Towards a model of the Samoan syntax-prosody interface

Introduction
Sentence-medial high edge tones
A uniﬁed source for H-’s?
Towards a model of the Samoan
syntax-prosody interface
Kristine M. Yu
[email protected]
Department of Linguistics, UMass Amherst
Yale Department of Linguistics Colloquium Series
January 23, 2017
Kristine M. Yu www.krisyu.org Towards a model of Samoan syntax-prosody 1

View Slide

Introduction
Collaborators
Matthew Frelinger Deniz Özyıldız Ed Stabler
UMass Amherst UMass Amherst UCLA/Nuance

View Slide

Introduction
Background
Bringing prosody into the grammar to inform the
parser
Challenges:
Many theories of syntax-prosody interface with many moving parts
Many interacting, conditioning factors on prosody that may obscure
informativity of prosodic information for syntax

View Slide

Introduction
Background
Bringing prosody into the grammar to inform the
parser
Strategy:
Deﬁne and compare computational models which capture
fundamental properties that distinguish proposed theories from one
another
Start with case studies where syntax is clearly the primary
determining factor for prosody
Today: a ﬁrst case study on Samoan

View Slide

Introduction
Background
A fine-grained model of the syntax-prosody
interface
Indirect and direct reference theories agree that phonological rules refer to
cross-categorical relationships rather than specific syntactic
categories. (Kaisse and Zwicky 1987, p. 7).
Here for Samoan interface: We place tones by individual rules that
refer to specific morphosyntactic constructions.
In Samoan, prosodic constituents do not appear to be unified by sharing
some common syntactic relation.

View Slide

Introduction
Background
Other ﬁne-grained models of interface
Similar examples from languages where word-level tone overlays apply in
speciﬁc morphosyntactic contexts: Dogon languages of Mali (Heath and
McPherson 2013, McPherson and Heath 2016), also Papuan languages
Awa and Usarafu, and Nigerian languages Nkoroo and Kalabari.
In Tommo So (Dogon):
gamma ‘cat’ HH; also in ‘three cats’ and ‘the cat’
but ‘black cat’, ‘one cat’, ‘Sana’s cat‘: LL

View Slide

Introduction
Background
Samoan’s ergative-absolutive case system
Subject of transitive clause receives “ergative” case, marked with
prepositional element [e]
Subject of intransitive clause and object of transitive clause receive
“absolutive” case, said to be unmarked
Prepositional element [i] marks “oblique” case on indirect objects,
locatives, etc.

View Slide

Introduction
Background
Basic sentences
(1) Transitive clause
na
past
lalaNa
weave
*(e)
erg
le
det
malini
marine
le
det
mamanu.
design
‘The marine wove the design.’

View Slide

Introduction
Background
Basic sentences
(1) Transitive clause
na
past
lalaNa
weave
*(e)
erg
le
det
malini
marine
le
det
mamanu.
design
(2) Intransitive clause
na
past
Nalue
work
le
det
malini
marine
(i
obl
le
det
mamanu).
design
‘The marine worked (on the design).’

View Slide

Introduction
Background
Word order
Primarily VSO word order in transitive clauses, although VOS
possible
Fronted arguments, with SVO or OVS word orders in clefted
structures

View Slide

Introduction
Background
Word order
Primarily VSO word order in transitive clauses, although VOS
possible
Fronted arguments, with SVO or OVS word orders in clefted
structures
(3) Fronted argument
[Po
[topic
le
det
malini]
marine]
na
past
lalaNa
weave
le
det
mamanu.
design
‘It was the marine that wove the design.’

