Beyond domain specific languages

Most of my friends, growing up in the late 70's and early 80's, have played Zork or some other kind of interactive fiction.

Although the earliest adventures used just a simple VERB NOUN syntax, the Infocom games had much more powerful parsers that could understand more complete and natural inputs such as "take all the rings but the silver one" or "put the red battery inside the lantern".

In high school (say, around 1982), I wrote a very stupid, very simple adventure game, "Obelisk Adventure", in PDP-11 Basic. It had pointless quests, a maze with a big nasty rat (the maze wasn't mappable, because it randomly moved your stuff around). I think I also remember an endless hallway (poorly implemented). I also implemented a few similar adventures on my home computer (first a Challenger 1P, later an Atari 800). This was my claim to fame for a while, as pathetic as it appears in retrospect.

These games didn't go away after Infocom and the other publishers disbanded due to the onslaught of more graphic games (such as Castle Wolfenstein and, eventually, Doom and Quake). The underlying virtual machine was reverse engineered and an open source community grew up around virtual machine implementations, as well as languages and tools that compiled down to that machine: Graham Nelson's Inform.

The Inform language is a mix of general purpose features, and a lot of very domain specific stuff -- specific to the idea of simulating a world of rooms, objects and actors, and specific to the hooks of the natural language parser. Here's some Inform 6 source, stolen from the Wikipedia entry:

Constant Story "HELLO WIKIPEDIA";
Constant Headline "^An Interactive Example^";

Include "Parser";
Include "VerbLib";

[ Initialise;
    location = Living_Room;
    "Hello World";
];

Object Kitchen "Kitchen";
Object Front_Door "Front Door";

Object Living_Room "Living Room"
    with
        description "A comfortably furnished living room.",
        n_to Kitchen,
        s_to Front_Door,
    has light;

Object -> Salesman "insurance salesman"
    with
        name 'insurance' 'salesman' 'man',
        description "An insurance salesman in a tacky polyester 
              suit.  He seems eager to speak to you.",
        before [;
            Listen:
                move Insurance_Paperwork to player;
                "The salesman bores you with a discussion
                 of life insurance policies.  From his
                 briefcase he pulls some paperwork which he
                 hands to you.";
        ],
    has animate;

Object -> -> Briefcase "briefcase"
    with
        name 'briefcase' 'case',
        description "A slightly worn, black briefcase.",
    has container;

Object -> -> -> Insurance_Paperwork "insurance paperwork"
    with
        name 'paperwork' 'papers' 'insurance' 'documents' 'forms',
        description "Page after page of small legalese.";

Include "Grammar";

And you can do a lot with that language, and I've occasionally looked into trying my hand at it. But it never seemed like a good use of my time (if I'm not working on Tapestry, I try to find things that get me away from my computer, and preferably, out of my house).

... so I was stunned and amazed when I saw the new version of Inform, Inform 7. The language above is gone and replaced with a natural language of tremendous power and flexibility. The same example, also off Wikipedia, in the new language:

"HELLO WIKIPEDIA" by A Wikipedia Contributor

The story headline is "An Interactive Example".

The Living Room is a room. "A comfortably furnished living room." The Kitchen is
north of the Living Room. The Front Door is south of the Living Room.

The insurance salesman is a man in the Living Room. The description is "An insurance
salesman in a tacky polyester suit. He seems eager to speak to you." Understand "man"
as the insurance salesman.

A briefcase is carried by the insurance salesman. The description is "A slightly
worn, black briefcase." Understand "case" as the briefcase.

The insurance paperwork is in the briefcase. The description is "Page after page of
small legalese." Understand "papers" or "documents" or "forms" as the paperwork.

Instead of listening to the insurance salesman:
    say "The salesman bores you with a discussion of life insurance policies. 
    From his briefcase he pulls some paperwork which he hands to you.";
    now the player carries the insurance paperwork.

Graham's goal was to allow non-programmers to write interactive fiction. The "code" is readable by human and compiler alike. Many of the examples in the documentation maintain this look, even when doing complex things like adding new verbs to the grammar (though some of the advanced examples are less natural language, and occasionally some Inform 6 code is injected as well).

What's important is that is not COBOL. It's not complex and rigid and verbose ... it is, in fact, very concise and the parser is very, very flexible and adaptive to the user. For example, you can easily talk about related nouns in a pretty random order. By editting the source, you can discuss the briefcase before describing the insurance salesman ... I tried this, juggling the order of the paragraphs in the example and the game still compiled and worked identically.

It also has a lot of cool features, such as instead of reading the letter the third time, say "Re-reading the letter isn't going to change what it says." The language is expressive enough to understand lots of dependent information, such as actions occuring in sequence, related to the "game clock", or repetition. Much of the game logic takes the form of "before" and "instead of" rules around normal actions, built into the language (and its standard library).

This language is partnered with a special purpose IDE and reams of excellent, hyperlinked, integrated documentation and examples. Further, the IDE has built in regression testing capabilities (you can play a game within the IDE and record the responses, then later play back the game after changing the source). The IDE also generates maps and indexes of the game.

Sure, he's been working on this for a long time (Inform 6 is nearly ten years old). But the possibilities of this, taken out of this specific domain, are impressive. Imagine writing complex behaviors in succinct natural language instead of verbose XML. I'm normally against MDA approaches, but I think this kind of natural language deserves a careful assessment to see where it can be used within the enterprise.