Feature geometry and markedness

Plan for today

• Motivation for feature geometry
• Geometry, visibility and activity
• Linking geometrical structure to markedness and underspecification

Feature geometry: what and why?

Autosegmental phonology: tone spreading

One-to-many tonal mapping in Shona

‘N’	‘with N’	Gloss
mbwá	né-mbwà	‘dog’
hóvé	né-hòvè	‘fish’
mbúndúdzí	né-mbùndùdzì	‘worm’
hákátà	né-hàkàtà	‘bones’
bénzíbvùnzá	né-bènzìbvùnzá	‘fool’
mùrúmé	né-mùrúmé	‘man’
bàdzá	né-bàdzá	‘hoe’

• All H tones in an initial sequence of H become L after a prefix H
• No lowering of non-initial H tones (*mùrúmè)
• We can’t rely on a rule lowering a H after another H: no other way of getting [né-mbùndùdzì]

Place assimilation

• Full place assimilation: [C +nas] → [αcor βant] / _[C αcor βant]
• No [αPlace] feature in standard theories

• Why?
  • Difficult to avoid with binary features
  • Hangover from Jakobson, Fant & Halle (1951), who had an acoustic justification
  • [±lab], [±dor], [±cor] predict rare or unattested phonological classes

Jakobson, Roman, Gunnar Fant & Morris Halle. 1951. Preliminaries to speech analysis. Cambridge, MA: MIT Press.

A bad prediction

• What about C → [αcor] / _[αcor]?
• Should be simple, but it’s a car crash

Input		Output	Input		Output
mt͡ʃ	→	nt͡ʃ	ŋt͡ʃ	→	ɲt͡ʃ
ɲp	→	ŋp	np	→	mp
ɲk	→	ŋk	nk	→	mk
ɲt	→	ɲt	nt͡ʃ	→	nt͡ʃ

A solution

• Assimilation: identity between two feature values

• How do we enforce identity?
  • Traditional approach: agreement (via alpha rules)
  • Autosegmental approach: one-to-many mapping between non-segmental units (tones) and segments

• Extension of autosegmentalism to features: feature geometry

• Features are not rows in a matrix
• They reside on a separate tier and associate to (possibly multiple) segments, just like tones
• Further refinement: nodes unifying several features

A possible geometry

Spreading of nodes

• A node that spreads brings all of its daughters: total place assimilation
• Subnodes may spread independently

Iterative processes: harmony

Turkish vowel harmony

NOM	NOM.PL	GEN.PL	Gloss
jel	jel-ler	jel-ler-in	‘wind’
diʃ	diʃ-ler	diʃ-ler-in	‘tooth’
ɡyl	ɡyl-ler	ɡyl-ler-in	‘rose’
ɡøl	ɡøl-ler	ɡøl-ler-in	‘sea’
kɯz	kɯz-lɑr	kɯz-lɑr-ɯn	‘girl’
dɑl	dɑl-lɑr	dɑl-lɑr-ɯn	‘branch’
kol	kol-lɑr	kol-lɑr-ɯn	‘arm’
kul	kul-lɑr	kul-lɑr-ɯn	‘slave’

• Iterative processes like harmony work exactly like the spreading of tone

The Line Crossing Condition

• Rounding harmony cannot skip non-round vowels, because the result would violate the Line Crossing Condition
• The LCC basically states that association lines do not cross
• More formally: if a segment follows the start of an association domain and precedes its end, it is within that domain

Subnode spreading: Meadow Mari

NOM.SG	POSS.3SG	Gloss
ola	olaʒɛ	‘city’
tɛr	tɛrʒɛ	‘sledge’
kit	kitʃɛ	‘hand’
lym	lymʒø	‘name’
pørt	pørtʃø	‘house’
oŋ	oŋʒo	‘breast’
myj	myjʒø	‘honey’
tul	tulʒo	‘fire’
kol	kolʒo	‘fish’

• [±bk] and [±rd] spreading together
• Could be analysed using a Colour node

Other uses

• Under-use of features
  • The feature [±anterior] is basically useful only for the alveolar/postalveolar contrast in coronals, so lives under the Coronal node
• Unified Feature Theory (Clements 1991; Clements & Hume 1995) and cognate theories (Morén 2003): single set of features for consonants and vowels
  • Dorsal = [+back]
  • Coronal = [-back] (Hume 1996)
  • Consonant-vowel place assimilation as spreading

Clements, G. Nick. 1991. Place of articulation in consonants and vowels: A unified theory. Working papers of the Cornell Phonetics Laboratory, vol. 5. Ithaca, NY.

Clements, G. Nick & Elizabeth V. Hume. 1995. The internal organization of speech sounds. In John Goldsmith (ed.), The handbook of phonological theory, 245–306. Oxford: Blackwell.

Morén, Bruce. 2003. The Parallel Structures model of feature geometry. Working Papers of the Cornell Phonetics Laboratory, vol. 15, 194–270. Ithaca, NY.

Hume, Elizabeth. 1996. Coronal consonant, front vowel parallels in Maltese. Natural Language & Linguistic Theory 14(1). 163–203.

Quick summary

• Separation of skeleton and subsegmental representation
• Many-to-one mappings
• Link between presence of autosegmental structure and phonological behaviour

Geometry, visibility and activity

A geometry for place features

The following is based on (Avery & Rice 1989).

