The systematic but errorful speech patterns of young toddlers (in addition to being adorable) serve as great windows into the workings of the developing child mind. In particular, the ordered trajectories that we observe in children’s phonological errors over time provide great clues as to how linguistic knowledge is acquired: what prompts children to change their grammars, what evidence they use, what representations they build, and how they eventually succeed in sounding like adults. 

One trajectory: the phrase ‘dog barking’ as produced by Trevor (data from Pater, 1995)

… [gɔgaɹkɪŋ] (age 1;8)  …  [gɔg ga:kɪŋ] (1;9.29)  …  [gɔg ba:ɹkɪŋ] (1;11.5)  …  [dɔg ba:ɹkɪŋ]

In this talk, I will focus on two types of developmental phenomena in child phonology: regressions, where children get worse at some aspect of speech production before they get better, and avoidance, where children fail to even attempt words or constructions that contain sound patterns they have not mastered. My goal will first be to demonstrate the subtle range of ways in which children do (and do not!) regress and avoid in their phonologies, using corpus data from my own joint work and from a fascinating case study in Donahue (1986). My second goal will be to demonstrate the extent to which existing constraint-based error-driven learning algorithms (as in Hayes, 2004; Prince and Tesar, 2004 interalia) are inherently equipped to derive the attested typology of child regressions and avoidance, and to discuss some implications of this grammatical approach to children’s messy production data.