If S₁ had been chosen to be a geometric distribution, no a priori choices would have been necessary. This distribution was effective in Suppes (1970), but it would provide a poor fit in this case. If the geometric distribution had been used here, the parameter S₁ would have appeared in every theoretical probability involving N, and whenever one or more adjectives were included, a corresponding number of S₂ = 1 - S₁ terms would have appeared in the probability.