<?xml version="1.0" standalone="yes"?>
<Paper uid="W01-1625">
  <Title>Melodic cues to turn-taking in English: evidence from perception</Title>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
3 Method
</SectionTitle>
    <Paragraph position="0"> In the first experiment, subjects were presented with a dialogue fragment ending at an IPU boundary. The subjects' task was to predict what happens next, i.e. whether the speaker holds the turn, or holds after a brief backchannel response, or cedes the turn. In the second experiment we again asked subjects to judge what the first speaker had intended - to continue or to cede the turn, but under slightly different conditions: this time subjects heard brief exchanges involving both speakers. This was to see if the presence of an actual response influenced subjects' judgement of the first speaker's intention. The same subjects took part in both experiments.</Paragraph>
    <Paragraph position="1"> They were 25 native speakers of Southern British English, 9 men and 16 women, aged between 19 and 54, only 7 of whom had some background in linguistics. No hearing difficulties were reported.</Paragraph>
  </Section>
  <Section position="5" start_page="0" end_page="2" type="metho">
    <SectionTitle>
4 Experiment One
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
4.1 Stimulus material
</SectionTitle>
      <Paragraph position="0"> For the first experiment, the stimuli consisted of dialogue fragments, around 8 to 13 seconds in length, and ending in an IPU. The fragments were chosen such that they ended according to the following four conditions: (i) turn exchange plus syntactic completion (ii) turn exchange minus syntactic completion (iii) turn hold plus syntactic completion (iv) turn hold minus syntactic completion The five contours chosen were as listed in paragraph 2 above. For all but the high rise (H* H%) two stimuli were chosen for each of the above conditions, giving 32 stimuli. As syntactic completion could, of course, include interrogatives, which would be highly likely to project a turn change, these were avoided for all but one stimulus for the fall-rise contour and three for the high rise. We found very few cases of the high rise in the English data, an interesting finding in itself, and it was not possible to find examples for each condition; only six cases were used altogether, four syntactically complete (two interrogatives and two declaratives) and two syntactically incomplete.</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
4.2 Procedure
</SectionTitle>
      <Paragraph position="0"> After three practice examples, the 38 randomised stimuli were each presented twice.</Paragraph>
      <Paragraph position="1"> Subjects were asked to predict whether (1) the current speaker would continue, (2) the current speaker would continue after a short, non-obligatory backchannel response, or (3) the second speaker would take over.</Paragraph>
    </Section>
    <Section position="3" start_page="0" end_page="2" type="sub_section">
      <SectionTitle>
4.3 Results
</SectionTitle>
      <Paragraph position="0"> The results for this experiment are given in Tables 1 and 2. Table 1 gives the frequency of responses per condition, and Table 2 shows whether the differences in number of turn-keeping responses between the contour types are significant. Note that in the latter table we conflate the responses 'hold' and backchannel', since, despite subtle pragmatic differences, we judged the prediction of a backchannel to entail the prediction of a turn continuation (cf. Koiso et al.). A hierarchical loglinear analysis performed on the factors response type, contour type and syntactic completion shows significant associations between response type and contour G87G92G83G72G3 G11G83G68G85G87G76G68G79G3  =288.3, p&lt;.0001), between syntactic completion and response type (partial  =288.3, p&lt;.0001), and interaction between the G87G75G85G72G72G3G73G68G70G87G82G85G86G3G11G51G72G68G85G86G82G81G3  =507.4, p&lt;.0001). This means that there are main effects as well as interaction effects of contour type and grammatical completion on the responses.</Paragraph>
      <Paragraph position="1"> Table 1 shows that subjects virtually never expect a turn change when the fragment is syntactically incomplete (2%). The only significant differences in the number of expected turn-keepings ('backchannel' plus 'hold') are found between contours H*L L% and H*L H% and between H*L L% and H*L %, but these effects are rather small (see Table 2). The main difference appears to be the degree to which contours invite a backchannel response. This  tendency is weak for both the fall (H*L L%) and the truncated fall (H*L %), but nearly half of the H*L H% contours in syntactically incomplete positions are judged to invite backchannel feedback.</Paragraph>
      <Paragraph position="2"> The syntactically complete utterances, on the other hand, show a clear effect of contour type: a rising pitch accent followed by a level boundary tone (H* %) leads to 89% expected 'hold' responses, supporting the hypothesis that this melodic configuration functions as a turn-keeping device. In this respect it differs strongly from all other contours, as is evident from the data presented in Table 2.</Paragraph>
      <Paragraph position="3"> The results for the syntactically complete H* H% stimuli reflect the utterance type, and should therefore be treated with caution. Not surprisingly, the two interrogatives attracted almost exclusively the judgement 'change'; the remaining two declaratives attracted almost exclusively the judgement 'backchannel'. The use of a high rise on declaratives is a recent and highly marked innovation in British English, and is assumed to have the function of eliciting hearer acknowledgment. Our results are consistent with this view.</Paragraph>
      <Paragraph position="4"> As Table 2 shows, there was an interesting and significant difference between the effect of the complete fall (H*L L%) and the truncated fall (H*L %). The truncated fall is much more likely to cue a turn hold (71% of responses compared with 41% for the complete fall) and correspondingly less likely to cue a turn change (29% compared with 59% for the complete fall).</Paragraph>
    </Section>
    <Section position="4" start_page="2" end_page="2" type="sub_section">
      <SectionTitle>
4.4 Discussion
</SectionTitle>
      <Paragraph position="0"> The results of this experiment suggest that, in this variety of English, incomplete syntax overrides any melodic cues. Only the high level tone appears to be a strong turn keeping device, regardless of syntax. On the other hand there appear to be no melodic contours which, when they occur in conjunction with syntactic completeness, can be said to predict a turn change. We thus find more evidence for the use of melody as a turn keeping device than as a turn ceding device. The second experiment was designed to investigate the degree to which such judgements of speaker intention were upheld in the presence of an actual speaker response.</Paragraph>
