Abstract/Summary

This thesis explores the visual similarity that underlies the coherence in the design of individual typefaces. Typeface designers aim to achieve a unifying coherence in their typefaces, so that characters can be identified individually as well as belonging together giving rise to an overall style. The objective is to determine whether the coherence perceived by readers differs from the coherence intended by designers. The research is cross-disciplinary, combining empirical studies of readers’ perceptions with a computational model that is based on relevant typeface design knowledge. Character similarity is studied in multiple different typefaces (fonts) intended for continuous reading in Cyrillic, Devanagari, and Latin scripts. The studies were conducted online to collect a large number of responses. The participants were presented with a sequence of character triplets. They were asked to identify the odd one out in each of these triplets judging by their visual similarity, thus making a statement about the similarity of the two complementary characters. This method studies the similarity in context, which provides more refined details about participants’ similarity judgements. The model interprets characters using two kinds of features: more specific parts and more general roles. The model learns the relative saliences of these features from a subset of the data collected in the studies. This allows the model to predict participants’ responses to the triplets from the studies and for other, unseen triplets. Additionally, the model can provide explanations of the criteria participants used in their similarity judgements and can generate similarity matrices. The model achieved high scores when predicting response probabilities and identifying the overall odd ones out. A view of coherence that is supported by readers’ perception can be used to assist designers in their creative process, help with fonts’ quality assessments, and contribute to readability research and multi-script typography.