Remarks on Autosegmental Representations Jeffrey Heinz and Adam - - PowerPoint PPT Presentation

remarks on autosegmental representations
SMART_READER_LITE
LIVE PREVIEW

Remarks on Autosegmental Representations Jeffrey Heinz and Adam - - PowerPoint PPT Presentation

Remarks on Autosegmental Representations Jeffrey Heinz and Adam Jardine (PhD expected Spring 2016) UC Berkeley, Phonology Phorum May 4, 2015 1 Questions about Autosegmental Representations (ASRs) 1. Phonologists often represent words with


slide-1
SLIDE 1

Remarks on Autosegmental Representations

Jeffrey Heinz and Adam Jardine (PhD expected Spring 2016)

UC Berkeley, Phonology Phorum May 4, 2015

1

slide-2
SLIDE 2

Questions about Autosegmental Representations (ASRs)

  • 1. Phonologists often represent words with graph structures, but

what kinds of graphs are they?

  • 2. What aspects of these representations are language-specific and

what are universal?

2

slide-3
SLIDE 3

Today’s Focus

  • 1. How are tonal melodies associated to the timing tier?

Different answers

  • 1. Left-to-right and right-to-left association conventions (Leben

1973, Goldsmith 1976, inter alia)

  • 2. Optimal satisfaction of universal, violable constraints whose

prioritization is language-specific (Zoll 2003)

3

slide-4
SLIDE 4

The proposal:

  • Constraints govern how tonal melodies associate to the timing
  • tier. These constraints are:
  • 1. language-specific,
  • 2. inviolable,
  • 3. and local.
  • We will illustrate with reference to Mende, Hausa, Kukuya,

and Northern Karanga (a Shona dialect).

4

slide-5
SLIDE 5

What ‘local’ means

The well-formedness of a structure can be determined solely on the basis of all of its local sub-structures.

What ‘local sub-structure’ means

A sub-structure is local if it fits inside a sphere whose diameter we fix in advance. (DRAW PICTURE HERE)

5

slide-6
SLIDE 6

Strictly Local constraints for strings

When words are represented as strings, local sub-structures are sub-strings of a certain size. Here is the string abab. If we fix a diameter of 2, we have to check these substrings.

  • k?
  • k?
  • k?
  • k? ok?

a ⋊ a b b a a b b ⋉ An ill-formed sub-structure is forbidden. (Rogers and Pullum 2011, Rogers et al. 2013)

6

slide-7
SLIDE 7

Strictly Local constraints for strings

When words are represented as strings, local sub-structures are sub-strings of a certain size.

  • We can imagine examining each of the local-substructures,

checking to see if it is forbidden or not. The whole structure is well-formed only if each local sub-structure is. b a b a b a a a a b

... ...

b (Rogers and Pullum 2011, Rogers et al. 2013)

7

slide-8
SLIDE 8

Strictly Local constraints for strings

When words are represented as strings, local sub-structures are sub-strings of a certain size.

  • We can imagine examining each of the local-substructures,

checking to see if it is well-formed. The whole structure is well-formed only if each local sub-structure is. b a b a b a a a a b

... ...

b (Rogers and Pullum 2011, Rogers et al. 2013)

8

slide-9
SLIDE 9

Strictly Local constraints for strings

When words are represented as strings, local sub-structures are sub-strings of a certain size.

  • We can imagine examining each of the local-substructures,

checking to see if it is well-formed. The whole structure is well-formed only if each local sub-structure is. b a b a b a a a a b

... ...

b (Rogers and Pullum 2011, Rogers et al. 2013)

9

slide-10
SLIDE 10

Examples of Strictly Local constraints for strings

  • *aa
  • *ab
  • *NC

˚

  • NoCoda

Examples of Non-Strictly Local constraints

  • *s. . . S (Hansson 2001, Rose and Walker 2004, Hansson 2010,

inter alia)

  • *#s. . . S# (Lai 2012, to appear, LI)
  • Obligatoriness: Words must contain one primary stress (Hayes

1995, Hyman 2011, inter alia).

10

slide-11
SLIDE 11

Simple Mende

(1) Mende word tone (Leben, 1973, 1978)

  • a. k´

O ‘war’

  • b. p´

El´ E ‘house’

  • c. h´

aw´ am´ a ‘waist’

  • d. kp`

a ‘debt’

  • e. b`

El` E ‘pants’

  • f. kp`

ak` al` ı ‘three-legged chair’

  • g. mbˆ

u ‘owl’

  • h. ng´

ıl` a ‘dog’

  • i. f´

el` am` a ‘junction’

  • j. mbˇ

a ‘rice’

  • k. n`

ık´ a ‘cow’

  • l. nd`

av´ ul´ a ‘sling’

  • m. mb˝

a ‘companion’

  • n. ny`

ahˆ a ‘woman’

  • . n`

ık´ ıl` ı ‘groundnut’

11

slide-12
SLIDE 12

Simple Mende surface tone patterns

(2) H HH HHH L LL LLL F HL HLL R LH LHH R-F LF LHL

Left to right association (Leben 1973)

(ILLUSTRATE WITH HL MELODY AND σσσ) The analysis, among other things, accounts for the absence of surface HHL (in Simple Mende).

