Class 3: Contrast and neutralization (part 2) Adam Albright - - PowerPoint PPT Presentation

Class 3: Contrast and neutralization (part 2)

Adam Albright (albright@mit.edu)

LSA 2017 Phonology University of Kentucky

Goals for today

▶ Continue discussion of markedness and contrast

▶ Licensing by cue: another example where neutralization

correlates with diminished cues

▶ Constraining perceptual distance directly: dispersion theory

▶ Reminders

▶ Assignment 1 for “option 1” due on Monday (7/17) by PDF on

Canvas

▶ My office hrs: Thurs 2-3pm (library basement, the Hub) and by

appointment

▶ I’m happy to meet and talk about class or about your work!

References 1/29

So far…

▶ A framework characterizing phonological restrictions

▶ Markedness constraints: *[+spread glottis] (no aspirated

segments), or *[spread glottis] (no s.g. specification)

▶ Contextual markedness: *[+spread glottis]/

[−sonorant]

▶ Faithfulness: Ident([±spread glottis])

▶ A result: with ranking and candidate competition, this system

allows us to characterize phonological distributions

▶ Contrast ▶ Lack of contrast ▶ Contextual neutralization

▶ An assumption (to be discussed more in class 8)

▶ All human grammars contain or are drawn from the same

constraints

▶ Defines a space of possible languages ▶ Reasoning about the constraint set: undergeneration,

• vergeneration

References 2/29

The wide world of phonological restrictions

Languages of the world exhibit numerous types of phonological restrictions

▶ Contrast vs. lack of contrast ▶ Positional neutralization

▶ No voiced obstruents word-finally (“final devoicing”) ▶ No voiceless obstruents after nasals (“post-nasal voicing”) ▶ No vowels other than [ə] in stresless syllables (“vowel reduction”)

▶ Assimilation and harmony

▶ No voiceless obstruents before a voiced obstruent, and vice versa

(“voicing assimilation”) ▶ Prosody: stress, tone, etc.

It would not be possible to do justice to surveying such restrictions in a class of this length.1 Instead, we will focus here on a few selected restrictions, and how analysts haved reasoned about their analysis.

1A nice set of summary articles can be found in the Blackwell Companion to

Phonology.

References 3/29

Reminder: licensing by cue

▶ Last time: two restrictions on voicing in Lithuanian

▶ Positional restriction: contextual neutralization word-finally ▶ Assimilation: obstruents must agree in voicing with a following

• bstruent

▶ What unifies these two restrictions

▶ Contrast is limited to pre-sonorant positions

References 4/29

Reminder: licensing by cue

▶ The ‘licensing by cue’ hypothesis grounds the set of markedness

constraints in considerations of perception (Steriade, 1997)

▶ Pay-off: derive asymmetries (contrast in a perceptually

disadvantageous context implies contrast in perceptually more advantageous context)

▶ Interpretation of markedness as penalizing any specification

directly captures insight that the markedness eliminates contrasts (not specific feature values)

▶ Caveat: both of these hypotheses are controversial ▶ Alternative: find other ways to limit set of markedness

constraints, or simply allow a wide range of markedness constraints and seek typological explanations elsewhere (class 8)

References 5/29

Contextual neutralization: place

Navajo also shows neutralization for place of articulation in final position

▶ Final t, ʔ but no *p, *k ▶ Same questions as above

▶ Why does neutralization target some contexts

(word-final/pre-pausal), but not others (word-initial/pre-vocalic)?

▶ When place is neutralized, why are t,ʔ favored?

