Varietal Differences in the Grammaticalization of Keep V-ing
In this project, Michael Wolfman and I investigated the use of the keep V-ing construction (where V is any other verb) in British and American English. This construction functions as an aspect marker for both continuative and iterative aspects, meaning it can either indicate that the V verb is being continued after having been started earlier or that the V verb is being performed repeatedly. Background research indicated that the interpreted aspect is heavily influenced by the lexical properties of the V verb.
As a preliminary study for the purposes of proposing a larger study for the full project, we hand-annotated 50 examples of the keep V-ing construction from each corpus, the Corpus of Contemporary American English (COCA) and the British National Corpus (BNC), annotated for morphological tense and aspect, Aktionsart or lexical aspect, and interpreted lexical aspect: whether it was continuative or iterative (or ambiguous). Chi-squared testing on this preliminary data revealed a statistically significant difference between the British and American usage, with the British samples using the iterative meaning more often.
Thus, for the final project, we hand-annotated 3,000 tokens of keep V-ing from each corpus for tense, subject person, V-ing, and interpreted aspect. Chi-squared testing revealed statistically significant interactions between dialect and all four annotation fields. However, we were most interested in the interaction between dialect and interpreted aspect, and these results matched our hypothesis that the British corpus used iterative aspect more often than the American corpus. In fact, British use of the iterative construction was about 7 percent higher. Finally, we also performed some additional analysis of the verbs used most often in each aspect, by creating rank-to-rank graphs which compare the ranks of verbs in each aspect to their corresponding ranks in the other aspect. They show that verbs ranked highly in one rank tend to have much lower ranks in the other aspect, which indicates that each aspect is associated with different sets of verbs. This is one way of pointing out that aspect is highly influenced by the lexical properties of the V verb without needing to manually annotate 6,000 tokens for Aktionsarten.
You can view the full poster presented for the course here, and you can read the final paper here.
This project was accepted to be presented at the University of North Carolina Chapel Hill's Spring 2020 Linguistics Colloquium before it was cancelled due to COVID-19.