12

slide-13
SLIDE 13

It is 1973. A Hyperbolic Theory of Tone (Version 1)

  • In ALL languages, tonal melodies associate in left-to-right

fashion as in Simple Mende.

13

slide-14
SLIDE 14

Overview of Issues

  • 1. Right-to-left association in Hausa
  • 2. Quality-dependent spreading in Kukuya
  • 3. Accent-like effects (actual Mende)
  • 4. Edge-in association in Northeren Karanga

(Zoll 2003)

14

slide-15
SLIDE 15

Graphs

Graphs are labeled nodes connected by labeled edges.

H L µ µ a b c

Earlier work

  • Previous research took a ‘top-down’, axiomatic approach to

understanding ASRs (Goldsmith, 1976; Bird and Klein, 1990; Coleman and Local, 1991; Kornai, 1995).

  • Jardine and Heinz (to appear, MOL) provide a ‘bottom-up’

approach to ASRs by concatenating graph primitives (cf. Engelfriet and Vereijken 1997).

15

slide-16
SLIDE 16

String graphs

String graphs are “chains” of labeled nodes, where the edge represents the successor relation. a b a b

Axioms for ASRs

  • 1. ASRs contain two string graphs (for melodic and timing tiers),

with association edges connecting nodes of one string graph to another.

  • 2. NCC: If x precedes y on the timing tier than the elements

associated to x on the melodic tier must precede elements associated to y on the melodic tier.

  • 3. OCP: Elements in the successor relation on the melodic tier

cannot be identical.

16

slide-17
SLIDE 17

Example ASRs

H σ σ σ H L σ σ σ L H L σ σ

Example graphs that are not ASRs

H L µ µ a b c H H σ σ σ H L σ σ σ

17

slide-18
SLIDE 18

ASRs for Simple Mende

(3) g(H) =

H σ

g(HH) =

H σ σ

g(HHH) =

H σ σ σ

g(F) =

H L σ

g(HL) =

H L σ σ

g(HLL) =

H L σ σ σ

18

slide-19
SLIDE 19

More ASRs in Simple Mende

(4) g(C) =

L H L σ

g(LF) =

L H L σ σ

g(LHL) =

L H L σ σ σ

19

slide-20
SLIDE 20

Forbidding HLH

(5) φHLH =

H L H

Example

*

L H L H σ σ σ σ

*g(LHLH)

20

slide-21
SLIDE 21

Forbidding nonfinal contours

(6) φNF-Cont =

H L σ σ

L H σ σ

Example

*

L H L σ σ σ σ σ σ

*g(RL) *

H L σ σ σ σ σ σ σ

*g(FLL)

21

slide-22
SLIDE 22

The last piece: directionality in Simple Mende

(7) a.φNF-H2 =

H L σ σ

b.φNF-L2 =

L H σ σ

(cf. Zhang 2000)

Example

(8)

  • a. *

H L σ σ σ σ σ σ σ σ σ

*g(HHHLL)

(Later on we shall see with Kukuya that languages may pick one or the

  • ther—this corresponds to Zoll (2003)’s notion of spreading that is

‘dependent on tone quality’.)

22

slide-23
SLIDE 23

Summary of Simple Mende Analysis

¬φHLH ∧ ¬φNF-Cont ∧ ¬φNF-H2 ∧ ¬φNF-L2 Interpret ¬φ as “The structure φ is forbidden.”

23

slide-24
SLIDE 24

Right-to-left association in Hausa

(9) Hausa word tone

  • a. j´

aa ‘pull’

  • b. j´

ır´ aa ‘wait for’

  • c. b´

eeb´ ıy´ aa ‘deaf mute (fe

  • c. w`

aa ‘who?’

  • d. m`

ac` e ‘woman’

  • e. z`

amf` ar` a ‘Zamfara’

  • f. j`

aak´ ıi ‘donkey’

  • g. j`

ım` ın´ uu ‘ostriches’

  • h. b`

abb` abb` ak´ u ‘be well roaste

  • i. f´

aad` ı ‘fall’

  • j. h´

ant´ un` aa ‘noses’

  • k. b´

uh´ unh´ un` aa ‘sacks’

  • l. mˆ

ant´ a ‘forget’

  • m. k´

ar` ant´ a ‘read’

  • n. k´

akk´ ar` ant´ a ‘reread’

24

slide-25
SLIDE 25

Hausa surface tone patterns

(10) H HH HHH L LL LLL LH LLH LLLH HL HHL HHHL FH HLH HHLH

Right-to-left association

(ILLUSTRATE WITH HL MELODY AND σσσ)

25

slide-26
SLIDE 26

The Hyperbolic Theory of Tone (Version 2)

  • In ALL languages, tonal melodies associate either left-to-right

OR right-to-left.