View Slide

Introduction
Background
Basic word prosody
Described as a non-tonal language in a non-tonal language
family
Basic primary stress pattern: moraic trochee at right edge
(Zuraw, Yu, and Orﬁtelli 2014)
Primary stress on ﬁnal vowel if long, otherwise on penultimate vowel
◦ ("manu) ‘bird’ , ma("lini)
◦ la("va:) ‘energized’

View Slide

Introduction
Background
Basic sentence-level prosody
Rising pitch accent associated to primary stressed syllable (“LH*”)
(Orﬁtelli and Yu 2009, Calhoun 2015) appears on every content
word, presence seems insensitive to discourse conditions
Sentence-medial high edge tones (“H-”) including before an
absolutive (Yu 2011, 2016)

View Slide

Introduction
Background
Fieldwork/data
All consultants from (Western) Samoa
Primary consultant: born in Samoa, moved to United States at age
15, age 19-23 during elicitations
◦ Data collected in 2008-2009, and later in additional elicitation
sessions through 2016

View Slide

Introduction
Background
Fieldwork/data
Work with 5 additional speakers in trips to Apia, Samoa in
November 2011 and to Carson, CA in January 2012
Work with 4 additional speakers in trips to Auckland, New Zealand
in July 2015

View Slide

Introduction
Background
Fieldwork/data
Work with 5 additional speakers in trips to Apia, Samoa in
November 2011 and to Carson, CA in January 2012
Work with 4 additional speakers in trips to Auckland, New Zealand
in July 2015
Data primarily elicited in tautala lelei (formal language)
Data primarily elicited out-of-the-blue, unless pronominal elements
present, in which case referent introduced

View Slide

Introduction
Background
Fundamental frequency contour: transitive clause
100
125
150
175
Fundamental frequency (Hz)
Time (s)
0 0.5 1 1.5
LH* LH*
H-
LH*
L-L%
LH* LH*
H-
LH*
L-L%
na lalaNa e le malini le mamanu
na la la Na e le ma li ni le ma ma nu
F0 contour for transitive clause in (1)

View Slide

Introduction
Background
F0 contour: intransitive clause
100
125
150
175
Time (s)
0 0.5 1 1.5
LH*
H- LH*
LH*
L-L%
LH*
H-
LH*
LH*
L-L%
na Nalue le malini i le mamanu
na Na lue le ma li ni i le ma ma nu
F0 contour for intransitive clause in (2)

View Slide

Introduction
The phonetic realization of the H-
Invariability in the presence of H-’s
Where H- tones invariably occur
Absolutives: at right edge of word preceding absolutive (Yu 2011,
2016)

View Slide

Introduction
2016)
Coordination: at right edge of ﬁrst coordinate (Orﬁtelli and Yu
2009, Yu 2011, 2016)

View Slide

Introduction
2016)
2009, Yu 2011, 2016)
Clefts: at right edge of fronted argument (Orﬁtelli and Yu 2009, Yu
2011, 2016; Calhoun 2015)

View Slide

Introduction
2016)
2009, Yu 2011, 2016)
Clefts: at right edge of fronted argument (Orﬁtelli and Yu 2009, Yu
2011, 2016; Calhoun 2015)
None of these syntactic environments can coincide: an H- can be
uniquely traced back to its syntactic source

View Slide

Introduction
(4) Absolutive
na
past
lalaNa
weave
*(e)
erg
le
det
malini
marine
H-
abs
le
det
mamanu.
design

View Slide

Introduction
(4) Absolutive
na
past
lalaNa
weave
*(e)
erg
le
det
malini
marine
H-
abs
le
det
mamanu.
design
(5) Coordination
na
past
lalaNa
weave
[*(e)
[erg
le
det
malini
marine
H-
coord
ma
conj
Malu]
Malu]
H-
abs
le
det
mamanu.
design
‘The marine and Malu worked on the design.’

