What’s Hard? Quantitative Evidence for Difficult Constructions in German Learner Data

1. Introduction Our study is concerned with the identification of 'difficult' structures in the acquisition of a foreign language, which will shed light on theoretical considerations of L2 processing. We argue that – compared to simple vocabulary items or abstract syntactic patterns – structures that contain lexical material as well as categorial variables are especially difficult to acquire. The difficulty level for particular patterns is shown to depend on surface invariability but not on the syntactic categories within which target patterns are embedded. As an example we study the distribution of certain structures which are underused by L2 German learners. The question " what is difficult for a language learner? " can be addressed using several kinds of data, including learner corpora (e.g. error analysis and over/underuse data, for an overview see Granger et al. 2002), elicitation data, or psycholinguistic studies. Here we focus on corpus data. Previous corpus studies focusing on learner difficulties have examined token and type frequencies in order to calculate vocabulary richness measures, such as lexical density as an index of learner competence (Halliday 1989, Laufer & Nation 1999, and many others). However, lexical frequencies do not tell us what constructions are difficult for learners beyond individual lexemes, nor why. Many other studies (examples are Borin & Prütz 2004 or Westergren-Axelsson & Hahn 2001) focus on interference errors due to the learners' native language (or other learned languages) by comparing learners with a certain L1 to native speakers. Yet in order to establish explanations for difficulties in L2 acquisition independent of a learner's native tongue, we must examine the distributions in native and learner data of e.g. lexemes, collocations, colligations (cf. Stefanowitsch & Gries 2003) and syntactic structures, across learners' linguistic backgrounds. We take the stance that L1-independent underuse phenomena are due to learners either not acquiring patterns, or else avoiding their use despite familiarity with them, in both cases indicating increased difficulty. 2. Data The data for this study comes from the Falko corpus (Fehlerannotiertes Lernerkorpus des Deutschen als Fremdsprache), which consists of texts from advanced learners of German and control data from German L1-speakers (Lüdeling et al. 2008), allowing contrastive interlanguage analyses. The corpus is stored in a multi-layer model searchable at various levels of annotation. In order to diminish the possibility that the learners are simply unfamiliar with the items in question, we examine only advanced learners and focus on frequent, prevalent patterns. …