Speakers and hearers use prosody to disambiguate scopally ambiguous sentences

Kristen Syrett, Georgia Simon, Kirsten Nisula

Background  

Production Experiment

Perception Experiments

Selected References Baltazani  2002  The  prosodic  structure  of  quan4fica4onal  sentences  in  Greek;  Fodor  2002  Psycholinguis4cs  cannot  escape  prosody;  Hirschberg  &  Avesani  2000  Prosodic  disambigua4on  in  English  and  Italian;  Kadmon  &  Roberts  1986  Prosody  and  scope:  The  role  of  discourse  structure;    Jackendoff  1972  Seman&c  interpreta&on  in  genera&ve  grammar;  Koizumi  2009  Processing  the  not-­‐because  ambiguity  in  English:  The  role  of  pragma&cs  and  prosody;  Ladd  1980  Intona4on  and  Grammar;  McMahon,  Lidz,  &  Pierrehumbert  2004  Suprasegmental  cues  to  meaning  in  child-­‐directed  speech  

Conclusions and Implications 1.  Speakers can distinguish between interpretations using surface cues, but these cues (a) are

highly variable within and across speakers & contexts, (b) depend on the quantifier and its position, (c) are not restricted to sentence-final contour, (d) are not linked to scope alone.

2. Hearers reliably recruit such cues to arrive at the speaker’s intended interpretation, in most, but not all cases. Success depends on the quantifier, speaker, and scope.

3.  Psycholinguistic (including acquisition) studies that present participants with scopally ambiguous sentences should control for prosody, taking into account these patterns. It is not enough to say that prosody was “kept neutral” or was delivered “naturally.”

NELS  2012  

Purpose Determine  whether  certain  scopal  rela&ons  elicit  a  par&cular  prosodic  contour,  and  whether  any  surface-­‐level  signature  varies  with  the  context.  

Theoretical Jackendoff  (1972):  prosody  is  4ed  to  scopal  rela4ons  ‘A  accent’  (fall):  ¬  is  part  of  the  presupposi&on,  not  focus;  ‘B  accent’  (non-­‐fall):  ¬  is  associated  with  the  focus  (asser&on)  Objec4ons:  Liberman  &  Sag  1974;  Ladd  1980;  Ward  &  Hirschberg  1985;  Kadmon  &  Roberts  1986  Emerging  picture:  While  it  is  possible  that  a  fall-­‐rise  contour  may  (strongly)  favor  a  ‘¬  wide  scope’  reading,  this  contour  is  neither  necessary  nor  sufficient,  and  should  vary  with  the  context  (informa4on  structure,  QUD,  speaker  knowledge  of  salient  scalar  alterna4ves).        

ALL  the  men  didn’t  go.  

Participants 26  undergraduates  (data  from  19  used)  

Procedure Par4cipants  were  recorded  reading  a  series  of  short  passages:  

Stimuli Target  sentences  were  all  ambiguous,  with  controls  clearly  able  to  be  disambiguated  prosodically.  

Sentences  were  presented  in  a  brief  discourse  context  favoring  one  possible  interpreta4on.  

Purpose Determine  whether  in  the  best  case  scenario,  hearers  can  match  a  prosodic  rendi&on  of  a  sentence  to  its  intended  interpreta&on.  

Experiment 1 Participants

44  undergraduates  

Stimuli and Procedure 24  minimal  pairs  selected  from  4  speakers  (3  from  produc4on  experiment,  1  experimenter)  6  pairs  each:  

 2  all,  ¬;  1  ¬,  many/most;  2  ¬,  because;  1  only/even      

     

       

 context  1    (M  >  ¬)  context  2  (¬  >  M)                        

Experiment 2 Participants

37  undergraduates  

Stimuli and Procedure 18  minimal  pairs  selected  from  3  speakers  (2  F  from  previous  experiment,  1  experimenter)  6  pairs  each:  

 2  all,  ¬;  1  ¬,  many/most;  2  ¬,  because;  1  only/even          

                   

 context  1  (M  >  ¬)  context  2  (¬  >  M)  

Acknowledgments This work benefitted from a Rutgers startup grant to K. Syrett and funding from the Aresty Research Center at Rutgers University, as well as technical assistance from Paul de Lacy and the Rutgers Phonology lab, and discussions with Shigeto Kawahara.

Corresponding author: Kristen Syrett k-syrett@ruccs.rutgers.edu

Experimental Speakers  and  hearers  use  prosodic  cues  to  disambiguate  a  wide  variety  of  syntac4c  structures  and  presupposi4on/focus  dis4nc4ons  (Speer,  et  al,  1993;  Price  et  al.  1991),  as  well  as  non-­‐quan4fica4onal  ¬/because  ambiguity  (Cooper  &  Paccia-­‐Cooper  1986;  Hirschberg  &  Avesani  1997,  2000;  Koizumi  2009),  although  there  is  considerable  varia4on  in  produc4ve  strategies.  Greek  speakers  and  hearers  do  recruit  prosodic  cues  for  scopal  disambigua4on  (Baltazani  2002/3),  but  previous  studies  in  English  leave  the  ques4on  open  (McMahon  et  al.  2004;  Jackson  2006).        

