Minimising the storage

By a theory combining the work of Diana Archangeli in the 1980s, that of Carole Paradis and Jean-Francois Prunet and contributors in the 1990s, summarised by Michael Kenstowicz (1995), further developed in Nunes (2002), multiple phenomena are best explained by assuming that what the brain stores of the sound structure of words is reduced to the barest minimum. Rather than storing every detail that defines the actual pronunciation, the brain stores something like a highly compressed version, consisting of single features or even just a bare ‘slot’ for a consonant or vowel. The storage is ‘underspecified’. 

Archangeli (1984 and 1988) calls her version of this approach ‘Radical underspecification Theory’, RUT. But she soon abandonned RUT in the face of criticism that RUT seemed not to explain some data from some languages, in one case a dying language, investigated by only one scholar. I personally think Archangeli was premature in abandoning RUT. Prunet and Paradis (1991) just address the special status of those consonants articulated with just the tongue tip against the fleshy ridge ,just behind the upper front teeth, in English T, D, N, S and Z. Kenstowicz (1995) calls these approaches ‘extrinsic’ and ‘intrinsic’ underspecification.

Extrinsic and intrinisic underspecification are complemented by an older idea known as ‘Morpheme Structure Conditions’, or MSCs. In the case of English, if there are three consonants before a vowel, the first has to be S, the second has to be K or P, the third has to R or W after K, as in scream or square, or R or L after P, as in spring or splash. S in these cases need be represented only by a one segment slot before two others before the vowel.

On the strength of the patterning of speech errors in English, Nunes (2002) argues for a combination of extrinsic and intrinsic underspecification. The key data is very robustly agreed by speech and language pathologists across the English-speaking world. Children stop, front, and lisp (See Processes). The data is vastly larger than that adduced against RUT. And there may be ways of addresssing the problematic data, in at least one case, by proposing that underspecification works in slightly different ways at different levels. (See Nunes (2002)). On this basis storage is minimised. But it is required that every phoneme has to be derived in order to be pronounced. There is an obvious trade-off in one direction or the other, minimising storage or shortening derivation. By the proposal is here, the issue is not one of abstractly weighing up the empirical or theoretical advantages either way, but considering the logically possible sequencing of evolution. Underspecification is not an advantageous economy or disadvantageous penalty but just a consequence of the first steps towards modern speech, from which there was no going back.

A next move?

The obvious next move might be called ‘Extreme Underspecification Theory’, EUT, combining underspecification theories and MSC theory. But there is no such thing as EUT. Making it work in full and in detail was half a dozen bridges too far for Nunes (2002), as it still is. While the MSCs for any one language may seem clear enough, a human-language-universal theory of MSCs so far defies phonology. But a fully combined EUT would however seem possible in principle by invoking speech pathology data collected in a niform way from a small sample of demographics speaking unrelated alnguages.