References 6/29

Contexts for neutralization

Jun (1995, 2004)

▶ Place contrasts are typically maximal before vowels and

sonorants, reduced before obstruents and word-finally E.g., English #C

▶ Stop place fully contrastive before vowels and ɹ, limited before l,

banned before nasals and obstruents

▶ Fricative place fully contrastive before vowels, limited before ɹ and

l, only [s] before nasals and obstruents ▶ Cues to place

▶ Formant transitions in preceding, following vowel (V2 privilege;

Fujimura et al., 1978)

▶ Stops and nasals: release (strongest before V, also prominent

finally in some lgs, weak or absent before C)

▶ Nasals: frequency of nasal resonances ▶ Liquids, glides: formants

References 7/29

References 8/29

Example: nasal place contrasts

Nasal place contrasts: a mini-typology (de Lacy 2002) V # C (V)mV∼(V)nV Vm∼Vn VmtV∼VntV Japanese, Spanish

✓

* (neutr.) * (assim.) Latin, Diola Fogny

✓ ✓

* (assim.) Russian, English

✓ ✓ ✓

▶ Neutralization targets worse-cued positions over better-cued

positions

▶ Worst: no following transitions and no release ▶ Better: no following transitions, but audible release ▶ Best: following transitions and release

References 9/29

Not all C2’s are equal

▶ Although consonants/

C2 are generally prone to place assimilation, some places of articulation are more likely trigger assimilation than others

▶ Dorsals > Labials > Coronals ▶ Korean place assimilation

▶ Coronals assimilate to labials and dorsals, and not vice versa ▶ Labials assimilate to Dorsals (for some speakers, esp. in casual

speech) and not vice versa

+ e + to + pota + kwa

‘loc’ ‘too’ ‘more than’ ‘and’ /mitʰ-/ ‘bottom’ mitʰe mi(t)t’o mipp’oda mikkwa /apʰ-/ ‘front’ apʰe apt’o app’oda akk’wa /sok-/ ‘inside’ soge sokt’o sokp’oda sokk’wa

+ ə/a + ta + ko

‘Inf’ ‘decl’ ‘and’ /mit/ ‘believe’ midə mi(t)t’a mikk’o /ip-/ ‘wear’ ibə ipt’a ikk’o /mək-/ ‘eat’ məɡə məkt’a məkk’o

References 10/29

Latin place assimilation

▶ Coronals assimilate to labials and dorsals

ad- ‘towards’

/ [lab] ap-par-ere ‘gain in addition’ ap-pend-ere ‘hang upon’ ap-plaud-ere ‘strike upon’ ap-prim-ere ‘press close’ ab-brewi-aːre ‘shorten’ / [dors] ak-kept-aːre ‘take’ ak-klaːm-aːre ‘call to’ ak-kliːn-aːre ‘lean on’ ag-gluːtin-aːre ‘glue to’ ag-ger-ere ‘carry towards’

▶ Labials assimilate to dorsals (but not coronals)

sub- ‘under’

/ [cor] sub-teg-ere ‘cover beneath’ sub-tend-ere ‘stretch beneath ’ sub-deːlig-ere ‘choose’ sub-dubit-aːre ‘be a little doubtful’ sub-dok-eːre ‘teach as an assistant’ / [dors] suk-kend-ere ‘kindle beneath’ suk-kiːd-ere ‘cut below’ sug-ger-ere ‘carry below’ sug-gluːt-iːre ‘hiccup a little’ sug-grunn-iːre ‘grunt a little’

▶ Not just verbs: /sub-grundaːri-um/ → suggrundaːrium ‘grave of a

child less than 40 days old’

References 11/29

Jun’s hypothesis

▶ Susceptibility to assimilation correlates with gestural duration

▶ Dorsal > Labial > Coronal

▶ T

argets: shorter gestures are more significantly affected/eclipsed by adjacent gestures

▶ Triggers: longer gestures are more likely to affect/eclipse an

adjacent gesture

References 12/29

The consequence for markedness constraints

▶ A (non-necessary) assumption about features

▶ [PLACE] is a feature, can have values [dorsal], [labial], [coronal] ▶ Equivalent: feature geometric interpretation (cf. LARYNGEAL)

▶ Fixed rankings reflecting cue availability

▶ *PLACE/

C ≫ *PLACE/ # ≫ *PLACE/ [+son] ≫ *PLACE/ V

▶ *PLACE/

[−syl,dors] ≫ *PLACE/ [−syl,lab] ≫ *PLACE/ [−syl,cor] ▶ Implicational relations

▶ Ident(PLACE) may be ranked at various points along hierarchy ▶ Assimilation before labials implies assimilation before dorsals

▶ Why assimilation?

