Return of the minimal pairs (when is a morpheme not a morpheme?)

[Robert Whiting]

On Mon, 21 May 2001, Larry Trask wrote:

> --On Thursday, May 17, 2001 2:17 pm +0300 Robert Whiting
> <whiting at cc.helsinki.fi> wrote:

>> I'm perfectly happy to accept 'thy' as a ModE word.  But
>> 'thigh' and 'thy' are as perfect a minimal pair as German
>> 'Kuhchen' and 'Kuchen' or 'Tauchen' and 'tauchen'.  If [T] and
>> [D] contrast in 'thigh' and 'thy', then so do [c,] and [x] in
>> German.

> No.  The German contrast depends crucially on the presence of a morpheme
> boundary.  The English one does not.

> [on my minimal pair 'Than' (for 'Thandy') and 'than']

>> It is not a minimal pair as long as there is a morpheme
>> boundary involved.

> But there isn't.

>> Unless, of course, you claim that morpheme boundaries are not
>> phonological conditions.

> I do not.

> [on my list of near-minimal pairs like 'this' / 'thistle']

>> And every one of your near minimal pairs involves a morpheme
>> boundary after [D].

> Certainly not.

>> You stick by your contention that morpheme boundaries are not
>> phonological conditions then?

> No.  I never said any such thing.  It is beyond dispute that
> morpheme boundaries can have phonological consequences.
> Accordingly, it seems to be in order for our theories of
> phonology to take morpheme boundaries into account.

> [LT]

>>> The observation that initial [D] occurs only in
>>> "grammatical" words and initial [T] only in "lexical" words is
>>> just that: an observation.

>> Quite true.  But the observation that morphemes are units of
>> meaning is, I think, more than an observation.  It is a basic
>> principle of how human language works.  Duality of patterning
>> says that morphemes are made up of phonemes and that morphemes
>> have meaning (iconically, indexically, or symbolically) but that
>> phonemes do not have any inherent meaning.  Initial [D] in
>> English is a morpheme because it has meaning associated with it.
>> Admittedly, it is grammatical meaning rather than lexical
>> meaning, but meaning nonetheless.

> I'm sorry, but I must disagree flatly.  I have no problem with
> grammatical morphemes, or even with semantically and functionally
> empty morphemes.  But I cannot see that word-initial [D] is ever
> a morpheme at all.

Then please read Z. Harris, _Methods in Structural Linguistics_ (Chicago
1951), 192-93 and G. Trager, _Language and Languages_, (1972), 76-79, and
tell me why their analyses are wrong.

> First, word-initial [D] has no recognizable meaning or function.

Okay, since you refuse to read it, let me read it to you.  Z.
Harris, _Methods in Structural Linguistics_ (1951), p. 192:

   Similarly, in _there_, _then_, _thither_, _this_, _that_,
etc., we obtain by the Appendix to 12.22 a segment /D/ with
demonstrative meaning, plus various residue elements with unique
meanings (/is/ 'near', /aet/ 'yonder', etc.).

> Second, if we segment it away as a morpheme, then we are left with all
> those unanalyzed residues like <-is>, <-en>, <-ere> and <-e>.  These must
> now also be analyzed as morphemes, or as sequences of morphemes, and the
> position is impossible.  Remember, a morpheme boundary has to have a
> morpheme on each side of it -- not just on one side.

Gee, Larry, have you ever heard of case endings?  We're not
talking about productive morphology here, but perceptual
morphemes -- the ability of naive native speakers to recognize
units of meaning and base their perceptions of the structure of
their language on them.

