Andrew Radford - Linguistics An Introduction [Second Edition]

processes. The process just illustrated is therefore often called velar harmony.

In adult languages, harmony processes tend to affect vowels rather than consonants, i.e. vowel harmony is more widely attested than is consonant harmony.

Thus, in Finnish, Hungarian, Turkish and a variety of other languages, essentially all the vowels of a word have to be either front vowels (such as [i e œ y]) or back (such as [u o ɔ ɑ]). If an ending is added to a word containing front vowels, then the vowels of the ending will be [−back], but if the same ending is added to a word containing back vowels, then the vowels of the ending will be [+back]. For

instance, the plural ending in Turkish is -ler (with a front vowel) when added to the words ev ‘house’, or ip ‘rope’ (which contain front vowels), so we get evler

100

sounds

‘houses’, ipler ‘ropes’. However, it is -lar (with a back vowel) when added to the words oda ‘room’, or pul ‘stamp’ (which contain back vowels), giving odalar

‘rooms’, pullar ‘stamps’.

Vowel and consonant harmony are themselves examples of assimilation pro-

cesses (see Introduction, p. 5). In such a process, one set of sounds, the target of the assimilation, becomes more similar to another set of sounds, the trigger for the assimilation, by acquiring a specification for some feature or set of features from the trigger. Thus, in the vowel harmony of Turkish, endings acquire the specification [−back] from words with [−back] vowels and the specification [+back] from words with [+back] vowels. In general, the target of an assimilation process only acquires some of its features from the trigger, giving partial assimilation. Thus, the Turkish plural ending alternates only with respect to the feature [back], it doesn’t become *-lir after ip or *-lur after pul, which would be the case if it were also taking on the height and rounding characteristics of the preceding vowel.

There are cases of assimilation in other languages, though, in which the trigger does become identical to the target, in which case we speak of total assimilation.

We can also see instances in Amahl’s speech where more than one process

applies. Thus, in socks, pronounced as [gɔk], the initial /s/ is stopped to [t] and it also harmonises with the following /k/ to give [k]. In addition, it is voiced to [g].

Sometimes, a sequence of processes acting in this way can give rise to sounds or sound sequences that are not found in English. Thus, Amahl’s pronunciation of the word snake was [ŋeɪk], and we have already observed that [ŋ] never occurs

initially in an English word. Work out which two processes of those mentioned above give rise to this form (exercises 1, 2 and 3).

Perception, production and a dual-lexicon model

While the simple model in figure 29 can account for a wide range of data and also acknowledges the discrepancy between child perception and production (URs correspond to what is perceived, whereas SRs correspond to what is produced), there are acquisition phenomena which suggest that it must be elaborated. We shall now consider one such phenomenon in some detail.

A very frequent production problem for children is the pronunciation of the

approximants [w l r j] (because of its familiarity, throughout this discussion, we will use [r] for the English ‘r’ sound, although, as observed in section 2, it would be more accurate to use [ɹ]). Amahl Smith, for instance, couldn’t pronounce [r j] if there was an [l] elsewhere in the word. Thus, yoyo was pronounced [joujou] but yellow and lorry were pronounced [lɛlou] and [lɒlɪ], and there was no distinction between his pronunciations of lorry and lolly – both were pronounced [lɒlɪ]. However, he could distinguish red and led in his production even at a time when he pronounced lorry and lolly identically. How can we account for this set of observations?

From section 2, we know that the sounds involved can all be described as

coronal approximants. We also know that a characteristic distinguishing [l] from

Child phonology

101

[r j] is that it is produced by passing air round the side of the tongue, i.e. it is a lateral sound, a distinction which is captured by the feature [±lat] in the feature system of appendix 2. Thus, [l] is [+lat] while [r j] are [−lat]. What happens in Amahl’s pronunciation is that the non-lateral sounds come to share the same

feature specification for the lateral feature as the neighbouring /l/ sound. Of course, this is just another example of harmony, so we can call Amahl’s process lateral harmony.

A consequence of the existence of lateral harmony is that there can be no

contrast between /l/ and either /r/ or /j/ when there is already an occurrence of /l/ in the word. This means that the feature [±lat] cannot be distinctive in such a word. In section 5, we said that when a feature is never distinctive, as in the case of the feature [aspirated] in English, we give that feature a specification of zero in the UR. This means that we ought to give the feature lateral a zero specification ([0lat]) in words like lorry for Amahl. Indeed, this is a common way of handling such harmony processes in adult grammars. However, we must also acknowledge

that both the /l/ and the /r/ of adult lorry are pronounced by the child as [l]. Hence, while we wish to maintain that these segments are [0lat] in the child’s UR, we must somehow also ensure that they are [+lat] in the SR.

At first sight, it might appear that the obvious way to approach this problem is to treat it like the cases of velar harmony mentioned above. There we suggested that initial coronals harmonise with final velars, and it is easy to see how this could be expressed as a rule along the lines of (73):

(73)

[Coronal] → [Dorsal]/ # __ V [Dorsal]#

(Here, the symbol # indicates a word boundary)

Recall that what appears after the slash is a specification of the context in which the rule applies, so (73) says that the place feature [Coronal] is changed to [Dorsal]

when it occurs initially and precedes an arbitrary vowel (V) and a final sound with the place feature [Dorsal]. In order to be effective, (73) requires the presence of the place feature value [Dorsal] in a word’s UR, and we can immediately see an

important difference between this situation and the case of lateral harmony we are considering. For the latter, we are supposing that both crucial segments are [0lat]

in the relevant representation of lorry, i.e. there is no lateral segment in this representation to trigger the harmony, since [lat] is not distinctive in such words for Amahl at this stage in his development.

A way of dealing with this is to say that the UR of lorry has a ‘floating’ feature

[+lat], which in a sense is a property of the whole word. This [+lat] feature is then anchored to specific segments, namely those which correspond to non-labial

approximants, /r j l/ in the adult words. This is achieved by spreading the [+lateral]

feature to those segments, as illustrated for yellow and lorry in figure 30. Note that, strictly speaking, underspecified segments don’t correspond to a single phoneme, so we’ll represent them using capital letters R, J and L. The dotted boxes here are simply to indicate that while [+lat] is not attached to anything on the left-hand side of the arrows, it is nonetheless an integral part of the representations.

102

sounds

Figure 30 Lateral harmony as feature spreading

Now, an intriguing aspect of this analysis is that it doesn’t fit the simple model of figure 29 in which the child is assumed to have representations that correspond to the adult URs. In these latter, distinctive features including [±lat] are fully specified and there is no place for [0lat] or other underspecified values ([0asp] is, of course, a different case, as [asp] is not distinctive in adult English).