▶ Steriade’s proposal for voicing is useful here too:

underspecification

▶ Consonant with no independent place specification is realized with

articulatorily simplest interpolation: same gesture as adjacent consonant

References 13/29

Additional asymmetries

Place assimilation is often limited to particular classes of segments, confirming that less well cued contrasts are dispreferred

▶ Nasals more likely to assimilate than stops

▶ Formant transitions in preceding vowel are less clear when it is

nasalized

▶ Better cues for oral stops than for nasal stops ▶ *[PLACE,+nas]/

X ≫ *[PLACE,−nas]/ X ▶ Stops more likely to assimilate than Fricatives

▶ Both stops and fricatives have external cue of preceding formant

transitions

▶ Fricatives have additional internal cue: frequency distribution of

aperiodic noise

▶ *[PLACE,−continuant]/

X ≫ *[PLACE,+continuant]/ X

References 14/29

More generally: eliminating some places, favoring others

What about cases where we see a reduced range of contrasts, but not total assimilation to a following consonant? Which places of articulation are more marked?

▶ *[PLACE] won’t do the trick

▶ Just yields one outcome, perhaps [ʔ] (no oral stricture)2

▶ Markedness constraints for specific places?

▶ *[+labial], *[+coronal], *[+dorsal], etc. ▶ Does not predict any implicational asymmetries ▶ Is it a coincidence that only t, ʔ are tolerated?

▶ Cue-based constraints?

▶ Favor places with longer/more robust place cues, such as dorsals? ▶ Favor contrasts that are more distinct from each other, such as

dorsal vs. labial? ▶ Potential sources of evidence for asymmetries

▶ Language-internal asymmetries ▶ Cross-linguistic asymmetries

2Or, sometimes assumed to yield default coronal… References 15/29

Language-internal place asymmetries

Three relevant configurations

▶ Languages with limited sets of place contrasts

▶ T

agalog contrasts p,t,k,ʔ in initial, medial, and final positions

▶ English contrasts p,t,k (i.e., *ʔ) ▶ Seneca contrasts t,k (*p, *ʔ) ▶ Hawaiian has only k,ʔ (*p, *t)

(We’ve seen how to handle these with markedness constraints)

▶ Within a language, place contrasts may be restricted in specific

contexts

▶ Phonological: Navajo lacks *p, *k in final position (We’ve seen how

to handle these with contextual markedness constraints)

▶ Phonological: epenthesis ▶ Morphological: restrictions in affixes, etc. (next time)

The hope: comparing these restrictions might reveal systematic and consistent markedness asymmetries

References 16/29

Epenthetic segments

▶ Epenthetic segments violate Dep ▶ Since there is no underlying segment, they don’t violate Ident

(i.e., dont change the place of any segment in the input)

▶ Choice of epenthetic segment therefore falls to markedness

▶ The Emergence of the Unmarked (TETU)

▶ Example: epenthetic [ʔ] before word-initial vowels

/ap/ *#V Ident(place) *[coronal] *[glottal] a. ap *! W L b. tap *! W L

☞

c. ʔap * /ta/ *#V Ident(place) *[coronal] *[glottal]

☞

a. ta * b. ʔa *! W L * W

▶ Epenthetic ʔ = *other ≫ *glottal

References 17/29

Inventory and contextual asymmetries

Some contradictions for a general hierarchy of place markedness

▶ Hawaiian stops: only k,ʔ

▶ *[labial] ,*[coronal] ≫ Ident(place) ≫ *[dorsal]

▶ Spanish stops: p,t,k

▶ *ʔ ≫ *[labial] ,*[coronal], *[dorsal]

▶ German epenthetic stops: ʔ

▶ *[labial] ,*[coronal], *[dorsal] ≫ *ʔ

▶ Navajo word-final stops: only t,ʔ

▶ *[labial]#, *[dorsal]# ≫ Ident(place) ≫ *[coronal]#, *ʔ#

No simple convergence! Even within inventories, there are contradictions, and epenthesis favors segments that are often disfavored (h, ʔ)

References 18/29

Contextual restrictions

A further example, involving nasal place neutralization

▶ Neutralization before C results in assimilation ▶ Neutralization of nasal place in final position may favor [ɴ]

(Japanese), [ŋ] (Dominican Spanish), [n] (Castilian Spanish), perhaps even [m]?