Now what you are saying is that a word like 'boysenberry' must be
monomorphemic because if you segment off the morpheme '-berry'
then what you have left is the unanalyzable residue 'boysen-',
which cannot be a morpheme because it has no meaning in English.
Pish and tosh.  What remains is a morpheme precisely because of
what you said above:  "a morpheme boundary has to have a morpheme
on each side of it."  If you take an utterance like 'give Tom a
boysenberry' and identify the morphemes, you have 'give', 'Tom',
'a', '-berry' ('berry' is a morpheme because if we replace
'boysenberry' with 'berry' in the original utterance we still
have the same sense, just a more general statement).  Since
speech is composed of morphemes, what is left over (viz.
'boysen-') is also a morpheme by your very rule.  It is called a
"unique residue" and its meaning is that it differentiates
boysenberries from other kinds of berries (e.g., blueberry,
blackberry, raspberry, cranberry, huckleberry, gooseberry, etc.).

> It is reasonable to take initial [D] as a *marker* of
> closed-class membership, perhaps.  But it is not reasonable to
> take it as a morpheme.

More pish and more tosh.  It is as reasonable to take initial [D]
as a morpheme as it is to take initial 're-' or 'con-' as
morphemes.  According to your analysis, 're-' and 'con-' could be
markers of class membership but not morphemes because if you take
pairs like 'resist'/'consist', 'remit'/'commit',
'revert'/'convert', 'refer'/'confer', etc., then if you segment
off 're-' and 'con-' then you have unanalyzed residues like
'sist', 'mit', 'vert', and 'fer', none of which have any
independent existence in English.  Therefore, 're-' is a class
marker of verbs that have a sense of "again, back, against" and
'con-' is a class marker of verbs that have a sense of "with,
together" according to this analysis.

To go back to native words, '-less' cannot be considered a
morpheme used to create adjectives from a noun X meaning "without
X", but can only be considered a class marker of certain
adjectives, because if we consider 'feckless' or 'gormless' then
the residues 'feck' and 'gorm' have no meaning in English and
therefore can't be morphemes and therefore '-less' can't be
a morpheme because "a morpheme boundary has to have a morpheme on
each side of it."  It is exactly the other way around:  'feck'
and 'gorm' must be morphemes because '-less' is a morpheme (as
shown by such things as 'hopeless', 'helpless, 'fearless',
'witless', 'senseless', 'ruthless', 'peerless', etc., etc.) and
"a morpheme boundary has to have a morpheme on each side of it."
But your analysis requires 'feckless' and 'gormless' to be
monomorphemic.

On the other hand, by your analysis, there must be an English
morpheme 'n-' with negative meaning since pairs like
'either'/'neither', 'or'/'nor', 'ever'/'never' leave a
recognizable morpheme when initial <n> is extracted.

What makes me so sure that initial [D] is a morpheme (perceptual,
not productive)?  After all, in the examples cited above, there
are always clear examples of the morphemic use of one of the
morphemes involved which allows us to say "once a morpheme,
always a morpheme," or there are sufficient other examples to
make an incontrovertible case (e.g., not only 'resist' and
'consist', but also 'assist', 'insist', 'desist', 'persist', and
so on).  But what is needed to identify morphemes is a pattern of
sounds and a corresponding pattern of meanings.  When these
things co-occur, speakers will tend to identify the corresponding
sounds and meanings as morphemes.  Perhaps not consciously (after
all, the average NNS doesn't know what a morpheme is) and perhaps
not even correctly from a historical perspective (this is the
basis of folk etymology), but as part of the way that the NNS
organizes his knowledge of his language.

Showing how this patterning works is easier done than said
("longum iter est per precepta, breve et efficax per exempla"
Seneca, Letters vi.5).  Let us take as an example three deictic
("pointing" or "demonstrating") words: 'there', 'here', 'where'.
Now Larry says that at most initial [D] can be a class marker,
not a morpheme, because if one segments initial <th> then one is
left with an unanalyzed <ere>.  Now unlike the 'boysen-' in
'boysenberry', this <ere> actually appears elsewhere in English;
in fact, in appears in the three deictic words that we are
discussing.  Therefore, if initial [D] (<th>) is a class marker,
then so must initial <h> and initial <wh> be class markers.  It
is a characteristic of class markers that they do not contribute
to the meaning of a word, but only mark it as belonging to a
particular word class (in fact, they can't contribute to the
meaning or else they would be morphemes, wouldn't they?).

