POOT-yin/POOCH-in

Mark_Mandel at DRAGONSYS.COM Mark_Mandel at DRAGONSYS.COM
Fri Aug 11 14:48:50 UTC 2000


Thomas Paikeday <t.paikeday at SYMPATICO.CA> writes, with an unfortunate type
of encoding
 [his lines wrap at this point]-----------------------------------------|


>>>>>
   --------86E2E1A82CEDC765A4473F9B
   Content-Type: text/plain; charset=iso-8859-1
   Content-Transfer-Encoding: quoted-printable

Mark Mandel's comment gives me an opportunity to define "system"  and exp=
lain
how our keyless pronunciation system can handle a non-English word like
"Skoplje."

     [...]

Since the User's=99 pronunciation system is meant for English speakers (I=
 thought
I killed or at least scotched "the native speaker" circa 1985 in The Nati=
ve
Speaker Is Dead!), pronunciation respellings are given in anglicized form=
=2E This
system doesn't fail if its basic principles are adhered to.

First we ascertain the common Slavic pronunciation of "Skoplje" from an
authoritative source. Yugoslavs may say the word in two syllables, but th=
e
English speaker cannot handle the palatalized "l" sound with ease. An
acceptable English pronunciation has to be in three syllables, "-lje" bec=
oming
"-lee.ay." The first syllable, "Skop-," is heard as either short "o" (as =
in
"cop") or diphthong "o" (as in "cope").

The keyless respelling is then written as (SCOP.lee.ay) or (SCOPE.lee.ay)=
=2E I
would consider a respelling such as (SKOHP.lyeh) as seen in some encyclop=
edias
rather un-English and too abstract for the common English speaker.

I hope you find this keyless pronunciation clear enough for Americans, na=
tives
and non-natives.
<<<<<

That works for your dictionary, but wouldn't work for our purpose of speech
recognition. With most foreign words we assume that at least some speakers
who use them at all will pronounce them approximately as they're pronounced
in their source language, and then we try to approximate that pronunciation
with our English phoneme set. That creates clusters and syllables that
cannot occur in English, and your assumptions, not mentioned above but
implicit in the approach, fail because of the difference in our goals. The
same would be true for any dictionary that tried to give a closer
approximation than your of the source-language pronunciation of foreign
words.

   Mark A. Mandel : Dragon Systems, a Lernout & Hauspie company
          Mark_Mandel at dragonsys.com : Senior Linguist
 320 Nevada St., Newton, MA 02460, USA : http://www.dragonsys.com



More information about the Ads-l mailing list