience in Western societies, as in Germany, they are most of the time immersed in natural unscripted discourse. This means that one should be cautious in taking results from perception studies that employ highly controlled contexts, or results from reading studies, at face value when determining what the yardstick for children is in intonation.

The findings for the adult group mean that we have to recast the first research question (Do five- and seven-year-old children adhere to the information status–pitch accent mapping that has been suggested for adults?). Since the suggested information status–pitch accent type mapping does not even hold for adults (in unscripted speech), it is the wrong question to ask. We need to ask whether children mark the information status of discourse referents in the same way as adults do in unscripted speech, or if they diverge from adults in a systematic way that may tell us something about potential cognitive or linguistic developments between the ages of five and seven.

Adult-like marking of information status

In view of the complexity of intonation patterns found in unscripted speech, the findings regarding the children's production are remarkable. Children as young as five years old have learnt to use intonation to mark information status, and they do it in an adult-like fashion. Children signal newness by accenting. As there was only a single case of an unaccented new referent on both groups, the data strongly suggest that children know that new referents need to be accented. They do not seem to make ‘egocentric’ errors in intonation. Furthermore, five-year-olds (and seven-year-olds) know that givenness (previous mention) licenses deaccentuation. This is in line with what Wonnacott and Watson (Reference Wonnacott and Watson2008) reported for simple sentence pairs, and the results of the present study establish that this also extends to more complex discourse structures such as narrations, in which tracking of reference is more challenging. What is more, the results provide evidence that within such structures, children did not operate on a simple rule to treat every referent that has been mentioned as given. They realized accessible referents for the most part in the same way as new referents, just like the adult speakers. The differential treatment of given and accessible referents demonstrates that the children were sensitive to recency of mention, and used intonation to indicate this. The only difference between the children and the adults in information status marking concerns the use of L+H*. While the children used this accent more for new referents, adults used it more often as a marker for accessible referents. Given that L+H* is a perceptually more prominent accent than H*, it is possible that the adults wanted to draw the listener's attention more to the accessible referents than the children. Recall that the adults were instructed to tell the stories in a way that would enable the listener to retell them. The adults may have used L+H* to a greater extent to ‘remind’ the listener of the reappearing referent – to invest more activation cost, in Chafe's terms.

Importantly, the results show that five-year-old children have already acquired a sizeable inventory of different pitch accent types. While the Wonnacott and Watson (Reference Wonnacott and Watson2008) experiment provided global acoustic evidence that children manipulated pitch in nominal reference production, the present study reveals that this pitch manipulation is expressed in a number of different intonation patterns, and that these patterns are similar to the adult inventory.

This early intonational competence in information status marking is in contrast with the difficulty children of the same age have been found to experience in the morphosyntactic domain, where they tend to use definite articles and pronouns inappropriately to introduce new discourse referents (e.g. Kail & Hickmann, Reference Kail and Hickmann1992). When morphosyntactic marking is correct, children do not make errors in intonational marking. Children seem to have both the linguistic knowledge and the cognitive capacity to mark the information status of discourse referents correctly in narrative discourse. What may be the reason for the different trajectories in morphosyntactic and intonational marking? A possible explanation may be that intonation is generally earlier in place than morphosyntax. Numerous studies on infant perception have shown that children are very sensitive to the prosody of their native language (e.g. Jusczyk, Cutler & Redanz, Reference Jusczyk, Cutler and Redanz1993) and that they use prosody as a start to learn other aspects of grammar, like syntactic structure or the lexicon (e.g. Christophe, Gout, Peperkamp & Morgan, Reference Christophe, Gout, Peperkamp and Morgan2003). Already at the babbling stage, children's productions reflect the intonational patterns of their ambient language (e.g. Whalen, Levitt & Wang, Reference Whalen, Levitt and Wang1991), and there is mounting evidence that children seem to acquire most of their native language intonational inventory early, around the age of two (e.g. Prieto, Estrella, Thorson & Vanrell, Reference Prieto, Estrella, Thorson and Vanrell2012, for Catalan and Spanish). This means that at an age at which articles are typically omitted, children have already intonational means at their disposal to signal the givenness and newness of discourse referents. The present study shows that by the age of five, children have not only acquired the relevant intonation contours, but that they also use them appropriately. The study suggests also that there are no developmental changes between the age of five and seven in this respect.