▶ This is just like the problem we saw with voicing: a value that’s

marked in general may be favored in a specific context

▶ A possible start: *PLACE/

¬[+son] favors underspecification.

But then all languages should be like Japanese.

▶ Other outcomes must be favored by markedness contraints, but

no consistent ranking?

References 19/29

One solution: dispersion

▶ No general hierarchy that singles out specific places of

articulation as universally more marked than others

▶ Markedness penalizes perceptually indistinct contrasts

▶ i.e., [t] is difficult to distinguish from [k], or [ʔ] (or other stops

more generally), so may be marked in contrast with those

▶ But it is tolerated or even preferred when there are sufficient cues,

• r when there is no contrast

▶ This allows for a fairly wide range of ‘optimal’ small inventories.

Once you are down to just one or two places, it’s easy to be distinct. ▶ Faithfulness favors perceptually minimal changes

▶ Unlike small inventories and contextual neutralization, epenthetic

segments do show a fairly high degree of consistency in place (glottal h,ʔ; homorganic glides)

▶ We’ll come back to this next week, and attribute it to the nature of

Dep: insert perceptually unobtrusive segments

References 20/29

Dispersion theory

Flemming (2004, 2006)

▶ An observation about vowel inventories: vowels tend to be

spaced (more or less) regularly throughout vowel space

▶ Schematically: some 3 and 5 vowels inventories

2200 1800 1400 1000 700 600 500 400 F2 (Hz) F1 (Hz) i a u 1000 1400 1800 700 600 500 400 F2 (Hz) F1 (Hz) ɨ ə a 2200 1800 1400 1000 700 600 500 400 F2 (Hz) F1 (Hz) i e a

• u

▶ A recurring pattern: segments that are marked in some context

are preferred in others

▶ Inventories with [i] and [u] frequently lack [ɨ] ▶ Inventories without front/back contrast prefer [ɨ]3

3In such systems, [ɨ] typically varies in backness depending on context; but this is

typical of systems with few vowels.

References 21/29

Flemming (2004): schematic vowel space

▶ Discretely quantized dimensions

(6) a. F2 6 5 4 3 2 1 i i y   u 1    2 e ø 

• 

3 e ø  

• 4 F1

    5 æ   6 a a 7

▶ Grammar candidates of contrasting elements

▶ E.g., a vowel inventory: ɨ-ə-a

▶ MinDist constraints penalize contrasts that are too close on a

specified dimension: e.g., MinDist = F1:2

▶ Intrinsic ranking(?): smaller distances always penalized more

▶ MaximizeContrasts: penalizes neutralization (fewer elements in

candidate than input)

▶ Relative ranking determines size of inventory, and optimal

dispersion

References 22/29

MinDist constraints

▶ MinDist = dim:dist

▶ For all pairs of segments x,y: assess a violation if x,y are not at

least distance dist apart on dimension dim

▶ Specific values vs. stringency relations (dim≥dist)

▶ Dimensions: F1, F2, duration, etc.

