Underlying representations

We have seen how the final [t] sounds of Rad and Rat in German can be understood as underlyingly ‘different’ although pronounced identically. In many other cases too, we can only account for the facts of what a native speaker ‘knows’ about his/her language by appealing to an underlying representation of the words he/she utters. While phonologists (and linguists generally) are sometimes accused of over-abstraction and/or abstruse terminology, there are occasions where explanation at an abstract level can yield valuable insights, as the following two examples will demonstrate.

Most varieties of British and southern hemisphere English are non-rhotic, i.e. they only have /r/ before vowels. Rhotic areas like North America, or the west of England, by contrast, do allow /r/ in other positions. The normal pronunciation of the following words in non-rhotic Essex, therefore, differs from that of rhotic Somerset (note that West Country accents use the retroflex /r/ variant ):

In spite of the fact that Essex speakers do not pronounce the /r/ in these words, many phonologists would argue that their underlying representations of each of these words is the same as that of their Somerset counterparts, i.e. that even for an Essex speaker car is underlyingly , but Essex speakers have an /r/ deletion rule which Somerset speakers do not. But does this not simply defy common sense? How can it be helpful to posit an underlying /r/ which simply isn’t there?

A phonologist might respond that, even for non-rhotic speakers, the /r/ actually is there, in some circumstances: even for non-rhotic speakers, it reappears before a vowel:

car /r/ of the year

Arthur /r/ and Jim

beer /r/ and sandwiches

Non-rhotic Essex speakers may not use the /r/ when these words are pronounced in isolation but they ‘know’ which words have an underlying /r/ which appears intervocalically, and which do not: no one would say tree */r/ on the hill for example. Not only does the abstract representation and /r/-deletion rule analysis account for a fact about the speaker’s knowledge of his/her language, it also captures a diachronic fact about English: non-prevocalic /r/ used to be pronounced in all varieties, but a change occurred in some, but not others, which led to its deletion. To account for these data otherwise would require specification of a complex /r/ rule for non-rhotic speakers setting out in detail the environments in which it can and cannot occur: positing an underlying /r/ in the representation looks altogether neater and more elegant. Some phonologists, it should be pointed out, would reject the underlying /r/ analysis in favor of an insertion rule on the grounds that many speakers insert the consonant even where it is etymologically unjustified, a phenomenon known as ‘intrusive r’:

law /r/ and order

India /r/ and Pakistan

Disputes between linguists are more likely to be over the most elegant account of the data available than about the data themselves!