Two conditions, with 8 verbs in each condition, were selected. One condition contains verbs which have a legal cluster at the inflected verb-end, and the other contains verbs with an illegal cluster at the verb-end. All verbs have a two-consonant cluster in the inflected form, in order to maintain constant prosodic complexity. It proved impossible to balance the conditions for past tense frequency given the constraints on which verbs could be used - not only did the verbs have to be familiar to children of the ages taking part in the experiment, but they had to enter into stimulus sentences that were syntactically correct and pragmatically plausible. However, if a correlation is found between frequency and performance, then frequency can be partialled out of the analysis. Table 4.1 shows the characteristics of the stimuli. For the full list of verbs, see the Appendix A.1.
Frequency values are the raw frequencies augmented by 1 and In-transformed. Two different measures are used: CO-BUILD, and Francis and Kucera. The CO-BUILD measure is used because this is the one used in van der Lely and Ullman’s (2001) study, and because it is available in an electronic format which makes the calculation of cluster
frequencies possible. The Francis and Kucera measure is used because this is the database I use for the experiments in future chapters.
Which frequency measures one uses depends on how one thinks children produce past tense forms. If one assumes that they create regular past tense forms by rule, then the frequency measure to use is the sum of the frequencies of the base form and all its inflectional variants, i.e. for played this would be play, plays, played and playing. If one assumes that the past tense form is stored, then the relevant frequency measure is just the frequency of the past tense form (see Alegre & Gordon, 1999). Obviously, van der Lely and Ullman’s model is one whereby typically developing children use a rule for regulars (and therefore the full paradigm frequency is appropriate), while the G-SLI children store regulars, and all children store irregulars (and so the past tense frequency is appropriate). Being faced with having to choose one or the other, I decided for this experiment to use the frequency of the past tense form only.
Table 4.1. Characteristics of the stimuli
Condition9 Examples Past tense frequency Cluster frequency
CO-B* F&K** CO-B*
VC-D legal killed, wrapped 1.647 2.089 7.726
VC-D illegal .r ^ ^ ;
touched, robbed 1.175 1.425 4.785
voicing of the preceding consonant. * Frequencies obtained from CO-BUILD, CELEX database; Frequencies obtained from Francis & Kucera
4.2.2. Procedure: Elicitation task
The procedure was based on that used by van der Lely and Ullman (2001). In their task the lead in was of the form:- ‘Everyday I rob a bank. Yesterday, just like everyday, I
a bank. The format used here has been changed in an attempt to raise the level
of correct performance for the G-SLI children, who achieved a score of only 22.2% (for regulars) in van der Lely and Ullman’s task, and many of whom perform poorly on the Verb Agreement and Tense Task (VATT, see Section 2.2). The lead in for the task reported in this chapter presents both the past tense and bare stem form:- ‘Last week Kipper robbed a
post office. Everyday I rob a post office. Yesterday I______ ’. Hence the lead in primes
the syntactic form of the past tense, as does the inclusion of only regular stimuli. By raising the level of performance it was hoped to get a wider spread of scores in the different verb groups, hence avoiding any possible floor effects.
The experimenter introduces the child to two toy dogs. One of the dogs, Kipper, is likely to be familiar to the child through the popular children's books and television programmes. The experimenter reminds/tells the child that Kipper is a very adventurous dog which does all sorts of exciting things. The second dog, which the child won't know, is called Bean Dog. The experimenter tells the child that Bean Dog is one of Kipper's best friends, but that he gets very jealous of Kipper and all the adventures that Kipper has. The experimenter explains that Kipper has been busy doing lots of things recently. Bean Dog is jealous and wants to tell everyone that he has been doing them too. The experimenter asks the child to help her to be the voice of Bean Dog and tell everyone the things that he has been doing. There are 4 practice items using irregular verbs, and 16 experimental items which are listed in Appendix A.2. One pseudo-randomised list was created for all participants.
4.2.4. Participants
14 G-SLI children participated, and 28 typically developing children were selected as controls. 14 of the control children were matched on raw score (to within ±1 point) on a test of sentence comprehension, the Test for Reception of Grammar (TROG; Bishop, 1983) and 14 matched on raw score (±3) on the British Picture Vocabulary Scales (BPVS; Dunn
et al., 1997). These children were then divided into two groups according to age, in order
to give a picture of typical development. The Language Ability 1 (LA1) control group are aged 4;06-7;05, with a mean age of 6;00, and the Language Ability 2 (LA2) control group are aged 7;06-12;00, with a mean age of 9;06. Details of the participants are presented in Table 4.2.
