Representing and responding in the visual world: a new model of contextual cuing.

点击翻译按钮获取中文摘要

One of the most fundamental psychological functions humans possess is the ability to recognize a familiar scene and perform an action relevant to it or an action relevant to some internal goal. For example, we may want to search for our keys within a room in our house. We know that the actions we perform in completing this simple task will vary, depending on whether we are searching for the keys in the kitchen, the bathroom or the lounge. That is, the context in which we are situated is crucial to the order in which we search locations in the room. We rely on our memory for a specific scene to narrow down the options and make our search as efficient as possible.Of course, this form of learning is not restricted to locating household objects. The same cognitive processes are likely to play an important role when any organism performs an action within many different environments, from driving cars in crowded traffic, walking to the local shop, or even when an animal hunts its prey in a forest. In every case there is a need to process the different features, or cues, within the environment, and then generate predictions about where certain elements will be located using our memory of previous encounters with similar scenes.We currently have a very limited understanding of this fundamental aspect of human behaviour. In experimental tasks developed over the last 10 years, researchers have started to examine how we learn this type of information. In a laboratory task designed to invoke this behaviour, participants view a computer screen which on each trial displays a context of distractor objects (e.g., different coloured "L" shapes) and a unique target object (e.g. a "T" shape). Participants have the goal of locating the target and responding to a particular feature of the shape, such as its orientation. During the task, participant's reaction times are measured. Crucially, some distractor configurations are repeated throughout the task. It has been shown that participants are faster to locate the target in repeating configurations than in completely novel arrangements. This decrease in the time taken to detect targets must be due to participants storing the repeating patterns of context in long-term memory.The current research aims to test a newly proposed computational model of this type of learning. This model learns about repeating scenes by creating memories for how specific objects within the scene are arranged with respect to the target object. Taking the analogy of a familiar kitchen scene, the model predicts that we learn only about the spatial location of a target object relative to other objects in the room (e.g., the toaster is opposite the fridge and to the right of the cooker). This information is enough to explain why the toaster is located more quickly on successive searches, as each object in the kitchen provides some information as to the location of the toaster. However, it seems natural to suppose that we will also learn about other aspects of the kitchen that are not relevant to our search. That is, we will engage in "incidental learning" (unintentional or automatic learning) about the general layout of the room, the objects present and their positions relative to one another (e.g., the cooker is opposite the fridge, the sink is under the window). Recent evidence from our laboratory suggests that these relationships, which are irrelevant to the search task, are also learned. The current project takes these important findings as a starting point and will provide a thorough examination of scene learning processes, allied to the development of a new computational model. In a complimentary strand of research we will monitor and study eye movements during scene learning. This research will provide data which will determine whether attention plays a key role in controlling learning in this behavior. The findings of this research will also inform the development of our new model of scene learning.