The reason for assuming full specification was that children appear to perceive sound distinctions in an adult-like manner from a very early age, and there
is good reason to believe that this perceptual accuracy extends to words which include /l r j/. This means that the representations which reflect the child’s perceptions are fully specified. However, children’s production of words is
much less accurate than their perception at this age. We can therefore think of the underspecification of the lateral feature in words like lorry or lolly as the way in which the theoretical model reflects this inaccuracy in pronunciation. On the perceptual side, Amahl knows that lorry has an /r/. However, he doesn’t know how to pronounce that /r/ in a word of that shape. We can propose, therefore, that the initial set of representations in figure 30 (those on the left-hand side of the arrows) are representations of the child’s production ability, an indication that the child doesn’t know how to articulate the /j/ in yellow or /r/ in lorry. If this is correct, there are three representations we must consider: (a) what the
child actually says (the SRs in figure 29, the right-hand side of figure 30);
(b) the adult forms (the URs of figure 29 to which the child appears to have
Child phonology
103
access via perception; (c) forms which are relevant to the child for production (the left-hand side of figure 30).
What the above discussion suggests is that it is plausible to maintain that there are two phonological representations stored in the child’s mind, one for perception (b immediately above) and one for production (c). We call these input representations and output representations and there clearly has to be some relationship between them. In general, the output representations are similar to the input representations, but with certain aspects of the representation missing or simplified.
For instance, suppose we maintain that the child’s input representation (based on perception) for lorry corresponds to the adult representation /lɒlɪ/. In order to
‘derive’ an appropriate output representation (what we have on the left-hand side of figure 30), we have to assume two processes. First, the [−lateral] feature representation of /r/ is replaced by [0lateral]. This is called despecification. Then, the
[+lateral] feature is ‘floated’ or delinked from the /l/, so that the /l/ segment itself is also [0lateral]. This is illustrated in figure 31, where for clarity we have separated delinking (indicated by breaking the line between /l/ and [+lat] and ‘floating’.
The output representation in figure 31 can now serve as the UR for the process of lateral harmony illustrated in figure 30, and this UR can be referred to as the child’s output UR, i.e. the underlying representation, which is subject to phonological processes that determine the form of the child’s utterances. Thus, we are Figure 31 Lateral harmony: constructing the output UR
104
sounds
Figure 32 A dual-lexicon model of child phonology
now proposing two types of processes. The type of process represented by figure
31 is called a selection rule, and, taking the adult form as input, it gives rise to an output UR which contains a number of unspecified features and other aspects of representation which need to be filled in. This filling in is achieved by other processes, which we call pronunciation rules. The spreading of the floating
[+lateral] feature to give lateral harmony is an example of such a rule.
In the model that results from this type of analysis, children are credited with two types of phonological representation, one corresponding to their perception of the word and the other determined by the set of distinctive features, syllable templates and so on over which they have mastery in production. Because there are two distinct sets of lexical representations, we will call such a model a dual-lexicon model. The overall structure of the model is illustrated in figure 32.
It is possible to see how the model in figure 32 can account for other processes which we have so far assumed to be accommodated in the simpler model of
figure 29. For instance, we noted above that Amahl Smith’s most complex syllables were of the form CVC; that is, they included at most one onset consonant and at most one coda consonant. Obviously, this can be accommodated to the
model of figure 29 by relying on a rule such as (72) linking URs and SRs, but our dual-lexicon approach now provides us with an alternative way of dealing with this phenomenon, provided we take English syllable structure into account (as discussed in section 5, pp. 79ff.). Specifically, we can propose that Amahl operates with a syllable template over the input representations governed by the Sonority Hierarchy. Children in general find it easier to pronounce sounds and combinations which differ from each other maximally, so they tend to choose the least sonorous elements as onsets and codas and the most sonorous elements as nuclei. At early stages, only one consonant is allowed in the onset or the coda and so this has to be the least sonorant of the cluster. We know from section 5
(p. 81) that /s/ in clusters such as stay or string is exceptional in English, so this will not enter into early child templates even in clusters such as sm- or sl-, in which it is the less sonorant (though children tend to differ in the precise way they treat these clusters).
In the case of codas, there is rather more variability between children; in part, which item from a cluster is pronounced by a child depends on the language being
Child phonology
105
Figure 33 Matching input representation to syllable structure template
learned. However, Amahl treated the voiceless plosive in a coda cluster such as the -mp of stamp as we would predict from the Sonority Hierarchy. How this
works is illustrated in figure 33 for stamp, which at this stage Amahl pronounced as [dap].
This template, operating as a selection rule, produces the syllable [tap], along with an unattached [s] and an unattached [m]. What happens to these items that are not integrated into the child’s syllable via the matching process? The answer is that they are deleted by a general process which phonologists refer to as Stray Deletion or Stray Erasure. In terms of the model in figure 31, this is a pronunciation rule. In general, any material that is left over, because it has not been associated with some part of the template or has not been incorporated into the word by means of some pronunciation rule, is deleted by this process. In the case of lateral harmony discussed above, the floating [+lat] feature in the output representation in figure 31 is saved from extinction by the pronunciation process of lateral spreading. There are no comparable processes which will save the
unassociated /s/ or /m/ of stamp in Amahl’s system. Thus, the only segments
that survive to the level of the output SR for stamp are [tap]. Additionally, along the way the [t] is voiced to [d] by Prevocalic Voicing, another pronunciation rule, giving us the attested form [dap].
This concludes our brief survey of the nature of child phonology. We have,
of course, only scratched the surface of this developing and fascinating field.
However, consideration of what goes on when children acquire sound systems has
106
sounds
enabled us to draw attention to some important notions in theoretical phonology, notions which are regularly applied in the analysis of adult phonological systems.
Of particular importance is underspecification, especially as a way of formalising harmony processes, and another useful notion is that of the floating feature. We have also seen the fundamental importance of syllable structure, in understanding the nature of children’s forms, and the idea of associating segments to a template to filter out combinations which are not allowed in the phonological system is used widely. However, perhaps most important of all is the idea of distinct levels of representation, and especially the idea that there is at least a distinction between an underlying level and a surface level. Although the model of child phonology we have introduced here raises additional complications (because unlike adults, children can’t pronounce most of the words they can recognise), if we look at the output (right-hand) side of the model in figure 32, we see there the two-level system introduced in figure 29. The distinction between underlying and surface levels is one of the key ideas in phonology and indeed in linguistics generally, and even in widely different theoretical approaches, it tends to reappear in some guise or other (exercises 4 and 5).
Читать дальше