Universal properties of compositional structure in language

Language has hierarchical compositional structure; its smallest components-speech sounds, or temporal slices of signals in signed language (Brentari, 2002)-are meaningless in isolation, but they combine to create meaningful units like words, which further combine to create phrases, then sentences, and so on. In Anglo-American and Western European linguistics, we are trained to divide this structure into morphology, the internal structure of words, and syntax, the internal structure of phrases (e.g., Aronoff and Rees-Miller, 2003).

Crucially, this division depends on having a solid definition of "word"; otherwise, we cannot know where to draw the line between intra-word and inter-word structure. But despite decades of work aiming to formulate universal diagnostics of wordhood, no definition of "word" is valid for all the world's languages (cf. Dixon and Aikhenvald, 2003; Haspelmath, 2011; Pike, 1952; Tallman, 2020; Zwicky and Pullum, 1983).

It is therefore time to approach this old question from a new angle. I propose that there might be no way to define "word", and thus to distinguish morphology and syntax, because morphology and syntax might not be separate subsystems of language, but rather two levels of a singular compositional system. This idea is supported by the fundamental cognitive preference for simplicity (Chater and Vitányi, 2003): why have two compositional systems when one would do? Further, research on language evolution has shown that language evolves its compositional structure due to competing pressures for it to be both learnable and expressive (Kirby et al., 2014; Kirby et al., 2015), and there is no principled reason for these pressures to distinguish words from sentences.

Based on this idea, my project will take an entirely novel approach to the issue of wordhood. I will pioneer the use of experimental methods originally created for developmental psychology and language evolution, along with Bayesian cognitive modelling, to ask: what evidence is there for a cognitive distinction between levels of composition in language?

To answer this, I will investigate language universals and their cognitive basis. Universals are features that recur across many (unrelated) languages. They are often hypothesized to be explained by cognitive principles, since human cognition determines the form that language can take (Berent et al., 2008; Chomsky, 1988; Culbertson et al., 2012; Fedzechkina et al., 2012; Ferdinand et al., 2019; Hudson Kam and Newport, 2005; Kirby et al., 2015; Wilson, 2006). I will test whether language universals manifest the same way in morphology and syntax. If they do, this may be because a single compositional system underlies both levels. My findings will be significant for theoretical linguistics, since they will provide new evidence for a long-standing core issue in the field.