View Slide

Introduction
(4) Absolutive
na
past
lalaNa
weave
*(e)
erg
le
det
malini
marine
H-
abs
le
det
mamanu.
design
(5) Coordination
na
past
lalaNa
weave
[*(e)
[erg
le
det
malini
marine
H-
coord
ma
conj
Malu]
Malu]
H-
abs
le
det
mamanu.
design
‘The marine and Malu worked on the design.’
(6) Fronted argument
[Po
[topic
le
det
malini]
marine]
H-
front
na
past
Nalue
work
i
obl
le
det
mamanu.
design

View Slide

Introduction
A closer look at the phonetic realization of H-
150
165
180
195
210
Time (s)
0 0.2 0.4
ma li ni
No H-
[LH* not followed by H-]
150
165
180
195
210
Time (s)
0 0.2 0.4
ma li ni
[LH* followed by H-]

View Slide

Introduction
Evidence that H- tones are edge tones: Bach test
Is steady/rising pitch over the last mora in a H- correlated with
stress position or the word edge?
Problem: Samoan primary stress is realized no further left than on
penultimate mora → tonal crowding
Solution: Separate the inﬂuence of primary stress and boundary
tones by doing a Bach test
Use a nonce stress pattern, with primary stress aligned to the left edge
Introduce as English personal name to be codeswitched in

View Slide

Introduction
More on Lise Menn’s Bach test (Halle 1978)
Traditional wug test can’t rule out a list grammar over an alphabet
of segments

View Slide

Introduction
More on Lise Menn’s Bach test (Halle 1978)
Traditional wug test can’t rule out a list grammar over an alphabet
of segments
Bax test: use a nonce form that ends in a nonnative phoneme
x is [-voice]
/bax-z/ → segment-based rules
/bax-s/ → feature-based rules

View Slide

Introduction
The Bach test for intonation: task naturalness
Codeswitching between Samoan and English is a common everyday
occurrence
Loanwords may receive aberrant stress, cf. [
>
tsai´
ami
>
tsa] ‘diameter’
[Audio]
Great liberty and variety in personal names in Samoan, e.g. (in
Samoan orthography)
pese ‘song’, moa ‘chicken’, lagi ‘sky’
mauaituaolefaleolemalo ‘found behind the government house’, maua
for short
alofa | ile | lautusi | o | le | salamo ‘love the pages of the song’,
names split up among children in same family

View Slide

Introduction
Example: f0 contours and stress position
S1 S2 S3 Following
160
180
200
220
240
30 35 40 45 50 55 60 65 70
Time slice in word
Mean fundamental frequency (Hz)
Stress position
initial
penult
ﬁnal

View Slide

Introduction
Homophonous H-’s
No positive evidence that H-’s are realized diﬀerently depending on
syntactic structure
Lengthening at right edges with H-’s compared to baseline when no
H- is present

View Slide

Introduction
Homophonous H-’s
No positive evidence that H-’s are realized diﬀerently depending on
syntactic structure
Lengthening at right edges with H-’s compared to baseline when no
H- is present
Other examples of “homophonous” tones: overlapping tonal
inventory for lexical and grammatical tones in Bantu languages; high
tones can be edge tones or pitch accents in head/edge-prominence
languages, e.g. Turkish, French

View Slide

Introduction
Sensitivity of presence of to speech rate
(7) Sensitivity of prosodic phrasing to speech rate in Calcutta
Bengali (Hayes and Lahiri 1991)
a. (Omor)
Armor
(čador)
scarf
(tara-ke)
Tara-obj
(díečhe) deliberate speech
gave
‘Armor gave a scarf to Tara’
b. (Omor čador) (tara-ke) (díečhe) faster speech
c. (Omor) (čador tara-ke) (díečhe) faster speech
d. (Omor čador tara-ke) (díečhe) very rapid speech

View Slide

Introduction
Sample minimal pair from speech rate stimuli
(8) A sample minimal pair from speech rate data set
a. Transitive clause
na
past
laNona
hear
e
erg
malini
marine
H-
abs
le
det.sg
liona
lion
i
obl
le
det.sg
aoauli.
afternoon
‘The marines heard the lion in the afternoon.’
b. Intransitive clause
na
past
manoNi
smelly
H-
abs
malini
marine
i
obl
le
det.sg
liona
lion
i
obl
le
det.sg
aoauli.
afternoon
‘The marines were smelly to the lion in the afternoon.’

