At the weekend I attended Banterhack 2016, an amazing hackathon run by my friend and esteemed colleague Manoj. Banterhack is, as far as I can tell, a reaction to the swathes of hackathons that have big name corporate sponsors who judge entries and give out appealing prizes and even more appealing chances to work for them. What that means is that people chase the “dream” of winning a 3D printed Oculus Leap Nest with the BarclaySachsWaterhouseCooper Prize for Innovation, and don’t make cool shit any more. In contrast, I was one of the sponsors of Banterhack, where we gave these terrible China clone iWatches as top prize. My main judging criterion was maximum ratio of high cool to low commercial viability.

Keeping within the spirit of Banterhack, I wanted to create something fairly cool, but with basically zero practical application. Thus, StoryTime (potential name rethink pending) was born.


First, an example:

 A republican hears the democrat say nothing and the democrat hears the republican say wise things. Campaigning is where someone says promises they can't keep and the democrat hears them and they listen to the republican and the republican hears the democrat. I always campaign.

This is a valid program in StoryTime lang. Running it prints the first 20 fibonacci numbers

StoryTime is a language whose aim is to be easily writable by a human programmer, but also easily mistook for valid semantic English. The idea came from the concept of steganography (security by obscurity, often hiding a secret inside something innocuous, not to be confused with stenography, which is the art of being able to write something so quickly it’s instantly unreadable).

I actually had three criteria in mind while making this:

  • It should be possible to construct programs by hand
  • The program should look and read like valid English
  • The program should surprise and delight (shock and awe) even those who know it’s a program, by being non-obvious how the program’s function came from its structure

That last rule was pretty useful in making decisions about how to structure the language, because it ruled out doing things in the way other natural language-esque esolangs do, e.g. ~English.

Storytime is currently an interpreted language built in Haskell with Parsec for parsing. You can see the source on Github here



A storylang program is a series of sentences. Sentences are separated by full-stops . (translation for American English: periods). All ‘statements’ (linguistically equivalent to clauses) are expressions: every statement returns a value.


One thing I really wanted was arbitrary nouns and verbs to be used as variables. I also wanted proper nouns, because that really makes things a lot more fun. Anything preceded with a or the is a variable. Anything with a capital letter is also treated as a variable, apart from “A”, “The” and verbs at the beginning of method declarations.

When variables are instantiated with a number equal to their length, for example:

> A dog.  
[ returns 3 ]
> The mouse.
[ returns 5 ]
> Gandalf.
[ returns 7 ]

Variables can hold either string or integer values at a given time.


The verb said represents printing to the console. Other conjugations of the verb are valid, and they don’t need to make grammatical sense in the sentence (although that’s the fun of it).

A variable followed by a said and then a quoted string will print the value of the string, as well as returning that as the expression’s value.

> A cow said "moo".
[ returns moo ]

A variable followed by said and then anything else will print the current value of the variable.

> The judge said his verdict.
[ returns 5 ]


Conjugations of the verb to hear followed by an expression represent assignment, and return the value after the assignment.

> The man heard the woman say "hello".
[ returns "hello"; man's value is "hello" ]

Method declarations

Method declaration is of the form:

> Fooing is where someone says their name.

And is used as it would be in valid english:

> I foo. You foo. The horse foos.

Methods currently take a single argument, which must be a variable, which binds to “someone”, as well as “they”, “their” and “them”. Other variables are resolved to whatever exists in the environment at the time (i.e. not bound into a closure).

Actually, now I think about it, I don’t really know what method declaration returns. Maybe I should look into that.

Other builtins

X listens to Y is equivalent to X = X + Y

X always METHOD run METHOD 20 times (hella arbitrary, I know. It was for the demo).


The thing is interpreted, but is flakey as hell. Syntax errors, access errors, etc. will all basically just cause a runtime error. Needless to say, you shouldn’t run this anywhere near production, because that would be like letting a kid with a toy stethoscope perform heart surgery.

Building It

Making the parser in Parsec was a real pleasure. I know it’s no longer the most fashionable, but it definitely did the job.

I actually parse in two stages, once to roughly tokenize, and once to try to make sense of what’s a pretty ambiguous grammar. For that second stage I already have a stream of tokens, and would have ideally used parsec a second time, but since there were Nerf bullets flying inches from my nose, I decided to stop combing through the documentation and LIVE A LITTLE. So I basically coded it by hand and now probably regret that.

The evaluation uses the StateT monad to handle “the environment”. It’s the first time I’ve used a monad transform in anything other than a pet example, and it was great. I’ve read some great stuff about combining them with the Free monad in order to get true monadic composability.

Future Development


I’d like to use adjectives for some purpose slightly separate to verbs (i.e. methods).

I was thinking one obvious thing would be to make them modify the instantiation of the variables, so that it’s easier to get bigger numbers in the program. For example a big dog could equal 30, with big being a x 10 modifier.

Time travel

It would be nice if you could use methods etc. before they’re defined, or even better, use future “facts” to influence the computation before the reader gets to them. Anthony introduced me to the Tardis Monad, which basically allows state to flow forwards and backwards within an execution frame. I’d love to find a place to use it, and this feels like it could be good. But is this within the spirit of being “highly obfuscated” or is this just being “too fucking confusing to actually use”? Who knows.


Something like Sally's age or Obama's birth certificate could actually refer to valued properties. This would make it fairly object oriented – my worry with this is that it might move the semantics too close to just describing the world, rather than highly obfuscated programming.