Montreal-Ottawa-Laval-Toronto Phonology Workshop (MOLT) Heather Newell Carleton University, March 18-20 2016 UQAM 1 THE  PATHOLOGY  OF  LEVEL-­‐SPECIFIC  MORPHO-­‐PHONOLOGY     Heather  Newell   Université  du  Québec  à  Montréal     1.   Outline   2. Issues with current analyses of English Morpho- Phonology 3. What we need to account for: Semantics, Morphosyntax, Phonology 4. Size Matters: Why phonology is the only one who can get the job done 5. Outer Affixes, Phonological Domains, and Peripherality 6. Conclusion: The Morpho-Syntax determines cyclic domains, the Phonology masks them.   2.   Issues   with   current   analyses   of   English   Morpho-­‐   Phonology     2.1   In  a  Distributed  Morphology  +  Phases  account,  how  do  we   account  for  the  phonological  behaviour  of  affixes?       • The  prediction  is  that  all  words  will  behave,  phonologically,   like   :   ‘(ω’   here   indicates   a   domain   for   phonological   rule   application)     (1)   a.   [[[[[govern]√  ø  ]v  ment]n  less]a  ness]a  è   b. [[[[[góvern]ω  ment]ω  less]ω  ness]ω   …where   outer   cycles   are   independent   from   inner   cycles   and   the   phonological  form  built  in  cycle  1  persists.     • Current   theories   of   DM+Phases   (Embick   2010,   Marantz   2013)   propose   that   cycle   1   consists   of   the   root   +   the   1st   category  defining  head  (whether  overt  or  null),  each  outer   cat.   def.   head   is   interpreted   separately   from   cycle   1,   and   separately  from  each  other.  (modulo  phase  suspension  à  la   Bobaljik  &  Wurmbrand  2013)  (see  also  Lowenstamm  2010   for  a  discussion  and  an  alternate  analysis)     • A  word  like  (2),  where  an  outer  cat.  def.  affix  (-­‐al)  affects   inner  domains  (govern),  (-­‐ment),  is  therefore  problematic,   all  else  being  equal.,       (2)   a.   [[[govern]√  ø  ]v  ment]n  al]a  è b.   [gòvernméntal]ω    (cf.  [[govern]  ω  ment]  ω  )     2.2   2  alternatives:  Selective  Spell-­‐Out     • Spell-­‐out  is  not  triggered  by  every  category-­‐defining  head,   but   only   a   few.   These   cyclic/phase   heads   are   lexically   specified.  (see  a  detailed  discussion  of  the  following  options   in  Scheer  2011,  ch.  8  &  9)       2.2.1   Level  1  heads  are  cyclic:  Halle  &  Vergnaud  (1987)     • Phonological  rules  are  triggered  by   o the  root     o Level  1  affixes  (interior  to  cycle)   • Stress   (and   other   phonological   operations)   assigned   at   previous  cycles  is  overridden  (but  may  be  copied)       Montreal-Ottawa-Laval-Toronto Phonology Workshop (MOLT) Heather Newell Carleton University, March 18-20 2016 UQAM 2 (3)   a.   [[parent]root  al]level  1      è                    [paréntal]ω     b.   [parent]root  hoodlevel  2        è [párent]ω  hood                 2.2.2  Level  2  heads  are  cyclic:  Kaye  (1995):       • Phonological  rules  are  triggered  by   o Level  2  affixes  (exterior  to  the  cycle)   o the  word   • Stress  (and  other  phonological  operations)  implemented  at   previous  cycles  is  frozen/unmodifiable     (4)   a.   [[parent]  al]word      è                                              [paréntal]ω     b.   [[parent]Level  1  hood]word          è [[párent]ω  hood]ω               • But  these  types  of  accounts  are  problematic     3.   