View Slide

Introduction
Invariability in presence of H-: speech rate
(a) subject under slow speech rate (b) subject under normal speech rate

View Slide

Introduction
Invariability in presence of H-: speech rate
(c) subject under normal speech rate (d) subject under fast speech rate

View Slide

Introduction
Properties of Samoan registers
tautala lelei ‘good language’: used in literary contexts and and
Westernized contexts, with foreigners
tautala leaga ‘bad language’: used in traditional ceremonies and
meetings, with family and friends
Case markers often dropped (Mosel and Hovdhaugen 1992, Mayer
2001)
(9) Mergers from tautala lelei to tautala leaga
a. /t/ and /k/ → /k/
b. /n/ and /N/ → /N/

View Slide

Introduction
Invariability in presence of H-: register
(10) Transitive sentence minimal pair in tautala leaga
a. VSO word order
Na
past
laNoNa
hear
∅
erg
le
det.sg
lioNa
lion
H-
abs
le
det.sg
malie.
shark
‘The lion heard the shark.’
b. VOS word order
Na
past
laNoNa
hear
H-
abs
le
det.sg
lioNa
lion
∅
erg
le
det.sg
malie.
shark
‘The lion was heard by the shark.’

View Slide

Introduction
Invariability in presence of H-: register
(a) f0 over 3-syll verbs (b) f0 over 3-syll ARG 1

View Slide

Introduction
Other invariability in presence/placement of H-’s
Word order
Prosodic length (number of syllables/words in arguments)
Information structural manipulations (Contrastive and informational
focus)

View Slide

Introduction
A general syntactic relation?
A general phonological relation?
Deﬁning the syntax/prosody interface
Deﬁne beginnings of a formal Minimalist Grammar for Samoan (Yu
& Stabler to appear)

View Slide

Introduction
Adopt a particular syntactic perspective on Samoan case, but other
syntactic alternatives could be deﬁned as well

View Slide

Introduction
Adopt a particular syntactic perspective on Samoan case, but other
syntactic alternatives could be deﬁned as well
Why Minimalist Grammars? Regular derivation trees makes it
easy to interface with:
Regular (tree) transducer to do spellout
Regular transducers deﬁned for the prosodic/phonological grammar

View Slide

Introduction
What is the absolutive H-?
Following Collins (2014, 2015, 2016) (and Legate (2008)):
Absolutive is default, syncretic marking of nom and acc

View Slide

Introduction
Collins: S and P behave diﬀerent in nominalized clauses: S must be
genitively marked, while P can have same marking as in ﬁnite clauses

View Slide

Introduction
In nominalizations, H- possible for transitive object P but not for
intransitive subject S

View Slide

Introduction
Case markers adjoined to arguments in spellout, additional spellout
rules for coordination and clefted structures

View Slide

Introduction
Verb-initial ordering derived by fronting the VP to a function head F
below T after the arguments have been raised out of it (Collins)
Head movement moves T na to C (Collins)

View Slide

Introduction
Verb-initial ordering derived by fronting the VP to a function head F
below T after the arguments have been raised out of it (Collins)
Head movement moves T na to C (Collins)
Other ideas about abs: (1) material preceding abs fronted into high
Spec position, (2) abs arguments extrapose

View Slide

Introduction
Derived syntactic trees (Yu & Stabler to appear)
Absolutive Coordination Clefting

View Slide

Introduction
Absolutive H- is distinct from other H-’s
Anecdotal mentions of “always optional” particle, ia, that precedes
absolutives (Hovdhaugen 1987, Mosel and Hovdhaugen 1992, Vonen
1988)
Frequency of usage of absolutive ia currently seems very sporadic,
but speakers still have systematic intuitions about distribution

