Imagine a goddess that assigns sound systems to languages. She says, “English shall have 44 sounds, Hawaiian shall have only 13, and !Xu shall have 141.” And so it is. How might we understand the goddess’ ways? We can assume that she has access to all the sounds humans can produce and perceive, but she doesn’t randomly sample from these and assign them to different languages – she follows certain principles. She wants to ensure that the sounds are easy to articulate – linguists call this principle Markedness. She also makes sure that the sounds are not too similar to each other so that listeners can tell them apart – Dispersion. And finally, she wants to make the entire system easy to learn so that children can learn it efficiently – Economy. Moreover, since these considerations may conflict – e.g., making the sounds in a language more distinct may lead to the inclusion of more complicated sounds – the goddess must find a way of balancing these organizing principles.

Most research on such questions has focused on vowels, which have simple articulatory and acoustic properties. Languages tend to have dissimilar vowels so that listeners can tell them apart. That is, Dispersion above all else. Consonant systems, however, have much more complicated articulation and acoustics, and are not as well understood. To understand how these conflicting requirements may be balanced in "stops", arguably the most complicated consonants, I analyzed 3020 stop systems and quantified their Markedness, Dispersion and Economy. I then built a computational model to predict the relative contributions of these factors given the observed stop systems and comparing them to systems that could exist but don’t. The results revealed that unlike vowels, stop systems are not organized to be dissimilar, but to be articulatorily simple and efficient to learn.