So, following Larry's analysis, if <th>- is a class marker, then
<h>- and <wh> must also be class markers.  And if they are class
markers, then there must be an English word <ere> whose meaning
we can establish empirically as "in ___ place."  Now, again
empirically, when the class marker is <th>- ('there') the word
means "in that place", when it is <h>- ('here') the word means
"in this place", and when the class marker is <wh>- ('where') the
word means "in what/which place."  Therefore, if these are class
markers, English must have class markers that mark word-classes
as "far deixis" (<th>-), "near deixis" (<h>-) and
"relative/resumptive/interrogative deixis" (<wh>-).  We can check
this by looking at some of the other triplets where we find a
word <ence> that means "from ___ place/time" and the same class
markers produce 'thence' "from that place/time", 'hence' "from
this place/time', and 'whence' "from what/which place/time"; and
a word <ither> that means "to ___ place/time" and 'thither' "to
that place/time", 'hither' "to this place/time", and 'whither'
"to what/which place/time."

The existence of 'then' "at that time" and 'when' "at what/which
time" implies that there should be a '*hen' "at this time."
Historical investigation shows that indeed there was at one time
but it has been replaced by 'now'.  We know that it existed,
however, because 'hence' is derived from it (of course, if we
didn't have the historical information about the existence of
earlier 'hen(ne)' we could always say that 'hence' is an
analogical form based on the "class marker" <h>- and the forms
'thence' and 'whence').

So we have a (nearly) complete set of adverbs of time/place
comprising far deixis, near deixis, and relative/interrogative
forming as tight a pattern as anyone could ask for.  By simple
linear analysis there is no question where any part of the
meaning of these words comes from.  So let us turn to the adverbs
of manner.  Here, the pattern is not so tight, but the "class
markers" are still recognizable.  The deictic adverbs of manner
are 'thus' and 'how'.  The <h>- of 'how' is really a *<wh> which
has lost its <w> because of the other <w> in the word.  Perhaps
this has forced the far and near deixis into a single form with
<th>- because the near deictic <h>- could not be distinguished
from the <h>- (< <wh>) of the relative/interrogative (or perhaps
not).  In any case, after segmenting the "class markers" we have
two unique residues <us> and <ow> both with the same meaning "in
___ manner".  We can then establish that 'thus' means "in
this/that manner" and 'how' means "in what/which manner."

But wait a minute.  How does the meaning of <us>/<ow> differ from
the meaning of the adverbial marker '-ly'?  This marker is used
to make adverbs from adjectives and adds a sense of "in (a) ___
manner" to the word.  Thus from 'cold' we get 'cold+ly' = "in a
cold manner", from 'warm' we get 'warm+ly' "in a warm manner",
from 'quick' we get 'quick+ly' "in a quick manner", and so on
through hundreds (if not thousands) of such words.  Clearly, if
we follow Larry's analysis, the '-ly' is a class marker of
adverbs because '-ly' has no meaning of its own, it simply marks
adverbs of manner derived from adjectives.

But wait another minute.  In every case where we have '-ly' as an
adverbial marker, what comes before it is clearly a morpheme
('cold', 'warm', 'quick', etc., etc.).  Since '-ly' clearly marks
the sense "in (a) ___ manner" (one can't say that it has the
meaning "in (a) ___ manner", because if '-ly' had meaning it
would be a morpheme, wouldn't it?), then equally clearly is it
<us> and <ow> that mark the same sense in 'thus' and 'how'.
Since it is a morpheme that goes into the blank that completes
the sense (not meaning; can't be, can it?) of the '-ly' adverbs,
then it must be a morpheme that goes into the blank in the
<us>/<ow> adverbs to complete the sense.  And similarly it must
be a morpheme that goes into the blank that completes the sense
of the place/time adverbs.  Otherwise, the method of using
"slots" to show identity of grammatical function is a hoax.

