Trade-offs in Linguistics

As with most things, occasionally in linguistics you have a case where there's an obviously best solution - a solution that is superior to alternatives in all aspects (efficiency, complexity, clarity, etc.). More often than not, however, you have to make due with trade-offs between opposing factors. I just ran into a nice example of such a thing a few days ago, while thinking about Caia.

For quite some time I've been thinking about how to best use noun classes (the more general form of 'gender', which is not limited to simply gender, but may group nouns in a wide variety of ways, such as 'long things', 'dangerous things', etc.) and number so as to maximize the clarity of pronouns references. Obviously the goal of this is to minimize the number of potential nouns that a given pronoun could refer to.

To get an idea of what I didn't want, we can look at Trique. Trique mainly uses four pronouns: a first-person pronoun, a second-person pronoun, and two third-person pronouns (technically one is a 'fourth-person pronoun', but I'm not exactly sure how this works); single/dual/plural is then marked with auxiliary words. This makes for a very clean, simply pronoun system. Unfortunately, it also makes for a relatively ambiguous system, as well.

Getting back to my objectives. A couple days ago, I just happened to realize the optimal way of structuring noun classes and number, which also served to provide an excellent example of why it's usually not acceptable to optimize for just one variable. The optimal method would be to have a large number of noun classes structured that the distribution of nouns (more specifically, commonly-used nouns) in these classes is uniform.

This minimizes the potential for ambiguity, as it minimizes the probability that there will be more than one noun in a given class that a pronoun could potentially refer to. Unfortunately, this also maximizes complexity, requiring that you memorize the class of each noun. People would be even more of a pain, as to evenly distribute them among classes, you would need to separate them into noun classes as well, based on the specific attributes of the person (I'm not even sure what criteria you could use to evenly distribute people into classes like this).

So, while that provided some interesting food for thought, it ultimately only provided a good example of how not to do it in Caia.