Text Versus Icons

A fundamental decision to be made is whether the language of interaction should be textual or iconic. 


The greatest advantage of text is that it is the natural language that people already recognize and use. 

The greatest disadvantage of text is that it must either be written by a human or generated by an algorithm. The text written by a human is specific to a narrow set of circumstances; it cannot be generally applied to a wide variety of situations. Inasmuch as a storyworld should be able to generate zillions of situations, we won’t be able to prepare human-written text for story worlds.

Text generated by algorithms is mechanical and suffers from repetition. However, AI researchers have learned a great deal about how to generate text from conceptual information, and it might be possible to apply these methods to overcome this problem. Many years ago I developed a system called “Tinkertoy Text” that assembles sentences out of differing combinations of words. This gives some greater variability to the text, but it is a weak solution. 

Icons do not carry the semantic baggage that words in natural languages carry. Natural words have shades of meaning that can cause confusion in the reader. Normally the writer supplies contextual cues that disambiguate the words; such disambiguation would not be readily available in computer-generated text. But players can accept the meaning of an icon directly, recognizing its semantic breadth. Consider, for example, this icon:

thumbsup 80

It denotes the general concept of affirmation. It can be taken to mean ‘yes’, ‘agree’, ‘approve’, ‘OK’, or any number of other concepts related to affirmation. Moreover, an iconic language is intrinsically international. On the downside, iconic languages must be learned by the player; the learning curve required for a language is initially offputting. 

Can we actually build an iconic language that meets our needs?
I have researched various logographic language systems. Most are built around writing with a pen, using various combinations of pen strokes. We have a peculiar set of capabilities and requirements. For example, we can use more complex imagery utilizing resolutions of up to 80x80 and 24-bit color. Moreover, we can lay out sentences in two-dimensional structures — a huge advantage. We don’t have to use extended relative pronouns that link one part of a sentence to another; the connections between words can be geometrically designated.


But we also have some constraints to consider. We cannot use multi-part icons. Here is an iconic language that combines word components to make complete words:

multi-icon-words med hr

We cannot use this approach, because the player would not be able to select a word in two steps; doing so would require the player to visualize the word in advance, which in turn requires the player to memorize the language before playing. Our players will need proper prompting. 

Another constraint is that our sentence structures must be computable. Natural languages permit a wide range of sentence structures; to be computable, we must use a single, narrowly-defined sentence structure. The Deikto system, in which a sentence consists of a subject, a verb, and a variable number of objects, is simple and can handle our needs.