Towards the Automation of Book Typesetting: Experimentation and Discussion

Table of Links

• Abstract and 1. Introduction
• Computational Approaches in Editorial Design
• System
• Experimentation and Discussion
• Conclusion
• Acknowledgments and References

4. Experimentation and Discussion

We performed an experiment to assess if, and to what extent, the proposed system is able to automatically design book layouts with different purposes and styles. In particular, we are interested in studying the ability of the system to perform two design tasks: (i) create books with distinguishable layout designs, i.e. to generate a set of books that present varied visual characteristics; and (ii) create visually coherent books, i.e. to generate a collection of books that follow and share the same visual style between them. Therefore, we conducted a survey to assess the visual diversity and coherence of designs created with the proposed system.

The following subsections describe our experimentation process. First, we explain the conducted experiment. Then, we report and discuss the obtained results.

3.1 Experimental Method

For the evaluation of visual diversity, we performed the following actions. First, we selected one public domain book. Then, we input this content into the system and generated 15 books while not manually setting any visual or typography attribute, i.e. the attributes were defined at random by the system within the predefined ranges based on the typographic styles, rules and principles defined by default in the system (see Table 1). Lastly, we presented the 15 generated book designs to a group of 42 participants and asked them to assess the layout diversity and/or coherence of the set. The selected content was the book “Contos” written by the Portuguese author Eça de Queiróz and published in 1992.[2] This book comprises thirteen short stories, each one structured as a chapter, composed of about 73.330 words.

For the evaluation of visual coherence, we proceeded as follows. First, we selected a book from the set of 15 generated earlier to evaluate diversity. Then, we exported to file the settings used by the system to generate the selected book. Table 3 overviews the visual and typographic features of the chosen book design. Next, we input these settings into the system and created book designs for 5 different contents. Lastly, we presented the 5 generated designs to the same group of participants and asked them to evaluate the layout diversity and/or coherence of this set.

The two sets of books were evaluated by the same group of testing participants through a survey. First, we presented to the participants a set of 15 designs generated at random and then a new set of 5 designs generated using the settings of one design selected from the first set. After observing each set of designs, each participant was asked to classify its layout diversity and/or coherence on a scale between 1 (very coherent) and 5 (very diverse). As already mentioned, the testing group includes 42 individuals. The age of the participants ranged from 19 to 49 years old.

3.2 Results and Discussion

Figure 3 presents several pages composed automatically by the system for the first part of the experiment, where the visual settings are defined at random. Looking at the resulting designs, we noticed that most of the designs (12 out of 15) are portraitoriented. This is due to the automatic classification of the input content by the system as a long reading book and therefore the probability of selecting a portrait format is higher. Nevertheless, the format of the generated books exhibits slight variations between them. Concerning the text box, the grids and the typefaces used, we noticed the influence of the typographic principles and rules encoded in the system. Most designs present the body text typeset in the justified text (13 out of 15) over a onecolumn grid (12 out of 15), which are characteristics considered suitable for long reading books. Nevertheless, it is possible to observe diversity in other aspects. The dimensions of the margins and the size of the grid gutter, when it exists, vary between books. The used font, text size and leading and position of the running and page numbers also change. Additionally, it is possible to observe that different experimental features were used, alone or combined, in 6 of the generated designs.

Observing books created when the system settings are imported from a previously generated book, we can notice the share of visual characteristics among the resulting designs (e.g. book format and size, page margins, grid or used typefaces) and their similarity to the initial design. Figure 4 shows pages from the book that sourced the settings file (Figure 4a), which was selected from the set used in the first part of the experiment (Figure 3), along with pages of some books used in the second part of the experiment (Figure 4b). One should note that Figure 3 and Figure 4 depict only a small portion of the designs generated for the experiment, which can be all consulted in the supplementary files.

The conducted survey indicates that the system is able to generate both diverse and cohesive book designs. The chart in Figure 5 shows the distribution of answers obtained in the survey.

