Features in phonology and phonetics

OK, first let me apologize for the long absence. This was owing to many factors within a busy summer including poor internet access, but I am planning several posts for this month.

I'm going to discuss phonology and phonetics in the next few posts. I'll begin with the problem of features.

The theory of features has never been worked out to anyone's satisfaction, and there are many competing accounts. I am engaged in a project on "phonetic features" with my colleague Darin Flynn of the University of Calgary, and we are trying to sort things out. First of all, since Jakobson we have had "distinctive features" in phonology. These are generally assumed to be binary or unary-valued, either present or absent in some sense. They operate at the abstract level of phonological or phonemic representation (whatever you take that to be). They are not precisely indicative of phonetic properties, but are in some way derived from the phonetic properties of speech sounds. Their purpose is to include every feature whose presence/absence "can serve to differentiate one utterance from another in any language" (Anderson 1974). This is why they are called distinctive features.

So far, so good. The more problematic topic concerns whatever is "below" the abstract phonology, in the phonetic realm. Some models essentially propose that there are such things as "phonetic features" which specify sounds at some phonetic level. This dates back at least to Chomsky and Halle's SPE model. It is a pretty good idea, but there are problems with their proposal that the phonetic features are merely the "other side of the same coin", wherein phonetic features are scalar-valued objects that are linked with or somehow identical to the distinctive features of the phonology. Historically this subject has become terribly confused, because much literature has called the binary distinctive features of phonology by the name of "phonetic features" and has implied that these provide phonetic specifications.

Anderson (1974) gives some nice reasons why the phonetic features should really be different from the distinctive features. He states that phonetic features should include "every feature in terms of which one language differs systematically from another," and these may include features which are never distinctive, such as whether final stops are released. Anderson proposes that going from phonology to phonetics one uses "detail rules" which specify the scalar values of the phonetic features on a language-specific basis. This is really the phonetic component of the grammar, as against "universal phonetic implementation" which applies to all languages because it is derived from the human speech production mechanism.

Another reason (to allow phonetic features different from distinctive features) that Flynn and I have come up with comes from models of sound change in diachronic linguistics. We noticed that numerous sound changes are well-described as resulting from auditory confusions among sounds with similar values of acoustic phonetic features such as [grave] and [flat], while these features construed as binary-valued are not useful in distinctive feature theory for phonology.

A remaining question is whether the set of phonetic features should be disjoint from the set of distinctive features, or could they be somehow linked when they overlap in a useful way?