Results from a Sasak (Eastern Indonesia) language production experiment show that speakers' word order choices are shaped by both the animacy and relative length of arguments in the clause. Specifically, Sasak speakers exhibit an animate before inanimate, as well as a long before short, bias. The latter of these findings challenges assumptions that a universal feature of the language production system is to produce shorter elements before relatively longer ones. As such, the resulting puzzle is that data from some languages (e.g., English, German) reveal a short before long bias, while data from other languages (e.g., Sasak, Japanese) demonstrate a long before short bias. I argue that these different biases are not language-specific, and propose the Primacy of Semantic Richness before Event Structure Hypothesis, which states that "the semantic richness of a nominal element takes precedence over its increased demand on working memory at any temporal/linear position prior to the lexical verb." With this hypothesis, I propose that length biases can be understood as a function of where the shifted domain is relative to the lexical verb because of the demands that the verb places on form-oriented production processes. Foremost, the hypothesis accounts for a range of cross-linguistic data, and does so without stipulating language-specific parameters. Moreover, it makes several clear empirical predictions, one of which is supported with initial evidence from verb-initial languages.