Projecting non-lexical phonology from phonetic knowledge

1
Projecting non-lexical phonology from phonetic
knowledge

• Donca SteriadeMIT

2
Phonetics/phonology interactions

•  Phonological grammars reflect phonetic
  constraints
• articulatory and perceptual limitations
•  DD no voiced obstruent geminate
• Podesva 2002 Hayes Steriade 2003 Kawahara
  2004
•   Articulation
• sustained airflow across glottis requires more
  articulatory adjustments in longer stop closures
• Ohala 1981
•   Perception
• DD, if passively devoiced DD9, more similar to
  TT
• ? (D-T) gt ? (DD TT) if DD DD9

3
Phonetic asymmetries of this kind

•  are reflected in phonological systems
• DD relates to p????????????????????????d,
•  How do we get from phonetics to grammar?
• ????? ?(D-T) gt ?(DD TT) to grammar
•  What is the grammar like?
• Is DD the right constraint?

4
Option 1 link via phonetic knowledge

•  Speaker awareness of perceptual limitations
• e.g. knowledge that ?(D-T) gt ?(DD TT)
•  Projects a fragment of the grammar
• Ident voice/ _-long gtgt Ident
  voice/_long
• phonetic knowledge from Kingston Diehl
  (1994) awareness of the perceptual correlates of
  a contrast their articulatory implementation.

5
From perception to phonology
Correspondence rankings
Auditory comparisons
Similarity rankings

• Voicing ratio (T-D)
• gt DVoicing ratio (T-DD)

Ident voice/_,-long gtgt Ident
voice/_,long

• (T-D)
• gt ?(TT-DD)

P-map hypothesis
6
Voicing ratio

•  Total voiced closure duration
• Total closure duration
•  if T 100ms, 1ms of closure voicing VR .01
•  if D 100, 80ms of closure voicing VR .8
• ? (VR (T-D)) .79
•  if TT 200ms, 1ms of closure voicing VR
  .005
•  if DD 200ms, 180ms of closure voicing VR
  .9
• if DD9 200ms, 100ms of closure voicing VR
  .5
• ? (VR (TT-DD)) .895
• ? (VR (TT- DD)9) .495

7
Relevant here

•   ?(VR (TT- DD9)) lt ? (VR (T-D))
• If the speaker observes that the geminate DD
  is substantially devoiced ( DD9), he knows that
  the closure voicing contrast is less effective
  for TT- DD9 than for T-D.
• Barring further differences, he generalizes
•   ?(TT- DD9) lt ? (T-D)

8
From perception to phonology
Correspondence rankings
Auditory comparisons
Similarity rankings

• (VR (T-D)) gt
• DVR (TT- DD9 )

Ident voice/_,-long gtgt Ident
voice/_,long

• (T-D) gt
• ?(TT- DD9)

P-map hypothesis
9
Option 2 link via history

•  Confusion, without self-awareness
• ?(D-T) gt ?(DD TT)
•  Mistakes
• more TT-for-DD mistakes than D-for-T
•  Variation
• lots of words with DD acquire TT variants
•  Sound change All DD -gt TT
•  Lexicon no DDs left
•  Grammar as generalizations over lexicon DD
•  Speaker knows that DD
•  No basis of knowing that ?(D-T) gt ?(DD TT)

10
Option 2 cont.

•   It is enough for speakers to know that their
  language contain a given phonological pattern.
  They need not also know why.
• (Barnes 2002)
•   Since phonological systems have evolved
  diachronically, their properties reflect
  constraints on sound change . Explanations
  for phonological patterns may reside in
  synchronic analysis or diachronic evolution.
  Which explanation will emerge in any case is a
  matter to be resolved based on the evidence, but
  since historical accounts permit simpler
  grammatical models they are preferable wherever
  possible (Blevins Garrett 2002)
•  Do speakers know phonetics? Is their knowledge
  of phonology stored in phonetic terms? (Hyman
  2001)

11
What type of knowledge

•  do we choose to investigate?
•  Of existing words?
•  Of lexically manifest alternations?
•  Of lexically manifest distributional patterns?

