Empty category phenomena in LFG Nigel Vincent University of Manchester.

Empty category phenomena in LFG

Nigel Vincent

University of Manchester

Caveat

This presentation was prepared for use at the LFG Winter School held at the University of Canterbury, 4-8 July 2004. It was designed to follow on from the foregoing presentation by Kersti Börjars. Feel free to make use of it but please acknowledge the source.

Properties of LFG

• Non-derivational

• Parallel correspondence

• Monotonic

Therefore LFG eschews:

• movement• the (consequent) use of empty categories• the (consequent) use of uninterpretable features

(in particular Case and EPP)

Instead, new analytical tools consistent with LFG premisses need to be found

A typology of empty categories

Construction Empty category Overt category

Finite clause arguments

pro pronouns

Non-finite clause, ‘equi’ subjects

PRO –

Raising/passive NP/DP trace anaphors (herself)

Unbounded dependencies

wh-trace R expressions

The treatment of ‘pro-drop’

Italian: (Memo) canta

€

↑PRED = 'sing  <↑SUBJ >'

↑ SUBJ NUM = sg

↑ SUBJ PERS = 3

↑TENSE = present

↑ SUBJ PRED = 'pro'( )

canta

f-structure for canta ‘(s)he sings’

€

PRED 'sing <(SUBJ) >

SUBJ

PRED 'pro'

NUM sg

PERS 3

⎡

⎣

⎢ ⎢ ⎢

⎤

⎦

⎥ ⎥ ⎥

TENSE pres

⎡

⎣

⎢ ⎢ ⎢ ⎢ ⎢ ⎢

⎤

⎦

⎥ ⎥ ⎥ ⎥ ⎥ ⎥

English non-pro-drop

English: Bill sings vs *sings

€

↑PRED = 'sing ↑SUBJ '

↑ SUBJ NUM = sg

↑ SUBJ PERS = 3

↑TENSE = present

sings

Control and raising

• ‘Missing’ subject relatable to matrix verb• ‘Missing’ subject is a semantic argument of

both verbs = control (aka equi)e.g. Bill tried to dance

• ‘missing’ subject only a semantic argument of the infinitival verb = raisinge.g. Bill seemed to dance

equi vs raising

• Equi traditionally handled via a construction specific empty category with no overt analogue, viz PRO

Bill tried [PRO to dance]

• Raising handled via movement[[e] seemed [Bill to dance]

Obligatory (OC) vs non-obligatory (NOC) control

OC antecedents NOC antecedents

obligatory optional

local non-local

c-commanding not c-commanding

unique split

Bill tried to dance

• to dance requires a verb to introduce it

• introducing verb is in the next clause up

• therefore introducing verb c-commands inf.

• no split antecedence, so:

*Bill tried (*for him and Sally) to dance

f-control = OC

‘Let us first observe that Williams’ “obligatory control” corresponds to our functional control. That is, the central properties that Williams takes to be characteristic of obligatory control follow from our theory of functional control.’

(Bresnan 1982: 350)

Functional control

• Involves ‘structure sharing between SUBJ of matrix verb and SUBJ of embedded verb

• Structure sharing achieved by means of a new type of function, namely the ‘open function’ XCOMP

COMP vs XCOMP

• COMP

Bill said that Sally appointed Sue

COMP

• XCOMP

Bill persuaded Sally to appoint Sue

OBJ XCOMP

Lex entries: say, try, persuade

say ‘say <(SUBJ) (COMP)>’

try ‘try <(SUBJ) (XCOMP)>’

persuade

‘persuade <(SUBJ) (OBJ) (XCOMP)’

Lexical Rule of f-control

For any lexical form:

a) XCOMP SUBJ = OBJ if present

otherwise

b) XCOMP SUBJ = SUBJ

F-structure for try

€

PRED 'try < (SUBJ) (XCOMP) >'

TENSE PRES

SUBJ 'Bill'

XCOMP

PRED 'like < (SUBJ) (OBJ) >'

SUBJ −−−

OBJ 'strawberries'

⎡

⎣

⎢ ⎢ ⎢

⎤

⎦

⎥ ⎥ ⎥

⎡

⎣

⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢

⎤

⎦

⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥

try ‘try <(SUBJ), (XCOMP)>’(XCOMP SUBJ) = (SUBJ)

• ‘exhaustive’: same info referred to in two places in f-structure, so split antecedence impossible

• ‘local’: verb can only subcategorise for a clause contained in its own immediate constituent

• ‘obligatory’: control pattern can only be introduced via lexical entry of controlling verb

• ‘c-command’ (or f-command): controlling verb one clause up and thus subject/object/indirect object necessarily c-commands controllee

persuade vs promise

‘persuade <(SUBJ), (OBJ), (XCOMP)>’

(XCOMP SUBJ) = (OBJ)

‘promise <(SUBJ), (OBJ), (XCOMP)>’

(XCOMP SUBJ) = (SUBJ)

Lexical form for seem

seem ‘seem < (XCOMP) > (SUBJ)(XCOMP SUBJ) =

(SUBJ)

NB: (SUBJ) outside the angle brackets shows it is syntactically but not semantically selected

F-structure for seem

€

PRED 'seem < (XCOMP) > (SUBJ)'

