Goals

Language Level

Since most of my programming is currently done in languages like C++ and Java, sopl (Speech Oriented Programming Language) is at roughly the same level. Also, since interoperability with existing systems vastly increases the appeal of a language, a primary goal is easy translation from sopl to a commonly used language. Because Java fulfills more of the other goals listed below (lower perplexity, garbage collection, separate subtype/inheritance, etc.), I have tried to make the semantics easily translatable to Java where possible.

Perplexity

A measure of the usability of a programming language with a voice recognition system is the perplexity, a number which roughly corresponds to the average number of lexemes that can be said in an individual context. In a language like C++, the perplexity is very high. For example, any of the following lines could follow a semi-colon:

        #include <stdio.h>              //  Preprocessor commands
        int x;                          //  Type names
        x = 10;                         //  Local variables
        Screen::set_width(10);          //  Class names
        cout << x;                      //  Global variables
        printf("Hello, world");         //  Functions
        for (x = 1; x < 10; x++) {      //  Language commands

The first goal for the language is therefore to lower the overall perplexity.

Conventionalization

In natural languages, phrases which are used commonly become conventionalized. Their literal meaning gets replaced by the conventional usage. "Kick the bucket" is a good example, since we no longer have access to the original literal meaning. Although in programming there is a strong desire to generalize, in natural languages there seems to be a pressure towards specialization. Sopl conventionalizes some very common programming idioms in order to achieve lower perplexity. For example, when I look at existing code, 80% of the "for" loops are simple counters with static upper bounds and static increments. A conventionalized phrase can be used to express this semantic using a very low perplexity construct:

loop for a from 1 to 10 by 2

After the word "loop", only "for", "while" or "until" can follow. After "for", only the name of a variable can follow. After the variable, the word "from" is required, etc. Although a "keyboard programmer" may accuse this of chattiness, it is extremely easy to enunciate.

Isolation of New Names

Since new names in a programming language (e.g. variables, classes, etc.) can be just about anything, speech recognition engines have a very hard time recognizing them. To ease the introduction of new names for the speech recognizer, sopl lexically marks all such introductions with known keywords. Note that because of type, storage, and access qualifiers, C++ and Java do not provide this marking.

Ease of Utterance

The language should be easy to speak. At this point, "easy to speak" is highly subjective. Certainly, short English words are easier than punctuation or nonsense words. Further research is required.

Hoare

Where they do not conflict directly with the goals outlined above, I have tried to follow Hoare's advise: Simplicity, security, fast translation, efficient object code, and readability.

prev next