 Solutions to undocumented problems.
12
Onset of passive devoicing

•   Ohala Riordan 1979Westbury Keating 1986

13
The typology
14
The typology
Ossetic, Hindi, Hungarian, Japanese, Latin,
15
What does the speaker know about the status of DD

•  If she learns Fula, Japanese, Hungarian?
•  All DD types are attested in the lexicon.
•  No process targets DD differently from D or TT
  
•  No lexical evidence for any constraint on DD
•  No evidence for the ranking
• Ident voice/ _-long gtgt Ident
  voice/_long

16
But

•  If constraints are projected from phonetic
  knowledge,
•   if a substantial number of /DD/ DD9, then
  
•   the learner observes that
• ? (VR (D-T)) gt ? (VR (DD9 -TT))
• whence a similarity ranking
• ?(D-T) gt ? (DD TT)
• and thus a correspondence ranking
•  Ident voice/ _-long gtgt Ident
  voice/_long
•   Then no matter what the lexicon says, expect
• DD to be more prone to categorical devoicing
  than D

17
Note

•  Under this scenario, speakers learn a specific
  fact about DD
• its easier to devoice DD than D.
•  Not just that DD is worse than D or TT or NN.
•  Nor that DD is easier to modify than TT or NN
  for any feature other than voice
• (e.g. long, sonorant, continuant)

18
Then

•  The P-map hypothesis (Option 1) predicts
•  that exposure to a phonetic norm DD DD9
  teaches learners that they can modify voice in
  DD more than D, TT, NN.
•  but not that they can modify long,
  sonorant or other feature in DD more than in
  any other C type.
•  In contrast Option 2 predicts nothing.

19
Kawahara 2003

•  Japanese loanword gemination
• Vlax C(V) VCCV
• whip wippu
• web webbu
• kit kitto
• kid kiddo
• pick pikku
• pig piggo

Katayama 1998, Kato 2004
20
Both TT and DD

•  DD is recent
•  from European loans webbu
• Ito and Mester (1995)
•  from recent sound changes (dialects)
• itigo -gt iggo strawberry
• yakuba -gt yabba town hall
• Haraguchi (2002)

21
Motoori Norinaga (Lyman)s Law

•  DDstem
• No effect on non-Yamato lexicon, inc. loans
• Doug dagu
• dive daibu
• job dZobu
• bogey bogi
• Gabriel gaburieru
• dogma doguma
• this and more in Kawahara 2003

22
But

• In D//DD DD optionally devoiced
•  bag baggu bakku
• good guddo gutto
• bed beddo betto
• David Deibiddo Deibitto
• beverage bebiradZi bebirattSi
• debug debaggu debakku
• Budda Budda Butta
• judge dZaddZi dZattSi
• hit by pitch deddobooru dettoboru
• (deadbore)

Haraguchi 2002 Kawahara 2003
23
Kawaharas analysis

• the crucial observation is that devoicing seems
  to take place in reaction to the OCP and not in
  direct response to the markedness of the
  obstruents per se.
• Otherwise put DD does not explain it.

24
Kawaharas analysis

• Ident voice-long gtgt OCP Ident voice long

25
Kawaharas analysis

• Ident voice-long gtgt OCP Ident voice long

26
Kawaharas analysis (cont)

•  Ident voice/_, long gtgt DD

27
No devoicing without ranking of

• Ident voice/_-long gtgt Ident voice/_,long
•  Ident voice gtgtOCP bogey bogi, boki
•  Ident voice gtgt DD web webbu,weppu

28
Source of knowledge?

•   properties inherent in all DD-TT contrasts?
• ???(VR (D-T)) gt ?(VR (DD-TT))?
• Not universal GiovanardiDiBenedetto 2003
  (Italian)
• ??(clos.duration (D-T)) gt ?(clos.duration
  (DD-TT))?
• Not universal Samudravijaya 2003 (Hindi)
• ???(V1 duration (D-T)) gt ?(V1 duration (DD-TT))?
  
• Not universal Keating 198x (French, Polish)
• ???(V1/C ratio(D-T)) gt ??(V1/C ratio(DD-TT))?
• Prob. not universal TserdanelisArvaniti 2002
  (C.Greek)

29
D (clos. duration) - not always larger for D-T
wrt DD-TT
Closure duration differences between D, T, DD, TT
in Hindi based on data in Samudravijaya 2003
30
Language specific source?

