Don't touch my phonemes

[Douglas G Kilday]

Larry Trask writes:

>> pet - spin   : both have  /p/, even though one is aspirated, the other not

>More complicated.  The ordinary contrast between /p/ and /b/ is neutralized
>in this position.  Our orthography writes <p>, and our intuitions --
>at least among those of us who are literate -- is that this voiceless
>unaspirated [p] is still /p/.  But there is another analysis, in which
>this [p] is assigned instead to the phoneme /b/.  Hardly anybody has
>ever seen this second analysis as attractive, but I don't think we can
>simply dismiss it as plainly wrong.

Okay, so we transcribe <spin> /sbIn/, <stem> /sdEm/, <skin> /sgIn/, etc.
with the /s/ devoicing the following stop. Let's see how well this scheme
works. Consistency requires us to write <pin> [phIn] as /bhIn/ where the /h/
devoices the preceding /b/. No problem, since English doesn't use *[bhIn] or
*[pIn], only [phIn] and [bIn]. Likewise we have <tin> /dhIn/, <kin> /ghIn/,
etc.

In final clusters /s/ may devoice the stop from either side: <lisp> /lIsb/,
<lips> /lIbs/. As Bob Whiting recently pointed out, final unvoiced stops are
aspirated in English. Hence they may be treated as allophones of voiced
stops here also: lip /lIbh/, bent /bEndh/, tick /dhIgh/, etc.

English doesn't consist of monosyllables, so we can't discard /p/, /t/, and
/k/ just yet. Medial position presents some complications. The glottalized
"closed" pronunciation of <button>, <rotten>, etc. has no audible
aspiration. On the other hand no dental stop is enunciated either; the stop
heard is glottal: /ra?n/ in closed, /radh at n/ [rath at n] in open pronunciation.
No problem here.

Phonetic [sb], [fd], etc. occur medially in compounds and borrowings such as
<busboy>, <wolf-dog>, <asbestos>. Our scheme gives:

   grass-bin  /grAsbIn/
   grass-pin  /grAsbhIn/
   grass-spin  /grAssbIn/

When /sb/ (for example) begins a syllable the stop is devoiced and there is
no potential contrast of the form /sbIn/ ~ */sbhIn/. When /sb/ straddles a
syllabic boundary the stop is not automatically devoiced. Any phonemic
analysis of English will run into the effect of syllabic boundaries. Also,
the traditional phonemic scheme with /p/ requires /ss/ as well, or
<grass-spin> can't be separated from <grass-pin>.

Compounds like <bedtowel> and <bet-towel> 'a towel for covering bets' must
be distinguished. In the first, the [d] is clearly prolonged (with respect
to its normal length in <headache>) before the aspiration. In <bet-towel>
silence is substituted for that [d]. This is not a "silent phoneme", of
course, but a manifestation of the glottal stop in this position. In
<bed-dowel> the [d] is prolonged as in <bedtowel> without the aspiration. In
<bet-dowel> the <t> is again silent while the [d] has its normal length.
Hence:

   bedtowel  /bEddhaw at l/
   bet-towel  /bE?dhaw at l/
   bed-dowel  /bEddaw at l/
   bet-dowel  /bE?daw at l/

In the case of <naptime> aspiration is clearly heard between the stops, so
there is no trouble with transcribing /nAbhdhaym/. Similar considerations
apply to other medial clusters of stops.

The real difficulty arises when we deal with compounds like <hothead> and
<roadhouse>. For simplicity, compounds like those above are chosen. We can
transcribe

   bed-howl  /bEdhaw at l/
   bed-owl  /bEdaw at l/

without running over previous transcriptions, but we are out of options for
their unvoiced counterparts unless we introduce a tap /t'/:

   bet-howl  /bEt'haw at l/
   bet-owl  /bEt'aw at l/

Similar taps /p'/ and /k'/ are required for <uphill>, <backhand>, etc. Hence
our scheme of writing /sbIn/ provides no net economy of phonemes. It also
complicates English morphology. The regular plural of nouns ending in
aspirated stops cannot be formed by adding /-s/ to the singular; instead
/-h/ must be replaced by /-s/:

   top  /dhabh/  ~  tops  /dhabs/
   bit  /bIdh/  ~  bits  /bIds/
   neck  /nEgh/  ~  necks  /nEgs/

Other complications are apparent in the earlier analysis. For example <bet>
in isolation is /bEdh/, but in combination it may become /bE?/ or /bEt'/. In
the traditional scheme [?] and [t'] are arguably allophones of final /t/ in
combination.

I agree that there is nothing "plainly wrong" with considering unvoiced
stops to be allophones of voiced stops in English. Phonemic analysis,
however, is not an end in itself. We construct phonemic systems in order to
do something further. For most purposes, we are probably better off
retaining unvoiced stops as English phonemes.

Doug Kilday