EUROGRAPHICS 2019/ P. Cignoni and E. Miguel Short Paper Font Specificity Luther Power 1 and Manfred Lau 2 1 Lancaster University, UK 2 City University of Hong Kong Figure 1: Four fonts with increasing font specificity scores. The normalized specificity scores are 0.106, 0.359, 0.544, and 0.793 respectively. For each font, the top and bottom 15% of the number of unique words given by 24 participants to describe the font are shown (left and right columns and with corresponding number of participants). The notion of specificity can be seen in the distributions of words. Abstract We explore the concept of “image specificity” for fonts and introduce the notion of “font specificity”. The idea is that a font that elicits consistent descriptions from different people are more “specific”. We collect specificity-based data for fonts where participants are given each font and asked to describe it with words. We then analyze the data and characterize the qualitative features that make a font “specific”. Finally, we show that the notion of font specificity can be learned and demonstrate some specificity-guided applications. CCS Concepts • Computing methodologies → Perception; • Human-centered computing → Human computer interaction (HCI); 1. Introduction The work of “image specificity” [JP15] introduces the concept that images that elicit consistent descriptions from different people are more “specific”. In this paper, we explore this concept for fonts and introduce the notion of “font specificity”. Analogous to im- age specificity, the idea of font specificity is that a font that elicits consistent descriptions from different people are more “specific”. Fonts are different from images as an image has colors and usu- ally consists of a larger scene, whereas a font focuses on text and the style that it is written in. In general, what makes a font specific (e.g. bold font) is different from what makes an image specific (e.g. image with people). We believe that this notion of font specificity can lead to new ways of understanding and thinking about fonts. The contributions of this paper are: • We introduce the concept of font specificity. The consistency of the human descriptions of fonts (i.e. the concept of specificity) can be used as a feature or descriptor of fonts, and the concept gives us new ways of thinking about fonts that have not been explored before. • We collect specificity-based data for fonts. • We explore the characteristics (i.e. qualitative features and quan- titative image descriptors) that make a font more “specific”. • We show that the notion of font specificity can be learned and learn a function to predict specificity scores for new fonts. • We demonstrate applications with the specificity-guided visual- ization and specificity-guided search of fonts. 2. Related Work Image Captioning. There has been much work in computer vision in the problem of automatically generating captions or sentences to describe images [BCE * 16, KFF17], and a detailed review of them is beyond the scope of this paper. The key difference is that they au- tomatically generate sentences to describe images whereas we ask humans to provide words to describe fonts. More importantly, we care about the distribution of the provided words for the specificity concept and the actual words themselves are less significant. Crowdsourcing. There exists previous work in collecting data through crowdsourcing and then learning from such data to solve various graphics problems. Examples include learning a similar- ity measure for 2D clip art [GAGH14] and for fonts [OLAH14]. c  2019 The Author(s) Eurographics Proceedings c  2019 The Eurographics Association. https://diglib.eg.org https://www.eg.org DOI: 10.2312/egs.20191015