26

slide-27
SLIDE 27

Well-formed Hausa words

(11) g(HLH) =

L H L σ σ σ

g(HHL) =

H L σ σ σ

g(FL) =

H L H σ σ

g(HHLH) =

L H H σ σ σ σ

27

slide-28
SLIDE 28

Forbidding sub-structures in Hausa

(12) a.φNI-Cont =

H L σ σ

b.φNI-L2 =

H L σ σ

c.φNI-H2 =

L H σ σ

(13)

  • a. φHLHL =

H L H L

b.φLHLH =

L H L H

28

slide-29
SLIDE 29

Summary of Hausa Analysis

¬φNI-Cont ∧ ¬φNI-L2 ∧ ¬φNI-H2 ∧ ¬φHLHL ∧ ¬φLHLH Again, interpret ¬φ as “The structure φ is forbidden.”

29

slide-30
SLIDE 30

Kukuya

  • Zoll (2003) points out that direction-based analyses of

spreading predict that H and L tones will have identical behavior.

  • Zoll (2003) shows with Kukuya that H and L behave

independently.

  • This is captured straightforwardly by forbidding sub-structures

since these grammars may ban any combination of φNF-H2, φNF-L2, φNI-H2, and φNI-L2.

30

slide-31
SLIDE 31

Kukuya data

Kukuya is like Mende but LHH patterns are not possible.

(14) F HL HLL

  • a. kˆ

a ‘to pick’

  • b. s´

am` a ‘conversation’

  • c. k´

ar` ag` a ‘to be entangled’ R LH LLH (*LHH)

  • d. sˇ

a ‘ knot’

  • e. k`

ar´ a ‘paralytic’

  • f. mw`

ar` @g´ ı ‘younger brother’ R-F LF LHL

  • g. bv˝

i ‘falls’

  • h. p`

alˆ ı ‘goes out’

  • i. k`

al´ @g` ı ‘turns around’

Deriving LHH

(ILLUSTRATE WITH HL,LH MELODIES AND σσσ)

31

slide-32
SLIDE 32

Some approaches to Kukuya

Hyman (1987) and Archangeli and Pulleyblank (1994) achieve this effect through the imposition of additional association rules.

  • Hyman (1987) posits a L-Spreading rule that undoes the

leftmost association of a doubly associated H.

  • Archangeli and Pulleyblank (1994) instead stipulate there is a

right-to-left Final H Association rule that takes priority over general left-to-right association.

32

slide-33
SLIDE 33

Zoll (2003)

  • Zoll (2003) argues that both of these analyses miss the

generalization that a prosodically prominent H tone is not allowed to spread in Kukuya.

  • As previously mentioned, forbidding particular aforementioned

local sub-structures captures this generalization: ¬φHLH ∧ ¬φNF-Cont ∧ ¬φNF-H2 ∧ ¬φNI-H2

33

slide-34
SLIDE 34

Actual Mende

Dwyer (1978) and Leben (1978). (15) Melody 2σ 3σ HL HF HHL a. k´ Onyˆ

  • ‘friend’

b. s´ ew´ ul`

  • ‘rodent’

LH LR LLH c. (not attested) d. l` el` em´ a ‘mantis’

34

slide-35
SLIDE 35

Analysis of Mende

  • Actual Mende can also be described by forbidding particular

sub-structures.

35

slide-36
SLIDE 36

Northeren Karanga

  • Hewitt and Prince (1989) argue that this language shows

edge-in association (Yip 1988)

  • This complicated pattern is sensitive to morphological tense.
  • Nonetheless, it is also describable by forbidding sub-structures.

36

slide-37
SLIDE 37

A missing generalization

  • Former accounts miss the generalization that

association—including directionality—is fundamentally local.

  • All any grammar needs to do to evaluate the grammaticality of

an association is to check it for banned local sub-structures.

37

slide-38
SLIDE 38

Conclusion

Constraints govern how tonal melodies associate to the timing tier. These constraints are:

  • 1. language-specific,
  • 2. inviolable,
  • 3. and local.

38

slide-39
SLIDE 39

References

Archangeli, D. and Pulleyblank, D. (1994). Grounded Phonology. Cambridge: MIT Press. Dwyer, D. (1978). What sort of tone language is Mende? Studies in African Linguisics, 9. Hyman, L. (1987). Prosodic domains in Kukuya. Natural Language & Linguistic Theory, 5:311–333. Leben, W. R. (1973). Suprasegmental phonology. PhD thesis, Massachussets Institute of Technology. Leben, W. R. (1978). The representation of tone. In Fromkin, V., editor, Tone—A Linguistic Survey, pages 177–219. Academic Press. Zoll, C. (2003). Optimal tone mapping. LI, 34(2):225–268.

39