View Slide

Introduction
Absolutive H- is distinct from other H-’s
Anecdotal mentions of “always optional” particle, ia, that precedes
absolutives (Hovdhaugen 1987, Mosel and Hovdhaugen 1992, Vonen
1988)
Frequency of usage of absolutive ia currently seems very sporadic,
but speakers still have systematic intuitions about distribution
ia illicit in syntactic environments where other segmental case
markers also illicit (Yu and Özyıldız 2016)
ia licit only where absolutive H- occurs, not where cleft or
coordination H-’s occur (Yu & Özyıldız 2016)

View Slide

Introduction
Absolutive illicit in naPo constructions
Calhoun (2014) noted that no absolutive H- occurs after naPo.
My consultants don’t accept ia with naPo either:
(11) No absolutive ia, H-
na
past
laNona
hear
e
erg
Melina
Melina
*H-
*abs
*ia
*abs
naPo
only
*H-
*abs
*ia
*abs
le
det.sg
liona.
lion
‘Melina heard only the lion.’

View Slide

Introduction
NaPo ‘only’ cannot occur with any case morpheme
(12) No ergative e
na
past
laNona
hear
*e
*erg
naPo
only
Melina
Melina
*e
*erg
H-
abs
(ia)
(abs)
le
det.sg
liona.
lion
‘Only Melina heard the lion.’
(13) No oblique i
na
past
leaNa
bad
H-
abs
(ia)
(abs)
Melina
Melina
*i
*obl
naPo
only
*i
*obl
le
det.sg
liona.
lion
‘Melina was bad to only the lion.’

View Slide

Introduction
Hypothesis for origin of absolutive H-
Hypothesis: Absolutive H- emerged diachronically from segmental
[ia] particle: segmental deletion and reassociation of orphaned
tone
ia is bimoraic and receives penult stress; pitch accent provides source
tone for reassociation
H H
=
manu i a

View Slide

Introduction
A unifying syntactic relation?
Suppose we take an H- as a diagnostic for some phonological constituent.
What could be a general syntactic relation between the H-’s that
maps to a common phonological constituent?

View Slide

Introduction
Why is it that H-’s always appear, and don’t disappear with fast
speech rates or decreasing prosodic length (Yu & Stabler to appear)

View Slide

Introduction
Why is it that H-’s always appear, and don’t disappear with fast
speech rates or decreasing prosodic length (Yu & Stabler to appear)
In contrast, appearance of sentence-medial L edge tones variable
within/across speakers

View Slide

Introduction
Variable appearance of L- tones: example
sentences
(14) L- before erg
Po
top
le
det.sg
mamanu
design
H-
front
na
past
lalaNa-ina
weave-INA
L- e
erg
le
det.sg
malini
marine
i
obl
le
det.sg
aso:
day
‘It was the marine that wove the design today.’
(15) L- before obl
na
past
malaNa
journey
H-
abs
le
det.sg
malini
marine
L- i
obl
le
det.sg
moana
sea
i
obl
le
det.sg
aso:
day
‘The marine journeyed to the sea today.’

View Slide

Introduction
Variable appearance of L- tones: f0 contours
We don’t expect syntactically conditioned H-’s where these L- tones
showed up.
150
175
200
225
250
Time (s)
0 0.5 1 1.5 2 2.5 3
LH*
H-
plateau
*?
L-
resetLH* LH*
L-L%
LH*
H-
plateau
*?
L-
reset LH* LH*
L-L%
’o lemamanu na lalaNaina e le malini le aso:
150
175
200
225
250
Time (s)
0 0.5 1 1.5 2 2.5
LH*
H-
*?
L-
reset
LH* LH*
L-L%
LH*
H-
*?
L-
resetLH* LH*
L-L%
na malaNa le malini i le moana i le aso:
(a) L- in fronting, before ergative (b) L- in VSO, before oblique

View Slide

Introduction
Phonological constituency and succinctness
Why posit phonological constituents? To make phonological
grammars more succinct. (see also Yu submitted)
Here: does positing phonological constituents where H-’s appear help
make phonological grammars more succinct? Not clear.

