“…Most previous work relies on phonetic or phonological features (Kretzschmar, 1992 , 1996 ; Heeringa, 2004 ; Labov et al, 2005 ; Nerbonne, 2006 , 2009 ; Grieve et al, 2011 , 2013 ; Wieling and Nerbonne, 2011 , 2015 ; Grieve, 2013 ; Nerbonne and Kretzschmar, 2013 ; Kretzschmar et al, 2014 ; Kruger and van Rooy, 2018 ) for the simple reason that phonetic representations are relatively straight-forward: a vowel is a vowel and the measurements are the same across varieties and languages. Previous work on syntactic variation has focused on either (i) an incomplete set of language-specific variants, ranging from only a few features to hundreds (Sanders, 2007 , 2010 ; Szmrecsanyi, 2009 , 2013 , 2014 ; Grieve, 2011 , 2012 , 2016 ; Collins, 2012 ; Schilk and Schaub, 2016 ; Szmrecsanyi et al, 2016 ; Calle-Martin and Romero-Barranco, 2017 ; Grafmiller and Szmrecsanyi, 2018 ; Tamaredo, 2018 ) or (ii) language-independent representations such as function words (Argamon and Koppel, 2013 ) or sequences of part-of-speech labels (Hirst and Feiguina, 2007 ; Kroon et al, 2018 ). This forces a choice between either an ad hoc and incomplete syntactic representation or a reproducible but indirect syntactic representation.…”