Previous research in the field of perceptual dialectology aiming to identify perceptual isoglosses has traditionally been based on Preston’s original ‘draw-a-map’ task (1981). In this task, listeners indicate on a hand-drawn map where they believe dialect boundaries to exist. This approach has two methodological drawbacks, in (1) requiring complex aggregation techniques of the individual maps (see Montgomery & Stoeckle, 2013); and (2) not offering a (direct) measure of how well the listener can actually identify varieties in the first place. We propose a novel ‘map-based accent-recognition’ task which provides direct insight in perceptual isoglosses on the individual and group level simultaneously at any granularity, and quantifies listener-related factors in accent recognition.

We applied this task to – and studied listener-related factors in accent recognition in – Standard Dutch as spoken in the Netherlands. Dutch lends itself well to the study of accent recognition, due to allowing for notable regional accentual variation within an otherwise well-defined standard language (Pinget, Rotteveel, & Van de Velde, 2014). While decades of production-based dialectological research have provided a clear view of how linguistic variation is geographically structured in The Netherlands, our understanding of how regional variation is perceived is very limited.

Our online map-based accent-recognition experiment was taken by 1,578 native Dutch listeners, who were presented auditorily with geographically marked speech, and asked to indicate their predicted origin of the speaker by click on a map. The forty speakers (selected from the Spoken Dutch Corpus; Oostdijk 2002) came from twenty places in the Netherlands that represent different traditional dialectal areas and include large cities, smaller towns and rural areas.

We used a multinomial GAM (Wood, 2017) to investigate the extent to which listeners’ placements of the speakers correlated with the speakers’ true origins. Results showed significant differences in identifiability between the different speaker origins, with some being recognized very accurately and others being localized in completely the wrong place. We additionally used recursive partitioning along the twenty true speaker origins to identify which regional divisions were more vs. less important. We observe a major role for the North-South division in The Netherlands, a major sociolinguistic boundary, and subsequently identify additional differences that have heretofore not been observed.

In addition, we used a gamma location-scale GAM to determine the extent to which listeners’ “accent recognition errors” (defined as the great-circle distance between their clicks on the map and the true speaker origin) could be explained by sex, education, age, geographical knowledge, regional origin, distance to their own region. Results show significant effects for all of these barring sex, which we interpret in terms of differences in social and mobility profiles.

Our results confirm and extend established effects in dialect recognition, specifically periphery, broad-distinction and substrate-language effects. We discuss these in light of the Dutch language situation, and elaborate on future research in which map-based accent-recognition tasks could be useful.