LINGUIST List 2.755

Tue 05 Nov 1991

Disc: Phonology: R-linking, Invariance

Editor for this issue: <>


  1. David Stampe, 2.754 R-Linking
  2. , r-linking
  3. Ellen Kaisse, invariance
  4. , Re: 2.754 R-Linking

Message 1: 2.754 R-Linking

Date: Tue, 5 Nov 91 05:22:35 -1000
From: David Stampe <stampeuhccux.uhcc.Hawaii.Edu>
Subject: 2.754 R-Linking
		Then cried the American poet where she lay supine:
		`My name is Purrel; I was caast before swine."
						-- Stevie Smith
The analysis seems to need spelling out in more detail: So-called
linking or intrusive r is not due to a process or rule at all, but due
to the reanalysis of final lax V as /Vr/. This happens because the
phonetic process that changes final and codal /Vr/ (where V happens to
be lax) to [V] or [V:] or [V] (details differ by dialect) removes any
contrast between these V(:) and Vr except word finally before a
vowel-initial word, e.g. saw Alf vs ignore Alf.
However, for various reasons, the constrast is anything but clear.
Glottal stop [?] can be inserted before a vowel-initial word, esp. an
accented one, and since such insertions precede deletions, then, as
one respondent noted, r is often deleted especially before accented
vowels: ign[Or] Alf -> ign[Or] ?Alf -> ign[O:] ?Alf, exactly like
s[O:] ?Alf. If the ? is deleted, since deletions are simultaneous,
r-deletion occurs anyway: ign[Or] ?Alf ==> ign[O:] Alf.
If no contrast is observed between the lax vowels (V or V: or V) and
Vr, and if one can be predicted from the other by an exceptionless
natural process like r-deletion, then in any phonological theory,
there is no warrant for positing a lexical difference between them.
Naturally Vr must be taken as basic, and the lax vowels as derivative.
Only /Vr/, never (V or V: or V) occur in lexical representations. So
words like Cuba and saw are analyzed as /kyubr/ and /sOr/. Only from
a diachronic viewpoint does it seem that an r has been inserted, e.g.
in saw it [sOrit]. Synchronically it simply isn't deleted before V.
Respondents who argued that spellings like Eeyore or Marmie do not
prove that the underlying representations have /r/ missed the point:
traditional spellings like Cuba and saw are even less convincing that
that they DON'T have /r/, if they are treated exactly like copper and
soar. And what argument is there, except a complacent acceptance of
standard pronunciation and spelling, for the assumption, against every
objective principle of linguistic analysis, that r-insertion is
synchronic? Why, beside the natural process
		Vr -> V(:) before non-V
should there also be a superfluous and "crazy" rule
		V(:) -> Vr before V
when Vr and V(:) don't even contrast in the dialects in question!
To Alexis Manaster-Ramer's insistence that r-intrusion refutes "the
original, and interesting, claim of [natural phonology, which] was
precisely that automatic processes are exactly those which can be
INDEPENDENTLY shown to be natural (phonetically motivated, found in the
speech of children who ultimately lose them, attested in many different
languages and at different times, etc.)", I can only reply that the
equation of automatic alternations with natural processes is Alexis's
own idea, and that his arguments against it seem entirely convincing.
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue

Message 2: r-linking

Date: Tue, 05 Nov 91 09:24:02 CST
Subject: r-linking
Let me try to take another run at the claim that David (and
Patricia) are making. The claim is that there *is* no rule of
r-linking. Instead there is a process of r-deletion that operates
only in codas. This is a natural process not because it is
automatic but because speakers of r-less dialects find it hard,
if not impossible to make coda r's. When `linking r's' are heard
they are not inserted, they are simply resyllabified to the onset
of the next syllable. Etymologically incorrect linking r's are
also underlying--words like `Cuba' and `tuber' are perfect rhymes
in those dialects with `linking r-s'.
In answer to Alexis let me point out what I believe to be additional
evidence that r-deletion is a natural process:
 r-type approximants are extremely rare in the languages of
the world--Maddieson's collection lists 11 alveolars and 15 retroflexes
out of 317. In English (at least--someone perhaps could check for
Mandarin--the only easily available language with the same sound)
/r/ is the last or second-last acquired by most children, and absence
often seems to be a stereotyped marker of a speech defect.
 Thus at least two additional tests for naturalness--rarity of a
sound in general, and lateness in acquisition are met. It seems to
me that, for those who have it, r-deletion meets the best test for
a natural process--it's hard not to do it unless either 1) you are
a linguist or 2) you make a way-back post-retroflex arrr (the way
my RP-speaking uncle does when he's making fun of Amerrrricans).
 While I do not speak Dutch, I suspect that the coda n-deletion
process in (I believe) unstressed syllables referred to by Jack
Hoeksema may have the same status. Certainly loss of syllable-
final n is widely attested in the languages of the world, and
the situation may simply mirror that of English. Of course, the
rarity and lateness of acquisition arguments don't apply here.
 There may well be automatic *rules* as well, but, for example
a/an is a little suspicious--people are perfectly happy saying
`a sofa and chairs' so it would be interesting to see how hard
people find it to say `a apple'.
 Geoff Nathan
 Southern Illinois University
 at Carbondale
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue

