Kethod I 104
relevant in that it is related to the distribution of turns. This argument
also explains the extreme simplification of some categorizations
(variables): as an example, for a study aimed at a functional analysis of
utterances, the 26 mutually exclusive alternatives (values) would be
insufficient.
Studies in conversation analysis always differentiate between Structure,
Process and Principles of conversation. In this study the focal point has
been the sustaining of conversation because it was considered important
for the development of fluency in L2 practice (see 4.5).
The elaboration of the coding system for this study (see Codebook at page
107) brought the same problems that are usually connected with this type
of instrument, and which are systematically exposed in Veils,1985,
chapter 2. The principle of allowing only one coding per item (the
dominant value) has been adopted even if it is unnecessary in principle
(Veils,1985:87) and cumbersome in practice. A major limitation of the
analysis is that no independent reliability checking was possible. Despite
sustained efforts, I was unable to find a Tsonga and/or Portuguese-
speaking person with some experience in linguistics who would code part
of the transcripts in order to check my own coding. A limited checking
was done using the same computer programme employed for the elaboration
of the data: a sample of 300 utterances that had been given the same
values was examined and found sufficiently homogeneous.
The coded data were fed into the University of London mainframe computer,
and the S.P∙S∙S.(X) programme was used for the statistical elaboration.
This was kept as elementary as possible, considering the nature and
limitations of the data, and consisted in frequency counts, crosstabula-
tions and lag analysis.