<$F> nods without concurrent spoken backchannels (‘gesture alone’ backchanneling nods, see Evans et al., 2001) twice as many times as <$M> (see Appendix 4.4 for a breakdown of the frequencies of different nod types and functions, as evidenced in the case study data). In addition, nods that are used without backchannels were most likely to be of type A or C for this speaker. This is also true of <$M>, although the frequencies of use for these behaviours are significantly less for this speaker, as detailed in Figure 4.11.
Speaker TOTAL <$M> <$F> N o d T y p e A 8 25 33 B 7 6 13 C 6 15 21 D 1 0 1 E 2 2 4 24 48 72
Figure 4.11: The frequency counts of backchanneling nods occurring without
4.3.3.2. Nods with(out) concurrent spoken backchannels
Figure 4.12 provides a detailed breakdown of the different types of head nods that co-occurred with the spoken backchannels in the data. In comparison to Figure 4.11, this figure shows that backchanneling nods proved more likely to co-occur with spoken backchannels for both speakers than to be used alone, although the extent to which this is true differs across the speakers49. Indeed, in the vast majority of cases (aside from type D nods) the frequency with which these nods are used with concurrent spoken backchannels is far greater than the frequency with which they are used in isolation.
Speaker TOTAL <$M> <$F> N o d T y p e A 7 6 13 B 4 3 7 C 7 11 18 D 1 1 2 E 3 1 4 22 22 44
Figure 4.12: The frequency counts of different backchanneling nod types co-
occurring with spoken backchannels.
Figure 4.12 indicates that the type A nods were most likely to co-occur with spoken backchannels functioning as IR or CNV tokens (for both speakers, see Appendix 4.4 for further details). 8 of the 13 (31%) nods (4 from of each functional type) for type A were of this nature, consisting of 3 by <$M> (14% of the total number of spoken and non-verbal backchannels used by <$M>) and 5 by <$F> (23% of all instances). Type B nods proved most likely
49
to co-occur with spoken backchannels functioning as CNV tokens. 4 of the 7 (57%) type B nods seen were of this nature, with 2 by <$M> (9% of total) and 2 by <$F> (9% of total). Type D nods proved to be just as likely to co-occur with CON as CNV tokens, and type E nods were as likely to co-occur with either CON or ER tokens, although the frequency of these types was relatively small (1 occurrence for each).
Overall, more than half of the small nods (64%) of short duration (types A and C) enacted by <$F> co-occur with spoken backchannels adopting an IR function (9 from 14), while half of the small nods of a longer duration (type B) co-occurred with the spoken backchannels adopting the CNV functions across both speakers. The type C nods co-occurred with CNVs and CON on 39% (7 from 18) and 50% of occasions (6 from 12), respectively. This co-occurrence was shown to be far more likely for <$F> than <$M>50.
For 11 of the instances where <$F> used a type C nod (61% of the total), 100% co-occur with either a CNV token or a CON token. Whereas, for <$M>, only 6 of the total number of nods used were of type C (39%) and only 2 (22%) of these co-occur with either CNV tokens or CON backchannels. It is interesting to note that for all other types of nods no significant difference in the frequency of use exist, instead patterns of use are fairly consistent across the speakers.
4.4. Overview
The analysis above highlights a number of issues which need to be taken into account when embarking on a corpus-based approach to the analysis of gesture-in-talk. Some interesting relationships have begun to emerge following these basic frequency-based investigations, which are as follows:
Backchanneling head nods are used at the same rate or more frequently than spoken backchannels in conversation.
In general, the most common types of head nods used in discourse are of a short duration and/or intensity. Intense, complex and multiple nods are less frequently used.
Nods are used more frequently with concurrent spoken backchannels than alone. Similarly, spoken backchannels are used more frequently with concurrent nods than alone.
More engaged forms of spoken backchannels, those situated at the bottom of O’Keeffe and Adolphs functional model, tend to co-occur with longer and/or complex head nod sequences (types B, D or E), whereas the less engaged and more simple lexical forms of spoken backchannels most frequently co-occur with shorter nods (types A and C).
These preliminary findings have provided an insight into some of the pragmatic properties of backchanneling head nods and their relationship with spoken forms of this phenomenon. These findings are reformulated and utilised as specific premises (numbers 6-9) for further investigation as part of
the extensive analyses conducted in Chapter 6 (note that premises 1-5 are based on previous findings of backchannel research, cited in the literature review in Chapter 2).
While the case study has identified some interesting observations regarding the use of language and gesture in a dyadic communicative context, at this stage, it has not been possible to provide a graded taxonomy of gesture types and functions and their direct relationships to language use, form and function. Therefore, the initial observations made are in need of further qualification from the exploration of a larger and more varied data set, with more speakers and so on, before more detailed assumptions are made regarding the nature of spoken and non-verbal backchanneling behaviour, and the relationship between them.
4.5. Summary
This chapter has provided an outline of a corpus-based approach to the analysis of new MM datasets for CL enquiry. It has proposed how patterns of language and gesture use can best be examined in records of communicative episodes, providing a blueprint for the main study analysis which is undertaken in Chapter 6.
The case study has illustrated that the proposed methods are both effective and appropriate for tackling MM data, as some interesting results and observations have already been identified as a result of this analysis. However, since this chapter has investigated only ten minutes of data, no definitive conclusions about patterns of backchanneling behaviour have, as
yet, been drawn-up as a consequence of this. Instead the focus has been on illustrating and testing the analytical approaches discussed within.
No real problems, beyond those discussed in section 4.2, were encountered when investigating this data, and while it is understood that the subsequent analysis of the five-hour dataset in Chapter 6 is necessarily more time-consuming, the fact that a formal framework for analysis has been developed here means that methodological problems and challenges faced in this analysis should be kept to a minimum. This matter is re-addressed accordingly as part of Chapter 6.
The approach used in this chapter is essentially analyst-led, utilising a system for detecting and encoding features which is manual, inferential and potentially very time consuming. Prior to the main five-hour study, it is perhaps appropriate to look towards more automated methods to facilitate the analysis of larger scale and more varied corpora. Chapter 5 examines this notion of ‘automating the approach’ in more detail. It provides a critical review of an intelligent digital gesture tracking algorithm designed to facilitate the investigation of forms of backchanneling nods in discourse, examining the practicality of this system and the relative advantages and disadvantages of using this method in preference to the manual approach investigated in the present chapter.