What  we  need  to  account  for:  Semantics,  Morpho-­‐     syntax,  Phonology     3.1   Outer   level   1   affixes   can   influence   the   stress   of   an     inner  domain:       • Kaye  predicts  something  like:  [[góvern]ω  mèntal]ω     • H&V   can   do   governmental,   but   predict   a   single   phonological  domain:  [gòvernméntal]ω   • But   outer   level   1   affixes   do   not   induce   re-­‐footing   on   an   domain  interior  to  a  level  2  affix:     (5)   a.   [[[probable]Level  1  ist  ]  Level  2    ic]  Level  1  è     b.   [[(prò.ba).ble]ω  (ís.tic)]ω         ...  This  is  true  even  when  a  lapse  is  created:       c.   *[(pro(bàbl)(ístic)]ω  (cf.  arìstocrátic)     • H&V   appeal   to   stress   copying,   but   this   is   inconsistent   in   English  (cónd[ɛ]nsation  vs  cóns[ə]ltation)  when  the  overt   affix   is   level   1   (cóndem[n]ation),   but   consistent   when   an   overt  level  2  affix  intervenes.       3.2   Affixes  are  rampantly  dual-­‐affiliated     • Level  2  affixes  can  have  Level  1  behaviour     (6)   a.   [[[compare]√  able  ]a    è   b. [kɔ́mprəbl̩]ω     (7)   a.   [[[compare]√  ø  ]v  able]a    è   b. [[kəmpér]ω  əbl̩]ω     • See  Giegerich  (1999)  for  examples  involving  -­‐y,  -­‐ment,  -­‐ous,   -­‐ism,  ist,  -­‐ise,  -­‐ess,  -­‐ette,  -­‐esque,  -­‐(e)ry,  -­‐er,  -­‐ant/-­‐ent,  -­‐able/-­‐ ible   • No  level  2  affix  has  level  1  behaviour  when  affixed  outside   another  affix.  Having   2   affixes,   ex.   -­‐ableLevel1  and   -­‐ableLevel2   does  not  predict  this  pattern.   • In  other  words,  all  affixes  can  be  level  1  when  affixed  to  a   root.     (8)   a.   [[[sign]√  al  ]n  ø  ]v  er]n    è b.   [[sígnal]ω  er]ω     c.   *[signáler]ω   (in   a   universe   where   -­‐erLevel1   followed                           -­‐al)         Montreal-Ottawa-Laval-Toronto Phonology Workshop (MOLT) Heather Newell Carleton University, March 18-20 2016 UQAM 3 3.3   Neither  H&V  nor  Kaye  predict  the  correct  semantics     • For   H&V   there   is   always   a   cycle   on   the   root.   If   semantic   cycles   parallel   phonological   cycles,   then   they   predict   no   root  allosemy     (9)   a.   [[globe]√ al]Level  1         ‘pertaining  to  the  entire  domain/world’     b.   [globe]√ lessLevel  2             ‘without  a  globe  (*without  a  domain/world)’     • Kaye   predicts   allosemic   domains   consisting   of   the   root   +   any  number  of  Level  1  affixes.  This  is  not  borne  out.       c.   [[globe]√ al]Level  1  ity]Level  1                        ‘the  state  of  pertaining  to  the  entire  domain/world(/*a  globe)’     (c.f.  globe-­‐y-­‐ness)     • NB    there  are  some  words  that  appear  to  be  exceptions  (ex.   editorial   (Harley   2014)   personality  (Lowenstamm   2010))   See  Harley  (2014)  and  Marantz  (2013)  for  a  discussion  of   allosemy  vs.  idiomaticity.     o Is  editorial  like  global,  or  like  kick  the  bucket?   o For  the  moment  we  will  presume  the  latter.   o Also  see  Arad  (2003)  for  arguments  that  allosemy   in   Hebrew   is   restricted   to   the   root   +   1st   cat.   def.   head     3.4   Cross-­‐linguistic   Consistency   of   cycles:   inner/outer     domains     • vP  (agent)  is  the  domain  for  idiomaticity  (Marantz  1995)   • vP  is  a  landing  site  for  successive-­‐cyclic  movement   • xP  is  the  domain  for  allosemy  (Arad  2003,  Marantz  2013)   • lexical  vs.  syntactic  causatives   • alienable  vs.  inalienable  possession   • etc...     • If  cycles  are  determined  lexically  (H&V  and  Kaye)  then  we   predict  no  persistent  cross-­‐linguistic  patterns  with  regard   to  cycles.     Interim   Conclusion:   Morphologically   defined   affix   classes/   cycles/  phases  account  for  neither  the  English  pattern,  nor  the   cross-­‐linguistic  pattern.     