Analysis of the rhetorical structure of computer science abstracts in Portuguese

It is widely acknowledged that academic writing is a complex task even for native speakers, since it involves the complexities of the writing process as well as those specific to the academic genre (Sharples and Pemberton 1992). To make it worse, especially to novice writers, often the demands of the academic genre are not clear enough. A number of writing tools have been described in the literature whose ultimate goal is to improve the quality of academic English texts produced by novice and/or non-native writers, e.g. WE (Smith and Lansman 1988), Writer’s Assistant (Sharples et al. 1994), Composer (Pemberton et al. 1996), Academic Writer (Broady and Shurville 2000), Abstract Helper (Narita 2000), and Amadeus (Aluísio et al. 2001). These tools provide help during different stages of the writing process, from the generation and organization of ideas to post-processing tasks. For Portuguese, on the other hand, one can only find post-processing tools, such as spell-checkers, grammar checkers and online dictionaries and thesauri. With the ultimate goal of aiding academic writing, the project SciPo, currently being developed at NILC, aims at analyzing the rhetoric structure of Portuguese academic texts  in terms of schematic structure, rhetorical strategies and lexical patterns  to derive models for supporting the creation and evaluation of computational writing tools. To make the project feasible, the analysis has focused on specific sections of theses in Computer Science  abstract, introduction and conclusion, which are the most studied in the literature (Swales 1990, Weissberg 1990, Liddy 1991, Santos 1996) and also have been pointed out as the most difficult ones in a questionnaire applied to graduated students. The choice for this kind of text was made mainly for three reasons: theses having to be written in Portuguese, unlike research articles, which are preferably written in English; the high standardization of Computer Science texts, as in other scientific research areas; and SciPo’s developers’ familiarity with the Computer Science domain. A corpus of fifty-two (52) Computer Science theses was compiled and has been used both in the analysis of writing patterns specific to the focused research community and in the identification of their main writing problems. The whole corpus analysis is being carried out in a set of stages, the first of which has been accomplished, namely the analysis of the abstracts. This paper presents the results of this first stage and discusses the features of abstracts at a macro level of textual organization as well as comments on its annotation process. The next sections present the methodology used for the annotation and analysis of the abstracts (Section 2) as well as the results obtained (Section 3). Finally, features for a writing tool to assist novice academic writers are put forth based on what has been observed in the corpus (Section 4).