Therefore, the <th>-, <h>-, and <wh>- in the place/time adverbs
are morphemes in the same way that 'cold', 'warm', 'quick', etc.,
etc. are in the manner adverbs, and it is the <ere>, <en>,
<ence>, <ither>, that are the "class markers" in these words,
just as the <ly> and <us>/<ow> in the adverbs of manner are.  The
difference is that <th>-, <h>-, and <wh>- are grammatical
morphemes (have grammatical or functional meaning) while 'warm',
etc. are lexical morphemes.  Another highly significant
difference is that the <ly> marker is productive (or is an
open-class marker if you prefer) while <ere>, etc. are not
productive (or are closed-class markers).  Obviously, the <ere>,
etc. slots do not accept lexical morphemes (because they are
closed-class) and the <ly> slot does not accept grammatical
morphemes (an exception [there always seem to be a few] is the
secondary, doubly-marked 'thusly'; this is simply an example of
using a productive marker to clarify or reinforce a
non-productive marker [in much the same way a child learning
English will often say 'feets' or 'childrens']).

So Larry's analysis is exactly backwards.  Words (pronouns,
adjectives, and adverbs) with far deixis is not a word-class in
English.  It is adverbs (subclassed into adverbs of manner and
adverbs of place/time and further subdivided into adverbs of
far/near/asked-about place/time) that is the word-class.  Hence
initial [D] is not a class marker but is a morpheme with deictic
(demonstrative) meaning.  QED.

To check this analysis, let's take a typical English word-class,
say attributive adjectives, and see what happens through the use
of various "class markers" to modify the original class.  Now
primary attributive adjectives do not have any overt class
marker.  As an example we will take just two (to save space),
let's say 'sharp' meaning "keen, pointed" and 'soft' meaning "not
resisting, not hard."  These are morphemes because they are
sequences of sounds associated with meaning.  How do we change
this meaning through the use of "class markers"?

  Primary meaning + "class marker"   General sense of class marker

  sharp         soft

  sharp+ly      soft+ly               adverb of manner:  "in (a)
                                      ___ manner"
  sharp+en      soft+en               causative/factitive/inchoative
                                      verb:  "to make/become ___"
  sharp+ness    soft+ness             abstract noun:  "quality of
                                      being ___"
  sharp+ish     soft+ish              inexact/relational adjective:
                                      "sort of ___"; "more ___ than not"

So what does it look like?  Are sequence of sounds like 'sharp'
or 'soft' class markers of words having to do with sharpness or
'softness' and so on, or are the meaningless (but full of sense)
particles <ly>, <en>, <ness>, <ish> the real class markers (if
such there be in English)?  Of course it is possible the these
markers are really derivational morphemes (as most people seem to
consider them) and then we don't have to consider words like
'sharply' and 'softness' as monomorphemic.

>> No, the fact that these are all function words is not particularly
>> phonologically significant.  The fact that they all begin with a
>> functional morpheme is.

> I've just denied this.

Denial is not the same as disproof.

In order to disprove it you need to prove that it is completely
impossible for the speakers of English to perceive initial [D] as
a morpheme.  Now a morpheme is a sound or sequence of sounds that
has meaning associated with it.  It is particularly difficult to
prove that there is no meaning associated with initial [D] in
the face of an extremely tight pattern of sound and meaning
correspondences:

        Meaning of initial segment
 far deixis    near deixis     rel./interr.       General Meaning
    <th>-         <h>-            <wh>-         of residual segment
   "that"        "this"        "what/which"

   there       here              where          "in ___ place"
   then       *hen [OE henne]    when           "at ___ time"
   thence      hence             whence         "from ___ place/time"
   thither     hither            whither        "to ___ place/time"


Now if anyone told me that they couldn't see any pattern to the
initial sounds in the columns and the meanings in the rows, then
I would not conclude that there is no meaning associated with
these sounds, but would instead conclude that whoever said this
was not very good at pattern recognition.  I'm sorry, but I just
really don't see how anyone can look at this table and fail to
see a relationship between sound patterns and meaning.  This is
not (with the exception of '*hen') some reconstruction of
something that hasn't existed for centuries or millennia pieced
together from coincidental look-alikes from around the world and
through the ages.  This is the English as she is spoke today.

