minimal pairs (was: PIE e/o Ablaut)

[Robert Whiting]

[Sorry that I have been so long getting back to this, but I have had
to make two trips to the States in the interim and between being
jet-lagged and having to catch up on work, I just haven't had the
time  -- RMW]

On Tue, 7 Nov 2000 "Ross Clark (FOA LING)"
<r.clark at auckland.ac.nz> wrote:

<snip>

>On Sat, 21 Oct 2000 "Ross Clark (FOA LING)"
><r.clark at auckland.ac.nz> wrote:

<snip>

>RW>By "historical grammar" I mean the changes in grammar through
>RW>time, studied systematically.  By grammar I mean those features
>RW>of language that one will usually find in a standard grammar
>RW>book:  phonology, morphology, and syntax, but generally not
>RW>including pragmatics and lexicon.

<snip>

>  speakers     +     rules      -->      Old English
>                                               |
>  speakers     +     rules      -->      Middle English
>                                               |
>  speakers     +     rules      -->      Modern English

>RW>Synchronic (generative) grammar is concerned with the horizontal
>RW>rows.  Historical grammar is concerned only with the final
>RW>column.  Since we don't know what goes on in speaker's minds,
>RW>everything to the left of the arrows is an assumption.  We assume
>RW>that speakers start from a certain point and apply a certain set
>RW>of rules to arrive at the corpus on the right of the arrows.  But
>RW>everything to the right of the arrows is data.  When we move from
>RW>one set of data to another we also formulate rules to account for
>RW>the changes.  The difference is that that the historical rules
>RW>are not based on assumptions about how speakers produce language,
>RW>but on observed differences in the language that speakers have
>RW>produced.  Synchronic grammar is simply cut out of the loop in
>RW>historical grammar.

>I doubt that this description fits anything that we are familiar
>with under the name of historical grammar. A corpus is not a
>language, and I find it hard to imagine, say, historical syntax
>couched in terms of differences between a sentence found in an OE
>corpus and another one found in a ME corpus. (I don't mean
>examples, I mean the statements of historical grammar.) As soon
>as you even begin to talk about, say, cases, constituents or
>constructions, you are already moving left of the arrows.

I don't say we can't go to the left of the arrows.  All I say is
that as soon as we go to the left of the arrows everything is an
assumption -- constrained only by what is to the right of the
arrows (the data) and some vague notion of "universal grammar"
that we have generalized from trips across the arrows in many
different languages.  Unlike what is in the right hand column we
don't know (i.e., don't have any data) about what is on the other
side.

But although a corpus is not a language, it is an important tool
for linguistic work and an indispensable one for historical
linguistic work.  In most cases, there is simply no other record
of languages, particularly dead languages, on which to base
linguistic analysis.  For modern languages, a field linguist can
collect tapes or phonetic transcriptions of actual spoken
languages, but these are corpora too.  Field linguists can use
these to extract a lexicon and a grammar, and if native
informants are readily available can (relatively) easily develop
a tool that can be used by other researchers.  But if there are
no records of earlier stages of the language, then the only
possibility for historical work is through internal
reconstruction.

Without such corpora historical linguistics would get about as
far as paleontology would without fossils.

As for whether this looks like anything that you are familiar
with under the name of historical grammar, you might want to go
back and look at de Saussure again.  What I have presented is
simply a modification of his basic principles of historic
(diachronic) grammar as a constantly changing medium through time
and synchronic grammar as a cross section through this medium at
some particular point -- a "snapshot", as it were, of this
changing medium at some particular point in time.  Thus:

                      H G                      TIME
                       |                         |
                       |                         |
                       |                        \ /
                       |                         v
                       |
                       |
          S G -------------------- S G
                       |
                       |
                       |
                       |
                       |
                       |
                      H G


I am sure you will find this diagram or one similar to it in any
basic text on linguistics.  All I have done is present 3 cross
sections instead of one and broken down synchronic grammar into
its components.  What is to the left of the arrows corresponds to
what de Saussure termed "langue" (the language system) and what
is to the right of the arrows is what he called "parole" (the
speech act).  These two components correspond generally to what
Chomsky called "competence" and "performance."  The current
theory is that native speakers have internalized a set of rules
through something called a "language acquisition device" (de
Saussure's "langage") that allows them to produce their language.

Generative grammar is an attempt to determine these rules through
the linguistic analysis of patterns of grammatically correct
usage.  It is necessary to do this through linguistic analysis
because native speakers have not memorized these rules the way
they have memorized the multiplication tables or the formula for
the volume of a sphere, and so they cannot just reel off a list of
the rules.  They have acquired them at a time before they are
capable of serious memorization.  Native speakers have not
memorized the grammar of their language, they have just learned
it -- it is hardwired into their brains by the LAD.  Native
speakers are often taught prescriptive grammar in school, but
comprehensive education is a fairly recent thing.  Before
everyone was required to go to school, everyone still grew up
learning to speak the language.  Even today, those who for one
reason or another do not go to school will still learn to speak
their native language.  They may not learn to read or write, but
they will still be able to speak the language.

