On the phonological representation of timing and prosody Bert Vaux University of Cambridge CUNY Symposium on the Foot January 15, 2009 x x x O N C R.

On the phonological On the phonological representation of representation of timing and prosodytiming and prosody

Bert VauxUniversity of CambridgeCUNY Symposium on the FootJanuary 15, 2009

x x x

O N C

R

OutlineOutline Relevant conference Q: “What is the internal structure of the foot?” Mainstream opposition: MT vs CRONX

(we won’t consider the good evidence for Body-Coda in Korean and Japanese) MT: Hyman 1985, McCarthy and Prince 1986, Nespor and Vogel 1986, Hayes 1989, OT… CRONX: Kahn 1976, Steriade 1982, Levin 1985, Sloan 1991… OT still uses Onset and NoCoda labels but MT

Most arguments for sub-syllabic constituents (including moras) are weak or ineffective. Can more compelling arguments be found?

Overarching problems in investigating MT vs CRONX Problems with traditional arguments for sub-syllabic constituency

CRONX moras

CRONX structure is necessary more compelling arguments

Is CRONX structure sufficient? some evidence for moras

Conclusion: unfortunately, (C)R(O)NX seems to be required

Some overarching problems in Some overarching problems in investigating subsyllabicityinvestigating subsyllabicity

Significant increase in descriptive power with OT flaws in traditional argumentation (as we’ll see)

incomplete data (Hayes, McCarthy&Prince) not diagnostic of prosody necessary vs sufficient distributional stats not accounted for

• e.g. re <hat>: æt more frequent than hæ (Vitevitch and Luce 1999)• but O-R effects still emerge when frequency is matched (Lee and Goldrick 2008)

Ohalan concerns• diachronic explanation (Hale/Reiss, Garrett, Blevins…)• phonetic explanation (Steriade, Gordon…)

Problems with traditional Problems with traditional arguments for arguments for subsyllabic constituencysubsyllabic constituency

Problems for CRONXProblems for CRONX Traditional arguments:

don’t consider non-sub-syllabic alternatives, e.g.:• phonotactic constraints on onsets and codas in English

(Kenstowicz 1994, Blevins 1995) can be gotten just by reference to what can precede or follow a V within a syllable.

• A typical Onset process can be captured with (tautosyllabic) /_V or

[ _. Mirror image for Codas.• r-deletion example on following slide• Foot- rather than syllable-based (cf. Stuart Davis today)

don’t show the segments acting as a unit• Nice exception is Steriade 1982’s Pig Latin case, but this too can

be dealt with asyllabically (Raimy) Some desiderata:

(Non-Raimyan) constituent movement processes processes affecting e.g. C2 in [(C1)C2(C3)V…] These are key since they

don’t submit to edge analysis…

Case study: r in non-rhotic ECase study: r in non-rhotic E

typical description: delete postvocalic r “the everyday speech of [NY] city exhibits consistent vocalization of postvocalic /r/”

(Labov et al. 1997) “in certain dialects [of English], postvocalic /r/ undergoes a process of reduction”

(Gick 2002) doesn’t work: merry, etc

(unless you have Ons >> *Vr, but this is a different issue) other options

delete non-prevocalic r (the correct surface generalisation) (delete moraic r)

problematic predictions deletion with syllabic r and -Cr (Halle’s oeuvre…) processes targeting moraic Cs should also affect initial geminates that make weight

(and any other moraic Onset Cs) Conclusion: r-deletion doesn’t provide evidence for sub-syllabicity

MT and its problemsMT and its problems Sufficiency argument:

McCarthy and Prince 1986 etc. propose that all subsyllabic functions can be carried out by moras

• no need for CRON• X slot functions can be carried out by root nodes

But are moras sufficient? (Next 3 slides: no) Marquee arguments have subsequently been undermined

• Moraic consistency (see later slide)• Behavior of CL (see next slide)• Behavior of geminates (see later slide)• Behavior of “Onsets” (see also next 2 slides)

• “Moraic theorists have argued for their model on the grounds that the weightlessness of onset consonants does not have to be stipulated but is built into the theory…this argument is hard to accept. Both the moraic and the X-slot models exclude onset positions from counting for weight essentially by stipulation” (Kenstowicz 1994:431)

Compensatory Lengthening ICompensatory Lengthening I

Onset CL: Predicted not to exist by MT exists (Samothraki Greek (Newton 1972), Onondaga (Hayes 1989), Kasem