When we asked the survey participants to classify the diversity and/or coherence of the first set of books, which were generated at random by the system, the majority of participants (35 out of 42) considered them as diverse (19 participants) or very diverse (16 participants). Only a few participants considered that this set of books was very cohesive (2 participants) or cohesive (5 participants). When participants performed the same task for the second set of books, which were generated from a specific settings file, the majority of participants (38 out of 42) considered that the generated designs were visually coherent between them. Most of them (30 participants) considered that this second set of designs present a high-level coherence among them.

The survey results revealed that variables and properties defined by the system can create visually diverse layouts. Although the system engine determines the features of books based on a set of predefined typographic rules and principles, it also employs probabilistic mechanisms to define some attributes. This allows the occasional definition of attributes in an unexpected manner, promoting visual variation on the resulting layouts. This is visible, for instance, in the first part of the experiment, where books are created with the different formats even though the system tends to avoid the use of landscape format in long reading.

The obtained results expose the exploratory nature of the proposed system, which demonstrates high potential to stimulate and foster graphic designers' creativity and experimentation in the different stages of the design process. The system presents itself as a co-creativity tool which enables editorial designers to explore multiple conceptual and visual possibilities in an accessible, easy, and effortless manner. This is possible by enabling users to not only define values of the different attributes but also by enabling them to define the level of autonomy of the system and/or target the exploration of certain properties. Therefore, the system can be used in most stages of the design process, from the earlier and exploratory stages (when designers can take advantage of random generation to look for new types of layouts) to the final stages (when designers need to fine-tune one or more graphic attributes).

The system allows exporting data to settings files that encode the design of the book and, later, can be loaded into the system to create other books which are visually similar. As demonstrated by the survey results, the proposed system can automatically create highly coherent designs when the book properties and characteristics are prior determined and stored in these settings files. Users can also include in the system their desired book properties both directly on the system interface and/or by modifying the typographic principles and rules used by the system. The proposed system is also a useful tool to automate the design of books that need to follow a set of restricted typographic and visual attributes (e.g. when it is necessary to design new books that will be part of a collection or a series). Thus, besides its exploratory nature, the system also enables the automatisation of some editorial design tasks and routines.

With the proposed system, users do not require other software tools to manipulate and produce the resulting book designs since it operates inside of the popular editorial design software Adobe InDesign and the generated books are made available as editable documents. This way, it is fully integrated into the typical working environment of editorial designers. This allows its use both as an exploratory and as an automation tool, thus empowering designers to easily edit or fine-tune the output designs directly in a familiar environment.

The experimental features implemented by the system can be observed in some of the generated books and definitely contribute to their diversity and variation. Nevertheless, we noted that some generated designs exhibit certain graphic limitations. This may be related to the fact that the generated designs comply with the same default typographic rules and principles. Although users can manipulate these rules, this will primarily change the typeset of the book but it will not include new visual features. For this reason, the system facilitates the addition of new features as well as their control. In this sense, one can add new features to the system for exploration and automatisation purposes. We believe that this possibility may allow the system to solve some lack of distinguishable visual features of the output, including in the design of the covers.

In summary, the experimental results demonstrate the ability of the proposed system to automatically generate finished and functional designs from scratch. Furthermore, the results reveal the potential of the system as a useful exploratory tool in the context of book typesetting and editorial design. It may be operated by graphic designers when they are searching and exploring new conceptual and visual perspectives, finetune a book characteristic, and/or design books that must be coherent with a given set of typographic rules. In addition, we consider that the automation provided by the system has great potential in varied graphic design commercial scenarios, e.g. the automatic design and production of customised books, which is relevant for print-ondemand applications, or the effortless typeset of books for an existing one-book collection.

[2] The book “Contos” by Eça de Queirós was retrieved from the Project Gutenberg. One may download the book at the following address www.gutenberg.org/ebooks/31347 (visited: 26 July 2022).