4.   Size  Matters:  Why  phonology  is  the  only  one  who  can       get  the  job  done     • What  we  need  is  an  analysis  that  gives  us  the  phonological   domains   of   Kaye,   allows   for   stress-­‐shift   that   respects   internal   domains,   and   gives   us   the   correct   morpho-­‐ semantic  patterns.     • The  DM  account  ([root+x]  y]  z])  gives  us  :   o The  correct  semantics   o The  dual  affiliation  pattern:  the  1st  affix  affixed  to   the  root  is  always  ‘Level  1’   • It  does  not  get  us  :   o The  phonological  behaviour  of  outer  level  1  affixes     • The   answer   to   this   problem   is   phonological   and   a   misdiagnosis  of  some  affixes’  Level-­‐affiliation  holds  the  key             Montreal-Ottawa-Laval-Toronto Phonology Workshop (MOLT) Heather Newell Carleton University, March 18-20 2016 UQAM 4 4.1   Outer  stress-­‐attracting  Level  2  affixes     • As  noted,  (i)  many  affixes  are  dual-­‐affiliated,  and  (ii)  Level   2  affixes  only  behave  like  Level  1  affixes  when  attached  to  a   root.   • But,   some   affixes   diagnosed   as   Level   1   are   really   Level   2   affixes  that  attract  stress:     • NC/CN  clusters  are  not  repaired  when  followed  by  a  Level   1  affix.  The  cluster  is  heterosyllabic.     (10)     a.   (i)   si(g)n  ~  si[g.n]ature       (ii)   bom(b)  ~  bo[m.b]ard       (iii)   dam(n)  ~  da[m.n]ation       • They  are  repaired  before  level  2  affixes       (b)   (i)   si(g)ner       (ii)   bom(b)y       (iii)   dam(n)ing     • -­‐ology,  ography,  -­‐ee,  -­‐ese,  (-­‐esque)  can  all  attract  stress  and   behave  like  Level  2  affixes  at  the  same  time     (11)   a.   wom(b)ólogy,  wom(b)ésque     b.   gan(g)ógraphy,  gan(g)ése       c.   kin(g)ólogy,  kin(g)ée         • We   see   the   same   pattern   with   regards   to   sonorant   syllabification     (12)   puzz[l̩]ólogy,  puzz[l̩]ése,  butt[n̩]ógraphy,  butt[n̩]ée...       cf.  (Marvin  2002)     (13)   a.   [[[twinkl]√  ing]n      è b.   [twink.ling]ω     ‘a  short  moment’     (14)   a.   [[[twinkl]√  ø  ]v  ing]n    è b.   [[twink[l̩]]ω  ing]ω   ‘act  of  twinkling’       4.2   The  2  kinds  of  Level  2  affix     • English   stress   patterns   are   based   on   a   lexically-­‐specified   pattern  of  extrametricality   • The   final   rhymes   of   nouns   and   derived   adjectives   are   ignored  for  the  purposes  of  the  (trochaic)  stress  algorithm.     (Hayes  1982)   • Note   that   Hayes   (1982)   attributes   the   extrametricality   in   derived   adjectives   to   extrametricality   of   the   affixes.   This   type   of   non-­‐modular   morpho-­‐phonological   rule   is   unnecessary.       (15)   a.   réc  (noun)  (vs.  recórd  (verb))     b.   perús,  sénsu  (vs.  illícit,  divíne)     • If   we   take   all   of   the   level   2   affixes   not   in   the   –ology   list   (from   Lieber   1992,   Mohanan   1986,   Halle   &   Vergnaud   1987,  Fabb  1988)    extrametricality  of  the  final  overt  vowel   and  all  the  follows  it  leaves  nothing  to  be  footed.     • No   feet   means   there   is   nothing   to   stress   (this   analysis   builds  on  Hayes  1982).   Montreal-Ottawa-Laval-Toronto Phonology Workshop (MOLT) Heather Newell Carleton University, March 18-20 2016 UQAM 5 (16)   -­‐n,   -­‐m,   -­‐h,   -­‐l,   -­‐f