So you are free to deny it and I will accept your denial as an
admission that you can't see the pattern or are willing to accept
it as coincidence.  But if all you can do in the way of disproof
is to raise points that have already been discussed and dismissed
in the literature, then it is just your word against that of
people like Zellig Harris and George Trager.  What you need to do
to establish your point is to refute with evidence their analyses.
In short, it's your burden of proof.  Now I have heard that
linguists in Britain do not pay any attention to American
structuralism, but I was not aware that they would not even read
books by American structuralists.  But until you read your
homework assigment, there is little point in discussing it in
class.

> [on my /oi/ example]

>> But is the diphthong /oi/ a morpheme?  What morphemic function do you
>> claim it has?

> It is not a morpheme.  It is merely a marker -- in this case, a
> marker of non-native origin, as is the consonant [ezh].

I'd be careful about saying things like this, otherwise Ross
Clark will get you.  He will tell you that native speakers cannot
distinguish non-native words from native words.  They are all
just English words.  On the other hand, I quite agree with you
that there are phonological markers of non-native origin in
English words (and in most other languages as well).

> [on my hypothetical example of a Greek 'dhelta']

<snip>

>> What you need to establish your point is not answers to
>> questions like "What would the world be like if Napoleon had
>> conquered Russia?" but an example of an initial [D] in English
>> that is not a deictic or a pronoun (i.e., does not begin with the
>> pronominal/deictic morpheme).  It doesn't even have to contrast
>> with a word with initial [T].  Any word that begins with [D] in
>> which the [D] is not a morpheme would be sufficient to justify
>> your claim that the fact that all words in English that begin
>> with [D] are pronouns or deictics is mere coincidence and that
>> the distribution of initial [T] and [D] is not predictable by any
>> phonological rule.

> Indeed.  Let me clarify my point.

> Word-initial [eng] is prohibited by English phonology and is
> wholly unpronounceable by most English-speakers.  When we borrow
> foreign words or names containing initial [eng], it is always
> eliminated in one way or another.

Yes, and, as I say, we have done this before.

> But initial [D] is not unpronounceable in English at all, and I
> cannot see that it is prohibited by English phonology.  So, what
> we need as a test case is a noun or a name with initial [D]
> borrowed into English from a language that permits it.

No, it is not prohibited at all.  It only happens to occur in
pronouns and deictic words.  Nothing in the phonotactics of the
language prevents borrowing words with initial [D] or coining new
words with initial [D].  It just hasn't happened.

The situation with initial [D] is paralleled by English words
with initial [v].  There are very few native words with initial
[v] just as there are very few words with initial [D].  This is
because [v] and [D] were originally just allophones of [f] and
[T], respectively, occurring in intervocalic position.  Therefore
initial [f] and [T] were not affected by this allophony and the
voiced sounds in initial position had to find their way there
from somewhere else.  The few native words with initial [v] are
loans from the southern dialects of English, where initial [f]
was voiced to [v], taken into the standard dialect.  But there
are many words with initial [v] that have been borrowed from
Romance which had an [f]:[v] contrast for a long time and so
English has a large number of words with initial [v], almost all
of them loans.  But there are also new coinings with initial [v]
in English (e.g., 'vax', 'Velcro'), which is something that
initial [D] does not share.

So the fact that there have been no borrowed words with initial
[D] even from languages that have this sound (see below), and
especially the fact that there have been no new coinings with
initial [D] in English suggests -- at least to me -- that
speakers recognize some kind of restraint on the use of initial
[D].

