A cognitive approach to syntax: Dependency grammar
There are two popular approaches to treating grammatical phenomena: (1) phrase-structure, where words combine into abstract categories (e.g., noun phrase, verb phrase, etc.) which are subject to constrained displacement (“movement”) processes (Chomsky, 1957, 1965, 1981, 1986, 1993); and (2) a simpler formalism, dependency-grammar (e.g., Tesniere, 1959; Hays, 1964; Mel’cuk, 1988; Hudson, 1984, 2015; Osborne, 2019; Nefdt and Baggio, 2023), in which a word is simply connected to another word via a dependency arc to form a larger compositional meaning, in “what-you-see-is-what-you-get” (WYSIWYG) representations. Here I suggest that dependency-grammar offers a more transparent representation system for language processing and cross-linguistic word order phenomena. First, human language processing is extremely sensitive to distances between words that depend on one another: shorter connections are preferred (Gibson, 1998, 2000) and longer connections come with a greater cost as measured behaviorally (e.g., Grodner & Gibson, 2005) or neurally (Shain et al. 2022). Second, the world’s languages minimize syntactic dependency lengths in their grammars to some degree (Futrell, Mahowald & Gibson 2015; Futrell, Levy & Gibson, 2020; cf. Liu, 2008). These phenomena are more transparently represented in dependency grammar than phrase structure grammar. I then revisit the standard arguments in favor of phrase structure with movement: (1) the English auxiliary system (Chomsky 1957, 1965, 1971); (2) cross-linguistic syntactic “island” phenomena (Ross, 1967; Chomsky, 1973; 1977; 1986; Schutze et al. 2015); and (3) constituency. When we look at each of these in detail, none favors phrase structure with movement over dependency grammar. First, the English auxiliary system seems to be better analyzed as a series of usage-based constructions rather than movement (Sag et al. 2020). Second, it is now becoming clear that there are no generalizations across constructions for long-distance dependencies (Abeille et al. 2020), thus favoring a construction-based approach. And third, most constituency tests are tests for meaning, not syntactic phrases, which are easily explained within dependency grammar. I therefore argue that usage-based dependency grammar is not only simpler, but covers grammatical phenomena better and more transparently than phrase structure with movement.