View Slide

Introduction
Phonological constituency and succinctness
Why posit phonological constituents? To make phonological
grammars more succinct. (see also Yu submitted)
Here: does positing phonological constituents where H-’s appear help
make phonological grammars more succinct? Not clear.
Additional phonological constituency tests besides distribution of
H-’s?

View Slide

Introduction
Defining prosodic grammars with xfst
Why xfst (Beesley and Karttunen 2003)?
Finite state sequence transducers expressive enough to define diverse
phonological formalisms
Common formalism allows for assessments of which approaches are
the most succinct
Easy to interface with the defined minimalist grammars

View Slide

Introduction
Disﬂuencies and phonological constituency
Not clear whether we would want to say H-’s are placed at the right
or left edge of a phonological constituent.
Placement of abs H- to the left: could be explained if Samoan tends
to group case markers leftward, prosodically.

View Slide

Introduction
Disfluencies and phonological constituency
Not clear whether we would want to say H-’s are placed at the right
or left edge of a phonological constituent.
Placement of abs H- to the left: could be explained if Samoan tends
to group case markers leftward, prosodically.
Evidence for this: asymmetric chunking in disfluencies—higher
frequency of function words being “enclitic” rather than “proclitic” in
disfluencies/hesitations (Himmelman 2014)

View Slide

Introduction
f¯
agogo from The Archive of M¯
aori and Paciﬁc
Music

View Slide

Introduction
Hypothesis 1: H- does not demarcate a prosodic
constituent
(16) The absolutive H- is an edge tone but does not demarcate a
prosodic constituent.
(Verb H-)PWd ( . . . and (Verb e)PWd ( . . .
Length can appear to support contour tones, which are not necessarily
edge tones; length isn’t necessarily always a correlate of a prosodic
boundary.

View Slide

Introduction
Hypothesis 2: Prosodic phrasing of abs and erg
the same
(17) The prosodic phrasing of the absolutive and ergative arguments is
the same even though no H- appears before the ergative, and
there are non-tonal phonetic correlates of prosodic constituency.
(Verb H-)PPh ( . . . but (Verb e)PPh ( . . .

View Slide

Introduction
Hypothesis 3: H- demarcates a PPh
(18) The absolutive H- demarcates a prosodic constituent higher than
a prosodic word, e.g. a phonological phrase. That is, an
absolutive argument induces a prosodic domain higher than the
prosodic word. (And what about coordination, fronting?)
(Verb H-)PPh ( . . . but (Verb e)PWd ( . . .

View Slide

Introduction
Conclusion
In Samoan, placement of sentence-medial high edge tones entirely
predictable with knowledge of syntactic structure
But not clear that a cross-categorial syntactic relation uniﬁes
conﬁgurations where H- appears

View Slide

Introduction
Conclusion
Opportunity to carefully evaluate assumptions underlying why we
posit phonological constituents

View Slide

Introduction
Conclusion
Opportunity to carefully evaluate assumptions underlying why we
posit phonological constituents
Starting point for deﬁning and comparing computational models of
the interface which capture fundamental properties that distinguish
proposed theories from one another

View Slide

Introduction
Acknowledgements
Primary consultants: John Fruean, Kare’l Lokeni
Coordination of Samoan ﬁeldwork: Gladys Fuimaono, Peone
Fuimaono; Coordination of Auckland ﬁeldwork: Jason Brown
Adam Albright, Ryan Bennett, Rajesh Bhatt, Seth Cable, Sandy
Chung, James Collins, Lyn Frazier, Alice Harris, Larry Hyman,
Sun-Ah Jun, Diana Massam, Jim McCloskey, Masha Polinsky, Russ
Schuh, Lisa Selkirk, Donca Steriade, Ellen Woolford, Kie Zuraw
Funding: from UCLA Academic Senate’s Committee on Research,
Department of Linguistics at University of Maryland College Park
and University of Massachusetts Amherst

View Slide

Towards a model of the Samoan syntax-prosody interface

Towards a model of the Samoan syntax-prosody interface

krisyu

More Decks by krisyu

Other Decks in Research

Featured

Transcript