Avery, Peter & Keren Rice. 1989. Segmental structure and coronal underspecification. Phonology 6(2). 179–200.

The Node Activation Condition

If a secondary content node is the sole distinguishing feature between two segments, then the primary feature is activated for the segments distinguished. Active nodes must be present in underlying representation.

Catalan place assimilation

• Much like English, coronal nasals assimilate freely but non-coronal nasals resist assimilation

/son/	/som/	/tiŋ/	before…
[n]	[m]	[ŋ]	amics ‘friends’
[m]	[m]	[ŋ]	pocs ‘few’
[n̪]	[m]	[ŋ]	dos ‘two’
[n]	[m]	[ŋ]	sincers ‘sincere’
[ŋ]	[m]	[ŋ]	grans ‘big’

Analysis: major place

• Underlyingly, coronals are [Place], labials are [Place [Labial]], dorsals are [Place [Dorsal]]
• Standard autosegmental assumption: spreading only occurs into an empty position

Coronal → Dorsal assimilation

No Labial → Dorsal assimilation

Coronals are not triggers

Analysis: coronals

• Coronal is only absent underlyingly: /t/ and /s/ differ in [±dist] but distinguished by [Continuant]
• [Stop] → [+dist] by a postlexical default rule
• We now must have a Coronal node
• The two Coronal nodes undergo fusion

Coronal fusion

Pohnpeian

• Heterorganic clusters: epenthesis
  • /kitik-men/ → [kitikimen] ‘rat’
  • /ak-suwei/ → [akusuwei] ‘demonstrating boastfulness’
• Homorganic non-coronal clusters: nasal substitution
  • /kehp-mʷot/ → [kehmʷmʷot] ‘variety of yam’
  • /ak-keelail/ → [aŋkeelail] ‘demonstrate strength’

Pohnpeian: nasal substitution

Nasal substitution applies when one of the Place nodes is fused

Pohnpeian: no nasal substitution with coronals

• Coronal clusters: epenthesis
  • /weid-da/ → [weidida] ‘proceed upwards’
  • /lus-saŋ/ → [lusisaŋ] ‘jump from’
• Nothing to fuse below Place, because in Pohnpeian Coronal is never inserted, not even postlexically
  • /e kalap pahn soupisek/ ‘he will always be busy’ → [kalam pahn]
  • /ke meid daŋahŋa/ ‘aren’t you lazy!’ → *[mein daŋahŋa]

Sanskrit nati

Place
retroflex	ʈ	ʈʰ	ɖ	ɖʰ	ɳ	r	ʂ
dental	t	tʰ	d	dʰ	n	l	s
palatoalveolar	c	cʰ	ɟ	ɟʰ	ɲ		ʃ

• Sanskrit has a rich Coronal inventory, so that node must be present underlyingly
• Dentals are [Coronal], retroflexes and palatoalveolars are [Coronal + feature]

• n \(\rightarrow\) ɳ / {r, ʂ} _ across non-coronals
  • pur-āɳa- ‘filled’
  • kʂubʰ-āɳa- ‘quaked’ (non-coronal transparency)
  • marɟ-āna- ‘wiped’ (coronal blocking)

Analysis

Spreading is local

Non-coronals are transparent because [retroflex] must be on a Coronal tier

Blocked spreading

Marked coronals do block spreading, because of LCC

Interim summary

• Feature geometry expresses locality and visibility effects
• Presence of structure allows
  • Triggering of spreading
  • Non-targeting by spreading
  • Blocking in long-distance processes
• There are all markedness diagnostics

Size and markedness

A refresher: Yakkha place

Infinitive	3SG.PST	Gloss
lapma	labana	‘seize’
apma	abana	‘come’
jokma	joɡana	‘search’
pʰaʔma	pʰatana	‘help’
keʔma	ketana	‘bring up’
liʔma	litana	‘plant’
tʰuʔma	tʰurana	‘sew’
poʔma	porana	‘topple’

The coda inventory

labial = dorsal » coronal » glottal

An analysis

Based on the proposals in Causley (1999). The Peripheral node is justified by other kinds of data, notably from vowels (Rice 2002)

Rice, Keren. 2002. Vowel place contrasts. In Mengistu Amberber & Peter Collins (eds.), Language universals and variation, 239–270. Wesport, CT: Praeger.

• Coda neutralization: delink Place
• Preservation of the Marked: maintain Peripheral
• This works well in OT

Markedness and size

• ‘More marked’ literally means ‘bigger’
• Preservation of the Marked: more chances to single out a piece of structure for preservation
• Submergence of the Unmarked:
  • Cannot trigger if you don’t have structure
  • Easily targeted if you accept spreading
• Emergence of the Unmarked
  • Reduction produces lack of structure

Markedness orders and xo Theory

• xo Theory (Lacy 2006)
  • Glottal is [oooPlace]
  • Coronal is [xooPlace]
  • Dorsal is [xxoPlace]
  • Labial is [xxxPlace]
• Similar size effects emerge in OT, as Causley (1999) shows for the autosegmental option

Lacy, Paul de. 2006. Markedness: Reduction and preservation in phonology. Cambridge: Cambridge University Press.

Causley, Trisha. 1999. Complexity and markedness in Optimality Theory. Toronto: University of Toronto PhD thesis.

Summary

• We’re all the way back to the Merkmal
• Underspecification and size explain why markedness diagnostics work the way they do
• Tomorrow: where does geometrical structure come from?