LINGUIST List 2.669

Thu 17 Oct 1991

Disc: Phonology and Sound-Change

Editor for this issue: <>

Directory

Ellen Kaisse, autosegmental representation and sound change

Message 1: The Effects of r

Date: Tue, 15 Oct 91 11:38:14 EDT
From: Michel Jackson <jacksonshs.ohio-state.edu>
Subject: The Effects of r
>I don't think I need to look further than my own English to find similar >things to those Richard G is describing. All my [r] (voiced alveolar >approximants) are velarised and often rounded, and the vowels after them >are also retracted; but after my [l], the vowels have quite different >qualities. so: [ri-:d] but [li:d]. Wish I could use the full IPA! How >can autosegmental phonolgy explain this sort of thing? > >Richard Ogden It's not clear that you need or want autosegmental phonology to 'explain' this phenomenon in American English. It is simply a phonetic _fact_ that /r/ in American English is rounded (for some speakers or some dialects, etc.) It simply follows that adjacent segments are coarticulatorily rounded. Many phonologists (perhaps not all) would regard this phenomenon as a 'low-level phonetic process' since it exhibits the typical characteristics of such a process (demonstrably gradient in time, non-distinctive, etc. - c.f. Kiparksy's claims about post-lexical processes). It is analogous, perhaps, to the minor changes in place of articulation that /l/ undergoes in clusters such as /ln/, /lth/, /ls/, etc. A rather teleological explanation for this phenomenon goes back at least to Jakobson, Fant & Halle. Rounding, like retroflexion, and (certain) pharyngeal constrictions have the general effect of lowering formant frequencies. To somewhat anachronistically paraphrase JF&H, all three are articulatory mechanisms for implementing the acoustic property (or perceptual quality) [+grave]. Or, in a more modern framework (see Stevens, Keyser, & Kawasaki), we might describe rounding in American English /r/ as a 'helping feature'. ---michel jackson (jacksonshs.ohio-state.edu

Message 2: autosegmental representation and sound change

Date: Wed, 16 Oct 91 10:27:41 -0700
From: Ellen Kaisse <kaisseu.washington.edu>
Subject: autosegmental representation and sound change
This is a very partial reply to some queries of Alexis Manaster-Ramer (Hi, Alexis!), Richard Goerwitz, and Richard Ogden about the relation of feature-geometric/autosegmental spreading-type representations of phonological rules to possible historical changes and to assimilations involving non-distinctive features. It's nonetheless kind of *long*, so poise your fingers over those `q' keys, colleagues! I brought up the relative markedness of rules inserting features out of the blue in the new autosegmental spreading representations because Alexis had said all phonological models indeed *encouraged* writing context-free changes. In the more mechanistic versions of an SPE-style feature-counting evaluation metric, it is indeed true that the formalism should encourage such synchronically unnatural rules: X --> [alpha F] takes less symbols than either an assimilation X --> [alpha F] /__ [alpha F] or a non-assimilatory change X --> [alpha F] / __ [beta G]. While in a spreading model X Y | \ | [alpha F] (where that backslash should be a dotted line adding an association between F and X) is a natural adjustment, and inserting F is not. But of course you can allow your theory of representation latitude enough to write all kinds of highly marked phenomena. So feature geometry/spreading won't organically rule out operations that aren't adding and delinking of association lines, and it's not clear that we want it to be that restrictive, though there are current proposals to that effect. But I think Alexis will agree that these developments in phonological representation are in part directed to the questions of natural phonological changes he's worrying about. Richard G, echoed by Richard O, wonders if autosegmental representation can deal with the spreading of non-distinctive features and the partial assimilation of one segment to a neighbor. These questions are somewhat orthogonal to the question of how you represent phonological processes. That is, you tell the feature geometry what features are represented at some stage in the derivation and it will spread them for you. You need another part of your theory to tell you what features are available. The theory of Lexical Phonology, for instance, claims that the later in the phonology he rule is located, the more non-distinctive features it should have access to. So-called 'postlexical' rules (which, as Richard G has anticipated, bear a family resemblance to Stampean `processes', though they aren`t quite the same, and also sort of resemble rules of phonetic implementation), will refer to a wide range of redundant, non-distinctive features, while `lexical` rules (which, again, overlap to some degree with Stampean `rules') refer only to distinctive features. (I don't want to overemphasize the resemblance, but readers of our listserver probably don't want to be subjected to an entire discourse on lexical phonology). As you progress in the derivation from lexical to late-lexical to postlexical rules, you expect non-distinctive features like the rounding of a rhotic to start to participate in the phonology. (Richard - your Tiberian Hebrew case sounds pretty gnarly - I don't think LP will be able to handle it in some easy, summary fashion! I'll send you a forthcoming paper of mine on some historical changes in Cypriot Greek that will illustrate what I'm talking about.) As for spreading of a feature partway into its neighbor, that's easy enough: X Y / \ | [-F] [+F] (again, the backslash should be dotted, indicating spread of +F from Y back onto X). What I just wrote will give you a contour segment X which starts out with one value of a feature and ends up with another. You can find actual uses in recent literature - e.g. Nick Clements' 1987 CLS Parasession paper on intrusive stops. Kiparsky 1985 (_Phonology Yearbook_ 2) also talks about gradient phonological rules of this and other sorts in the LP model. None of this tells us what kind of telescoped rules you can get relating a grammar at one stage to that at another. The literature of the late 60's and early 70's on absolute neutralization does speak somewhat to this issue. As for why rules generalize (often beyond their phonetic motivation and sometimes so that their telescoped, diachronic statement has no environment at all, I leave to someone else or at least to a later message! P.S. Alexis - my favorite version of `what am I -- chopped liver?' comes, I must confess, from a remark of Oliver North's lawyer. He said to the congressional committee, `What am I -- a potted palm?'.