We evaluate corpus-based measures of linguistic complexity obtained using Universal Dependencies (UD) treebanks. We propose a method of estimating robustness of the complexity values obtained using a given measure and a given treebank. The results indicate that measures of syntactic complexity might be on average less robust than those of morphological complexity. We also estimate the validity of complexity measures by comparing the results for very similar languages and checking for unexpected differences. We show that some of those differences that arise can be diminished by using parallel treebanks and, more importantly from the practical point of view, by harmonizing the language-specific solutions in the UD annotation.

This chapter demonstrates how compression algorithms can be used to address morphological and syntactic complexity in detail by analysing the contribution of specific linguistic features to English texts. The point of departure is the ongoing complexity debate and quest for complexity metrics. After decades of adhering to the equal complexity axiom, recent research seeks to define and measure linguistic complexity (Dahl 2004; Kortmann and Szmrecsanyi 2012; Miestamo et al. 2008). Against this backdrop, I present a new flavour of the Juola-style compression technique (Juola 1998), targeted manipulation. Essentially, compression algorithms are used to measure linguistic complexity via the relative informativeness in text samples. Thus, I assess the contribution of morphs such as –ing or –ed, and functional constructions such as progressive (be + verb-ing) or perfect (have + verb past participle) to the syntactic and morphological complexity in a mixedgenre corpus of Alice’s Adventures in Wonderland, the Gospel of Mark and newspaper texts. I find that a higher number of marker types leads to higher amounts of morphological complexity in the corpus. Syntactic complexity is reduced because the presence of morphological markers enhances the algorithmic prediction of linguistic patterns. To conclude, I show that information-theoretic methods yield linguistically meaningful results and can be used to measure the complexity of specific linguistic features in naturalistic copora.