I suspect that the thing that you disagree with is my statement
that historical grammar is concerned only with the final column,
that is, with the data, and you are right to disagree, because
a synchronic grammar of Old English has to go to the left of the
arrows to produce an explanation of what is on the right of the
arrows.  My point was not that what is on the right of the arrows
exists in splendid isolation, but that without it, it is very
difficult to venture to the left of the arrows.  If, for example,
case endings have completely disappeared from a language, it is
very difficult to reconstruct them without examples from an
earlier stage of the language.  So I will agree that the data is
not the sole concern of historical grammar, but I will maintain
that you won't get very far in historical grammar without it, and
even with it, you can't be certain of what is on the other side
of the arrows.

<snip>

>RW>Realistic means based on realia -- things that actually exist
>RW>as opposed to mental constructs.

>This is not sufficient to make any discrimination among
>linguistic descriptions. Any linguist of any theoretical
>persuasion will claim that their description is "based on"
>"things that actually exist".

Any theorist will tell you that his theories are based on things
that actually exist.  The problem is that for any given set of
data there are virtually an infinite number of hypotheses that
will account for it.  Some will just be more plausible than
others.  However, it is usually not too difficult to see how far
beyond the data any particular theory has gone, and, more
important, whether it is falsifiable or not.

>However, I guess it might distinguish you from a radical
>Hocuspocusian.

No comment.

>RW>When you start talking about "mental reality" you are getting
>RW>into a murky area of philosophy. Is there such a thing as "mental
>RW>reality"?  This is something that can be discussed endlessly and
>RW>inevitably inconclusively.

>Well, that's what philosophers are paid to do.

Philosophers get paid to do that?  Maybe I should become a
professional philosopher. :)

>As linguists, I think we should simply admit that insofar as our
>synchronic descriptions are any more than mere transcription,
>they are attempts to capture some aspect of mental reality.

I don't say that we shouldn't; in fact, I think we pretty much
have to.  What I say is that such a thing is presently beyond our
reach.  Until we can get inside the human mind, there is no way
to verify whether what we have captured is "mental reality" or
not.  It may be a useful model to explain certain things (like
Niels Bohr's planetary model of the atom) but it also may have
nothing to do with reality (like Niels Bohr's planetary model of
the atom).  I don't say that we shouldn't try to get there, only
that there is no direct road there presently available.  So
whether all our rules of synchronic grammar actually reflect
something that actually goes on in the mind (which is what I
assume you mean by "mental reality") is something that remains to
be seen.

[snip]

>RC>> The rest of your post is entirely dependent on the further
>RC>> assumption that native speakers of modern English (in general,
>RC>> not just linguists) distinguish "foreign" from "native" words,
>RC>> and that the words I listed with /th/ in voiced environments are
>RC>> marked as "foreign". Since I don't share this assumption, I would
>RC>> like to know what evidence leads you to it. Do you have any such
>RC>> evidence, other than the fact that by excluding these hundreds of
>RC>> words you can arrive at a nice phonological generalization?

>RW> Let me answer this from back to front.

>RW> You ask what evidence I have that native speakers
>RW> distinguish by rule the [th] of loan words from the [dh] of
>RW> native words other than hundreds of examples and the fact that it
>RW> produces a nice phonological generalization.

RC>This is not quite what I said. I said that in order to arrive at
RC>your nice generalization you need to *exclude* hundreds of
RC>English words, and I asked what independent evidence you had that
RC>these words are somehow marginal to the structure of English?

RW>I don't remember saying that these words were marginal to the
RW>structure of English and I don't remember your saying it
RW>previously.  I didn't say these words were marginal to the
RW>structure of English.  I said that these words follow a different
RW>rule from native English words and that foreign words don't
RW>necessarily follow the same rules as native words.  So these
RW>words aren't marginal to English -- there are too many of them --
RW>they just follow different rules.

RW>But I can see that I shouldn't expect connections to be made when
RW>these connections have not been expressly stated.  So let's start
RW>again.  Grimm's Law says that certain PIE consonants shift in a
RW>certain way in Germanic.  But there are a large number of
RW>exceptions.  One exception is that unvoiced stops don't shift
RW>when the stop is the second part of a voiceless cluster (thus
RW>/sp/, /st/, and /sk/ remain unshifted).  However, even with this,
RW>there are still a large number of exceptions to Grimm's Law.
RW>Verner's Law says that when the PIE stress fell after the
RW>consonant then the consonant shifted in a different way.  So
RW>Verner's Law in effect *excludes* hundreds of Germanic words from
RW>the effects of Grimm's Law thereby producing a nice phonological
RW>generalization.

>This is the most peculiar interpretation of Verner's Law I have
>ever seen. The fact is that Verner made it possible to *include*
>many words which had previously simply been sidelined as
>"exceptions".

To include in what?  Certainly not Grimm's Law.  Verner's Law
justifies certain exceptions to Grimm's Law.  The consensus is
that Verner's Law operated after Grimm's Law (thus Grimm's Law
feeds Verner's Law by providing voiceless fricatives for it to
operate on).  The words "in effect" above indicate that what I
said is not what actually happened, but rather that this is the
result that it produced.