    </Section>
  </Section>
  <Section position="6" start_page="2" end_page="2" type="metho">
    <SectionTitle>
5 Experiment Two
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="2" end_page="2" type="sub_section">
      <SectionTitle>
5.1 Stimulus material
</SectionTitle>
      <Paragraph position="0"> The stimuli for this part of the experiment were drawn from the same material as in Part A. Each fragment that ended in the original data in a turn exchange was extended to include the turn exchange itself. This produced a sound file of around 8 to 12 seconds in length. The turn exchange was then excised as a short separate file of around 3 to 5 seconds. Regardless of contour, a speaker change at a syntactically incomplete point was hard to find in our data, and a number of these stimuli were created artificially by editing out intervening material.</Paragraph>
    </Section>
    <Section position="2" start_page="2" end_page="2" type="sub_section">
      <SectionTitle>
5.2 Procedure
</SectionTitle>
      <Paragraph position="0"> The same subjects participated in both parts of the experiment. They were first presented with the longer fragment containing the relevant turn exchange, and then heard the file containing only the turn exchange twice in succession. The 20 stimuli (4 for each contour) were preceded by three test stimuli. The subjects were asked to judge whether the first speaker had expected the turn exchange, had expected to continue, or whether it was unclear.</Paragraph>
    </Section>
    <Section position="3" start_page="2" end_page="2" type="sub_section">
      <SectionTitle>
5.3 Results
</SectionTitle>
      <Paragraph position="0"> Tables 3 and 4 contain the results for the second experiment. A hierarchical loglinear analysis performed on the factors response type, contour type and syntactic completion shows significant associations between response type and contour</Paragraph>
      <Paragraph position="2"> =143.8, p&lt;.0001), between syntactic completion and response type (partial  =200.5, p&lt;.0001), and interaction between the thG85G72G72G3G73G68G70G87G82G85G86G3G11G83G68G85G87G76G68G79G3  =282.6, p&lt;.0001). Again the biggest effects of contour type are found for the syntactically complete points: subjects do not think the original speaker wanted to yield his  his or her turn after a high level contour (there are only 8% expected changes after H* %), and Table 4 shows large differences between this contour type and all others. In contrast with the first experiment, however, there is a clear influence of contour type on the responses in the minus syntactic completion condition: in almost a third of the cases subjects feel that the original speaker had expected the turn to change after a default pitch accent (H*L) followed by a low (L%) or high (H%) boundary tone, that is, after a complete fall or after a fall-rise, and Table 4 shows that these two contour types differ significantly from all others (except from each other). The similarity between the complete fall and the fall-rise, which is also evident in the syntactic completion condition, suggests that both contours are perceived to have a similar function with respect to turn-taking and to be at least strong secondary cues to turn completion. In cases where there is a clear mismatch between syntax and contour (i.e. melodic completion but no syntactic completion) the actual presence of a speaker change makes subjects more likely to judge that this was the intention of the first speaker than they were in the first experiment, where they did not know what happened next.</Paragraph>
      <Paragraph position="3"> Although subjects were simply asked to judge what they thought the first speaker had intended, their judgements were probably to some extent based on a post hoc analysis of the whole exchange. It is a general principle of pragmatics that utterances will be assumed to be relevant unless proved otherwise, and that conversational interaction will be assumed to be cooperative unless proved otherwise. There is therefore a strong likelihood that subjects subconsciously sought a cooperative explanation for actual turn changes wherever possible.</Paragraph>
    </Section>
  </Section>
  <Section position="7" start_page="2" end_page="2" type="metho">
    <SectionTitle>
6 General Discussion
</SectionTitle>
    <Paragraph position="0"> The major finding of this study, especially of the first part, is that if an isolated utterance is syntactically incomplete, listeners are highly unlikely to predict a turn change, whatever the melodic contour used. Where the syntax is complete, none of the contours lead listeners to predict exclusively a turn change. This means that both hold and change are possible at this point. There is one exception, namely where the accompanying contour is a high level tone (H* %). This contour in English appears to signal a clear turn hold, regardless of syntax.</Paragraph>
    <Paragraph position="1"> We were also able to make some cross-linguistic comparisons. First, the similarities: it appears that in both Southern British English and Dutch the H* % contour signals the speaker's intention to keep the turn. This effect cannot be attributed to the absence of a 'real' boundary tone, since the truncated fall, which also ends in a %, does not behave as a cue to turn-keeping.</Paragraph>
    <Paragraph position="2"> We also observed two main differences between the languages. The first concerns the occurrence of high rise tones (H* H%): we had difficulty in finding any of these in the English data but not in the Dutch, which may indicate a general difference in contour distribution, or a difference in contour function in the two languages. This is an interesting question to pose in a larger-scale, corpus-based study.</Paragraph>
    <Paragraph position="3"> The second difference relates to our observation that some contours are more likely than others to suggest a subsequent backchannel response. This has important implications for the study of cooperation in interaction, both within and between languages (cf. Wichmann 2000).</Paragraph>
    <Paragraph position="4"> The number of 'backchannel' judgements given as responses to the stimuli ending in a high level tone H* % differs between Dutch and English: Caspers (2001) reports that in the Dutch study 56% of these contours suggest a backchannel response, compared to only 6% in the English study. This difference may have consequences for cross-cultural communication: if types of conversational behaviour are 'appropriate' in one language but not in the other there is potential for cross-cultural misunderstandings which may be perceived as 'attitudinal'.</Paragraph>
  </Section>
class="xml-element"></Paper>