As we learn our languages, we decompose continuous linguistic signals into discrete units and recombine them in novel ways. What constrains our ability to flexibly compose words together? Studies investigating the neural bases of composition reveal early sensitivity in the left anterior temporal lobe (LATL), regardless of whether the words being composed appear in their canonical order (“red cup” vs. “cup red”) or the words come from the same language (“icicles melt” vs. “icicles noga”). However, behavioral evidence from three different bilingual groups (Arabic/English, Korean/English, Spanish/English) suggests distributional norms around heads of expressions (e.g., nouns in nominal expressions) affect overall processability. I will argue that more work on acquisition (in particular, bilingual development) is required to further our understanding of how feature(s) that denote heads of projections are abstracted away as well as whether and, if so, when they play a role during language development.