Corpus data in usage-based linguistics: What's the right degree of granularity for the analysis of argument structure constructions?

The notion of usage-based linguistics is becoming more and more central in cognitively-inspired linguistics. One corollary of this is that there is an increasing number of studies that are based on authentic usage data from corpora, broadening the range of evidence against which theories are tested. This is generally a welcome improvement of methodological standards, but the choice of using corpus data also brings with it a host of methodological questions. One of these questions is concerned with the degree of granularity that provides the most insightful results. Many studies adopt an approach (implicitly) based on lemmas while some recent work distinguishes inflectional forms. Some studies are based on whole corpora or undifferentiated corpus parts whereas others distinguish between, say, spoken and written data. The present study investigates these two granularity issues – different inflectional forms and (register-)based corpus parts – on the basis of corpus data from English. First, I will compare the results of a lemma-based corpus analysis of an English argument structure construction to an inflectional-form-based corpus analysis to determine whether the two approach result in different suggestions concerning the semantics of the construction in point. Second, I will outline how to determine whether data from different corpus parts/registers result in different semantic generalizations of the same construction and how relevant corpus distinctions can be determined in an objective bottom-up manner. Finally I will discuss methodological implications of these results. Acknowledgments I thank the audiences at the Fourth International Conference on Construction Grammar and Conceptual Structure, Discourse, and Language 2006 for comments as well as John Newman for feedback and discussion. The usual disclaimers apply.