So once again:  Grimm's Law says (inter alia) that PIE voiceless
stops become Germanic voiceless fricatives:

            PIE           Germ.

            *p             f
            *t             T
            *k             x

But considering the dental, PIE *t results in Germanic T, d, and
t.  Now T is the normal outcome of Grimm's Law; t is the result
of Grimm's Law not acting on the second member of a voiceless
cluster (Latin 'sto:', Goth. 'standan' "stand"; this sound is
"in actuality" excluded from the effect of Grimm's Law rather
than "in effect" excluded).  Until Verner discovered his law, d
was just an unexplained exception to Grimm's Law.  Verner's Law
is:

   [+fric., -voice] > [+voice, (±stop)] / [+voice, -accent] ___
   {+voice, #}
   (the notation (±stop) is included because this rule apparently
   also voiced PIE *s in this environment; for those who may not
   be able to read the 8-bit character, this is a "plus or minus"
   sign)

This rule explains the shift of PIE *t to Germanic d (Latin
'pater', Goth. 'fadar').  Although the actual rule has shifted
T to d (where T was the result of Grimm's Law), the effect is to
exclude this word from having to have the expected outcome
(*faTar) if only Grimm's Law had operated.  This is why I said
that *in effect* Verner's Law excludes this word from the effects
of Grimm's Law.  It doesn't actually exclude it from the
operation of Grimm's Law (otherwise the fricative wouldn't have
been there for Verner's Law to convert).

>RW>What independent evidence is there that Verner's Law
>RW>correctly excludes these words?  Only the fact that when it can
>RW>be checked, the PIE stress does fall after the consonant in those
>RW>cases where the consonant does not shift according to Grimm's
>RW>Law.  Of course this might just be coincidence, and some people
>RW>might not accept Verner's Law and ask what independent evidence
>RW>there is for it other than that when you exclude these words
>RW>using this rule you get the nice phonological generalization
>RW>known as Grimm's Law.

>If someone asked this I would have to conclude that they had not
>understood either.

Seems to be a lot of that going around. :)

>RW> This seems rather like asking what evidence I have for
>RW> Grimm's Law or Verner's Law other than hundreds of examples and
>RW> the fact that they provide a nice phonological generalization.

>RC>No, because you are not presenting this as a historical law. As
>RC>a historical explanation of why /th/ occurs in certain places and
>RC>/dh/ in others, it is not in dispute.

>RW>Historical law - historical explanation -- What's the difference
>RW>in your mind?

>You misunderstood me. I was not proposing an antithesis.

I'm sorry.  It rather sounded like you were.  But if there is no
antithesis, why is it a valid historical explanation but not a
valid historical law?

RW>What happened happened.  If the explanation can be
RW>stated as a rule and is not in dispute, what is the problem?

>What I said was that the *historical* explanation of the
>distribution of [th] and [dh] in native English words was not in
>dispute. What we are arguing about is a *synchronic* account of
>[th] and [dh] in Modern English. This is why the comparison with
>Grimm/Verner was not appropriate.

The point is that sound laws (changes) are historical events that
occur at a certain time in a certain language under certain
conditions.  Once the change goes to completion, that's it.  If a
word got missed by the change, it stays.  If new words that would
have undergone the change if they had been in the language when
it took place come into the language after the change, they
stay.  These words don't have to be loans; they could be new
coinings.  So if you know of any new coinings in English that
have intervocalic [T] (and aren't neologisms based on Greek or
Latin), I'd be happy to know.  Otherwise, if you find an English
word with intervocalic [T] there is about a 98% chance that it
is a loanword borrowed after the change.  And all loanwords that
came into the language after the change with intervocalic [T]
still have it.

So the further point is that sound laws (changes) that took place
long, long ago are still present (and significant) in the
language until some other change eliminates their effects.
Therefore, you can't disqualify the sound patterns of the modern
language from being significant simply because they are based on
*historical* events that happened in the distant past.  The sound
law (change) that voiced intervocalic [T] in Old English is
qualitatively no different from Grimm's Law that operated in
Pre-Germanic to produce the [T] in the first place.

So the *synchronic* pattern is the same as the historical
explanation.  Native English words have intervocalic [D] (with a
few exceptions) and borrowed English words have intervocalic [T]
(if that's what they were borrowed with).

It seems to be very difficult for people to grasp the concept
that synchronic grammar and historical grammar have to have the
same result because they both describe the same entity (the
modern language as it is produced by its speakers).  The modern
language is not produced by historical events, it is produced by
its modern speakers.  A knowledge of history is only useful to
help explain *why* modern speakers produce the language that they
do rather than some other or some previous version.  The modern
speakers tell us *what* they produce.  What modern speakers
produce and what historical explanations tell us to expect them
to produce must be the same thing, or else our historical
reconstruction must be wrong (because the native speakers of the
language can't be).  But neither the native speakers nor
historical explanations can tell us *how* they produce their
language.

[snip]

>RW> But ultimately, it is unimportant whether speakers can still
>RW> recognize foreign words after several centuries or not. The
>RW> pronunciation rules are marked in the lexicon.  The native
>RW> speaker learns these rules and follows them.  It is these rules
>RW> that produce the pattern, not the speaker's perception of the
>RW> words as native or non-native.