    ,   -­‐(adj),   -­‐l,               -­‐,  -­‐   ,   -­‐,   -­‐,   -­‐s,   -­‐,   -­‐,   -­‐,             -­‐,  -­‐,     ab1     • -­‐able  retains   a   syllable,   but   we   know   English   has   a   word   minimality   requirement.   Stress   domains   must   be   minimally  bi-­‐moraic.       • -­‐ology,  ography,  -­‐ee,  -­‐ese2  are  all  large  enough  to  be  footed   even  given  extrametricality.   • Final  long  vowels  will  not  be  extrametrical  as  their  melody   is  doubly  linked  (geminate  integrity)     (17)   -­‐(ólo)g,  (ógra)ph,  -­‐(ée),  -­‐(é)   • English  stress  (ignoring  for  now  the  affixes  that  are  strong   or  weak  retractors  (Liberman  &  Prince  1977)  falls  on  the   rightmost  foot  in  the  word.   • Big   Level   2   affixes   add   a   foot   to   the   word,   and   therefore   attract  stress     (18)   a.     [[[gang]√  ø  ]v  ing]n    è   b.   [[(gáng)]ω  ]ω       (19)   a.   [[[gang]√  ø  ]n  ology]n    è b.   [[(gàng)]ω  (ólo)g]ω     1 One exception here is -like whose diphthong should attract stress. 2 -­‐esque  is  not  of  the  correct  phonological  size  to  be  in  this  list,  but  its     incorporation  in  to  English  through  the  borrowing  of  French  vocabulary   (ex.  grotesque)  would  explain  its  lexical  stress. • Phonological  rules  apply  when  their  environment  is  met  /   altered.   • Main  stress  is  sensitive  to  the  word  domain.  The  addition   of  a  foot  will  induce  re-­‐application  of  the  rule.   • In  (18)  there  is  no  ‘freezing’  of  the  previous  cycle,  there  is   just   no   motivation   to   re-­‐apply   the   Main   Stress   Rule.   (Kiparsky   1979   (as   quoted   in   Hayes   1982)):     metrical   structure   is   retained   as   long   as   it   is   not   affected   by   refooting.       5.   Outer   Affixes,   Phonological   Domains,   and     Peripherality   5.1   The  problem  of  outer  Level  1  affixes.     (20(1))  a.   [[[[[govern]√  ø  ]v  ment]n  less]a  ness]a  è     b.   [[[[[góvern]ω  ment]ω  less]ω  ness]ω     (21(2))  a.   [[[govern]√  ø  ]v  ment]n  al]a  è b.   [gòvernméntal]ω     • (21b)  is  not  the  structure  of  governmental.  Kaye  is  correct   about  the  domains,  but  not  because  of  selective  spell-­‐out.   • Note  that  all  Level  1  affixes  begin  with  a  vowel   • -­‐al,  -­‐ic,  -­‐(at)ion,  -­‐ous,  -­‐astic,  -­‐y(noun),  -­‐an,  -­‐ant,  -­‐ance,  -­‐ity,  ive   • (and  the  Level  1  stress  retractors)  ate,  -­‐ade,  -­‐ote,  -­‐ene,  -­‐use,  -­‐ ide,  -­‐ize,  -­‐ify,  -­‐ary,  -­‐ory,  ite,  -­‐oid,  ative     • Proposal:  Initial  vowels  of  Level  1  affixes  are  floating.     o Floating   vowels   will   link   to   the   Final   Empty   Nucleus  (FEN)  within  a  preceding  domain.   o Affixes   merging   into   the   phonological   domain   of   their  host  is  not  uncommon  (ex.  infixation)   Montreal-Ottawa-Laval-Toronto Phonology Workshop (MOLT) Heather Newell Carleton University, March 18-20 2016 UQAM 6   (22)   Phonological  Merger  [X  [......PWd]]  è  [X  [..  X....PWd]],  where     X  is  an  affix.  (Newell  &  Piggott  2014)     (23)   Derivation  of  governmental       a. [[govern]√  ø  ]v    è [(góvern)]ω   b.   [[[govern]√  ø  ]v  ment]n  è [[(góvern)]ω  m]  ω     b.   [[[govern]√  ø  ]v  ment]n  al]a  è             [[(gòvern)]ω  (mén)t]ω  al]  ω       (24)   a.   C  V  C  V  C  V  C  V  –  C  V  C  V  C  V  -­‐        C  V   è          |    |    |    |    |      |    |    |            |    |    |      |    |    |                |    |       g  o  v    e  r    ø n ø m e n ø t ø a l ø       b.   