Sensory correlation detection by children treated for congenital visual deprivation

Abstract

The temporal covariance of sensory signals provides critical information for determining the relationship between different entities in the sensorium. How quickly during the developmental timeline does the ability to detect such correlations become evident? Addressing this question is important for assessing whether this ability can help bootstrap the early stages of perceptual learning. Here we report work designed to assess the ability to detect temporal correlations of varying strengths within and across sensory modalities in 15 patients treated for congenital blindness as part of Project Prakash, a humanitarian and scientific effort focused on treating early blind children and through their help, understanding visual development. The performance of Prakash patients was compared with that of 21 normally-sighted blur-matched controls. In the intra-modal condition, participants were asked to determine which of two disks was blinking most in unison with a circumscribing ring. In the inter-modal condition, participants had to identify the disk blinking most congruently with a concurrent audio track of beeps and silences. This experimental design yielded three main results. First, we found that, while not fully reaching the level of normally-sighted controls, Prakash patients were able to detect correlations with markedly above-chance accuracy rapidly after sight onset. Second, for both groups, performance levels in the inter-modal and intra-modal conditions were comparable. Finally, the extent of the time series that participants observed before making a decision was similar between the two groups but markedly longer than would be required when using a pre-defined statistical decision criterion. These results help characterize a foundational process for detecting relationships between environmental entities, point to the resilience of acquiring this ability to early-onset, prolonged visual deprivation, and suggest that it could potentially serve as a bootstrapping mechanism for learning to extract environmental cliques.