The Role of Natural Language in a Computer Game
The Author Discovers the Problems Inherent in Natural Language Understanding
on Tuesday 27 January 2009
by popa author list print the content item create pdf file of the content item
in Editorials > Video Game Editorials
comments: 0
not rated -

In the dawn of the computer age, natural language understanding was a seductive idea that never fulfilled its promise. The problems and limitations of natural language understanding in the context of a game are discussed.

Several decades ago I had the dream of creating a computer game in which the player could converse with computer-generated characters. For example, a murder mystery in which the player could ask suspects questions. At the time this seemed like a reasonable idea. Little did I know that this simple idea was to send me down a long and frustrating road.

The first computer game I ever saw was on a CRT in a laboratory at White Sands Missile Range in the middle of the New Mexico desert. I was employed there while I was working on a doctorate degree in physics at New Mexico State University. It was not much of a game. It consisted of a triangle-shaped spacecraft that was captured by a simulated mass in the center of a circular CRT display. One button would fire the thrusters and the other would change the angle of the thrust. For its time, it was amazing. It was a way to directly experience the physics of being in orbit— in two dimensions anyway.

The second game I ever played was a text adventure coded in Fortran called ADVENTURE that ran on a mainframe the size of a small house. This was the game that opened my eyes to the possibility of using human language as an interface to the internal knowledge of a computer. That promise caused me to spend a lot of time developing that concept, eventually resulting in a U.S. patent in natural language computer comprehension.

But, as we all know, in the commercial world the language interface concept never went much beyond ADVENTURE. Today’s games are about moving through and performing actions in a simulated landscape. If language input is used at all, it’s usually in the form of a list of possibilities. My naive idea of using language as a means to communicate to a computer’s artificial intelligence has been relegated to the area of scientific study for simulating human understanding of language, a goal now defined by the so-called Turing Test.

The idea behind the Turing Test is a software program that can convince someone that the computer is actually a human. Although this test is not designed for the purpose of a game, one might argue that someone might find a game more enjoyable if the computer-driven character with which one is conversing is convincing in its role. That was the idea I originally wanted to develop.

After many years of study and programming, I now believe there’s a flaw in the assumptions behind this idea. There are two conclusion I came to: (1) the Turing Test goal is impractical and probably impossible (although some recent tries have been impressive) and (2) it doesn’t substantially advance the idea of a computer game.

First of all, the Turing test is anthropomorphic, meaning that it tests for human-like qualities rather than intelligence. Also, purely intelligent behavior can be inhuman. This means that there is an inherent contradiction that arises when we try to simulate a human being with computer intelligence.

Second, a game by its very nature has to be limited in scope. Human intelligence is marked by its seemingly unlimited and constructive memory. For the game designer, unlimited options mean an impossibly large set of consequences which are impossible to realize inside a finite game with current technologies.

What this means is that language, if it’s used at all, has to be relevant to the specific options of the game which severely limits the advantages of using human language in the first place. I believe that one day there will be games that mimic some elements of human language capability, but that will have to wait for a considerable leap in memory storage and processor speed, and even then it will probably never be the central mechanism of a mainstream game. Natural language will remain principally in the domain of human communication for the foreseeable future.

You must be logged in to make comments on this site - please log in, or if you are not registered click here to signup

Stock photos courtesy of stock.xchng
All other content is copyright © 2010