(Rialland 1991), Semitic yuqwalu > yuqa:lu, etc. (O’Leary 116), Bantu CyV CyV:, Romanesco Italian (Kavitskaya 2002), Piro (Lin 1997))

is in fact allowed by Stratal (moraic) OT (Kiparsky 2008) cf onsets can count for weight wrt stress phenomena

• Arabela (Topintzi 2005a), Pirahã (Everett and Everett 1984), Tümpisa Shoshone (Dayley 1989), Banawá (Buller et al. 1993), Nankina (Spaulding and Spaulding 1994), Aranda (Strehlow 1942), Alyawarra (Yallop 1977), Manam (Lichtenberk 1983), Bislama (Camden 1977), Júma (Abrahamson and Abrahamson 1984)

• NB weight systems for stress can have historical/perceptual explanations, as proposed by Gordon 2005—one needn’t build all of the machinery into the synchronic grammar.

Can be accommodated in MT by allowing Onset Cs to be moraic, but this removes a primary claimed advantage of MT and makes it difficult to distinguish Onset vs Coda behavior

Compensatory Lengthening IICompensatory Lengthening II Hasn’t been shown that CL is only triggered by moraic Cs in a

language where WBP is limited to a subset of the C inventory (Kenstowicz 1994:433)

other analyses of compensatory lengthening have been proposed that provide grounded explanations for this phenomenon and do not rely on moraic considerations (Fowler 1983, Blevins and Garrett 1998, Kavitskaya 2002, Campos-Astorkiza 2005)

Levin 1989 does Ponapean CL with X slots; shows it can’t be done with just moras

Conclusions wrt Compensatory Lengthening: Moras are neither necessary (Blevins/Garrett) nor sufficient (Levin) The empirical facts robustly falsify the predictions of MT Subtler predictions of MT haven’t been tested (see also next )

Relevant evidence available Onsets should be irrelevant for template shape (Broselow 1995:194)

• counterevidence: Sloan 1991 empty slots should all be moraic (Broselow 194)

• possible counterevidence: French, Onondaga, Seri geminates should always make weight (vs Tranel 1991, Muller 2001, Campos Astorkiza 2007)

• Selkup (Tranel 1991, Ringen and Vago 2002) has weight-sensitive stress and treats V: as heavy and syllables closed by a consonant or a geminate as light

• Tübatulabal and Malayalam (Tranel 1991), Cypriot Greek (Arvaniti 1991, Arvaniti and Rose 2003), Leti (Hume et al. 1997)• NB light geminates can’t be two root nodes, contra Broselow et al 1997, Ham 1998, Davis 1999 (Arvaniti and Rose 2003)

V: should never be lighter than VC (vs Lahiri and Koreman 1988 on Dutch) V: and VC heavy syllables treated identically (vs Sloan 1991, Shaw 1992) moraic consistency (Broselow 1995)

• Steriade 1991 (Lithuanian), Archangeli 1991, Crowhurst 1991, Lunden 2006 (Norwegian), Walker 1994 (Buriat); surveyed in Gordon 2004 Untested

CL triggered by moraic Cs in lg where WBP is limited to subset of C inventory (previous slide) Word-initial geminates should make their syllable heavy Arabic template priming (Boudelaa and Marslen-Wilson 2004) should treat non-final CVC and CVV syllables identically Long-distance weight shifts should only be possible in lgs where “Coda” Cs don’t make weight More

Some interesting predictions Some interesting predictions of Mora Theory Iof Mora Theory I

Some interesting predictions Some interesting predictions of Mora Theory IIof Mora Theory II

Alignment theory of Onset and NoCoda (M&P 1993, I&M 1994) makes problematic predictions no ? insertion with words beginning with a syllabic C in

English and other such languages problems with syllabification algorithm in languages with

syllabic consonants, such as ITB Hiatus should be allowed in V + syllabic-C sequences

syllabic status of offglides originating in triliteral lexical consonantal roots in Arabic (Watson 2002)

Interim conclusion:No compelling evidence presented for MT

CRONX is CRONX is necessarynecessary

• We’ve seen so far that we need to go back to the drawing board and examine the evidence with a fresh, more critical eye.

• We’ll now look at each putative sub-syllabic element in turn, moving down the structure from O/R to N/C to X/.

