Sam Kabo Ashwell, "Scents and Semiosis"

There are, among interactive fiction authors, many aficionados of procedurally generated text. And indeed many of the most famous -- I mention Nick Montfort, Emily Short and Aaron Reed. Personally, however, I've never really seen the point of procedural text generation. What's the advantage? What does it give you? As far as I can see, three things:

• Surprise: the algorithm for procedural text generation is complex enough that not even the author can oversee all the possibilities. Hence, the text that gets generated can surprise even the author.
• Quantity: the amount of possible texts generated by even a simple algorithm quickly rises above what a reader could ever read.
• Uniqueness: the text read by me is read by me alone; other readers will read different texts.

But none of these seems very advantageous from the perspective of the reader. Text generated by an algorithm can surprise me, certainly, but text written by a human author can also surprise me. When I'm looking for material to read, I'm interested quality and not in quantity. (Indeed, the longer a book is, the more reason I need to pick it up.) And it is surely irrelevant to my experience of the text whether or not it is different from the text that someone else will read; in fact, there's no detectable difference between text generated procedurally or deterministically. So... what's the point?

Which brings us to Scents and Semiosis by Sam Kabo Ashwell, a piece of IF written in Inform 7 + Vorple, and written specifically for Emily Short. In this text, we get to play the role of a perfumer who keeps a chest full of old scents. We unstopper a series of bottles, and are then met with procedurally generated texts like this:

Forlorn coffee drifts through earthy white pepper. One of your more unabashedly commercial efforts, a celebrity-branded scent for Berea, the architect, game developer and squatter. You didn't get credited for it, thankfully.

• white pepper feels like things violent 
• white pepper might mean dissolution 
• coffee feels like the flirtatious affection of decay 
• coffee might suggest practical neglect 
• None of these feel right. Reconsider.

There is, first, a descriptive passage that describes a scent and then puts it into some life-context; then, second, a list of interpretations, a menu of choices that define what this scent means for you as the character. Scents and semiosis, obviously. Choose an interpretation, choose a new bottle, repeat until you've generated a sort of composite scent-memory.

Sometimes, the generated scents are very simple indeed:

Plumeria limned by allspice. Made for you by a noble editor.

At other times they are more complex, and they might be tied to quite specific events too:

Dark sweet pea, golden orchid, brisk caraway, pale reposado tequila, and the memory of red dal. You gave this to your mother on her deathbed.

So... what does it all add up to? What is the point? As I played the game, my general feeling was: not much. Part of this may be because I'm not really into scents, and I can't form even the vaguest conception of what the second and third of these perfumes would smell like. (Coffee with pepper I can handle.) But much more important is that fact that none of it came together into a narrative, or even a coherent character. I read that last perfume, and I'm like: tell me more about the mother, tell me more about the deathbed, how did she react to the scent, how did my character deal with the loss... but instead, I'm off to the next scent, and some new vaguely alluded to anecdote.

Of course, not all texts need to be narrative texts. I can enjoy a good short poem as much as the next person, and those certainly don't need to tell a story. An impression of a mood can be enough, as in this amazing little poem of Dickinson:

As if I asked a common alms -
And in my wondering hand,
A stranger pressed a kingdom -
And I - bewildered stand -
As if I asked the Orient
Had it for me a morn?
And it should lift its purple dikes
And shatter me with Dawn!

But that's not quite what Scents and Semiosis is doing, is it? Nor is it clear that algorithms that procedurally generate texts would be particularly good at this, since it requires all the skills of a great writer.

Having been left mostly unmoved by the piece, and in a sense disappointed, since I think highly of its author, I decided to peruse the source code of the piece. And that's when things started to click for me. I'm not sure it's the same click that other people are looking for when they study procedurally generated texts, but I'll take any click I can find. So here's my new and updated take: the beauty and fascination of a piece like Scents and Semiosis is not so much in the texts it generates, although these of course need to have certain qualities; rather, it is in the machinery of the generation itself. It is in the thoughtfulness and creativity exhibited by the author when he constructed the algorithm.

A windmill is a tool for turning grain into flour. But admiring the flour is not the best way of appreciating the windmill. Rather, you should get inside and check out the amazing technology, the giant wheels turning, the thoughtfulness and creativity that went into its design. This analogy fails at a couple of points, but then, so do all analogies.

As soon as I started reading the source code, I started appreciating the piece. I found out that scents fall into certain themes, like "goth" or "lewd" or "summer", all with their own carefully chosen vocabulary. There's a complicated system in place for picking several elements of the scent; for linking them to specific meanings; for linking them to seasons and anecdotes. There's a great attention to detail, to the point where the game explicitly ensures that no sexual overtones are connected to anecdotes involving teacher-pupil relationships. And so on, and so forth. What we're looking at is a semiotic machine, a machine that doesn't so much generate meaning, as it embodies it. There is of course a sense in which it spews out texts that the reader gets to interpret; but even more crucially, it is itself the result of thousands of acts of interpretation of its author. And knowing about those acts of interpretation actually gives more meaning to the texts that are generated. The fascination of the piece -- at least for me, and one would suspect for the author as well -- lies in seeing how the choices of the author end up generating the texts that end up being generated.

(Also, there is a Rabelesian list of butts hidden in the source, and you have not experienced this piece until you've read about

intense butts.

elusive butts.

butts.

butts.

unattainable butts.

butts which causes defeat.)

So I'm wondering... was this the point all along? Of procedurally generated text in general? To be appreciated always with the source code in hand, or at the very least to be appreciated and thought about while also contemplating the means of generation? Is it just a reductive mistake to think of the generated text as the text? I'm far from sure, but at the very least Scents and Semiosis has given me something to think about.