C  V  C  V  C  V  C  V  –  C  V  C  V  C  V  C  V            |    |    |    |    |      |    |    |            |    |    |      |    |    |    |    |       g  o  v    e  r    ø n ø m e n ø t a l ø     • The   ‘-­‐‘   in   (24)   and   (26)   represent   spell-­‐out   domains,   not   morpho-­‐phonological  objects.     • Linking   of   the   vowel   to   the   FEN   causes   the   fusion   of   the   phonological   domains   of   the   affixes.   This   causes   (i)   the   final  rhyme  of  -­‐ment  to  lose  its  status  as  extrametrical,  and   (ii)  the  consequent  footing  of  (men)       (25)   Peripherality  condition:     [X][+ex]  →  [-­‐ex]  /  ____Y]D     Where  Y  ≠  Φ  and  D  is  the  domain  of  stress  rules       (Hayes    1982:270)     5.2   The  problem  of  outer  Level  2  affixes     • Level   2   affixes   do   not   have   initial   underlying   floating   vowels   • They  will  never  merge  phonologically  with  the  preceding   domain     (25)   Derivation  of  governmentless     a. [[govern]√  ø  ]v    è [(góvern)]ω   b.   [[[govern]√  ø  ]v  ment]n  è [[(góvern)]ω  m]  ω     b.   [[[govern]√  ø  ]v  ment]n    less]a  è             [[(gòvern)]ω  m]  ω  l]  ω       (26)   a.   C  V  C  V  C  V  C  V  –  C  V  C  V  C  V  -­‐      C  V  C  V            |    |    |    |    |      |    |    |            |    |    |      |    |    |              |    |    |    |       g  o  v    e  r    ø n ø m e n ø t ø l e s ø     • The  domain  of  stress  rules  is  the  word.   • Here  we  have  non-­‐peripheral  extrametricality3     (27)   Revised  Peripherality  condition:     [X][+ex]  →  [-­‐ex]  /  ____Y]D     Where  D  is  the  cycle/phase         3 See  Newell  &  Piggott  (2014)  for  evidence  from  Ojibwe  for  multiple,  non-­‐ interacting   foot   domains   within   a   single   word/domain   for   stress   assignment.     Montreal-Ottawa-Laval-Toronto Phonology Workshop (MOLT) Heather Newell Carleton University, March 18-20 2016 UQAM 7 • We  know  that  the  domain  between  -­‐ment  and  -­‐less  is  there   because  non-­‐repaired  word-­‐internal  sequences  like  ntl  are   disallowed  in  English  (Giegerich  1999).     • There  is  also  no  re-­‐syllabification  to  create  complex  onsets   across  Level  2  boundaries  (ex.  hope.less,  *ho.pless)   • Here   both   m   and   l   are   extrametrical.   As   -­‐less   does   not   merge   phonologically   with   -­‐ment   (it   has   not   floating   V),   these   two   cycles   do   not   interact,   the   extrametricality   determined   on   a   previous   cycle   is   not   modified   • Neither   affix   gets   footed,   consonant   clusters   are   not   repaired,  and  stress  is  not  pulled  to  the  right.   6.   Conclusion:   The   Morpho-­‐Syntax   determines   cyclic     domains,  the  Phonology  masks  them.   • Spell-­‐out   must   occur   upon   the   merger   of   each   category-­‐ defining  head   o This  accounts  for  the  Level  1  behaviour  of  Level  2   affixes   being   restricted   to   the   domain   of   root-­‐ attachment   o This  explains  the  special  status  of  the  first  cycle  for   allosemy     • Level   1   affixes   undergo   Phonological   Merger   into   the   domain  to  their  left  due  to  the  status  of  their  initial  vowels   as  floating.   o This   accounts   for   their   consistent   phonological   behaviour  regardless  of  attachment  site   o This   accounts   for   the   stress   lapses   as   in   (5).   Re-­‐ footing  is  restricted  to  the  domain  of  the  affix  with   which  the  floating  vowel  merges.     • Extrametricality  is  independently  necessary  to  account  for   English  stress.   