Wim Gombert

78 CHAPTER 5 beginner to native speaker, identified and defined by Bartning and Schlyter (2004) and the results were validated by Granfeld and Ågren (2014). ANALYTICAL TEXT MEASURES To support the holistic and overall DP judgments mentioned above analytically, a number of text measures was used. Di erent broad CAF measures that showed almost linear change across pro ciency levels from beginner up to intermediate pro ciency levels (CEFR level B1) in English (cf. Verspoor et al. 2012) were chosen to support the ndings of human-mediated and machine-mediated ratings. As coherence and cohesion already played an important role in the holistic judgments (see Appendix A) and the writing samples were too short to provide useful data, this text measure was not included in the analysis. For complexity Tense Use, Guiraud Index and Sentence Length were chosen. For Tense Use, the relative use of tenses other than the Present Tense was seen as a measure of verb phrase complexity (Granfeldt & Ågren, 2014). Tense Use was computed on a 10-point scale ranging from 0 (only Present Tense) to 10 (only other tenses), where a score of 6 implied that 6 out of 10 tenses were tenses other than the present tense. e Guiraud Index was chosen as measure of lexical diversity for texts containing more than 200 words (Van Hout & Vermeer, 2007). Although Biber et al. (2011) claim that phrasal complexity (i.e., the use of di erent modi ers in a noun phrase) is a better indicator of writing pro ciency than clausal complexity, average sentence length (full sentence, including coordinate and subordinate clauses) was taken as the third complexity measure (Norris & Ortega, 2009; Oh, 2006; Yoon, 2017). e data investigated by Biber et al. (2011) consists of research articles, written by pro cient (academic) writers, while the participants in the Rousse-Malpat (2019) study were beginners (CEFR level A1-A2) and the participants in this study had an intermediate level (CEFR level B1). Consequently, the language investigated in both studies can be considered as conversational French and sentence length is an appropriate measure for examining the linguistic quality of writings. Accuracy measures were Subject-Verb Agreement (SVA) and Determiner-Noun Agreement (DNA) as they are expected to contribute signi cantly to accuracy in FL French (Ågren et al., 2012). Both SVA and DNA were calculated on a scale ranging from 0 (no agreement) to 10 (100% agreement). Finally, text length, operationalized as the total number of tokens in the text, was taken as a uency measure. Direct Pro l provided information on tense use and accuracy. Vocabpro lers (Cobb, 2018) was used to calculate the Guiraud Index, average sentence length and text length.