Late emergence of other functions of intonation

Nevertheless, this study also revealed some age-related differences in the intonation patterns: first, five-year-olds did not use phrase-final rising intonation (L*+H/L* accents together with high boundary tones) as much as the seven-year-olds and the adults; second, five-year-olds produced more downstepped !H* accents; and third, both child groups used the accent type H+L* less often than adults did. These findings do not bear directly on the issue of information status marking, but they tell us something about how speakers integrate different functions of intonation when producing speech in a less restrictive setting, something that children have to learn in order to be a competent speaker of their acquired language. In what follows, the three age-related differences observed in the data will be discussed in turn.

As mentioned above, several researchers have suggested that high pitch at the end of the utterance (continuation intonation) is also a ‘turn-keeping’ device (e.g. Couper-Kuhlen & Selting, Reference Couper-Kuhlen, Selting, Couper-Kuhlen and Selting1996; Schegloff, Reference Schegloff, Ochs, Schegloff and Thompson1996), that is, a signal that the speaker is about to say more. The fact that the five-year-olds did not produce this pattern as often as the older speakers may be an indication that children learn this discourse-structuring or interactional function of intonation only later. This is in line with a study by Potamianos and Narayanan (Reference Potamianos and Narayanan1998), who found that compared to older speakers (eleven- to fourteen-year-olds), even eight- to ten-year-old children produce fewer filled pauses in dialogue, which are an important device to signal delays in responding, thinking, etc. in adult discourse (e.g. Clark & Fox Tree, Reference Clark and Fox Tree2002; Schegloff, Jefferson & Sacks, Reference Schegloff, Jefferson and Sacks1977). Potamianos and Narayanan's finding is corroborative evidence that the acquisition of certain interactional devices (including continuation intonation) may be a late development.

In this context, it is interesting to note that, during testing, the experimenters had the impression that five-year-olds often did not signal when they were thinking or searching for words, which sometimes resulted in cross-talk between child and experimenter. The seven-year-olds, on the other hand, appeared to produce signals such as ähm ‘uhm’ (filled pauses) much more often. In order to see whether this impression could be backed up by the data, an additional analysis looked at the number of utterances that had been transcribed as ‘mmh’, ‘äh’, and ‘ähm’ in the narrations of the three age groups (excluding utterances that were clearly intended as feedback to the interlocutor's talk in the sense of ‘I understand’). Although this was only a cursory investigation, interesting differences across age groups emerged: in the corpus of the five-year-olds, there were 324 such instances (on average 1·1 filled pauses per 100 words), compared to 613 in the seven-year-old group (on average 2·2 filled pauses per 100 words). Interestingly, the number of these markers dropped again in the adults, who produced 327 (on average 1·1 filled pauses per 100 words). A z-test for proportions showed that both the difference between five-year-olds and seven-year-olds and that between seven-year-olds and adults were significant (p=·0004). The decreasing trend in adults seems at first surprising. However, it was also clear that compared to the children, the adults were more fluent and had much less difficulty with verbalizing the events in the pictures as they were going along, so that there were fewer delays in speech production that needed to be signalled. It is also possible that the adults perceived the situation as more formal (after all, they were aware that their narrations were being recorded), and therefore made an effort to produce fewer disfluencies. Potamianos and Narayanan unfortunately do not provide information about the number of filled pauses in their adult group, so it is unclear whether they observed a similar decrease.