Onset-RimeOnset-Rime

Yip 2003: no Onset-Rime boundary Evidence for Onset-Rime structure:

Extensive evidence from reading/processing (Barton et al 1980, Bowey and Francis 1991, Penney 2002…)

Verlan with monosyllables? (e.g. moi wam)Treiman 1983…

BlendsBlends Experiment 1

Question• Do Onsets and Rimes exist (as suggested by e.g. brunch

vs. *blunch)? Method

• Train subjects to combine pairs of well-formed English nonce monosyllables (such as krint and glupth) into a new monosyllable that contains parts of both.

Results• responses like krupth (Onset kr- of the first syllable and

Rime -upth of the second) were produced far more often than any other possible combination.

Conclusion• The natural break within English syllables is immediately

before the vowel (i.e. Onset vs. Rime).

σ σ

O R O R

N C N C

k r i n t g l u p th

Experiments from Treiman 1983

BlendsBlends Experiment 2

Hypothesis• If a syllable is composed of Onset + Rime, then artificial

games that keep these units intact should be easier to learn than games that break up the syllables in a different way.

Method• Subjects taught 2 types of word games:

1.Blend the Onset of a nonce CCVCC syllable with the Rime of another• e.g. fl-irz + gr-uns fl-uns

2.Combine non-constituents (f-runs, flins, flir-s). Results

• Game 1 was learned with fewer errors than was Games 2.Conclusion

• Speakers have access to the constituents O and R.Experiments from Treiman 1983

Interim ConclusionsInterim Conclusions

Clear evidence for distinction between pre-nuclear bit vs the remainder.

But does this require O/R, or can it just be gotten from reference to Vowel or Nucleus? Reference to V of syllable may run afoul

of glides, e.g. wa might be expected to always be treated as complex Nucleus

OnsetOnset Levin 1985, Steriade 1988: the Onset is not a constituent

Its status as a constituent (as revealed by e.g. movement) is of particular interest since this is the one CRONX element not easily capturable in MT

Traditional evidence for Onset constituent is problematic: Onset identification tasks (Treiman 1985, Treiman and Zukowski 1991, Squires

2004) could be affected by coarticulation Onset-Rime effects could be targeting Rime Greek w deletion (Attic, Ionic, Aeolic; Steriade 1982:117-118)

• $_V (woikos ‘house’ > oikos; newos ‘new’ > neos, ksenwos ‘stranger’ > (Att) ksenos)• $_C (wri:nos ‘skin’ > ri:nos, wri:zda ‘root’ > ri:zda)• $C_ (dweyos ‘fear’ > deos, swekuros ‘father-in-law’ > hekuros)• “preserved w only in the weak position of the nucleus’ (basileus ‘king’, trauma ‘wound’)• Good for ruling out pre/post-V and pre/post-C, but one could still say “delete non-

moraic u/w”• prediction: geminate w should survive

Potential evidence: Scobbie 2000 on gestural timing in sty vs dye (cf. Kim 1970 on [sg]) English t/d retroflexion by r (tree, street vs hat rack, intér)

Onset: Interim ConclusionOnset: Interim Conclusion

The Onset constituent can technically be done without, but having it makes our lives much easierCf Steriade 1988 on O/R in later slide

RimeRime Rime more easily detachable than Body (Fowler et al. 1993)

could be attributed to phonetic factors, but this wouldn’t explain the contrast with performance by Japanese and Korean speakers on similar tasks

Klamath deglottalisation: delete [+cg] in Rime (Kingston 1985, Steriade 1988)a. wenwy’tk ‘widow’ wenwitkb. lodgn’aksga lotganksgac. sill’s ‘sickness’ silalsd. gatdk’s ‘cold’ gatdakse. sle?-ca-a ‘goes to see’ sle:caf. hes-sle? ‘show!’ hesle:g. sle?-a ‘sees’ sle?ah. ?o-y’o:q’-a ‘shaves someone’ ?oy’o:q’a Could be “*/delete moraic [+cg]”, assuming that “onset” Cs in Klamath attach

directly to (and all post-vocalic [+cg] to

Rime: English /L/Rime: English /L/ “Thus, in many dialects of English, the L in strings like pill, peal, elm, curl, and boil is dark,