>RW> For other evidence that native speakers distinguish by rule
>RW> the [th] of loan words from the [dh] of native words I offer the
>RW> pattern created by the presence of intervocalic [th] and [dh] in
>RW> English words.

RC>This is not "other evidence". This *is* your evidence.

RW>Then my position is unassailable.

>I think you should be worried about this. "Self-validating"
>might be another way of putting it.

Are you now denying the existence of the pattern?  Or are you
still just saying that this pattern can't have any significance?
Because if the pattern is there, then this is not self-validation;
it is validation through evidence.

>RW>As with Verner's Law, the fact that the rule by which the
>RW>exclusions are made is valid and leads to a valid generalization
>RW>is sufficient.

>Not in the slightest like Verner's Law, as I've pointed out.
>Consider: For any pair of English phonemes (say /p/ and /b/) it
>is possible to extract a subset of the words in which they occur
>where the two are in "complementary distribution". (Say [p]
>initially and [b] medially, so we include pig, nibble and rubber,
>but exclude big, nipple and supper.) We then state that there is
>only one phoneme, but that the excluded words "follow a different
>rule". What principle would exclude such a ridiculous analysis?

Surely the principle that the "different rule" would have to be
in direct conflict with (the opposite of) the first rule would
reveal that this analysis is self-contradictory.  But why make
your ridiculous analysis so complicated?  Why not just extract a
subset of English words that begin with /b/ and say that all
English words begin with /b/ and that all the excluded English
words "follow different rules"?  The fact that the different
rules must be in conflict with the original rule makes such an
analysis impossible because P and ~P ("a proposition and its
opposite" or "P and not-P") cannot both be true.  The principle
is known as contradiction or mutual exclusivity.

But are you trying to say that you consider this to be in any way
parallel to saying that native English words have intervocalic
[D] and borrowed words have intervocalic [T]?  If so, then you
are clearly not even close to grasping what I have been talking
about.  It is one thing to present analyses that violate the
cherished rules of phonology (such as "no non-phonetic conditions
can affect phonology" or "rules that have to do with how words
are pronounced aren't necessarily phonological"), but it is quite
another to make up analyses that violate the rules of logic.

[snip]

RC>As I understand your position, you are claiming that there is a
RC>single phoneme, say /th/, with a realization rule that says
RC>(among other things) that /th/ is realized as [dh]
RC>intervocalically, except in certain exceptional words.

RW>Close.  I claim that there was once a single phoneme /th/ in
RW>English.  Then that this phoneme became voiced to [dh]
RW>intervocalically without creating any new contrasts (/th/ now
RW>has allophones [th, dh]).  Then that any words that came into the
RW>language with intervocalic [th] after this sound change operated
RW>retain [th] intervocalically (this is not necessarily restricted
RW>to loanwords; it could also include new coinings -- there just
RW>don't seem to be any with intervocalic [th]).  There is nothing
RW>particularly exceptional about words with intervocalic [th]
RW>except that the vast majority of them came into the language
RW>after the sound change.  There are a few native words with
RW>intervocalic [th] (for reasons that can be accounted for), but
RW>there are no loanwords that originally had intervocalic [th] that
RW>now have [dh].  This is exceptional, at least to the extent that
RW>there are no exceptions.

[snip]

RC>Meanwhile, how do we know that children and the less educated
RC>actually formulate the rule this way? If they did, one would
RC>expect that errors consisting of pronouncing [dh]
RC>intervocalically in words which should have [th] would be common.
RC>I don't recall any such tendency from my experience as a native
RC>speaker of English, but surely it's you that should be looking
RC>for such evidence.

RW>My point is that we don't know how any speaker actually
RW>formulates the rule.  What goes on on the left side of the arrows
RW>is a mystery.  All we know is that there is a pattern, the
RW>historical reason for the pattern, and that, synchronically,
RW>patterns must be accounted for by rules.  My further point is
RW>that it doesn't matter that we can't formulate the synchronic
RW>rule.  There must be one, or the pattern wouldn't be there.

>Since you are sure there is a rule, and are even willing to state
>it, I don't see why you claim it is such a mystery.

This is just your same philosophical question in a different
guise.  As I said earlier this is something that can be discussed
indefinitely (in both senses of the word), and I don't have time
to go into it here, nor do I consider it appropriate to this list
(it belongs on a list devoted to the philosophy of science).  A
good (philosophical) discussion on hypothesis confirmation will
be found in S. F. Barker, _Induction and Hypothesis_,
Contemporary Philosophy (Ithaca, 1957).  An extensive discussion
of Linguistic applications (including a bibliography of more
recent general studies) will be found in D. Wunderlich,
_Foundations of Linguistics_, Cambridge Studies in Linguistics
(Cambridge, 1979) [translation of the German version, _Grundlage
der Linguistik_, published in 1974].

