tv.um.es

As a carrier of scientific information and academic dissemination, research
article (RA) abstracts succinctly and skillfully epitomize the core information of the
full text, and have attracted the attention of a number of scholars. The previous studies
of this genre mainly focused on the rhetorical structures, meta-discursive features and
lexico-grammatical features, while very few have made explorations from the
perspective of information theory. The present study adopted a corpus-based approach
and conducted an entropy-based analysis of 150 RA abstracts across the natural
sciences, social sciences and humanities (50 abstracts from each discipline), in order
to explore the distribution pattern of information content across moves and the
variations across disciplines. The most typical IMRD
(Introduction-Methods-Results-Discussion) model from Swales (1990) was adopted
for the identification of RA abstract moves. Three quantitative indices, i.e. 1-/ 2-/
3-gram entropies, were used to examine if different indices yield different linguistic
features of RA abstracts. It was found that the information content was unevenly
distributed across IMRD four moves in an abstract, where the Results Move and
Introduction Move were more informative than the Methods and Discussion Moves. It
was also found that the information content of RA abstracts was the highest as
measured by 1-gram entropies and the lowest when measured by 3-gram entropies. In
addition, there exhibited significant cross-disciplinary variations in terms of
informative distributions across moves. To be specific, in the Methods Move, the
entropy values of natural sciences were significantly higher than those of social
sciences, which were significantly higher than the entropy values of humanities. In the
Results and Discussion Moves, the entropy values of natural sciences were constantly
higher than those of social sciences and humanities, while social sciences and humanities demonstrated no disciplinary difference. In the Introduction Move, the
entropy values showed no cross-disciplinary difference. The aforementioned results
could be accounted for the functions that each move bears on, the different aspects
possibly reflected by the different indices, and the very nature of different disciplines.
This study demonstrates the promising prospects of using a corpus-based approach
and quantitative linguistics indices, such as the entropy from information theory, to
unravel the characteristics of RA abstracts and other academic genres, where
traditional qualitative methods and basic statistics still prevail. This study also sheds
light on EAP (English for Academic Purposes) pedagogy, especially on the instruction
and practice of RA abstract writing, in that instructors should guide students into an
awareness of the structure of an abstract and the prominence of different moves, as
well as the training of corresponding writing skills. Besides, this study could
contribute to the automatic move recognition of abstracts in the field of NLP (natural
language processing) by adding entropies into the feature set, which may enhance the accuracy of a move recognizer. Such a recognizer, if put into application, can be extremely beneficial to academic writing and practice.