Establishing the morphological and phonological deficits
5.4. Positional markedness effects for onset clusters: An Optimality Theoretic account
The aim of this section is to begin exploring the impact that metrical structure, specifically the presence of an initial unfooted syllable, has on the realisation of non-words. I focus on an extreme case found in one G-SLI child, GD, who generally realises onset clusters correctly when they are word-initial, but never does so when they occur after an initial weak syllable, i.e. in word-medial position. I show that this pattern of onset realisation is also found to a less striking degree in typically developing children, and in some of the other children in the G-SLI group. The typically developing children whose data are used in this analysis are the same 20 who are described in Section 5.2.2.
In the TOPhS stimulus set, 48 non-words contain an onset cluster. In 24 of these the cluster occurs word-initially, e.g. drepa, and in the other 24 it occurs word-medially, e.g.
badrepa. I make a distinction between onset clusters which are produced segmentally
faithfully and those where a cluster is produced in the target position but which is segmentally unfaithful, e.g. dafrip -> dapripf. The data for GD and the two control groups are presented in Table 5.13.
Table 5.13. Mean (SD) % of onset clusters produced word-initially and word-medially
Onset cluster position GD LA1 LA2
Word-initial Segmentally correct 75.00 88.33 (9.78) 95.83 (6.51) Segmentally incorrect 4.17 0.83 (2.64) 0.00 (0.00)
Total 79.17 89.17(10.05) 95.83 (6.51)
Word-medial Segmentally correct 0.00 75.83 (18.40) 88.33 (11.59) Segmentally incorrect 12.50 2.92 (5.22) 0.42(1.32)
Table 5.13 shows very clearly that GD is much more likely to retain complex onsets word-initially (e.g. drepa)than word-medially (badrepa).Complex onsets are rarely realised word-medially, and on those occasions the segmental material is invariably altered.
For the control children, a 2 (Group: LA1, LA2) x 2 (Condition: word-initial, word- medial) ANOVA reveals a significant main effect of condition, F (1,18) = 12.345, p = 0.002, but no significant effect of group or interaction. The main effect of condition results from better performance on word-initial clusters than word-medial clusters. Hence the typically developing children show the same effect of cluster position as GD does, but in a less extreme form.
Now I consider the types of errors made in word-medial position which result in cluster simplification. GD makes three types of errors:-
1) Cluster reduction e.g. dafrimp —»dafimp, dafrimpala —»darempfala.
2) Vowel epenthesis e.g. badrepa —» darepa, fakletala —» kaletala. Note that the overall metrical structure of the non-word is unchanged when epenthesis takes place: there are no examples such as badrepa -» badarepa.
3) Unclassified e.g. sapnfi —> dafifi, badrepari —»dapifari.The first example looks like a case of possible coalescence of Iprlto Iflbut may just be harmony with the Ifl of
the third syllable. The second example looks like deletion of the Irlin the complex onset and the metathesis of Idland Ibl. However, it is not dear that these are the correct explanations, and so these errors will be considered undassified.
The control children’s errors are also dassified using this scheme, with the indusion of no responses in the unclassified category. The means and standard deviations for the three types of errors, expressed as a percentage of total responses, are shown in Table 5.14.
Table 5.14. Mean (SD) % of response types where a word-medial onset duster is simplified
Error response type GD LA1 LA2
Cluster reduction 25.00 12.50 (9.62) 9.17(8.96)
Vowel epenthesis 33.33 2.08 (4.05) 0.83(1.76)
Unclassified 29.17 6.67 (8.15) 1.25(2.01)
For GD, errors are fairly evenly distributed amongst the different error types. For the controls, a 2 (Group: LA1, LA2) x 3 (Error type: cluster reduction, vowel epenthesis, unclassified) revealed a significant effect of error type, F (2,36) = 14.262, p < 0.001, but no
significant effect of group or interaction. Paired t-tests revealed that cluster reduction errors are significantly more common than epenthesis errors and unclassified errors, t (19) = 4.682, p < 0.001 and t (19) = 3.676, p = 0.002. Unclassified errors are not significantly more frequent than vowel epenthesis errors. Therefore GD shows a different pattern of errors to the control children. While GD produces relatively equal numbers of errors, for the LA1 controls, reduction errors are the most common.
Some of the LA controls show additional evidence of pressure to create an output with the complex onset in word-initial position. On occasion, the cluster is realised instead in the onset of the initial weak syllable, e.g. dafrimpala —»drafimpala, fakleta —»flaketa, badrepa —> bradepa. On other occasions, the original cluster is retained and a further cluster is created word-initially, e.g. faklestala —> flaklestala and badrep —»bradrep. There are also occasions when the initial weak syllable is deleted so that the complex onset is now word-initial: faklesta —> klesta. GD, however, does not make these types of errors. These data show that even though word-position markedness effects are not as strong in the LA children as they are for GD, they are still in evidence.
Why are GD and typically developing children more likely to realise a complex onset correctly when it is word-initial? Here I present a positional markedness account of the data within an OT framework. The aspects of the data that the account needs to capture are:-
• Both GD and the LA children generally realise word-initial onset clusters correctly. • GD simplifies word-medial onset clusters by either reduction or vowel epenthesis. • LA children also simplify word-medial onsets, but also realise a large proportion
correctly.
The account I present here makes use of two markedness constraints that refer to complex onsets: a general markedness constraint, ‘ Co m p l e xOn s e t, and a specific
markedness constraint, Li c e n s eCo m p l e xOn s e t. I define these constraints as follows:-
* Co m p l e xOn s e t- onset clusters are not licensed.