TENSE PRES

SUBJ 'Bill'

XCOMP

PRED 'like < (SUBJ) (OBJ) >'

SUBJ −−−

OBJ 'strawberries'

⎡

⎣

⎢ ⎢ ⎢

⎤

⎦

⎥ ⎥ ⎥

⎡

⎣

⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢

⎤

⎦

⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥

believe

‘believe <(SUBJ), (XCOMP)> (OBJ)’

XCOMP SUBJ = OBJ

a-control vs f-control

Functional control(models oblig control)

Anaphoric control(models non-oblig control)

Structure sharing Co-reference

Open functions Closed functions

Corresponds to PRO Corresponds to pro

Keep + –ing

i) Susan discussed visiting Fred (anaphoric)

ii) Susan kept visiting Fred (functional)

Passive:

• Visiting Fred was discussed/*kept by Susan

Cleft:

• It was visiting Fred that Susan discussed/*kept

‘Tough’:

• Visiting Fred is unpleasant for Susan to discuss/*keep

Gen subj:

• Susan discussed/*kept our visiting Fred

Mechanism of a-control

Add the optional equation

( GF PRED) = ‘pro’

to the lexical entry of a non-finite verb

To visit Fred will annoy Susan

€

PRED 'annoy <(SUBJ) (OBJ) >

SUBJ

PRED 'visit <(SUBJ) (OBJ) >'

SUBJ 'pro'

OBJ 'Fred'

⎡

⎣

⎢ ⎢ ⎢

⎤

⎦

⎥ ⎥ ⎥

OBJ 'Sally'

TENSE future

⎡

⎣

⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢

⎤

⎦

⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥

Obviation:English want vs Italian volere

• Bill wanted to visit Fred• Bill wanted Susan to visit Fred• Memovoleva visitare Federico

Bill wanted visit.INF Fred• Memo voleva [che Susanna visitasse

Federico]

Bill wanted [that Susan visited Fred]

Wh-movement

• Involves link between a ‘filler’ and a ‘gap’

• What did Bill put [e] in the box?

filler gap

Unboundedness vs islands

• Potentially infinite distance between filler and gap

Who did Bill want Sally to try to invite [e]?

• Yet certain close dependencies are not OK

*What did Bill believe the report Sally said?

(Complex NP Constraint)

Wh-constructions:the challenge for LFG

• Can we avoid recourse to empty categories?

• The construction seems to refer to categories/positions not functions:

a) all categories except VP front

b) categories move to a specific c-structure position

DFs vs GFs

A functional account needs to identify a function for the wh-element:

TOPIC: old information; relatives; topics

FOCUS: new information; questions

SUBJ: grammaticalized DF; default topic

Functional dependencies:outside-in

Who did Bill visit? (FOCUS) = (OBJ)

Who did Bill try to visit? (FOCUS) =

( XCOMP OBJ)

Who did Bill say that Susan visited?

(FOCUS) =

( COMP OBJ)

Who did Bill say that Susan tried to visit?

(FOCUS) =

( COMP XCOMP OBJ)

etc etc

Functional dependencies:inside-out

Who did Bill visit? (OBJ) = (FOCUS)

Who did Bill try to visit? (OBJ) =

((XCOMP FOCUS)

Who did Bill say that Susan visited?

(OBJ) =

((COMP FOCUS)

Who did Bill say that Susan tried to visit?

(OBJ) =

((COMP XCOMP FOCUS)

etc etc

Functional uncertainty

• The infinite set of possible dependencies requires a means of selecting the right one for the sentence in question

• (DF) = ( GF* GF) (Outside-in)

• (GF) = ((GF* DF) (Inside-out)

Outside-in functional uncertainty

• filler-gap relation expressed solely at f-structure with no empty c-structure

• Island constraints statable as conditions on the path from filler function to gap function

• ( DF) = ({COMP, XCOMP}* (GF–COMP))

Off-path constraints

( DF) = ({COMP, XCOMP}* (GF))

• Only COMP and XCOMP can intervene between filler and gap

• So Complex NP Constraint follows since NPs cannot be COMPs or XCOMPs

Inside out functional uncertainty(IOFU)

• there is an empty node in c-structure• the empty node is annotated with the equation:

(GF) = ((GF* DF)• provided there is a legitimate path from the gap to

the required focus or topic function the equations can be solved and the structure is allowed

Why IOFU?

• f- /c-structure correspondences• weak crossover effects• wh- in situ and scope

Canonical structural realization

• SUBJ and OBJ must be realized as nominals (NP or DP) (Bresnan 2001)

a) That he would be late, I never would have believed. (That he would be late = COMP)

b) That he would be late was widely predicted.(That he would be late = ?)

c) Under the bed, we said they would find him.(Under the bed = ADJ)

d) Under the bed is where they found him.(Under the bed = ?)

CSR (cont.)

If that he would be late in (b) is COMP,

and if under the bed in (d) is PP, then CSR is violated.

So, assume a null expletive subject [e]

Weak Crossover

Who does his mother like [e]? (who ≠ his)

• f-precedence: a piece of f-structure f f-precedes a piece of f-structure g if the rightmost node associated with f precedes the rightmost node associated with g.

• A pronominal P cannot f-precede a constituent on which P is referentially dependent.