Karen Jesney (USC)

Positions are defined on the input: Evidence from repairs in child phonology Overview: Work in constraint-based child phonology has found that marked structures in privileged positions are typically acquired earlier than the same structures in non-privileged positions (e.g., Goad & Rose 2004, Marshall & van der Lely 2009, Revithiadou & Tzakosta 2007, Rose 2000, Tessier 2007). This paper argues that the effects of privilege are also visible in the patterns of selected repair, with onset clusters being repaired via epenthesis at a higher rate than comparable coda clusters, and absolute-initial input segments being preferentially preserved when deletion applies. These patterns cannot be reduced to an effect of sonority. I show that these data require that privileged positions be defined with respect to the input, rather than the output as has been commonly argued throughout the adult literature (Beckman 1998). Data: The data in this study come from Trevor, an Englishσ acquiring child whose utterances were transcribed by his speech pathologist mother (Compton & Streeter 1977, Pater 1997). All Onset Rhyme forms containing a target onset cluster in stressed, utteranceinitial position (n = 997) or a target coda cluster in stressed Nuc Coda utterance-final position (n = 636) were extracted. Target | clusters created through affixation were excluded. Data were M1 M2 P M2 M1 coded for age in months (range: 11 – 37), sonority profile of the target cluster, voicing of the target consonant in M1 position (i.e., the initial consonant of an onset cluster or the final consonant of a coda cluster – Baertsch 2002), accuracy, and repair type – epenthesis, deletion of M1, or deletion of M2. Clusters that underwent only segmental repairs were coded as accurate. Results: As expected, logistic regression showed that the probability of clusters being realized accurately increases significantly with age (β = 0.346, SE = 0.023, p < .001). Additionally, for Trevor coda clusters were realized accurately at significantly higher rate than onset clusters (see Figure 1 – β = 3.520, SE = 0.184, p < .001). Data for the repairs analysis was limited to rising sonority onset clusters and falling sonority coda clusters to avoid potential of appendix parses in /s+stop/ /stop+s/ targets. Deletion of M1, deletion of M2, and epenthesis were all attested in both onset and coda position. In onset position, however, epenthesis was over-represented and M1 deletion was under-represented (χ2(2) = 64.786, p < .001). This effect cannot be attributed to age, as there was no significant change in preferred repairs over time.

Figure 1 – Accurate cluster realization (dark grey) vs. repair (light grey) in onset and coda

Positions are defined in the input: Traditionally, positional faithfulness constraints are formalized as in (1a), with positions defined relative to the output. On this understanding, no distinction is expected between Figure 2 – M1 deletion (dark grey) vs. M2 clusters in onset and coda position with respect to deletion (mid grey) vs. epenthesis (light grey) epenthesis repairs; output-based constraints cannot in onset and coda distinguish between competing repair candidates (see 2). If positions are defined with respect to the input, as in (1b), however, the attested distinction in repairs is predicted. Epenthesizing ensures that both segments in the input onset are preserved without violating the input-based positional faithfulness constraint; no similar preference is expected for input codas. (1) a. MAX-C/ONSEToutput: Assign a violation mark to any consonant that lacks a correspondent in the onset of an output syllable

b. MAX-C/ONSETinput: Assign a violation mark to any consonant in the onset of an input syllable that lacks a correspondent in the output (2) MAX-C/ONSinput prefers epenthesis for onset cluster candidates; MAX-C/ONSoutput does not /blu/ bu lu b#.lu

MAX-C/ONSoutput

MAX-C/ONSinput * *

/ænt/ æn æt æn.t#

MAX-C/ONSoutput * * *

MAX-C/ONSinput

The same logic holds in modeling the under-representation of M1 deletion in onset clusters. Here, the relevant constraint references input word-initial position, as defined in (3). This constraint consistently prefers deletion of M2 to deletion of M1, but only when M1 originates in initial position – i.e., is in onset. (3) MAX-C/INITIALinput: Assign a violation mark to any word-initial input consonant that lacks a correspondent in the output Voicing and sonority effects: These repair effects cannot be attributed to differences in the sonority preferences of onsets vs. codas. This is clear in the subset of the data where voicing is contrastive in M1 position and sonority profiles are ideal (i.e., stop + liquid onset clusters, liquid + obstruent coda clusters, and nasal + stop coda clusters). Among coda clusters in this set, the probability of M1 being deleted increases significantly when M1 is voiced rather than voiceless (χ2(1) = 51.052, p < .001). This is not Figure 3 – Deletion in coda clusters with true in onset position. There, there are no instances of voiced M1 (dark grey) vs. voiceless M1 voiced M1 being deleted, even though this would allow (light grey) violations of *COMPLEXONS and *VOICEDOBSTRUENT to be simultaneously avoided. Instead, in onset clusters M2 deletion is strongly preferred in every case, ensuring that MAX-C/INITIALinput is respected. If it were only sonority at play in determining whether M1 or M2 is deleted, we would expect the effects of voicing to be equally evident among onset and coda clusters (cf. Pater & Barlow 2003). Sum: The effects of privilege on accuracy in child phonology are well known. To date, however, the role of privilege in determining repair preferences has been largely ignored. This study suggests that at least some of the effects that have been attributed to sonority and other factors in previous studies are best understood in terms of positional privilege. These findings argue for a rich representation of positions in the input and the ability for these to be referenced by positional faithfulness constraints. References: Baertsch, K. 2002. An Optimality Theoretic Approach to Syllable Structure: The Split Margin Hierarchy. PhD Dissertation. Indiana University. Beckman, J. 1998. Positional Faithfulness. PhD dissertation. University of Massachusetts Amherst. Compton, A.J. & M. Streeter. 1977. Child phonology: data collection and preliminary analyses. Papers and Reports on Child Language Development 7: 99–109. Goad, H. & Y. Rose. 2004. Input elaboration, head faithfulness and evidence for representation in the acquisition of left-edge clusters in West Germanic. In Fixing Priorities: Constraints in Phonological Acquisition, ed. R. Kager, J. Pater & W. Zonneveld, 109-157. Cambridge: CUP. Marshall, C. & H. van der Lely. 2009. Effects of word position and stress and onset cluster production: evidence from typical development, specific language impairment, and dyslexia. Language 85: 39-57. Pater, J. 1997. Minimal violation in phonological development. Language Acquisition 6(3): 201-253. Pater, J. & J.A. Barlow. 2003. Constraint conflict in cluster reduction. Journal of Child Language 30: 487-526. Revithiadou, A. & M. Tzakosta. 2004. Markedness hierarchies vs. positional faithfulness and the role of multiple grammars in the acquisition of Greek. In Proceedings of Generative Approaches to Language Acquisition (GALA). Utrecht: LOT. Rose, Y. 2000. Headedness and Prosodic Licensing in the L1 Acquisition of Phonology. PhD dissertation. McGill University. Tessier, A.-M. 2007. Biases and Stages in Phonological Acquisition. PhD dissertation. University of Massachusetts Amherst.