Because allophonic variation (excluding free variation due to stylistic considerations) is predictable, it is theoretically redundant to list such variation in the lexicon.1 So, at the systematic phonemic level of representation, allophonic variation is omitted. However, speakers do have to use the correct allophones in order to sound natural, and the phonology therefore has to supply this information somewhere: at the systematic phonetic level. How do we derive the surface pronunciation (the systematic phonetic level of representation) from the underlying systematic phonemic level? We can illustrate this by showing how generative phonology would describe the process whereby vowels become nasalized before nasal consonants in English. Although English (unlike French and Portuguese, among others) does not have nasalized vowel phonemes, a certain amount of nasalization does occur when vowels are in the context of a following nasal consonant. This is because the velum starts lowering to make the nasal consonant during the production of the preceding vowel (the velum cannot lower instantaneously, and so needs a certain amount of time to get into position for the nasal consonant). This results in vowel phonemes having a nasalized allophone in this context only. Generative phonology shows this process through the use of phonological rules (i.e., descriptive formalisms) to link the two levels of representation. The

input to the rule (in front of the arrow) is the underlying level (systematic phonemic); the output (after the arrow) is the realization (systematic phonetic).2 Rules use distinctive features, and the context that a rule applies in is shown after the output following a slash line. Here is the vowel nasalization rule:

4.1) +syll cons

+nasal / +cons +nasal−


 

 

  

 

 

One difference between generative and traditional phonemic approaches to phonological representation is that generative rules extend from simple allophonic variation to the area traditionally termed morphophonemics: the intersection between phonology and morphology. To illustrate this intersection, we can return to the regular plural suffixes in English. There are three phonological forms for this regular suffix: /-s/, /-z/, and /-/. Examples of their use can be seen in (4.4):

4.2) cat+s /kæts/ dog+s // horse+s //

Extrapolating from these examples, we can state that the /-s/ plural is added to stems ending in a voiceless consonant; /-z/ is added to stems ending in a voiced consonant; and /-/ is added to stems ending in sibilant consonants to avoid a cluster of two sibilants, which is perceptually and articulatorily unsatisfactory. Ball et al (2010) discuss how to decide in this instance which of the three forms is the underlying one (of course, even with allophonic variation, it is not always clear which variant should be considered the underlying form; but normally, as in the case of vowel nasalization, one variant is clearly the most commonly occurring one). A metric based on which form requires the least number of operations of rules suggests that an underlying form of /-z/ for the plural suffix is the best motivated. Because generative rules can change feature values (e.g., from voiced to voiceless or vice versa) and can add in or delete segments (e.g., the /-/ vowel in the /-/ suffix), both of the other forms can be derived easily from the underlying /-z/.