But in brief, theories do not eliminate mysteries, they provide a
plausible explanation for them.  A theory is just an explanation
put forward to account for some observation or set of
observations (the evidence).  A theory can account for all the
observations (evidence) in a plausible manner and still be wrong.
Until direct observation and objective verification of cause and
effect are possible, a theory remains a theory.  Now there are
those, apparently including you, who believe that once we have a
theory to account for something, there is no longer any mystery.
But a theory is just a plausible story.  It is not evidence.
Plausible stories are what we use to bridge the gaps where there
is no evidence.  Innocent people have been executed on the basis
of plausible stories well told.  But once the plausible story
breaks down, the mystery reappears, implying that it never really
went away.  So plausible stories (theories) don't really
eliminate mysteries, they just give them something to hide behind
so that people who don't look beyond the explanations can't see
them.

<snip>

>We may be working from fundamentally different conceptions of how
>languages are organized here. I repeat your schema from the
>discussion of historical grammar:

> speakers     +     rules      -->      Old English

>>From this it would appear that, for you, "rules" includes all
>the information commonly thought of as being part of the lexicon.

No, rules are generally marked in the lexicon only if they are
not the productive rule.  That is what a markedness rule means:
the word has to be marked for the rule in the lexicon.  Thus
nouns that have a regular plural in (some form of) /s/ will not
have the plural marked in the lexicon and verbs that have regular
principal parts in (some form of) /d/ will generally not have the
principal parts marked in the lexicon (unless there is some other
peculiarity like 'sleep' / 'slept' that makes the form
unpredictable from the productive rule).  The lexicon will (or
should) tell you things like whether nouns are plurale tantum or
collectives or plural in form but singular in construct and
whether verbs are transitive or intransitive or are state verbs
or action verbs or if the verb requires an animate subject or if
it takes a dative object rather than an accusative.  For
languages that have such things, the lexicon will tell you which
declension a noun belongs to (and its gender if this is not
correlated with declensions) or which conjugation a verb follows.

But to find out what such things mean and how they are applied in
a particular language you will have to look at a grammar of the
language.  Rules (both productive and markedness rules, as well as
paradigms) are found in grammar books, generally not in lexicons.
The lexicon usually just marks which rules particular words
follow.  That is why when you attack a language that you are not
familiar with you need both a grammar and a lexicon (dictionary).
One without the other won't do you much good because language is
an integrated system.  The individual subsystems don't give you
enough information to use or comprehend the system adequately.

"Speakers + rules" is just my way of characterizing what de
Saussure called "langue" and Chomsky called "competence."  The
rules are the rules of generative grammar which is the current
theory of how speakers produce language ("parole" or
"performance").  Needless to say, there are a number of competing
theories on how this is actually accomplished.

>Just above, you stated that "The pronunciation rules are marked
>in the lexicon", which makes me wonder whether you share an
>assumption that I had been taking as given, namely that there are
>two different types of information/knowledge relevant here:

>   (i) information about the phonological shape of particular
>lexical items -- for example, the fact that the English word for
>"fish" has an /f/ in initial position.

>   (ii) general information about the realization of particular
>segments and configurations (such as allophonic and
>morphophonemic rules).

>In my experience, most people would use the term "rules" of the
>latter type of information, but not the former.

Yes, rules are generalized statements of repeated (and
repeatable) observations.  What you give in (i) above is a single
instantiation of a more general rule (graphically, initial <f> in
English represents /f/ or, as a phonotactical rule, /f/ is a
permissible initial segment in English).  Now this single
instantiation would not be sufficient to formulate the general
rule (there is even an aphorism about this: "one swallow does not
a summer make"; or, as my professor always said: "testis unus,
testis nullus").

Basically, information about the phonological shape of lexemes
needs to be given in the lexicon only if it is not transparent
from the orthography.  Finnish, for example, usually does not
mark pronunciation separately in the lexicon; there is no need.
If you see a Finnish word spelled you know how to pronounce it.
If you hear a Finnish word pronounced you know how to spell it.
Finnish children learn to write from dictation at about the age
of 7.  By contrast, English speaking children never do.  English
orthography is far from transparent.  For this reason, there must
be rules to relate English orthography to English pronunciation.
In addition it is important to mark the position of the stress
(and sometimes the position of syllable boundaries) in English
because this information is not predictable and is sometimes
lexically significant (per'-fect [adj.] vs. per-fect' [v.],
reb'-el [n., adj.] vs. re-bel' [v.], etc.).  Finnish on the other
hand does not need to mark either stress or syllable boundaries
because both are fixed and never lexically significant.

So the rules of pronunciation that I spoke of are the rules for
relating the orthography of English words to their pronunciation.
English has words that are spelled differently but pronounced the
same as well as words that are spelled the same but pronounced
differently.  Now many parts of English orthography are
transparent.  Transparent (and even semi-transparent)
relationships could be given as general rules in a separate
section of the dictionary.  In your example (i), a general
statement that "initial <f> (and <ph>) represents /f/" would make
a pronunciation rule for <fish> unnecessary.  Similarly, anyone
who can generalize the transparent rules of English orthography
would know that <phlox> and <flocks> were pronounced the same.

