Collocation and the rhetoric of scientific ideas. Corpus linguistics as a methodology for genre analysis.

Abstract
This paper attempts to integrate the ethnographic
approach of genre analysis (Swales 1990) with the
large scale computational analysis of phraseology
in the field of corpus linguistics (Sinclair 1987a
inter alia). In particular the author attempts to
describe how language is used in hard science,
how scientists create new science in their writing
and how language functions in extremely specialised circumstances. The paper describes the working context of cancer research articles at Aston
University’s Pharmaceutical Sciences Department and uses the statistical analysis of lexis in
rhetorical sections of research articles to characterise these sections in terms of their collocational
and discoursal properties.
Extended abstract
One hypothesis is that new science is actually
embodied in research articles by a process of reformulating concepts within the text. To test this,
the scientific claims of a sample of ten texts are
analysed in terms of reformulation of grammatical
metaphor, discourse signalling and posture (Halliday 1985, Sinclair 1981). A second hypothesis is
that new science is founded on a system of preferred expressions, and that collocation is a fundamental mechanism that allows for new formulations to take place throughout the text. A corpus
analysis of 150 cancer research articles (The pharmaceutical sciences corpus: 500,000 tokens) is
undertaken to characterise the phraseology of
grammatical items in research articles and in the
various rhetorical sections of research articles namely Titles, Abstracts, Introductions, Methods,
Results and Discussion sections.
The paper finds that research articles use language
to create new science by reformulating data as
research models and by altering the established
patterns of phraseology. Collocation is seen to
vary systematically in rhetorical sections, and the
concept of phraseology is postulated as a preferred
way of expressing a delimited set of semantic and
communicative roles. Science should therefore
not be seen as a body of facts transmitted via
language, but as a special linguistic construct,
mediated by the mechanisms of textual reformulation and phraseological innovation.
Abstracts written by authors have have been characterised in terms of morpho-syntactic features,
especially verb tense and modality (Hanania and
Akhtar 1985, Malcolm 1987, Gunawardena 1989,
Salager-Meyer 1992). From a more rhetorical approach, discourse and analysis of abstracts has
involved comparison of rhetorical moves between
abstracts and articles (Nwogu 1989, EndresNiggemeyer 1985, Salager-Meyer 1992) and thematic choice between successful and non-successful abstracts (Gibson 1992, Drury 1991). This
paper applies a phraseological methodology to an
area that has been relatively well documented in
information science and text linguistics, but less
so in genre analysis: the collocational properties
of abstracts as they compare to those of the research article. Some phraseological features, such
as explicit discourse signals, have already been
identified in the genre analysis of scientific articles
(Oster 1981, Tadros 1985, Master 1987, Brett
1994) and to a lesser extent in abstracts (Diodato
1982, Zambrano 1987), but to my knowledge there has been no general phraseological comparison
of articles and abstracts as yet (see Gledhill 1995a
for a more balanced picture).
In this article I shall exploit the notion of phraseology, defining it as: a system of preferred expressions differentiated by the rhetorical aims of a
discourse community. I shall treat phraseology as
a lexical and a discoursal phenomenon. In terms
of lexis, the concept of collocation has been used
in the analysis of the intermediate level of language between syntax and lexis (the lexico-grammar,
Halliday 1993). Recurrent word patterns have also
been instrumental in recent developments in lexicography and the description of English, as in the
Cobuild project (Sinclair 1987, Francis 1993). On
the level of discourse, phraseology plays an important role in rhetorical choice, and idioms have
been claimed to constitute important stages in the
rhetorical development of texts (Moon 1992,
McCarthy and Carter 1993). Another aspect of
phraseology involves tracing the development of
expressions within texts where deviation from the
norm implies innovation and neology in the scientific community (Pavel 1993). This aspect of
textual development touches on the concept of
logogenesis which is the subject of ongoing research (Gledhill 1995 and forthcoming).
Biber and Finegan (1986) have been primary ex-
ponents of computer-based register analysis, an
approach that measures variation in texts by the
occurence of linguistc features. They identify dimensions such as ‘abstractness’ and ‘explicit information’ that emerge from the co-occurrence of
grammatical features such as clause complexes,
it-clefts, adverbials, and more recently, lexical
chains and deictic anaphora (Biber 1992). Kretzenbacher (1990) uses a similar methodology in
his analysis of academic abstracts and articles.
Essentially, this register-analysis approach maintains that text-types exist on a continuum, and that
their differences can be explained by the analysis
of internal linguistic properties of texts as opposed
to external social and rhetorical features. Typically, register studies depend on computer corpora
that are grammatically marked up or ‘tagged’.
Genre analysis, on the other hand, works on a
smaller scale of language than register, putting
more emphasis on the specific context of a professionally or socially recognised discourse type. As
such, genre analysis attempts to find patterns of
conventional formulations which are accounted
for by processes of use and production. Thus these
formulations are not necessarily grammatical or
cohesive features. Instead, they involve the norms
of a community which shares certain values, and
preferred textual structures expressed in rhetorical
moves such as: ‘establishing a territory’, ‘establishing a niche’ and ‘occupying the niche’ (Swales
1990:141). An important difference is that while
the register approach claims that a set of co-occurrent features has similar functions throughout the
language, the genre approach assumes that grammatical forms have different functions for different discourse settings, and that within a genre a
rhetorical move may be realised differently depending on what linguistic features can be adapted
according to the practices of the discourse community. Swales (1990:42) therefore distinguishes
between genre as a conventionally recognised instance of language in a discourse community, and
register as the ‘language of’ a certain field, such
as science or journalism.
While Atkinson (1992) has joined Biber’s approach to the genre analysis of research articles, this
paper attempts to contextualise corpus analysis,
firstly by setting out the context of use of a particular genre and secondly by analysing a large
corpus of the genre in collaboration with its users.
In this paper we limit our analysis to typical writing strategies in cancer research abstracts. For this
purpose, 150 papers (500,000 words) were collected with the collaboration of 15 expert informants
from Aston University’s Pharmaceutical Sciences
Department. The papers were scanned by an electronic optical reader and placed on a IBM PC hard
disk for automatic analysis. A decision was made
to analyse only high frequency grammatical items
which were significantly more frequent in abstracts than in research articles. These items were
then analysed for collocational properties using
OUP’s concordancer Microconcord (Johns-Scott
1991). The results suggest that when computational analysis of high frequency grammatical items
is carred out with a view to taking discourse features and the context of production into account,
the computational approach provides the genre
analyst with a replicable, powerful tool of analysis