Performance prediction strengths of noun and verb phrases in L2 writing: Comparison of density and complexity variables

This study aimed to compare different noun and verb phrase density and complexity variables regarding if and how strongly they could predict L2 writing performance. 183 short literary analysis essays with analytically-obtained scores, written in an English as a Foreign Language context, were used as the corpus of the study. The scores were divided as low, mediocre and high scores by K-means cluster analysis. Separate regression models were built for each group using Complex Nominal per T-unit, Complex Nominal per Clause, Noun Phrase per Sentence, Noun Phrase per 1000 Words, Verb Phrase per T-unit, Verb Phrase per Sentence and Verb Phrase per 1000 Word as predictor variables. A multivariate model of noun phrase complexity was also tested. The results showed that Complex Nominal per T-unit was the strongest predictor of high performance. Verb Phrase per 1000 Words was the only significant verb phrase-based (negative) predictor of high performance. Mediocre performance could only be predicted by Complex Nominal per Clause. Low performance could not be predicted by any of the noun or verb phrase-based variables. The results confirm that noun phrase complexity develops across performance levels; however, more traditional (length based) complexity indices may be more useful to explain low performance.