But some English orthographies are completely opaque.  One such
is the graphic sequence <ough>.  If one sees <cough>, one has no
idea whether is should be pronounced as if it were <coo>, <cow>,
<cuff>, or <coe>.  And this does not even include the correct
pronunciation.  Therefore every word with <ough> requires a
pronunciation rule.  Essentially such a rule simply says "this is
how you pronounce this graphic sequence in this particular word."
Now this is only true for people who can read.  This rule relates
the graphic sequence to a pronunciation.  Those who can't read
(or whose native language is not written), have no need for such
rules.  For them each word is simply pronounced the way it is
pronounced and there is no way for graphic sequences to interfere
with this.  They must have each word stored in their memory or
they must create it by analogy when they need it because there is
nowhere else for them to store it.  This creates less of a problem
because the only source of unfamiliar words for these people is
from hearing them, and presumably when they hear them they learn
them with the correct (received) pronunciation.

Another opaque graphic sequence is <th>.  Nothing about the
graphic environment of this symbol tells you whether is to be
read as [T], [D], or even [t].  Although final <the> is usually
realized as [D], it is not invariably so (e.g. some
pronunciations of 'blithe', 'lithe') and final <th> is not always
realized as [T] but often also as [D] ('smooth', 'mouth' [v.],
Brit. 'booth', some pronunciations of 'with').  Therefore, <th>,
like <ough>, should need a pronunciation rule for every occurrence
of this symbol.

Now there are those who say that they have no rules for the
pronunciation of <th>.  Presumably, when they encounter in print
an unfamiliar word with <th> in it they have to go to the
dictionary to find out how to pronounce it.  I happen to have
generalized rules for the pronunciation of <th> so I can usually
make a guess that will be right about 98 times out of 100 about
how <th> should be pronounced in an unfamiliar word.  I came
across one the other day:  'poikilothermic'.  Without going to
the dictionary I knew two things just by looking at it:  a) it is
not a native English word, and therefore, b) the <th> is
pronounced [T].

[snip]

>RW> The best evidence that the pattern is created by rule by the
>RW> speakers of the language is the fact that the pattern exists.
>RW> For if there is a pattern in a language, synchronically it must
>RW> be created by its speakers because that is the only place that
>RW> language comes from.  If the pattern is not created by rule, then
>RW> it is simple coincidence.

>RC>No, this is where you go wrong.

>RW>I presume that "this is where you go wrong" is your way of saying
>RW>that your opinion is different from mine. :)

>Isn't that what "wrong" usually means? :)

In linguistics, almost invariably. :)

RC>The pattern has been created by historical changes in the
RC>language. That is what creates the distribution of [dh] and [th]
RC>that is the input to each speaker's language-learning task. There
RC>is no general principle that speakers must recognize (consciously
RC>or unconsciously) any such pattern, or make use of it in their
RC>rule-governed language behaviour.

RW>No, the modern language is not created by the historical events
RW>that brought it about.

>There is obviously a distinction in your mind between "create"
>and "bring about" which I am not picking up. Otherwise you have
>contradicted yourself. Or is it the tense distinction between "is
>created" (by speakers,now) and "brought it about" (in the past).
>In either case, I am happy to re-state my claim above as:

>The historical changes *brought about* the present distribution
>of [dh] and [th].

My point about "create" is that history doesn't create things.
History is just a record of what happened -- a record of people
and events.  People and events create things (sometimes people
create events as well), but the record of people and events
doesn't create things.  The modern language (in the form of
modern speech and modern writing) is "created" by its modern
speakers.  And at any point in time the "modern" language is the
one currently in use.  So this "historical" change was "created"
by the speakers of the language at that time, not by history.
The fact that it is still present in the language of modern
speakers just shows that it has been successfully passed on from
generation to generation.  If it hadn't, it wouldn't still be in
the language.

The fact that these are historical changes just means that they
happened in the past.  But that doesn't necessarily mean anything
to the modern speakers of the language because for the most part
they don't know anything about the history of the language they
speak.  And this doesn't make any difference because the pattern
is present in the language *synchronically*.  You claim that
there is no dispute about this and have even suggested that my
demonstration of the existence of this pattern in the modern
language was unnecessary.  But now you seem to be claiming that
the fact that these are "historical" changes means that this
pattern has no significance for the modern language.  It may not
be significant in the "why" sense, but it must be in the "what"
sense.  But this doesn't matter because, by and large, speakers
don't know the "why" of their language, only the "what."

RW>The modern language is created by what
RW>goes on on the left side of the arrows in the diagram above.
RW>This is the assumption of generative grammar.  And it is a basic
RW>premise of generative grammar that patterns must be accounted for
RW>by rules.  If you have evidence, rather than just an intuition,
RW>that linguistic patterns are not created by rules, then please
RW>share it.  It will revolutionize synchronic grammar.

>No, I don't think it will. My view is one you have probably heard
>before: That there are two distinct phonemes /th/ and /dh/, which
>occur in various words in various positions. A subset of the
>English vocabulary shows a pattern of complementary distribution
>between these phonemes. The pattern is there in the vocabularies
>of English speakers because it is there in the corpus of
>presently spoken English on which they have based their
>internalized knowledge of English. And it is in the corpus as a
>consequence of changes which took place long ago in the history
>of the language.

