Novel scene understanding, from gist to elaboration

ABSTRACT

We examined the cognitive experience of novel pictorial scenes, using observers’ words. On each of two critical trials, a single novel scene (a photo) was presented briefly, after which observers described what they saw. The reports were highly valid, assessed against details of the stimuli. The most frequent concepts used by the observers defined scene gist empirically – there was wide agreement on the basic scene interpretation. The average extent of the reports, or cognitive bandwidth, was fairly high overall: a median of 14.4 correct meaningful words per scene. However, there was a wide range of report lengths across observers. With length, the reports became increasingly elaborative and unique (and still correct). The longer reports document the richer and more unique side of a gist-to-elaboration continuum. In addition, a manipulation of immediately prior experience (priming with other scenes) caused an increase in the reporting of prime-related information, conveyed by a wide range of concepts, including relational concepts. The results are consistent with cognitive theories of perception and the claim of substantial cognitive bandwidth. Furthermore, the highly elaborative reports document a measured diversity of powerful scene understanding.

KEYWORDS:

• Scene perception
• scene understanding
• context effects
• priming
• full report

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Researchers often present a pattern mask after the stimulus to terminate visual processing, including sensory memory (including Fei-Fei et al., 2007; Greene et al., 2015). However, a mask can enter consciousness as a distinct object that competes with the target scene (e.g., Breitmeyer, 2008; Harrison et al., 2016). The mask could discourage full engagement with the scene, and cause an underestimation of cognitive bandwidth. The assumption that a short term memory representation emerges from the stimulus scene is parsimonious and consistent with current research (e.g., Hollingworth et al., 2008; Robinson & Irwin, 2019).

2 We used a conventional ANOVA to estimate how reliable the variation in priming with quartile was. We analysed the proportions of people words as a function of quartile and prime group. The interaction of quartile and prime condition did not approach reliability even for a planned test, F[3,50]<1.