The mechanisms of phonological change have been known for some time in generative phonology. Essentially, underlying forms remain unchanged forever, while more and more rules are added to the end of the phonological component to make the surface forms increasingly distant from the underlying forms. More recently, many linguists (e.g. Hooper 1976) have objected to the high degree of abstractness involved here, and have had the nerve to suggest that underlying forms are actually based on surface forms available to the child learning the language. To settle this issue, we really need more empirical evidence. Fromkin (1971) considers speech errors as a good test for things of this sort. As she notes, if things are really so abstract, we may expect to see nonapplications of rules occasionally, revealing more abstract underlying segments; if we find such errors, the standard view of phonological change is confirmed, but if not, it is disconfirmed. Out of a corpus of 1900 errors, I have found ample evidence for abstractness, all in the form of rules failing to apply.
The first rule that fails to apply is Grimm’s Law, revealing aspects of underlying forms in essentially PIE form: pairly ‘fairly’, happen’t ‘haven’t’, tats ‘that’s’, dake ‘take’, dalk ‘talk’, donight ‘tonight’, indegrated ‘integrated’, kavv ‘have’, kair-
The second rule that fails is the palatalization of velars, revealing 1500 year old intermediate forms: boat-
Another rule from the same period that can fail is the deletion of nasals before voiceless fricatives: under ‘other’.
The rule that changes g into w (see Comrie 1978) can also fail, the only example I have involving the word woman, originally a loan from Old Norse gumman ‘the old lady’: [gʊmn̩].
One other rule created in English due to Old Norse can also fail. When the OE word sweoster ‘sister’ was replaced by the ON loan syster, obviously what really happened was the addition of a rule changing weo into y. When this rule failed, the output was sweater (where s was deleted in a further error); the whole sentence in which this error occurred was I knitted my sweater Michelle a scarf.
Many errors involved the nonapplication of Vowel Shift, a 500 year old rule: sames ‘seems’, raybate ‘rebate’, taych ‘teach’, narer ‘nearer’, [khæym] ‘came’, [mæyd] ‘made’, exclode ‘exclude’, Donahoe ‘Donahue’, reroated ‘rerouted’, [stʊŋ] ‘stung’.
Schwa Deletion, a 350 year old rule, failed in one error: [smu:ðəz] ‘smooths’.
My last error actually shows the accretion of rules in morphology. When ME hin, acc. of he, was replaced with dat. him, obviously hin was still present underlyingly, and an allomorphy rule was added to produce him. This rule failed in one error: bribe ‘in ‘bribe ‘im’.
These errors provide almost irrefutable evidence that the generative view of phonological change is correct. They provide yet another demonstration of the great value that the study of speech errors has for linguistic theory.
Comrie, Bernard. 1978. On the go ~ went alternation: A contribution (?) to the generative phonology of English. Lingua Pranca 1.59-63.
Fromkin, Victoria. 1971. The non-
Hooper, Joan. 1976. An Introduction to Natural Generative Phonology. New York: Academic Press.