Word order in a grammarless language: A 'small-data' information-theoretic approach

David Gil has argued that Riau Indonesian (Sumatra. Indonesia) has no syntax, or at least not much. This controversial analysis undermines all current models of grammar, especially those describing acquisition and on-line processing. To test the strength of this analysis, we computed the information gain holding between unigram and bigram models of regular and randomized samples of English and Riau Indonesian. English samples were included as a relatively syntax-heavy baseline. We then correlated information gain values with language (English vs. Riau Indonesian), text type (original vs. randomized), and their interaction within a linear mixed-effects regression. The results suggest (a) that English and Riau Indonesian have the same amount of bigram informativity and (b) that randomization eliminates this effect in both languages. These findings do not support Gil’s syntax-free analysis; rather, they point to some kind of productive constraints on Riau Indonesian word order.