Document Sample
TB_MSPIL_HO_emb Powered By Docstoc
					                 Butterpillar or Caterfly?

          The Bangla Passive in a Minimalist Parser

                        Tanmoy Bhattacharya
                      Department of Linguistics
                         University of Delhi

2-4 April 2006              MSPIL06, IIT Mumbai       1
                 What is the talk about
      Passive template historically is a result
       of a certain tension within the body of
       the clause
      Incorporation, verb-shell, “smuggling”,
       of P&P can capture the tension
      Invites treatment in a Minimalist Parser
      Importing syntactic analysis as it is does
       not work

2-4 April 2006         MSPIL06, IIT Mumbai    2
         The Place of Passive in the
             P&P Frameworks
      Later GB: severance between the active
       and the passive form 
      Different derivational histories
      Misses the generalization about how we
       think of passives not out of the blue
      Passive is more „surfacey‟
         Can we capture this in Minimalism?

2-4 April 2006     MSPIL06, IIT Mumbai     3
           Evaluation Matrix and the
      Evaluation matrix (EM) is a collection of
       Economy Principles (Last Resort, Least Effort,
       Procrastinate, etc.)
      Evaluation is of only convergent derivations
      Passive and Active are comparable, passive
       „wins out‟ later, iff speaker‟s intention had
       dethematicization of subject
       Passive/ active are one until EM acts

2-4 April 2006        MSPIL06, IIT Mumbai          4
                  The Passive in Bangla
      Very prolific
      Analytic Passive: pass ppl+aux v
            a. ama-ke dEkha jay
              me-dat seen goes
            b. dEkha jay (impersonal)
      Agent, if expressed, is marked by a P:
                  jim dara bagh-Ta mar-a gEche
                 Jim by tiger-cla kill-pass go.ppl.3
                 „The tiger has been killed by Jim.‟

2-4 April 2006                MSPIL06, IIT Mumbai      5
                 Idiomatic Passive forms
      a. mar-a pOre                   b. mar-a gElo
          hit-pass fall.3                hit-pass went.3
         „gets killed‟                  „got killed‟
       c. kha-wa cOle                 d. ama-ke dEkh-a hOy
          eat-pass walk.3                I-dat see-pass be
         „can be eaten‟                 „I am seen‟(=she/They
                                      see me)

2-4 April 2006          MSPIL06, IIT Mumbai                6
            The Revised Passive Template

      GEN subject:
          ama-r dara bagh mar-a hObe
          I-gen by tiger kill-pass be.fut
          „Tiger will be killed by me.‟
      Revised Passive template:
              [(NP-gen by) NP V-a be V

2-4 April 2006      MSPIL06, IIT Mumbai     7
                 Similarity with the Gerund
      Gerunds have GEN subject too:
           ama-r boi pOR-a
           I-gen book read-ger
           „my reading book.‟
      The P dara in passive can be dropped:
           amar kOfi ken-a holo
           I-gen book read-pass became
           „Coffee was bought by me.‟
      Norwegian:
           Det vart kjøpte kaffe
           it became bought coffee
           „There was bought coffee.‟ (Åfarli 1992)
2-4 April 2006           MSPIL06, IIT Mumbai          8
      LR Parsing and the Passive-Gerund
   (1)           a.   (jOn dara) boi pORa hoeche
                      J-(gen) by book read.pass be.ppl.3
                 b.   joner boi pORa hoeche
                      .gen book read.pass be.ppl.3
                      „the book has been read by J.‟
           The VPs are identical
   Difference: (i) Non-optionality of the subject (of the
   DP)in (b), and (ii) GEN on the subject in (b)
           GEN cased DP can‟t be recovered

2-4 April 2006              MSPIL06, IIT Mumbai             9
        First Parse of the Gerund/Passive
      (A) If GEN, mark –a on V as GER
          Parsing Question: How is the next V analysed?
      (a) If zero N, select T and check [NOM] on N
      (b) When V is scanned, -a triggers a PASS vP
      (c) PASS selects an unaccusative VP
       Rule: If 2 NPs, the V is not PASS, or if (A), then:??
      (d) GEN triggers: (i) POSS DP, or (ii) GER
      (e) If the next V is –a, (i) is rejected, parser
       backtracks to (ii)
      (f) When another V, (ii) is also rejected ??

2-4 April 2006           MSPIL06, IIT Mumbai               10
             How to Recover the POSS DP
         PASS with POSS DP:
            joner bagh mara gEche
            John.gen tiger killed go.ppl.3
            „John‟s tiger has been killed.‟

2-4 April 2006            MSPIL06, IIT Mumbai   11
        Algorithm for both Types of
                Light Verbs
   i. [joner bagh]
   ii. + mara
        A: ger expects N/ø and main V
        B: pass expects LV
   iii + gEche C: rejects (iiA)
        D: proceeds as pass
   iii. + hoeche
         then apply R1
        R1: [NP+gen… a]-> no pass
   iv. reject (iiB)  ger tree projected

2-4 April 2006              MSPIL06, IIT Mumbai   12
                 Algorithm for POSS DP
      But still no POSS DP parse!
   ii. + mara
             A: as before
             B1:[joner bagh] ø mara
             B2: :[joner bagh]dat mara
   iii. + gEche
             C: rejects (iiA)
             D: proceeds with B
               d1: rejects B1, apply R2: gEche takes nom
               d2: accept B2
   iii. + hoeche
             E: accept A  generates poss tree
             F: reject B