▶ Approach is compatible with any dimension that can distinguish

distances (auditory, abstract phonological)

▶ Flemming (2004) chooses auditory distance to account for

dispersion along multiple ‘featural’ dimensions (height, backness, rounding)

▶ However, it is also possible to specify distances in multiple

dimensions (which we’ll do here)

References 23/29

Defining an inventory: i, a, u

▶ ‘Point’ vowels differ maximally in F1, F2 ▶ Criterion: MinDist:F1=6 or F2=5 ▶ Assessing violations

▶ If we just give the grammar single forms or vowels, we won’t know

whether there’s another vowel out there that’s too close

▶ Solution: evaluate entire inventories

▶ Example

MinDist:F1=6∨F2=5

☞ a. i a u

✓

• b. i e a o u

*!***

i,e:∆F1=3,∆F2=1 e,a:∆F1=3,∆F2=2 a,o:∆F1=3,∆F2=2

• ,u:∆F1=3,∆F2=0
• c. i ɐ u

*!*

i,ɐ:∆F1=5,∆F2=3 ɐ,u:∆F1=5,∆F2=2

(6) a. F2 6 5 4 3 2 1 i i y   u 1    2 e ø 

• 

3 e ø  

• 4 F1

    5 æ   6 a a 7

References 24/29

Defining an inventory: i, e, a, o, u

▶ By itself, this system won’t ever let us choose an inventory with

contrasts that are closer than the maximally dispersed ones MinDist: MinDist: F1=3∨F2=4 F1=6∨F2=5

☞ a. i a u

✓ ✓

• b. i e a o u

✓

*!***

(6) a. F2 6 5 4 3 2 1 i i y   u 1    2 e ø 

• 

3 e ø  

• 4 F1

    5 æ   6 a a 7

▶ Larger inventories are favored by MaxContrasts

▶ Given: an input of logically possible contrasts ▶ Ranking determines how many are preserved

/i e ɛ a ɔ o u/ MinDist: Max MinDist: F1=3∨F2=4 Contrasts F1=6∨F2=5

• a. i a u

✓

***!*

✓

☞ b. i e a o u

✓

** ****

• c. i e ɛ a ɔ o u

****

✓

********

References 25/29

Which dimensions?

▶ Languages may differ in which MinDist constraints are highest

ranked

▶ F2 distinctness ≫ F1 distinctness: vertical vowel system

▶ I.e., no front/back distinction at all

▶ F1 distinctness ≫ F2 distinctness: ‘point’ vowel system

▶ Systems like i,u,a

▶ Side note: no strictly horizontal systems?

References 26/29

Application to place contrasts

▶ Acoustic dimensions for stops

▶ Preceding (VC), following (CV) formant transitions (if present) ▶ Release burst (if present)

▶ Possible restrictions:

▶ MinDist:CV transitions≥x

Requires following formant transitions; bans place contrasts before a C or word-finally

▶ MinDist:CV transitions≥x or VC transitions≥x+1

Allows place contrasts before or after a vowel, but more stringent condition after a vowel (recall Navajo)

References 27/29

Summing up

▶ Three different ways of formulating restrictions

▶ Markedness constraints on feature values: *[+voice] ▶ Markedness constraints on feature specifications in perceptually

weak positions (Licensing by cue)

▶ Markedness constraints contrasts (Dispersion: MinDist)

▶ Arguments against marked feature values

▶ No intrinsic predictions about contexts where a particular feature

value will be marked

▶ Value that’s marked in some contexts is preferred in other

contexts ▶ Perceptually grounded accounts make testable predictions about

contexts for neutralization

▶ Implicational asymmetries: contrast in weakly cued position

implies contrast in strongly cued position ▶ Dispersion account helps to resolve contradictions in which

values are favored

References 28/29

References

Flemming, E. (2004). Contrast and perceptual distinctiveness. In B. Hayes,

• R. Kirchner, and D. Steriade (Eds.), Phonetically-Based Phonology, pp. 232–276.

Cambridge University Press. Fujimura, O., M. J. Macchi, and L. A. Streeter (1978). Perception of stop consonants with conflicting transitional cues: A cross-linguistic study. Language and Speech, 337–346. Jun, J. (1995). Perceptual and articulatory factors in place assimilation: an Optimality Theoretic approach:. Ph. D. thesis, UCLA. Jun, J. (2004). Place assimilation. In B. Hayes, R. Kirchner, and D. Steriade (Eds.), Phonetically-Based Phonology, pp. 58–86. Cambridge University Press. Steriade, D. (1997). Phonetics in phonology: The case of laryngeal neutralization. UCLA ms.

References 29/29