No three productions alike

Situated language use is influenced by a number of dynamic phenomena that introduce lexical variability and path dependence, such as fluid discourse granularity, priming, and alignment in dialogue. The empirical tradition of usage-based lexicology does not account for such variability. In fact, its primary theoretical approaches appear to presuppose high population convergence on particular lexemes in language production. This is implied in several key concepts of phraseological and constructionist models, notably entrenchment, the principle of no synonymy, and the idiom principle, as well as the dominance of the statistical paradigm in the field. In spite of its relevance for linguistic theory and corpus methodology, this assumption appears to be untested. This study provides an analysis of inter-individual lexical overlap of verbs and nouns in five task-based corpora of (mostly) German. Results indicate that speakers are maximally variable and highly divergent in their lexical use in spite of narrow communicative constraints and group homogeneity. A qualitative analysis links this variability to situational (cognitive, socio-pragmatic, and discourse-level) engagement with the task and material, which results in referential diversity, spontaneous meaning mapping, and abundant word formation. The degree of observed variability raises questions with respect to the stochastic properties and functional mechanics of entrenchment and the role of repetition of identical material, such as lexicalized chunks, in conventionalization. It further emphasizes the need for a better understanding of the distributions that underlie pooled data, without which the validity of frequential extrapolation to individual behavior and system status stands to question. This is of particular relevance to language assessment and other practices of contrastive analysis of speaker productions, e.g., in multilingualism studies.