Message 3: invariance

Date: Tue, 5 Nov 91 09:22:56 -0800
From: Ellen Kaisse <>
Subject: invariance
One of the nicer arguments for using the same phonological feature for
segments that differ noticeably among languages is Pat Keating's
discussion of the feature [voice]. It's in Language, in the mid-
eighties, I believe. She argues that phonologically [+voiced]
segments in different languages tend to induce and undergo the
same phonological processes, even if they differ appreciably
in Voicing Onset Time. Thus, the [+voice] segments of a language
might induce lengthening in a preceding vowel, even if these
segments overlap in VOT with theVoice segments of some other
language. The very same VOT in that second language, being
phonologically classified as [-voice], will not induce lengthening.
-ellen kaisse
university of washington (
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue

Message 4: Re: 2.754 R-Linking

Date: Tue, 5 Nov 91 10:01:46 PST
From: <>
Subject: Re: 2.754 R-Linking
Alexis and I do have disagreements, but it sometimes takes a while to discover
just what they are. I criticized him for claiming that Stampe's r-linking
analysis had anything to do with automaticity. Alexis responded:
> OBVIOUSLY, Stampe does not raise this issue, because there is no
> issue. If r-linking were NOT automatic, there would be no difficultu
> fitting it into the framework of Natural Phonology...
Well then, the automaticity criterion is one Alexis imposed on Stampe's logic.
Processes are indeed automatic, but there is no stipulation in the theory that
Rules cannot be. In fact, Stampe has always put caveats on his statements
about how Rules behave. Natural Phonology is a theory about Processes, not
about Rules. The problem with David's position--that he doesn't have to say
much about the nature of Rules--is that people judge his theory on the basis
of how well they understand the dichotomy. And to understand the Rule/Process
dichotomy, you need a well-grounded theory of morphonological Rules. The two
go hand-in-hand, as I believe this little tempest on R-linking demonstrates.
I agree with Alexis that R-linking has been used a counter-example to the
naturalness of Processes:
> My point was and is that Natural Phonology ceases to be a testable
> theory as soon as we start claiming that a process is "natural"
> the moment we discover that it is automatic...
(I would use 'operation' rather than 'process' as a neutral term here.) And
my point to Alexis was that *he* was the one raising this issue, not Stampe.
I agree with him that others have tended to construe Natural Phonology in
this way and that this has caused them to believe that a counterexample existed
in these cases.
> The original, and interesting, claim of this theory was
> precisely that automatic processes are exactly those which can
> be INDEPENDENTLY shown to be natural...
> R-linking (and a number of other examples) have been cited for
> years as counterexamples to this universal. There has been no
> response, beyond the current proposal to emasculate the theory
> by saying that we can set up whatever underlying representations
> we want, so as to get the process in question to come out looking
> natural...
First of all, Stampe's derhoticization analysis constitutes an answer to the
R-linking criticism. Alexis criticizes him for making no response for years
and then blows up at him when he provides the analysis. Stampe showed how
one would handle R-linking in his theory. The confusing thing about his
explanation, in my mind, is what he means by "underlying". Here is my
interpretation: the /r/ either exists in the lexicon or it is placed in the
speech stream by an automatic morphonological operation. As I pointed out in
my last note, it is possible to consider the /r/ "derived" in a morphonological
sense. All Stampe is saying is that R-insertion, whether it exists or not,
is not phonological. Derhoticization is phonological.
					-Rick Wojcik (
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue