N PTA dB HL (M, SD) GFTA-II SS (M, SD) EVT-II SS (M, SD) Hearing Loss (N = 17) 4 year olds 9 45 (20) 103 (8) 104 (13) 7 year olds 8 41 (11) 95 (21) 107 (20) Normal Hearing (N = 30) 4 year olds 15 screened at < 15 109 (10) 111 (15) 7 year olds 15 screened at < 15 102 (10) 114 (10) all pooled 47 — 103 (12) 110 (14) Summary of results Repetition accuracy: 1. ...improves with increased expressive vocabulary scores. 2. ...improves with increased speech production ability scores. 3. ...reveals sensitivity to phonological and lexical density for all children. 4. ...for older children (7 yr olds) with NH is at ceiling, but older children with HI perform more like younger children. 5. ...improves at older ages for children with NH, but is about the same for children with HI. 6. ...seems to distinguish those children with hearing loss in 7 yr olds but not 4 yr olds. Conclusions 1. For children with NH, repetition accuracy is at ceiling at older ages and for high-density words at younger ages. 2. For children with HL, repetition accuracy is poorer for low-density words at both test ages (in contrast to children with NH who improve accuracy of low-density words at the older test age). 3. The effects of (1) and (2) may be explained by more impoverished representations in younger children with HL (evidenced by the relationship between EVT and percent-correct). Future directions 1. Collect more data. 2. Look at independent effects of density, phonotactic probability, lexical frequency, and lexicality. (See Garlock, Walley, & Metsala, 2001; Bailey & Hahn, 2001; Luce & Large, 2001; Munsen, Swensen, Manthei, 2005; Storkel, Armbrüster, & Hogan, 2006; Storkel & Hoover, 2010). 3a. Look at response latency as additional dependent factor. 3b. Look at phoneme-level error patterns (already coded). 4. Look at acoustic characteristics of the children's productions (with reference to child and stimulus acoustic features). 5. Improve statistical modeling procedure to better predict real performance; more data will help inform priors. 6. Investigate results with respect to lexical processing models (e.g., NAM, TRACE, PARSYN, Shortlist, MERGE). Method Participants Materials ► 100 monosyllabic word forms (68 words, 32 non-words) ► stimuli = citation-style, natural production, adult female voice ► to confirm naturalness of stimuli, open-set identification task (N=7 adult judges): 98.7% accuracy ► stimuli do not vary by overall size or lexical usage frequency ► stimuli selected are high probability/high density (confounded) ► phonological level of access is intended target of stimuli ► comparisons between Hoosier Mental Lexicon and the Child Corpus Child Corpus is described in Storkel & Hoover (2010): Kolson (1960) + Moe, Hopkins, & Rush (1982) Procedure ► All stimuli randomized as part of computer game with participants ► Listen-and-repeat task: "say what the lady on the computer says" ► Stimuli played in open field at about 68dB SPL ► Speech production was assessed via Goldman Fristoe-II ► Lexical probe was assessed via Expressive Vocabulary Test-II Data analysis, Bayesian modeling ► Dependent variable 1. spoken word reproduction accuracy (correct, incorrect) ► Independent variables 1. Age (4 yrs, 7 yrs) 2. Hearing status (HL, NH) 3. Neighborhood density (dense, sparse) 4. Productive phonology (GFTA standard score) 5. Expressive vocabulary (EVT standard score) Main research questions 1. How is lexical access affected by hearing loss in young children? Are young hearing aid users sensitive to the phonological neighbors of words? 2. Does lexical access develop similarly in children with hearing loss and children with normal hearing? 