in contrast to the prenuclear L in, say, leg or lawn (Jones 1967, Simpson 1979). If we recognize the onset/rime division, the context where L becomes dark is the same as that where French /n/ is lost: in the rime… One may think that the problem raised by the derivation of dark L's could be obviated by reversing the direction of the rule: we could start with underlying dark L, and derive its light allophone in prevocalic position. Such a statement would, however, run into comparable difficulties: the prevocalic position is itself insufficient to distinguish the dark (nuclear) L of cycling from the light L of cyclic. Nor is it possible to restate the rule by reference to syllable-initial position: the non-initial L of syllables such as play is as light as the initial one in lay. What we need, no matter what allophone we choose to derive by rule, is the ability to refer to the onset/rime distinction: onset L is light, rime L is dark… This is not to say that such facts cannot be described by C&K: the context of the English dark L rule need not be a prosodic category such as Nucleus. It could be a condition, such as: L is tautosyllabic with and preceded by (though not necessarily adjacent to) the vowel. Such a condition, which combines precedence with tautosyllabicity, can replace any reference to Onset or Rime. But it does so at considerable cost.” (Steriade 1988)

NucleusNucleus

Shaw 1992: N required even in MT to distinguish between monomoraic, bimoraic, and variable-weight reduplicative templatesnon-lexical quantitative transfer (Afar, Nisgha)variable template weight (Nootka, Nitinaht, Ojibwa)CV:CV:C vs CVC:V:C templates (Arabic)

Parallel argumentsSloan 1991 (Southern Sierra Miwok)old CVC/CVV templates

Non-lexical quantitative Non-lexical quantitative transfer in Afartransfer in Afar

Shaw 1992 (from M&P 1986 from Bliese 1981)a. usuul u-sus-suul laughb. biyaak bi-yay-yaak hurtc. idigil idi-gig-gil breakd. ‘amm ‘am-‘amm throwe. ess ess-ess take out

analysis: “coda” Cs are moraic word-final moraic C is extraprosodic circumscribe final base failure of distinctive V length of bases in (a-b) results from

template specifying monomoraic Nucleus

CodaCoda

typically replaceable by other generalisations: Rime (Steriade 1988)/moraic (English r-deletion) post-Nucleus (could also be Rime)

• Amahl’s L-deletion (Smith 1973:15)• Spanish s debuccalisation

-final (could also be Rime)• NB this probably doesn’t work for Spanish s debuccalisation, contra

Peperkamp 1997, Morris 1998, Lipski 1999, Wiltshire 1999: vals, Mayans, Felix, carnets

Align-L• Piñeros 2001 on Sp s debuccalisation• has same potential problem with Mayans, etc.

X slotsX slots

X-slot theoryLevin 1983, Lowenstamm and Kaye 1986, Sloan

1991… Evidence:

Polish czy vs trzy (Clements and Keyser 1983, Broselow 1995:179)

Korean poa vs oa (Ahn 2002)SSM templates (Sloan 1991)…

Sloan 1991Sloan 1991

Southern Sierra Miwok contrasts three LH templatesCVCVC - CVCV: - CVCVXMT wrongly predicts all LH templates should behave as

her CVCVX (i.e. [) SSM has -:V and -:C suffixes with floating initial X slots

V-:V V??VC-:C CəəCcan’t be derived with just moras

Is CRONX Is CRONX sufficient?sufficient?

Evidence for morasEvidence for moras mora-counting meters in Sanskrit, Japanese

Sanskrit mātrika meters: Āryā, Udgāthā, Udgīti, Upagīti, and Vaitālīya• e.g. Upagīti: a&c 12, b&d 15• Bhāskara, Lilāvatī 217.2 (NB e and o are heavy):

• kşetra-phalam samam evam / vedha-hatam ghana-phalam spaşţam// Japanese haiku, etc. Can probably be dealt with via brackets a la Halle/Idsardi/Raimy. Better

evidence: certain types of CL

Kihehe and Kimatuumbi iV- yV:- (Odden 1996) Musaler (next slide )

Homshetsma /t:-/ ‘hit’ (slide after next )

Musaler CLMusaler CL

doesn’t make sense with CRONX, but works tolerably with moras/weight tier

Word-final geminates (e.g. vədd ‘foot’)

AnalysisAnalysis

Homshetsma /t:-/ ‘hit’Homshetsma /t:-/ ‘hit’

AnalysisAnalysis

ConclusionsConclusions

Though few of the arguments are profoundly compelling, in order to account for the relevant range of phonological effects we need (C)R(O)NX in the form depicted on the title slideN.B. weight has to be encoded on an autonomous

tier—moras are not the mediators between syllables and segments

• (open Q: encoding of weight via moras vs brackets)