Proceedings of the Annual Meeting of the Cognitive Science Society

Language production requires speakers to resolve the problem of linearization, where they must choose what to talk about and in what order. Previous work has shown that scene descriptions while viewing the scene show evidence of clustering. Objects which are physically close together are mentioned close together and objects which are semantically similar are mentioned close together. Additionally, the transition time from one object to the next is longer when jumping across pre-defined physical and semantic clusters compared to when staying in the same cluster. We compared previous results with a new experiment where participants produced scene descriptions from memory. We find that descriptions from memory show evidence of clustering using physical and semantic relationships. Results also show that they rely more heavily on semantic relationships than descriptions generated while viewing the image. The work highlights the importance of task demands on linearization strategies in multiutterance production.