2-4 April 2006             MSPIL06, IIT Mumbai             13
                 Algorithm for the PASS Parse
      For this, we need yet another rule:
   Rule 3: ho can take NP-nom at [Spec,T] and NP-dat at [Spec,v]
   Now, step (iii) above becomes:
   (iii) + hoeche
            E: accepts A  ger
            F: rejects B1
            G: accepts B2  apply R3 pass (23b)
   We needed 3 ad hoc rules (Rule 1-3) to resolve the
    passive/ gerund ambiguity

2-4 April 2006           MSPIL06, IIT Mumbai                   14
                   Butterpillar/ Caterfly
      Trapped energy, caterpillar waiting to
       burst into a butterfly (CB)
      Opposite view: butterfly shrinking to a
       caterpillar (BC)
      Both possibilities in Passive:
            Clipping the wings of EA (BC)
            History and synchrony (CB)

2-4 April 2006          MSPIL06, IIT Mumbai   15
                 History of the Bangla Passive
      -a < denominative –aya
      Obscured by causative –aw
       a. daMR „stick‟ > daMRay „stands‟
       b. tOl „bottom‟ > tOlay „goes to the bottom‟
      Distinction between DENOM and CAUS is lost
    Verbalise (NV)      C B
      Both find syntactic analogues:
         incorporation (shelf  shelve)

         V  v

      feed (example par excellence)

2-4 April 2006            MSPIL06, IIT Mumbai         16
                 History of the Passive Agent
      N+Case P
      Loss of Case in MIA  N+Aff P+Case
      Skt extended P-use to verbal forms (pass ppl,
       prs ppl) influenced by Dravidian (IE
       regarding, during, concerning)
       a. kore „having done‟
       b. diye „having given‟
       c. dara inst of dvar „through the instrumentality of‟
      PV (a and b); PN (c)      BC

2-4 April 2006           MSPIL06, IIT Mumbai               17
                 Syntax of the Caterfly Effect
      Surfacing of v = Bypassing v
      Collins (2005) “Smuggling”
                 voice    vP
                         PP 2
                          v    <PartP>

2-4 April 2006                 MSPIL06, IIT Mumbai   18
                      Smuggling in Bangla
                          VcP    T
                           vP   Vc
                 amar dara 2
                         PrtP    v
                         2 hoeche
                         VP Prt
                      2 -a
                   boi      V
2-4 April 2006                 MSPIL06, IIT Mumbai   19
          Minimalist Parser and Minimalism
   Similarity: Incremental Processing
   Differences with Minimalism:
         Unavailability of Lexical Array (LA)
         No place for Merge/ Move in a LR parser since they
          are bottom up; ETs are the alternatives to them
         Move Box to capture effects of -theory
         Probe Box to capture Case and PIC

2-4 April 2006            MSPIL06, IIT Mumbai             20
                       Lexicon of a MP
       LI           properties                   -intp features

                                         phi-features     other

       v* (trans)   select(V)            per(p)           (epp)
                    spec(select(N))      num(n)
                    value(case(acc))     gen(g)
       v (unacc)    select(V)

       v(unacc)     select(V)
       v(unerg)     select(V)

       PRT(pass)    select(V)            per(p)           case(_)

2-4 April 2006                  MSPIL06, IIT Mumbai                 21
                         Elementary Trees
                   v*/v/v#                V       V
                                          (unerg) (trans/
         v           v         v               V        V
         1           1         1                        1
          v                      v                  V
          1                     1
        v*                       v#
             (e)       (f)       (g)          (h)           (i)

2-4 April 2006               MSPIL06, IIT Mumbai                  22
                 Move Box and Probe ox
   Move Box Preference Rule
       When filling open positions, always prefer the Move
       Box over the input
   Elements involving Agree are picked from the
   most current Probe stored in the Probe Box:
       Agree(p,g) if
       a. Match(p,g) holds. Then:
       b. Value(p,g) for matching features
       c. Value(p,g) for property value(f)

2-4 April 2006          MSPIL06, IIT Mumbai             23
   a. Given a category X, pick an ET headed by X
   b. From the Move Box or input:
      i. Fill in the Spec
      ii. Run Agree(p,g) if both p,g are non-empty
      ii. Fill in the Head
      iv. Copy h to Probe Box if h is a probe
      iii. Fill in the complement by recursively calling parse
      with X‟ where X has lexical property select(X‟)

2-4 April 2006          MSPIL06, IIT Mumbai                24
     “Smuggling” in a Minimalist Parser?
   Action         Parse                       Input                   MBox PBox

    :select(T)    [T_[T]]                       [joner dara] boi pORa empty   empty

   Fill Spec-T    [T [joner dara][T]]                 boi pORa hoeche [Pn]    empty

   Fill Head-T    [T [joner dara]                     boi pORa hoeche [Pn]    [past+]
                  [T past(+)]]
   T:select(Vc)   [T[joner dara] [Vc_[Vc]]            boi pORa hoeche [Pn]    [past+]
                  [T past(+)]]

   Fill Spec-Vc   [T[joner dara [Vc?? [Vc]                 ?? hoeche [Pn]     Agree
                  [T past(+)]]

2-4 April 2006                      MSPIL06, IIT Mumbai                           25
       Failure of Pass Parse with/ without
    Agree(T,Spec-Vc) will not take whole
     PRT, but only the Obj
    Obj wrongly valued nom

    If Obj moves alone, again Agree will

     value Case as nom, wrongly
    Movement of Obj not possible

    Vc is not required

2-4 April 2006     MSPIL06, IIT Mumbai      26
      A Minimalist Parsing algorithm cannot
       mimic syntactic object movement
       outside the VP shell
      Voice Phrase is unnecessary
      Probe-Goal Syntax in Minimalist
       Inquiries finds support from the Parser
      Movement to any higher functional
       position (Agro) is unimplementable

2-4 April 2006      MSPIL06, IIT Mumbai      27

Shared By: