Use this URL to cite or link to this record in EThOS:
Title: Detecting human comprehension from nonverbal behaviour using artificial neural networks
Author: Buckingham, Fiona Jane
ISNI:       0000 0004 7659 006X
Awarding Body: Manchester Metropolitan University
Current Institution: Manchester Metropolitan University
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Access from Institution:
Every day, communication between humans is abundant with an array of nonverbal behaviours. Nonverbal behaviours are signals emitted without using words such as facial expressions, eye gaze and body movement. Nonverbal behaviours have been used to identify a person's emotional state in previous research. With nonverbal behaviour being continuously available and almost unconscious, it provides a potentially rich source of knowledge once decoded. Humans are weak decoders of nonverbal behaviour due to being error prone, susceptible to fatigue and poor at simultaneously monitoring numerous nonverbal behaviours. Human comprehension is primarily assessed from written and spoken language. Existing comprehension assessments tools are inhibited by inconsistencies and are often time-consuming with feedback delay. Therefore, there is a niche for attempting to detect human comprehension from nonverbal behaviour using artificially intelligent computational models such as Artificial Neural Networks (ANN), which are inspired by the structure and behaviour of biological neural networks such as those found within the human brain. This Thesis presents a novel adaptable system known as FATHOM, which has been developed to detect human comprehension and non-comprehension from monitoring multiple nonverbal behaviours using ANNs. FATHOM's Comprehension Classifier ANN was trained and validated on human comprehension detection using the errorbackpropagation learning algorithm and cross-validation in a series of experiments with nonverbal datasets extracted from two independent comprehension studies where each participant was digitally video recorded: (1) during a mock informed consent field study and (2) in a learning environment. The Comprehension Classifier ANN repeatedly achieved averaged testing classification accuracies (CA) above 84% in the first phase of the mock informed consent field study. In the learning environment study, the optimised Comprehension Classifier ANN achieved a 91.385% averaged testing CA. Overall, the findings revealed that human comprehension and noncomprehension patterns can be automatically detected from multiple nonverbal behaviours using ANNs.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available