Well, I can't say that I've heard the idea that changes that took
place long ago in the history of a language have no significance
for the modern language before.  If the results of changes are
still part of the language, it doesn't matter when they happened.
Speakers of the modern language don't know and don't care when
they happened.  They just learn them as part of the language.

>Your analysis (again, correct me if I'm wrong) postulates a
>single phoneme /th/ with two different "pronunciation rules".
>Some words follow a rule that says intervocalic /th/ is
>pronounced [th], and others a rule that says it is pronounced
>[dh]. Which words follow which rule presumably has to be marked
>on individual lexical items.

Well, the first thing to correct is that "my analysis" does not
say that there is only a single phoneme /T/.  I have no basis
for saying this.  My analysis is only about the validity of the
evidence that is used to "prove" that there are two phonemes /T/
and /D/.  The fact that I consider this evidence inconclusive is
not proof that there is only one phoneme, nor does "my analysis"
require that there be only one phoneme.

My analysis is based on a number of observations.  These are
based on data, and, I think, are not questioned.  The first is
that there was originally a single phoneme /T/ in English.  This
phoneme is the normal reflex of PIE *t in Germanic.  At some
point there was a change that voiced intervocalic [T] to [D] (and
voiced intervocalic /s/ and /f/ as well).  This change did not
initially create any new contrasts but only a voiced allophone of
/T/ in intervocalic position.  This event obviously took place
before the introduction of the Latin alphabet into England in the
late 6th - early 7th century.  But probably not long before,
because the voiced fricatives were clearly considered allophones
of the voiceless ones through Old English.

The next set of observations is about the distribution of these
two sounds ([T] and [D]) in modern English.  The original
allophonic status of [D] as an intervocalic variant of [T] has
obviously changed.  Both [T] and [D] now appear in initial,
intervocalic, and final position.  However, the distribution of
these sounds in these positions is generally not arbitrary but is
determined by word classes.

In initial position, [D] is found in function words
(specifically, pronouns and deictics) while [T] is found in all
other words (whether native or borrowed). There are two minimal
contrasts that arise from this distribution:  'thigh' [Tai] and
'thy' [Dai], and 'thistle' [Tis at l] and 'this'll' [Dis at l]
(contraction of 'this' and 'will').

In intervocalic position, [D] appears in native English words
that were in the language and considered native when the voicing
rule operated.  There are a few instances of intervocalic [T] in
native words that did not voice for various reasons.  There are
no minimal contrasts between [T] and [D] in these two groups.
A much larger group of words with intervocalic [T] is made up of
loanwords that have come into the language with intervocalic [T]
after the voicing rule operated.  Although there are hundreds (if
not thousands) of such words, there is exactly one minimal
contrast between [T] and [D] between these two groups:  'ether'
[i:T at r] and 'either' [i:Der] (the latter also realized as
[aiD at r] in some dialects or by some speakers and hence not a
minimal contrast).

In final position, both [T] and [D] can occur.  There are,
however, two major groups where [T] and [D] are distributed by
meaning class:  a) nouns in final [T] often have a corresponding
verb in final [D],  b) nouns in final [T] often have a plural in
[Dz].  These groups are relics of the earlier voicing rule that
voiced intervocalic [T] but have since lost the conditioning
environment that triggered the initial change.  The conditioning
environment that preserves these distinctions is now grammatical
rather than phonological.  There are several minimal contrasts
between [T] and [D] in these groups:  'teeth' [ti:T] (plural of
'tooth') and 'teethe' [ti:D] (verb derived from 'tooth'),
'wreath' [ri:T] and 'wreathe' [ri:D] (verb based partly on
'wreath' and partly on a by-form of 'writhe'), and 'sooth' [suT]
("truth") and 'soothe' [suD] (originally meaning "verify"; the
modern meaning "calm, placate" is a semantic shift).  Outside of
these two groups, final [T] and [D] do not contrast, and, in
fact, are often found in free variation in the same word (this
also occurs within the groups, particularly among plurals).

This then, is the basis for my analysis.  As I said, I think that
there is nothing here that can be questioned.  If anyone sees
something that is not an empirically verifiable observation based
on data, please let me know.

Now my analysis says that you can't be certain that a contrast is
caused by different phonemes when there may be some other basis
for establishing the contrast.  My analysis says that you can
prove phonemicity using minimal pairs only through arbitrary
contrasts, contrasts where the presence of one phoneme or the
other doesn't tell you anything about the words involved except
that they are different words.  Thus if the presence of one sound
always tells you that the word is a pronoun or a deictic and the
presence of the other always tells you that the word is not a
pronoun or a deictic, then the contrast of these two sounds is
not necessarily the basis for the contrast between the words and
this contrast should not be considered evidence of the
phonemicity of the two sounds.  Please note, however, that this
is not the same as proving or even claiming that the two sounds
are not phonemes.

Similarly, if the presence of one sound tells you that the word
is a substantive and the presence of the other tells you that the
word is a verb, then the contrast predicts meaning (or function)
class.  This is not a phonemic contrast but a morphophonemic
alternation. While morphophonemic alternations do not preclude
the sounds involved from being phonemes as well, using
morphophonemic alternations to prove phonemicity is not
acceptable in my analysis.

