Transcript Document

Welsh soft mutation and Word
Grammar
Dick Hudson
LAGB Edinburgh 2009
Welsh data 1
• TELYN = ‘harp’ normally = telyn
• Changes by ‘soft mutation’ (SM) to delyn
• Lexical SM:
– e.g. ei telyn ‘her harp’
– but ei delyn ‘his harp’
• Syntactic SM (SSM):
– e.g. Prynodd y ddynes delyn
bought the woman harp
‘The woman bought a harp’
Questions
• What are the syntactic triggers for SSM?
– What do they tell us about syntactic structure?
• Two competing theories here:
– XP Trigger Hypothesis (XPTH)
• 1977: Borsley, Tallerman, …
– Dependency Distance Hypothesis (DDH)
• 2004: me
• 2009: Excellent evaluation by Tallerman
XP Trigger Hypothesis (XPTH)
• Based on phrase structure
– “A complement bears SM if it is immediately
preceded by a c-commanding phrase.” (Borsley
via Tallerman)
• [Prynodd [y ddynes]XP [delyn]]
bought the woman harp
• As with lexical SM, triggered by
immediately preceding environment
trigger
target
Dependency Distance Hypothesis
(DDH)
• Based on dependency structure (Word Grammar)
• ‘Valent’: subject or complement trigger
• SSM applies to any valent D2 which is:
target
– after the head word H
– and separated by another dependent of H, D1
• (so ‘dependency distance’ of D2 > 0).
H
Prynodd
dependency
bought
distance = 0
D1
y ddynes
the woman
D2
delyn
harp
dependency
distance = 2
Word Grammar Theory
• Classified dependencies
– dependent = valent or adjunct
– valent = subject or complement
– complement = object or predicative
• Landmarks, for word order
– landmark = is-before or is-after
– by default, a word’s landmarks are:
• its parent
• its (nearest) sisters
For example, …
dependent
valent
adjunct
complement
subject
predicative
H
Prynodd
bought
after
object
D1
D2
y ddynes
the woman
after
delyn
harp
Towards a grammar for SSM
• SM words are syntactically distinct
– cf Steedman, pace Hannahs, Neeleman
•
•
•
•
Any mutated token (‘Doken’) is-a SM word
A Doken = D2 is a valent of H
D2 is-after H and D1, another dependent of H
I.e. SM applies to non-initial valents
– pace Steedman: valents have SM except when adjacent
SM word
non-initial
valent
valent
D1
is-after
H
is-after
doken = D2
Why mutate non-initial valents?
• To guide hearers through potentially long
dependencies.
H
D1
…
…
…
…
…
• To trigger extra activation???
…
Doken
Coordination
• SM only applies to first conjuncts
Prynodd y ddynes delyn, c/*gorn a ffidil.
bought the woman harp horn
and fiddle
• Because only first conjuncts take the shared
head as landmark
– Other conjuncts take their position from the
first or from the conjunction
– so they’re not ‘non-initial valents’
Hard problems for DDH
1. If Doken isn’t the first word in its phrase
Dw i lawn mor grac â chi.
am I full as angry as you
‘I’m just as angry as you.’
2.
If D1 doesn’t separate H from Doken
Pwy brynodd delyn? Prynodd delyn.
who bought harp?
bought harp
D1
H
D2
H
D2
Problem 1:
‘the first word in a phrase’
• Hard to express without phrases
– but most phrases are head-initial
– so D2 = Doken
• But also hard with phrases,
– because the phrase is only indirectly linked to its first
word.
– [Dw [i] [[[lawn mor] grac][â chi]]].
– Worse still, the phrase is always remote from its first
word.
A solution for Word Grammar
• Distinguish Doken from D2 in syntax.
• Let Doken depend on both D1 and D2.
– so it must stand between them
– and not be mixed up with their dependents
• Doken shares its realization with the next
word.
– they merge at the level of form or phonology
– cf “you + are” realized jointly as {you’re}
For example, …
Dw
i Doken llawn
mor grac
â
chi.
realization
{dw} {i}
{lawn}
{mor} {grac} {â} {chi}
Problem 2:
Missing subjects
• D1 doesn’t separate H from D2
– Pwy brynodd delyn? Prynodd delyn.
who bought harp?
bought harp
• Generalisation: complements always have
SM.
– to be revised shortly
– better than empty categories
Against empty categories
“… there are good reasons to try to eliminate
empty categories from any theory of
syntactic competence. There is something of
the night to the whole idea of phonetically
invisible but syntactically present entities.
They greatly complicate the problem of
natural language processing, because
(unlike real words) they could be
anywhere…” (Steedman)
Why mutate complements?
• To distinguish V+O from V+S
– cf Roberts
• Gwelodd gath. = V + O
saw
cat ‘He/she saw a cat’
• Gwelodd cath. = V + S
saw
cat ‘A cat saw.’
But no SSM for complements of …
o
… non-finite
verbs.
s
p
Dymunodd y ddynes weld cath
Vnon-fin
Vfin
s
wanted the woman see
cat
‘The woman wanted to see a cat’
… impersonal verbs.
Gwelir ci. ‘Someone sees a dog.’
A better grammar for SSM
• Trigger 1: any valent has SM if it is-after:
– its parent and
– a sister of its parent.
• Trigger 2: any complement of a finite personal
verb has SM if it is-after its parent.
• Both cases are exceptional
– so trigger extra activation ????
– cf Iosad; pace Steedman
In a network
dependent
valent
finite
complement
•
adjunct
non-initial
D1
•
comp of finite
D2
Doken
•
Impersonals: tba
Doken
But ddim, ….
• ddim, ‘not’, blocks SM
– Gwnaeth Megan gysgu.
did
Megan sleep ‘Megan slept’
– Gwnaeth Megan ddim cysgu.
did
Megan not sleep
‘Megan didn’t sleep’
• Unlike other negators, e.g. eriodd ‘never’
• Why?
and elliptical coordination
Aeth i weld eglwys Llandaff yn y bore
went to see church Llandaff in the morning
ac yn y prynhawn g/castell Caerdydd.
and in the afternoon castle Cardiff
‘He went to see Llandaff church in the
morning and in the afternoon, Cardiff castle.
’
Why is SM optional?
• MT: there’s an optional hiatus which blocks
SM
– but why is this hiatus linked to coordination?
• But if not that, what?
Conclusion
• SSM signals two exceptional cases:
– a non-subject valent
– a non-initial valent
• Both of these signals help the hearer.
• Classified dependencies work.
– They avoid empty categories.
– They’re motivated.