Table 4.2. Participant details
Measure G-SLI N = 14 LA1 controls N = 14 LA2 controls N s 14 Age Mean 12;03 6;00 9;06 Range 9;09 - 16;08 4;06 - 7;05 7;06 - 12;00
TROG Raw, mean 12.86 10.76 16.43
Raw, range 6 - 1 7 6 - 1 6 1 2 -1 9
z-score, mean -1.67 -0.14 0.12
BPVS Raw, mean 79.93 60.00 94.21
Raw, range 4 7 -1 0 4 33-81 69-120
A series of independent samples t-tests revealed no significant difference between the G-SLI group and the LA1 control group on the TROG, t (26) = 1.062, p = 0.298, but a significant difference between these two groups on the BPVS, t (26) = 3.172, p = 0.004. The difference between the G-SLI and LA2 groups was significant for both the TROG and the BPVS, t (26) = -3.364, p = 0.002 and t (26) = -2.243, p = 0.034, respectively. The LA1 control group therefore provides a grammar age match for the G-SLI group. In terms of vocabulary ability, the G-SLI group falls between the LA1 and LA2 groups.
4.2.5. Predictions
Predictions for the elicitation task are different for the G-SU and the control groups. For the G-SLI group I predict that accuracy will be lower for verbs with illegal clusters compared to those with legal clusters. For typically developing children I predict no effect of duster phonotactics, although if there is an effect I expect higher levels of accuracy on verbs with illegal dusters compared to those with legal clusters, i.e. the opposite direction to the G-SLI group. The majority of errors for all groups are predicted to be bare stem errors.
4.2.6. Coding of responses Responses were coded as follows
• Correct correct inflection, e.g. hop -» hopped • Bare stem inflection missing, e.g. ro b-* rob • Other responses e.g. wash -> washing; judge -* jumped
On the rare occasions when a child corrected himself, the first response was accepted for analysis.
4.3. Results
Data from one partidpant in the LA1 group were discarded because her answers were at times muffled and therefore difficult to transcribe accurately. Correct responses to the elicitation task are shown in Table 4.3 and illustrated in Figure 4.1.
Table 4.3. % correct responses
Condition G-SU (N=14) LA1 (N=13) LA2 (N=14)
VC-D legal Mean (SD) VC-D illegal Mean (SD) 78.57 (23.22) 65.18 (38.97) 96.15(7.93) 94.23 (10.96) 99.12 (3.34) 95.54 (10.52)
Figure 4.2. % correct responses according to phonotactics
■ VC-D legal □ VC-D Illegal
G-SLI LA1 LA2
Note that a correlation between correct performance and verb frequency (both measures) revealed no significant or even marginally significant effects of frequency on the performance of any of the three participant groups. Therefore frequency is not used as a covariable in any of the analyses performed on these data.
Correct performance was analysed using a 3 (Group: G-SLI, LA1, LA2) x 2 (Condition: VC-D legal, VC-D illegal) ANOVA. This revealed significant main effects of group, F (2, 38) = 7.745, p = 0.002, and condition, F (1, 38) = 9.079, p = 0.005. The interaction between group and condition was marginally significant, F (2, 38) = 2.957, p = 0.064. T-tests showed that for the G-SLI group, performance on VC-D legal verbs was significantly higher than performance on VC-D illegal verbs, t (13) = 2.446, p = 0.029. For the LA1 and LA2 groups, however, the pairwise comparisons between the two conditions did not reach significance, t (12) = 1.000, p = 0.337, and t (13) = 1.749, p = 0.104 respectively. The pattern of correct performance is therefore VC-D legal > VC-D illegal for the G-SLI group, but VC-D legal = VC-D illegal for the controls.
The majority of errors for the all groups comprise bare stem responses. The proportion of these errors expressed as a percentage of the total number of responses is shown in Table 4.3.
Table 4.3. % Bare stem errors
Stimulus G-SLI (N = 14) LA1 (N = 13) LA2 (N = 14)
VC-D legal 16.96(18.74) 3.85 (7.88) 0.89 (3.34)
4. Discussion