•  Japanese DD are substantially devoiced DD9

From Kawahara 2003
31
To repeat

•  the Japanese phonetic norm is
• /DD/ DD9
•   so learners can observe that
• ? (VR (D-T)) gt ? (VR (DD9 -TT))
•   they generalize this to
• ?(D-T) gt ? (DD TT)
•   and project a correspondence ranking
• Identvoice/_-long gtgt Ident voice/_long
•   this explains, in part, why Lymans Law is
  satisfied in loans, by devoicing DD not D.

32
The total ranking
Ident sonorant
Ident voice/_, -long
Ident long
Ident voice/_, long
OCP voice
DD
33
What the lexicon says
tobta -gt tonda, not todda, and not totta
34
What this shows

•  tob-ta should have yielded toda, but
•  toda excluded by DD, so
•  toda surfaces as tonda, not tota, suggesting

DD Ident voice Ident sonorant
35
So, either

•  tob-ta -gttonda doesnt show how to satisfy DD
•  or it suggests DD, Ident voice gtgt Ident
  sonorant
•  not the ranking that emerges in loan
  adaptation
• Ident voice/_-long Ident sonorant
• Ident voice/_-long
• DD

36
Either way

•  the ranking adopted in new loans is not learned
  from exposure to the existing lexicon.
•  its source the phonetic partial devoicing.

37
Whence DD?
 fully or substantially voiced DD in ref
 larynx lowering, nasal leakage, pharynx
expansion

• Suomi 1980 Westbury Keating 1986

 then a substantially voiced DD is a more
complex articulation than either D or TT
or NN
38
ZZ

•  Ohala on Z

For optimal voicing, Po must be as low as
possible (to keep DP high). For optimal
frication noise, Po must be as high as possible
(to keep the pressure differential across the
oral constriction high). Both of these actions on
Po cannot be done simultaneously the result is
that voiced fricatives with strong frication
(e.g., z, Z have a tendency to devoice those
with strong voicing (e.g., B, D tend to have
weak, if any, frication.
39
Then

•  The P-map hypothesis (Option 1) predicts
•  producing DD DD teaches speakers that DD is
  harder than D, TT, NN.
• i.e. that DD gtgt D, TT, NN
•   producing ZZ ZZ teaches speakers that ZZ
  is harder than DD
• Here too Option 2 predicts nothing.

40
Does phonology know or care?

•  Hungarian loan gemination (Nádasdy 1989)
•  Voiceless obstruents
• suite szvitt sv9it
• sweater szvetter sv9EtEr
• gallop galopp gOlop
• couche (Fr.) kuss kuS
• mafia maffia mOfiO
•  Sonorants
• Feuer (Gm.) Feuer fOjjEr
• jam dzsem dzsemm dZEm
• Wren Wren rEn

41
DD-Gemination inhibited

•  Voiced stops generally do not geminate
• club klubbclubklub
• tubing tübingtübbing
• snob sznob sznobb
• smog szmog szmogg
• tuba tuba tubba
• video video viddeo
•  Voiced fricatives never do
• Oz óz oz ozz
• naive naív nOiv naivv
• lava lava lavva
• garaZ (Fr.) garázs gOraZ, aZ

42
Orthography

•  ltCCgt always transferred as C for TT, SS, NN
• stress sztressz strEs
• match meccs mEtS
• witz vicc vits
•   ltCCgt frequently not transferred for DD, ZZ
• jiddis jidis 3190/11 Google hits
• pizza piza 1850/34
• kabbala kabala 1320/1380
• Or ltDDgt, ltZZgt transferred as TT, SS
• drab drapp
• jazz dzessz

43
What does the native lexicon say?

• All consonants contrast for length
• ráz he shakes rázz shake!
• lobog blaze lobban flare
• ad give add you give
• magad time reggel morning
• No process provides evidence for DD or ZZ.
• All geminates prohibited next to T gt N gt L.
• All Cs geminate in Instr/Transl.

44
To repeat

•  the Hungarian phonetic norm includes
• /DD/ D and /ZZ/ Z
•   learners can observe producing target voice
  value in DD requires more complex adjustments
  than in D, TT, NN
• DD gtgt D, NN, TT
•   producing target voice value in ZZ
  requires more complex adjustments than in DD
• ZZ gtgt DD gtgt D, NN, TT
•   this is why loan gemination distinguishes NN,
  TT from DD and DD from ZZ.

45
The relevant ranking
Idet long
Ident voice
Ident son
I/T gemination
ZZ
DD
Foreign word gemination
TT, NN, LL