skip to content

Processing German modal particles and predicting referents in incremental interpretation

Vinicius Macuch Silva1, Timo Roettger2, Michael Franke3

1University of Birmingham, 2University of Oslo, 3University of Tübingen

We present a study focused on whether modal particles serve as predictive cues to interpretation in the incremental processing of language. We investigated the processing of the German particles eigentlich/tatsächlich, which signal either contrast or agreement between their host utterance and previous information in the discourse context. We report the results of a mouse tracking task where participants (n=50) listened to sentences like "Das ist eigentlich/tatsächlich ein Bild von einem Wolf" ("That’s actually/indeed a picture of a wolf") in response to a polar question like “Is this a picture of a wolf?”. Their task was to listen to the exchange while looking at two pictures displayed on a screen (e.g. wolf and dog) and to move their mouse and click on the picture depicting the most likely referent of the sentence containing the particle. We recorded participants mouse movements, clustering them in terms of similarly-shaped trajectories, as shown in Figure 1. Our results show that the trajectories - which index moment-by-moment update of expectations - are modulated by the usage of the particles in context. More concretely, we find that upon hearing eigentlich participants shift their trajectories towards the referent not mentioned in the question, whereas upon hearing tatsächlich they continue their trajectories towards the mentioned referent without subsequent change of course. Our results also show that when the particles are used unreliably there are both spatial and temporal modulations to the trajectories compared to pragmatically coherent usages.

Figure 1: Mean mouse trajectories. Top: Mean item trajectories aggregated over participants. Bottom: Clusters of mean trajectories aggregated over items and participants, with mean item trajectories shown in the background. The shaded areas in the graphs show the (average) time windows between the onsets of the discourse particle and the disambiguating noun. The left column shows data from the reliable group, while the right column shows data from the unreliable group. Tatsächlich items are shown in green while eigentlich items are shown in orange.
*