Transcript PowerPoint-presentation

Intonation, Word Order, and the Brain
Mikael Roll and Merle Horne
Lund University
Background
Hypotheses
In Swedish, there is a difference between main clauses and subordinate clauses regarding
the linear order of sentence adverbs and inflected verbs. Sentence adverbs follow the
inflected verb in main clauses, as shown in (1), whereas they canonically precede it in
subordinate clauses, as in (2).
In the current study, we investigate whether the brain uses the intonation rise
described by Roll (2004, 2006) for predicting a certain syntactic and semantic
structure. If it does, an intonation rise at the beginning of a subordinate clause would
make it easier to process a subsequent sentence adverb indicating main clause word
order.
(1) Ölen rinner inte
the.beer flows not
I
wonder if
(2) Jag undrar om ölen
the.beer not flows
On the other hand, the rise itself could cause a semantic anomaly if it followed a
verb that cannot take an assertion as its complement, like ‘want’ or ‘hope’.
inte rinner
In spoken Swedish, subordinate clauses introduced by the conjunction att ‘that’ may have
main clause word order, with sentence adverbs following the inflected verb. This is
illustrated in (3). Subordinate clauses with main clause word order are called ‘Embedded
Main Clauses’ (EMC). They have their own Force, and are typically interpreted as
assertions. Therefore, they cannot function as complements to verbs that put the truthvalue of their complements in question, as vill ‘want’ in (4).
I
(3) Jag sa att ölen
said that the.beer flows not
(4) *Jag vill
I
want
rinner inte
att ölen
rinner inte
that the.beer flows not
In a production study, Roll (2004, 2006) found an intonation rise at the left boundary of
EMCs in East Swedish. The intonation rise leads to a high tone at the first syllable
following the syllable that would have been associated with a high tone were the first
word focused (see Figure 1 and 2).
Prosody-related ERP effects
Steinhauer, Alter, and Friederici (1999) found a N400-P600 pattern for violation of
prosody-induced syntactic expectations. N400 is a broadly distributed negativity
reflecting semantic integration of stimuli. It often peaks at around 400 ms after
stimulus onset. P600 is a positivity that is usually attributed to structural integratioin
cost. They also found a positive deflection after prosodic boundaries, that they
denominated ‘Closure Positive Shift’ (CPS).
Heim and Alter (2006) found an N400-like effect immediately after the detection
point of unexpected pitch accents.
Material
To test the hypotheses, we created 8 different conditions:
1. Say-type matrix verb
2.
3.
4.
5. Want-type matrix verb
6.
7.
8.
+ EMC
+ SC
+ EMC
+ SC
with boundary tone
without boundary tone
with boundary tone
without boundary tone
with boundary tone
without boundary tone
with boundary tone
without boundary tone
1, 2: Besökaren menar alltså
att familjen
The.visitor means accordinglythat the.family
3, 4: Besökaren menar alltså
att familjen
The.visitor means accordinglythat the.family
5, 6: Besökaren hoppas alltså
att familjen
The.visitor hopes accordinglythat the.family
7, 8: Besökaren hoppas alltså
att familjen
The.visitor hopes accordinglythat the.family
känner
feel
känner
feel
känner
feel
känner
feel
ju
of.course
det nu
it now
ju
of.course
det nu
it now
det
it
på
in
det
it
på
in
på kvällen
in the.evening
kvällen
the.evening
på kvällen
in the.evening
kvällen
the.evening
In order to avoid influence from other prosodic parameters, manipulated speech was used.
The stimulus sentences were put together from three different parts, partly recorded
separately. In half the cases, the subordinate clauses were recorded without an initial high
tone, and a tone was created for the boundary conditions, in the other half, they were
recorded with an initial high tone, which was removed for the non boundary conditions
(see Figure 3 and 4). The same segments were used for the part immediately preceding
the subordinate clause across conditions, in order to avoid influence from phrase-final
prosodic features on att ’that’. For a similar reason, half of the time the main clauses were
paired with EMCs, and half of the time with SCs in the recordings, regardless of whether
they contained say-like or want-like verbs.
Possible ERP effects
Figure 1. Spectrogram and F0 contour of an utterance
containing an EMC: Jag sa att ölen rinner inte ’(lit.)
I said that the beer flows not’. There is a high tone in
the second syllable of ölen, the first word of the
subordinate clause. Compare to Figure 2, where the
high tone is missing.
Figure 2. Spectrogram and F0 contour of an utterance
containing an SC: Jag vill att ölen rinner lite ’(lit.) I
want that the beer flows a.little’. There is no high
tone in the second syllable of ölen.
In accordance with the findings of Heim and Alter (2006), an N400-like effect might
be expected at the first word of the subordinate clause in condition 5 and 7, as
compared to 1 and 3, since a boundary tone indicating an assertive utterance would
be unexpected after a ‘want’-type verb.
An N400-effect would be expected at the sentence adverb ju ‘of course’ in
conditions 2 and 6 as compared to 1 and 5, since no phrase-initial tone has
announced the upcoming main clause word order indicated by the sentence adverb.
A CPS might also be expected in conditions 1, 3, 5, and 7, as compared to
conditions 2, 4, 6, and 8, since a prosodic boundary is expected to be perceived in
these conditions.
Figure 3. Spectrogram and F0 contour of the EMC in
condition 1 and 5 above, familjen känner ju det på
kvällen ’(lit.) the family feel of course it in the
evening’. There is a naturally produced initial high
tone in the EMC
Figure 4. Same as Figure 3, but the initial high tone
has been removed in order to obtain the prosodyword order mismatch of condition 2 and 6.