This study addresses the problem of inconsistency in rating IELTS exams and, further, the need for valid criteria for rating levels of written performance.
Determining written performance involves a series of complex semiotic events or processes. On the part of examination candidates these include the interpretation of prompts and the composition of written responses. On the part of raters, the process includes interpreting criteria for rating, making sense of candidates' scripts and attributing rating criteria to text features in the scripts. Each of these events contributes to variation in scoring individual scripts. Scoring is a linguistically mediated activity and for this reason I planned the study as an investigation into the language features of candidates' scripts at different levels of written performance, and whether those language features might delineate performance level.
The data for this study came from students who were non-native speakers of English studying general English and English for academic purposes. The students wrote responses to Tasks 1 and 2 of the IELTS General Training Writing Module. Their texts were graded into three performance levels: basic, intermediate and upper intermediate.
The study examined language features of the subjects' texts as possible indicators of written language development at different levels of performance. The analysis identified numerous linguistic options writers chose in response to the prompts. Some of these stemmed from misunderstandings of cultural knowledge implicit in the topics of the prompts. Different interpretations of the task resulted in observable differences in responses. However, the analysis revealed that less developed texts expressed the interpersonal function with difficulty, using familiar and personal terminology when more formal linguistic choices would have been appropriate. Lower level texts were limited in the organisation of actual information, which was expressed with limited technical terminology. More developed texts demonstrated fluency through the use of clearly linked concepts. Even though upper intermediate texts contained structural errors, these did not interfere with understanding candidates' meanings.
The analysis indicates that it is not easy to identify specific lexicogrammatical features that distinguish different levels of performance. It is the sum of language features integrated textually which create successful scripts. This suggests that assessment, which focuses on isolated language elements such as vocabulary or sentence structures, detracts from the semiotic processes of composing texts and interpreting and therefore rating texts. The study proposes that as rating is a complex, meaning making activity, raters respond first and foremost to texts as a whole rather than to individual components. A possible explanation for this is that raters have expectations which stem from sociocultural conventions in terms of text types and social purposes of texts. Assessment is a response to the conventional combinations of linguistic elements in texts. These elements are numerous and vary considerably across texts.
The analysis of lexicogrammatical elements of texts provides evidence of how social purposes are realised in texts and raises questions about how these influence rating processes. The study suggests further research into the use of holistic descriptors of texts for scoring written texts.