Coherence and Cohesion in English Discourse

4 Corpus and methodology
The relatively recent language data discussed in this chapter are taken from 
two specialized corpora, one representing research articles (RAs) (Swales 2004) 
and written by experienced native speakers of English (amounting to about 
78,000 words) and the other comprising articles produced by non-native speakers 

---

35
of English, notably experienced Czech writers (amounting to about 58,000 
words). The former corpus comprises ten RAs selected from the journal 
Applied 
Linguistics
published between the years 2001 and 2008 (6 single-authored and 
4 co-authored RAs), while the thirteen RAs of the latter corpus were all written 
for the linguistics journal 
Discourse and Interaction
in the years 2008 to 2011, 
namely by ten Czech writers, some of whom are the author’s colleagues. As 
regards the average length of the RAs under investigation, it is much bigger in 
the former corpus, namely 7,753 words, which is the reason why thirteen, not 
ten, RAs are included in the latter corpus, in which the average length is only 
4,447 words.
Although relatively small in size, the two specialized corpora described above 
are considered sufficient and useful for the present analysis, because despite 
certain limitations in terms of size, representativeness and generalizability of 
their results, specialized corpora are more appropriate than large general corpora 
for a comparative study of academic written discourse (Flowerdew 2004: 18), 
especially for an analysis of particular language features such as conjuncts 
when studied in one particular genre – the genre of research articles. Since 
“corpus-based methodologies have been informed by genre principles of text 
analysis, while at the same time it has been shown that 
genre theories can profit 
from corpus-based methodologies” (Flowerdew 2005: 329-330), both of these 
approaches have been applied in the research discussed in this chapter.
It remains to be stated that in order to get comparable data for the comparative 
analysis it has been necessary to exclude from both the corpora all parts of texts 
which comprise tables, figures, graphs, references, sources, examples, and long 
quotations. All the results discussed and exemplified in this chapter have been 
normalized for the frequency of occurrence of conjuncts per 1,000 words, actual 
numbers being mentioned only occasionally.
As for the methods applied during the investigation, all the texts were 
first computer-processed using the AntConc concordancer and then examined 
manually in order to obtain both qualitative and quantitative results, since some 
of the language items under examination can perform functions other than those 
of conjuncts in written discourse.