Computer-Aided Rhetoric

This page collects ideas for using computer systems to automate the production of documents that create feelings and beliefs in a reader.

Word Processor as Integrated Development Environment

I believe that as grammar checkers gradually improve, they will become able to parse meaning. As one types, a list of the ideas and nuances one is expressing will build up beside the text, with a graphical indication of how strongly or effectively each idea is expressed. Other attributes of the text might also be graphically indicated, such as clarity, mood, speaker, dialect or period.
Alternatively, one might express target ideas in a language that is English-like but regularly structured (and at least partially graphical), then let the computer score one's efforts to communicate those ideas in English. It would be like having a teaching assistant looking over your shoulder all the time.

Over time these tools willl enhance an author's ability to express himself. As elegance and grace become measurable, and the impact of each potential utterance on a reader becomes predictable, a word processor will become like the tool-rich development environments used by programmers. By participating in an interactive creative process in which the computer is both easily accessible reference and simultaneous critic of all levels of a work, any reasonably bright, competent author will one day become able to write modern Shakespearian English.

Live Computer-Aided Rhetoric HUD

Although the IDE model will be more effective for the production of lasting works, conversation could also be augmented with a heads-up display. For example, an on-line rhetoric analyzer would be valuable in any sales or negotiation context.

Sentence Structure

It's now possible to determine the structure of a sentence as an author types it, using the technology behind the grammar checkers found in popular word processing programs. While editing, sentence structure could be made clear with different colours for different parts of speech.

When reading, the different parts of a sentence, e.g. noun phrase, subject, verb phrase, preposition, could be made typographically distinct in a subtle way, so a reader could subconsciously tell what each word's role was visually before having to understand it. For example, dense texts such as Paradise Lost or the Iliad would be clearer if the text gave one visual cues associating roles with words.
Once this association had been set up in a reader's mind, one could generate interesting emotional or meaningful responses in a reader by "colouring" words or phrases against their natural role in a discourse.

The grammar parser might provide a variety of hints to the typesetting module.

When typesetting, phrases and related words should be kept together. Each adjacent pair of characters in a document should have a number representing how important it is that they be kept together. If a word must be split, it's better to do so at a syllable boundary. In germanic-style words, it's better to break at some syllables than others to preserve meaning. It's more important that some words be kept with other words than that some individual words be kept whole.
When laying out a paragraph on a page, try to keep phrases together if it's possible to do so without making the whitespace too large or small.

Each pair of adjacent characters in a document should have a number indicating how important it is not to divide them at the end of a line.

From easiest to break to hardest:
Between one sentence and the next.
Between one phrase and the next
Between words within a phrase
Between sub-words in germanic-style words.
Between syllables within a word
Between other characters.

Mood-Inducing Typeface Manipulation

Looking at the typesetting for various books over the years it's become apparent that typeface can be used to support the mood of a message. An author should be able to select a section of text in a word processor and choose from menu items like Tense, Suspenseful, Angry, or Serene and have that supported by things like amount of kerning, serifs, even-ness of character spacing, line spacing, italicization, line weight, and character shape. The author should only have to worry about mood, not how the mood is achieved.

  I suspect that the effect of typeface on mood is a transient effect. Once the reader gets into a long passage of text he or she filters out the effect of typeface. It's only when things change that the reader's mood is strongly affected. Thus the typesetting system would have to make an ongoing series of changes to keep renewing the desired transient effect.

  Implementation detail: this might be achieved through the use of multiple-master fonts, although for complete effectiveness the (single) font should have easily-twiddled parameters.

  Idea: Once the author has selected an area of the text and marked it as melancholy, say, the specific character manipulations employed will vary depending on the location of each character within its surrounding sentence, paragraph, or discourse. It should be possible to use an ATN to determine what parts of each sentence are nouns, noun phrases, subject and object, subordinate clauses, etc. Different parts of each sentence or paragraph should be morphed in different ways to support the mood of the whole communication.

  Expansion of idea: Some effects might be indicated by the author in multiple parts, such as marking part of the text as anticipatory, part as climax, and part as denoument. More generally, any emotional effect might be indicated as rising or falling over a swatch of text, e.g. rising serenity.

Cadence and play on syllables.

  Unlike sentence structure, which can often be difficult to parse, the way each English word breaks into syllables is well-known, as is the way the word should be stressed as it is spoken. Much of the mood of a spoken utterance is conveyed through the tone, volume and speed envelopes of the utterance. Here an "envelope" describes the changes in a quantity over the duration of the utterance.

  If the tone, volume and speed of each syllable could be indicated typographically, emotional impact and meaning would become clearer.

Rhetoric Toolkit

The holy grail of Computational Semiotics would have to be a rhetoric generator. You input a desired meaning such as Vote Conservative, and the program comes up wth a convincing argument for doing so. The software involves a model of the user that can be interrogated and programmed with the desired imperative.

  Here rhetoric can be broadened to mean actions inducing a person to believe something or do something. The more possible actions the software has at its disposal, the greater the likelihood of success.

  The base case is for the software to come up with a textual argument supporting the desired proposition.

  Given more resources, the software might start rumours among the target's friends, run TV commercials and billboard ads, reprogram the target's sleep-course on conversational Italian, or have its secret police haul the target user down to the station and enroll him in a course of sleep deprivation and other cult inculcation tactics.

Research Directions and Tools

There are two main problems whose solution is implicit in the construction of a Shakespeare IDE or Rhetoric HUD. These are an emotional, motivational, and situational model of a person, which could be applied to characters in the narrative and also to the reader or interlocutor; and a model of the information in the domain of discourse. Once standards for these models have been agreed on, a computer will be able to apply them recursively. Thus, a computer will eventually be able to mark the correctness of a story in which "Alice", a nineteenth century schoolgirl, pretends to be a banker in order to convince the Mad Hatter to invest in her South Seas trading company.

Another research direction is determining the emotional effect of typeface on message. How long does it last? Do different effects last different amounts of time before they need renewing? Which moods are easy to achieve and which are more difficult? How does all this vary depending on the base typeface - i.e. do variations on Times Roman give better suspense than Helvetica? Does it matter if the reader is aware of the changes? Must the changes be noticeable to be effective?

As a starting point, one might see if an unholy lashup of the open source Eclipse IDE and the new SVG editor Inkscape might be possible.