3. Are standardized language tests (i.e., outcome measures) predictive in modeling the influence of lexical access performance? Background Neighborhood density (aka: lexical density. lexical similarity; see also frequency weighted neighborhood density) is the number of real words with similarity to a target word, varying by adding, subtracting, or substituting one phone. Effects on language have been shown to be inhibitory at the lexical level, facilitatory at the phonological level, and dependent on the task (Vitevitch & Luce, 1998; Vitevitch, 2002; Newman & German, 2002; Garlock, Walley, & Matsala 2001). That is, words from denser phonological neighborhoods may be less accurate when lexical competition is engaged, but more accurate when (only) phonological competition is engaged (possibly restricted to production; Storkel, Armbrüster, & Hogan, 2006). Phonotactic probability is the frequency of bi/phone occurrence. It has facilitatory effects on lexical access (e.g., spoken word recognition, repetition latency, identification errors, resolution in noise, etc.). That is, higher phonotactic probability words are more accurate, faster, easier, etc., in a variety of language-use contexts. **Probability & density are correlated, and in this study confounded. Age is almost always a facilitatory factor—but note that an older child's lexicon may be different in a number of respects that have been shown to affect accuracy (e.g., robustness of representations, fluency, number of lexical items and associated neighborhoods, etc.; Storkel, 2002; Edwards, Beckman, & Munson, 2004; Munson, Edwards, & Beckman, 2005). Lexical frequency is the estimated usage frequency of a form. It is not possible for non-words. It is not correlated with phonological density or phonotactic probability in the lexical items used here. It has a facilitatory effect in many domains of perception and production, including accuracy (Vitevitch, 1997, 2002). Higher lexical usage frequency means more accurate reproduction. Hearing loss (i.e., audibility) affects the lexicon: ► The lexicon may develop slower in children with HL (Mayne, Yoshinaga-Itano, & Sedey, 2000; Mayne, Yoshinaga-Itano, Sedey, & Carey, 2000; Moeller, Hoover, Putman, et al, 2007a, 2007b). ► Lexical competition may reduce lexical access ability (Jerger, Lai, & Marchman, 2002) ► Nonword repetition and phonological awareness performance may be decreased (Briscoe, Bishop, & Norbury, 2001; Moeller, Tomblin, Yoshinaga-Itano, Connor & Jerger, 2007) ♦ Little is known about how childhood hearing loss affects the structure of, or access to, words from dense versus sparse neighborhoods in the lexicon. Repetition of words from dense and sparse phonological neighborhoods in children with hearing loss and normal hearing Mark VanDam 1 Noah H. Silbert 2 Mary Pat Moeller 1 1 Language Development Laboratory, Boys Town National Research Hospital 2 University of Maryland mark.vandam@boystown.org www.VanDamMark.com Acoustical Society of America Seattle, WA | 26 May, 2011 4pSC14 STIMULUS WORDS (Munson, Swenson, Manthei, 2005; Storkel & Hoover, 2010) dense sparse Pr(dense ≠ sparse) z-val Word size number of phonemes 3.0 (0.2) 3.1 (0.3) > 0.05 -1.007 Lexical frequency (N dense = 34) (N sparse = 34) word frequency – CC 2.3 (1.7) 1.8 (1.5) > 0.05 -0.488 word frequency – HML 2.4 (1.0) 1.8 (0.9) > 0.05 0.896 Phonotactic probability (N dense = 50) (N sparse = 50) segment frequency – CC 0.060 (0.011) 0.050 (0.014) ** < 0.01 3.422 segment frequency – HML 0.056 (0.012) 0.048 (0.016) ** < 0.01 2.617 biphone frequency – CC 0.0051 (0.0031) 0.0035 (0.0026) ** < 0.01 3.203 biphone frequency – HML 0.0041 (0.0034) 0.0032 (0.0027) ** < 0.01 2.581 Neighborhood density (N dense = 50) (N sparse = 50) number of neighbors – CC 15.9 (4.9) 9.8 (6.0) ** < 0.01 -4.935 number of neighbors – HML 23.3 (5.9) 16.1 (7.4) ** < 0.01 -4.973 Results Phonological density, hearing impairment, and age R01DC006681 P30DC04662 T32DC00013 The model: 1. …predicts Hearing and Age effects reasonably well. 2. …consistently predicts accuracy as a function of both EVT and GFTA scores, and shows the interaction between hearing loss and lexical access. 3. …may be improved with more data (better informed weighting parameters). Speech production Expressive vocabulary EVT GFTA ρ SPEARMAN p ρ SPEARMAN p All pooled 0.556 0.001 0.383 0.001 HL 0.559 0.019 0.661 0.003 NH 0.376 0.040 0.022 0.907 4 yr old 0.579 0.003 0.680 0.001 7 yr old 0.490 0.017 0.459 0.007 HL 4yo 0.343 0.364 0.633 0.076 NH 4yo 0.670 0.006 0.593 0.019 HL 7yo 0.795 0.022 0.728 0.046 NH 7yo 0.218 0.434 0.311 0.258 Correlations between outcome measures (EVT & GFTA) and Hearing status and Age