o Its   relativization   to   cycles/phases   does   not   contradict   the   data   that   motivated   the   Peripherality  Condition  (outer  Level  1  affixation)   o It  is  lexically  specified:  Level  1  -­‐al  is  extrametrical,   but  Level  1  -­‐ic  is  not.     • The   distinction   between   outer   Level   1   and   2   affixes   can   only  be  phonological.   o The   behaviour   of   root-­‐attached   morphemes   is   uniform.     o The   behaviour   of   outer   affixes   is   divergent   phonologically,  but  not  semantically.     o The  divergent  behaviour  is  due  to  both  (i)  floating   vowels,  and  (i)  phonological  size       References     Arad,  M.  (2003).  Locality  constraints  on  the  interpretation  of  roots:  The  case  of     Hebrew  denominal  verbs.  Natural  Language  &    Linguistic   Theory,   21(4),     737-­‐778.   Bobaljik,  J.  D.,  &  Wurmbrand,  S.  (2013).  Suspension  across  domains.  Distributed     Morphology  Today:  Morphemes  for  Morris  Halle,  185-­‐198.   Embick,  D.  (2010).  Localism  versus  globalism  in  morphology  and  phonology  (Vol.     60).  MIT  Press.Marantz  2013   Fabb,  N.  (1988).  English  suffixation  is  constrained  only  by  selectional  restrictions.     Natural  Language  &  Linguistic  Theory,  6(4),  527-­‐539.   Giegerich,  H.  J.  (1999).  Lexical  strata  in  English:  Morphological  causes,  phonological     effects  (Vol.  89).  Cambridge  University  Press.     Halle,  M.,  &  Vergnaud,  J.  R.  (1987).  An  essay  on  stress.  MIT  press.     Harley,  H.  (2014).    On  the  identity  of  roots.  Theoretical  linguistics,    40(3-­‐4),   225-­‐   276.   Hayes,   B.   (1982).   Extrametricality   and   English   stress.   Linguistic   Inquiry,   13(2),     227-­‐276.     Kaye,  J.  (1995).  Derivations  and  interfaces.  Frontiers  of  phonology:    Atoms,     structures,  derivations,  289-­‐332.     Montreal-Ottawa-Laval-Toronto Phonology Workshop (MOLT) Heather Newell Carleton University, March 18-20 2016 UQAM 8 Kiparsky,   P.   (1979).   Metrical   structure   assignment   is   cyclic.   Linguistic   inquiry,     10(3),  421-­‐441.   Liberman,   M.,   &   Prince,   A.   (1977).   On   stress   and   linguistic   rhythm.     Linguistic  inquiry,  8(2),  249-­‐336.   Lieber,  R.  (1992).  Deconstructing  morphology:  Word  formation  in     syntactic     theory.  University  of  Chicago  Press.   Lowenstamm,   J.   (2010).   Derivational   affixes   as   roots   (phasal   spellout   meets     English  stress  shift).  Manuscipt,  Université  Paris,  7.     Marantz,  A.  (1995).  Cat  as  a  phrasal  idiom:  Consequences  of  late     insertion   in     Distributed  Morphology.  ms.,  MIT.   Marantz,  A.  (2013).  Locality  domains  for  contextual  allomorphy     across   the     interfaces.   Distributed   Morphology   today:   Morphemes   for   Morris   Halle,     95-­‐115.   Marvin,   T.   (2002).   Topics   in   the   stress   and   syntax   of   words   (Doctoral     dissertation,  Massachusetts  Institute  of  Technology).   Marvin,   T.   (2013).   5   Is   Word   Structure   Relevant   for   Stress   Assignment?.     Distributed  Morphology  Today:  Morphemes  for  Morris  Halle,  79-­‐93.   Mohanan,  K.  P.  (1986).  The  Theory  of  Lexical  Phonology,  Reidel,  Dordrecht.   Newell,  H.,  &  Piggott,  G.  (2014).  Interactions  at  the  syntax–phonology  interface:     Evidence  from  Ojibwe.  Lingua,  150,  332-­‐362.   Scheer,   T.   (2011).   A   guide   to   morphosyntax-­‐phonology   interface   theories:   how     extra-­‐phonological  information  is  treated  in  phonology  since  Trubetzkoy’s     Grenzsignale.  Walter  de  Gruyter.