Returning to the subject of continuation intonation, it is also possible that five-year-old children do use it, but only in contexts in which they already know what the global structure of their contribution/discourse is going to be. In the present setting, the stories were presented picture-by-picture, which could have discouraged them from indicating they were about to say more, since they did not know what would happen next. Looking at children's intonation in stories they are familiar with could test this hypothesis. These could be achieved either by giving them the chance to work through the stories beforehand, or letting them narrate personal stories. Alternatively, the relative scarcity of final rising contours – in particular L*+H H% – in the youngest age group may have physiological reasons. Several studies have reported that younger children have difficulty producing rising patterns (e.g. Behrens & Gut, Reference Behrens and Gut2005; Snow, Reference Snow1998), and it has been suggested that this is due to increased physiological effort (Snow, Reference Snow1998). However, in the light of the increasing number of studies that have observed an early mastery of the intonational ‘lexicon’, it seems unlikely that the children were physically not able to produce this contour. It is more plausible that the children have either not yet acquired the use of the contour (L*+H H%) as a turn-holding cue or that they did not use it in this setting for the reasons discussed above.

Another interesting difference was that five-year-olds used the downstepped accent !H* significantly more often than the other two age groups. The phenomenon of downstep has been the topic of a number of investigations, but the majority of these are concerned with questions of speech production planning and phonetic realization (e.g. Liberman & Pierrehumbert, Reference Liberman, Pierrehumbert, Aronoff and Oehrle1984). The function and meaning of downstep, on the other hand, has not been discussed much. Ladd (Reference Ladd1996) states that ‘downstepping adds a nuance like finality or completeness, but does not otherwise seem to affect the meaning of the contour’ (p. 76). If we follow this interpretation, the five-year-olds' increased use of !H* – or rather, the other age groups' decreased use of it – fits in nicely with the pattern found for the boundary tones and rising contours in general. Rising contours (which were produced more often by the older speakers) indicate continuity. Finality and continuity are opposites and cannot be signalled at the same time. If the older speakers were more often stressing the continuative aspect of their narrations, it follows that they would not use intonation that indicates finality as much; the five-year-olds' patterns would be the mirror image of this.

The third difference between the child and the adult speakers was the frequency of H+L*, which occurred only three times in the child data. Although it was also relatively scarce in the adult data, this difference between age groups was significant. Recall that there was no indication that H+L* is used to signal accessibility of discourse referents. There appear to be two possible explanations for differences between adult speakers on the one hand and children on the other. First, adult speakers may use this accent type in spontaneous speech primarily to communicate a particular attitude (rather than a particular information status). As speculated above, it is conceivable that pitch accent types fulfil different functions in different contexts. Perception experiments have shown that H+L* has a ‘matter-of-fact’ flavour to it and conveys an impression of ‘knowingness’ (Kohler, Reference Kohler1991). Given the simplicity of the story plots, it is not implausible that some adult speakers used this accent type to indicate that the event they were describing was to be expected, or obvious. Note that H+L* occurred most often in the given condition (cf. Table 2), in which the main protagonist of the story performed an action on the target referent, as a consequence of the target referent's preceding action. For example, in the given story presented before (see Figure 4) the seal is biting the seagull after the seagull had stolen a fish from the seal. It is conceivable that adults used intonation to mark this consequence as being ‘obvious’. It is possible that some paralinguistic uses of intonation to convey certain attitudinal meanings may be another aspect that is acquired only later. Findings that even seven- to nine-year-old children are unsuccessful in using prosody to detect irony, for example, point in this direction (Filippova & Astington, Reference Filippova and Astington2008). However, it is also possible that children do use intonation to signal attitudes, but that the materials and the discourse situation were not conducive to eliciting such patterns. Investigations of intonation production under different conditions (e.g. having children communicate with peers instead of adults) could shed more light on this issue.

An alternative yet related hypothesis is that H+L* accents are part of a certain register, such as ‘narrating’, which some of the adults made use of in this task, but children may not yet have acquired. In fact, one could imagine that precisely because H+L* does convey knowingness on behalf of the speaker, it is also employed in narrations, since, in telling a story, the speaker often takes the position of an omniscient narrator. However, in view of the low numbers, the explanations offered here could only be speculative. More research is needed to determine the function of H+L* accents in different discourse situations, and, if it is register-specific, at which age children come to use this.