> Unfortunately, there appear to be few such languages, and
> English seldom borrows anything from most of them.  But I've
> pinned my hopes on Greek, the one contemporary language which
> both permits initial [D] and lends words into English.  But Greek
> is not ideal, because of our long-standing tradition of
> converting Greek words and names into English by conventions
> which are more appropriate to classical Greek than to the modern
> language. The chief exception is food and drink, in which modern
> Greek pronunciations are usually the basis of the English form,
> as with 'souvlaki' and 'tzatziki'.  So, what we need is for the
> Greeks to come up with a prince of a dish which we
> English-speakers will delightedly take up -- and which has a name
> starting with [D].  Sadly, Greek chefs have not so far obliged us
> on this matter -- at least as far as I can think.  C'mon, Greek
> chefs: we need you! ;-)

You might do better to pin your hopes on Arabic, which has a
phonemic contrast between [d] and [D] (in fact, Arabic has dental
contrasts eight ways to Sunday).  Unfortunately, words with
initial [D] in Arabic don't come into English with initial [D].
An example is 'dahabeah' which is a kind of houseboat used on the
Nile (variants 'dahabeeyah', 'dahabiya', etc., but never with
[D]).

The only loan words that I know of with [D] are two Islamic month
names 'Dhu al-Hijja' and 'Dhu al-Qa`dah' (` for `ayn; dhu is an
indefinite pronoun so 'Dhu al-Hijja' is literally "the one of the
Hajj") and these are clearly more technical terms than everyday
English words.  They will only be known by those who study or
have some interest in Islam (or possibly in menology).  Those who
speak Arabic will pronounce them with [D] as in Arabic, those who
do not (and yes, there are those who study Islam without knowing
Arabic, just as there are those who study Christianity or the
Bible without knowing Hebrew, Aramaic, or Greek) will pronounce
them with [d].  Other than that there is only 'dhal', the name of
the letter in Arabic.  Anyone who knows this word is likely to
know its correct pronunciation (this occupies the same niche as
the name of the letter 'edh' in English; i.e., the only place
where <dh> represents [D]).  But this word, like 'edh', has no
existence apart from the sound that it represents.  [D] in these
words is not a symbol, but an icon.  Otherwise, <dh> usually
represent the phryngeal [d.] (sometimes referred to as
"emphatic") in Arabic transcription, as in the place name
'Riyadh'.

While we're on the subject of Greek and the names of letters and
people not being able to pronounce sounds that are not part of
their phonetic inventory (well sort of), there is an interesting
example of the effects of the loss of a sound on an alphabet.
Ancient Greek had a [w] sound and, naturally, a sign to write it
called wau.  Then the [w] sound was lost, and, normally, the
letter would have been lost also.  But the letter couldn't be
gotten rid of because it was part of the number system
(representing 6).  However, it was no longer possible to talk
about the letter using its original name because the sound needed
to say its name no longer existed in the language.  So the letter
had to be renamed, and it was called digamma because that's what
it looked like:  a gamma with an extra stroke.  Thus it is the
only Greek letter whose name reflects its shape rather than its
sound.

Now for the interesting part.  The Latin alphabet did not have a
sign for the [w] sound (mostly because it disappeared from Greek)
and used the <u/v> graph instead.  Much later, the medieval
Norman scribes began writing the u/v graph doubled (uu/vv) to
indicate the English [w] sound because they were unfamiliar with
the English letter wyn or wen.  This Norman double-u (or to the
French, double-vay) became the normal sign for the [w] sound and
once again, this is the only letter whose name reflects its shape
rather than its sound.  Outrageous coincidence that the old <w>
and the new <w> are the only letters in their alphabets named for
their shapes?  Not really -- if the [w] sound hadn't disappeared
from Greek, the letter would probably have made its way into the
Latin, and thence into the English alphabet with a name
reflecting its sound and wouldn't have had to be reinvented.
What it shows is that what goes around, comes around.

<snip>

Bob Whiting
whiting at cc.helsinki.fi