Let Me Choose: From Verbal Context to Font Selection

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 8607–8613 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 8607 Let Me Choose: From Verbal Context to Font Selection Amirreza Shirani†, Franck Dernoncourt‡, Jose Echevarria‡, Paul Asente‡, Nedim Lipka‡ and Thamar Solorio† †University of Houston ‡Adobe Research †{ashirani,tsolorio}@uh.edu ‡{franck.dernoncourt,echevarr,asente,lipka}@adobe.com Abstract In this paper, we aim to learn associations be- tween visual attributes of fonts and the verbal context of the texts they are typically applied to. Compared to related work leveraging the surrounding visual context, we choose to focus only on the input text as this can enable new applications for which the text is the only vi- sual element in the document. We introduce a new dataset, containing examples of different topics in social media posts and ads, labeled through crowd-sourcing. Due to the subjective nature of the task, multiple fonts might be per- ceived as acceptable for an input text, which makes this problem challenging. To this end, we investigate different end-to-end models to learn label distributions on crowd-sourced data and capture inter-subjectivity across all annota- tions. 1 Introduction In visual designs, textual information requires the use of fonts with different properties. Whether it is books, magazines, flyers, ads or social media posts, different typefaces are commonly used to express non-verbal information and add more dimensions to the text. An appropriate font usually embodies information about character, context and usage of the design (Doyle and Bottomley, 2006). This mo- tivates us to explore font associations with regular users in a crowd-sourced setting. In other words, we investigate how users relate fonts to different characteristics of the input text. Current font selection interfaces such as O’Donovan et al. (2014) and commercial online services (e.g., MyFonts1 and Typekit2) assist users in selecting fonts by taking into account font sim- ilarity. However, they do not consider the verbal 1www.myfonts.com 2https://fonts.adobe.com/ context of the input text. Having a better under- standing of the input text, users can benefit from a font recommendation system during authoring, saving time and avoiding tedious exploration of long lists of fonts. Most graphic designers agree that there is no strict or universally-accepted rule for choosing fonts. Different social and personal factors can be involved in typeface selection, which makes this process subjective. However, there seems to be enough agreement among human opinions to build reasonably effective models of font properties (O’Donovan et al., 2014; Shinahara et al., 2019). Several empirical studies have directly explored the relationship between fonts and texts (Shinahara et al., 2019; Henderson et al., 2004; Mackiewicz, 2007). For example, Brumberger (2003a) indicates that readers have strong opinions about the appro- priateness of particular typefaces for particular text passages, and they can differentiate typeface/text mismatches. In this study, we aim to model for the first time the associations between visual font attributes and textual context, with the final goal of better font rec- ommendation during text composition. Our main contributions are: 1) We propose and formulate a new task: “font …

Content truncated to 500 words