Realisations and variants of have to: what corpora can tell us about usage-based experience

This paper explores the potential of corpus data to account for language users’ mental representations of a high-frequency item that is prone to phonetic reduction. We present an analysis of the realisations of semi-modal have to in spoken American English, which is a candidate for to-contraction (compare wanna and gotta) but has no clearly established contracted form. The study therefore focusses on the potential reduction of to and its conditioning by speech rate and phonetic environment. Three variants are extracted from the data, a citation form [hævtʊ] (or [hæftʊ]), a ‘schwa variant’ [hævtə] (or [hæftə]), and a reduced form that is akin to to-contraction [hævɾə] (‘havda’). The first two can alternate freely, but are subject to preferences based on the following sound, while the contracted form is strongly tied to rapid speech. These results suggest that the citation form and schwa variant are generally stored exemplars of have to, while ‘havda’ is more weakly represented in the system (as an outcome of on-line articulatory reduction). In this it clearly differs from conventionalised contractions such as gotta and wanna. On the methodological level, the study shows that thorough analysis of spoken corpus data provides insights into exemplar representations, though experimental hypothesis testing is also necessary.