Abstract Since  Jackendoff  (1972)  first  claimed  that  sentences  such  as  (1)  can  be  reliably  disambiguated  via  a  sentence-­‐final  contour,  (2),  various  researchers  have  ques4oned  this  strength  of  this  rela4onship  and  the  nature  of  the  link  between  prosody  and  sentence  meaning,  arguing  for  a  pragma4c  account.  (1)  All  the  men  didn’t  go.    (2)  a.  ∀  >  ¬  (none)      ‘A  accent’  

 b.  ¬  >  ∀    (not  all)  ‘B  accent’  To  date,  systema4c  evidence  bearing  on  this  ques4on  has  been  lacking.  We  present  a  set  of  produc4on/percep4on  experiments  designed  to  inves4gate  whether  speakers  and  hearers  recruit  auditory  cues  (including,  but  not  limited  to  sentence-­‐final  contour)  to  disambiguate  such  sentences.  While  there  is  considerable  variability  in  speaker  produc4on,  there  are  surface-­‐level  cues  to  interpreta4on.  Moreover,  hearers  successfully  recruit  these  cues  to  arrive  at  the  correct  interpreta4on.  We  argue  that  psycholinguis4c  studies  inves4ga4ng  par4cipants’  ability  to  access  mul4ple  interpreta4ons  of  scopally  ambiguous  sentences  should  carefully  control  for  prosody.      

   

Sample  sentences  test  items  (n=28)      type  All  the  magnolias  won’t  bloom.  ∀,  ¬  Liam  doesn’t  know  many  alumni.    ¬,  many  Neil  doesn’t  enjoy  most  musicals.  ¬,  most  

control  items  (n=28)  type  They’re  not  late  because  of  his  driving.    ¬,  because  Warren  only  likes  the  Orioles.  focus  She  even  painted  the  garage.  focus  Alan  punched  Owen  and  then  he  kicked  him.  pronominal  ref.  

read  passage  silently   answered    

ques4on   re-­‐read  passage  aloud,  recorded  Sentences  blocked,  pseudorandomized  

Sonorance  and  presence  of  con&nua&on  sentence  controlled  for  Analysis:  comprehension  scores  of  1;  targets  excised  in  Praat  

A  few  years  ago,  the  township  decided  to  plant  magnolia  saplings  to  line  a  path  through  the  park.  The  saplings  on  the  north  side  were  planted  mainly  in  sand  and  haven’t  been  geLng  nearly  enough  nutrients.  However,  the  soil  near  the  south  side  is  rich,  and  the  magnolias  are  thriving  there.  All  the  magnolias  won’t  bloom.  But  I  bet  the  ones  on  the  south  side  will.  

Results Analysis  1:  Sentence-­‐Final  Contour  

             

Analysis  2:  AcousLc  Analysis  5  acousLc  measures  -­‐  quanLfier,  sentence-­‐final  word  max  F0,  max  F0  loca4on,  F0  st.  dev.,  word  dura4on    all:  most  delayed  max  F0  on  quan4fier  and  shortest  sentence-­‐final  word  in  context  4,  longest  sentence-­‐final  word  in  context  3;  many/most:  shortest  quan4fier  AND  longest  sentence-­‐final  word  in  context  1  

 

quantifier! context- scope-- scalar-alternative- %-falling-contour--

all- 1- all->-¬- quantity- 93.4-- 2- ¬->-all- which- 89.1-- 3- all->-¬- which- 71.1-- 4- ¬->-all! quantity! 95.5-many/most- 1- M->-¬- quantity- 91.3-- 2- ¬->-M! which- 65.1-- 3- ¬->-M- which-- 63.0-

!

Which  sentence  should  follow?  a.  He  really  has  to  make  more  connec4ons.  b.  But  the  ones  he  knows  are  well  

established  in  the  community.  

Speakers  blocked,  minimal  pairs  in  sub-­‐blocks  Sentences  presented  via  headphones  in  lab  using  Superlab  Sentence  presenta&on  within  blocks  randomized    Analysis:  comprehension  scores  of  1,  RTs  <15s  

Blocking  and  presenta&on  via  Superlab  as  before  Analysis:  comprehension  scores  of  1,  RTs  <3.5s  

Results Experiment  1  

                 

Experiment  2  

Neil  is  a  fan  of  serious  theatre.  I  bought  4ckets  to  the  musical  "Oklahoma!"  for  him  as  a  gio.  My  friend  Adam  was  concerned  about  the  choice.  I  realized  he  was  right.  <<Neil  doesn’t  enjoy  most  musicals.>>  He  thinks  they  are  very  cheesy.  

 A  B  

Liam  doesn’t  know  many  alumni.  

All  %  significantly  above  chance    4  x  2  x  3  ANOVA    (speaker,  scope  of  ¬,  lexical  item):  ME  of  speaker,  ME  of  scope  of  ¬,  significant  interac&ons  

x3  

All  %  significantly  above  chance,  with  excep&on  of  ¬  >  all    3  x  2  x  3  ANOVA  (speaker,  scope  of  ¬,  lexical  item):  ME  of  scope  of  ¬,  significant  ¬  x  lexical  item  interac&on