When we come to the difference between intervocalic [T] and [D],
the division of the words into classes is admittedly on less firm
ground.  While I don't have many qualms about using grammatical
information that is based on meaning as a means of distinguishing
words (because as Stefan Georg pointed out, words that can't
appear in the same context don't have to be otherwise
distinguished), using lexico-historical information is going
outside of the grammatical structure of the language.  Using such
distinctions does not involve meaning in any specific way.  But
I still say that when a certain sound is required in one clearly
defined set of words and another sound is required in another
clearly defined set of words then contrasts between those sounds
when they appear as required in members of those sets should not
be allowed as evidence of phonemic contrast.  Again, however,
this is not the same as saying that the two sounds can't be
phonemes.

But on the subject of phonological nativization of loan words, I
have a few questions about New Zealand English (NZE) that I
expect that you can answer.  I presume that NZE has a number of
Maori loan words, and that many of these loans have been in the
language for a couple of centuries (but none before 1740).  Now
based on what you have said about Greek and Latin loans
(neologisms) in English, I would expect that New Zealanders
consider loans from Maori to be just good English words, and that
they would see nothing unusual about Maori segments in contexts
in which they would not appear in English.  I realize that
initial /ng/ in words of Maori origin is not a realistic parallel
to intervocalic [T] in words of Greek or Latin origin because the
prohibition against initial /ng/ in English is a phonotactic one
while there is no such prohibition against intervocalic [T].  But
here are my questions:  Do native speakers of NZE recognize Maori
loanwords as Maori words or do they just consider them English?
If they do recognize them, on what basis do they do it?  Do
native speakers of NZE Anglicize the pronunciation of initial
/ng/ in any way or do they pronounce it as it would be in Maori?
Would native speakers of NZE be able to coin new words with
initial /ng/?  Have they actually done so?

>In the broad sense implicit in your schema, both analyses
>produce the [th]/[dh] pattern by "rules". (In the narrower sense,
>however, your analysis has "rules" that mine doesn't.)

>Now that we have two analyses to compare, perhaps you can
>explain why you think yours is superior.

Well, as far as I am concerned, both analyses have the same
linguistic rules.  Your analysis just ignores some of them.  It
excludes them by saying that they happened too long ago to have
any effect on the modern language, or it excludes them by saying
that modern speakers can't possibly be aware of the rules that
linguistic analysis uncovers, or it excludes them by saying that
some rules about how words are pronounced can't have any bearing
on phonology.

But again, my analysis is not about the distribution of [T]/[D]
in English, it is about what is acceptable evidence for the
distribution (and collaterally about such distributions in
general).  So if my analysis has rules that yours doesn't, one of
them is that evidence that is offered as unequivocal proof of
some condition should not have any other possible explanation.
If there is more than one possible interpretation of the
evidence, then that evidence should be discarded (or at least
weighted) in favor of more conclusive evidence (ideally, evidence
that can only result from the condition that the evidence offers
to prove).

Another concept that my analysis has that yours lacks is the
principle that patterns are created by rules, even if we don't
understand or can't formulate the rules.  This I think is a
general principle of science.  It is a specific assumption of
generative grammar that alternations must be accounted for by
rules (free variation is also a rule).

Another rule that my analysis has that yours apparently doesn't
is that evidence should be analyzed independently of whether it
confirms or conflicts with our expectations.  Regardless of
whether the evidence conflicts with or confirms our expectations,
we should always question both the evidence and our expectations.

Now I would expect [T] and [D] to be phonemes in English, if for
no other reason, because of the parallel with /z/ and /v/, which
were created at the same time and by the same process that
created [D] and which are clearly used as phonemes by native
speakers of English.  But my analysis says that I shouldn't
uncritically accept 'thy' and 'thigh' as proof of this just
because it confirms my expectation when there may be some other
basis for this contrast.  When I see that all words in English
with initial [D] belong to a clearly defined group of words
(pronouns and deictic words [or demonstratives]) and that all
words in English with initial [T] belong to the complementary
group, then I feel justified in questioning the validity of 'thy'
and 'thigh' as a minimal pair.  Now if there were some words with
initial [D] in the complementary group (even if there were no
contrasts) I would not feel justified in saying that 'thy' and
'thigh' should not be considered a minimal pair since the
contrast would not be completely predictable at the word-class
level.

As for which analysis is superior, that is not my call because I
am not an impartial observer.  However, were I called upon to
evaluate two analyses in which I was not involved, I would tend
to prefer the one that was more rigorous in its standards of
evidence, the one that accounted for more of the available
evidence and excluded evidence only on valid and objective
grounds, and the one that did all of this in the most consistent
and concise manner.

I would not accept any analysis that distorted the evidence to
fit a particular model, that used evidence that could as easily
or better be explained by some other model, or that accepted only
evidence that supported a particular model while ignoring any
other evidence of equal quality.

In short, the superior analysis is the one that accounts for the
greatest amount of evidence in the most consistent, coherent, and
concise manner.

<snip>

Bob Whiting
whiting at cc.helsinki.fi