Back to the Arabic PAPPI homepage

Notes on Arabic PAPPI:
VSO Word Order

1. Theory

1.1 Examples

Taken from [pg19ff, Chapter 2: Word Order, Agreement and Case, Fassi Fehri (1993)]
 
[5]   kataba r-rajul-u   r-risaalat-a   
      wrote  the-man-nom the-letter-acc 
      The man wrote the letter

[6]   Qarad-tu Qan  y-uqaabil-a r-rajul-u   l-mudiir-a
      wanted-I that 3-meet-subj the-man-nom the-director-acc
      I wanted the man to meet the director

[25]  Qa laa y-aQtii zayd-un
      Q  not 3-comes Zayd-nom
      Isn't Zayd coming?
[Notes: orthographical representation limited by ASCII. Q for ? (glottal plosive). Capitalization in general will indicate the presence of unrepresented diacritics.]

Need the following components:

2. Implementation

2.1 Parses

Two parses are reported by PAPPI for (6). Coindexation/non-coindexation of the two expletives.

2.2 V Raising to I

Head movement rules from file xbar.pl:
%% HEAD MOVEMENT
%% NB. requires rule for empty C for adjunction in periphery

rule v(V) moves_to i provided agr(strong), finite(V).
rule v(V) moves_to i provided agr(weak), V has_feature aux.

rule v(V) moves_to i moves_to c provided agr(strong), finite(V).
rule v(V) moves_to i moves_to c provided 
	agr(weak), V has_feature aux, finite(V).

rule i(I) moves_to v(V) provided agr(weak), \+ V has_feature aux,
	post_cond addFeature(noECP(_),I).
rule i(I) moves_to v(V) provided agr(strong), \+ finite(V),
	post_cond addFeature(noECP(_),I).

rule v doesnt_move provided v$vp subcat_by v.
rule v(V) moves_through neg provided_iff negationMoves, V has_feature neg.

rule i(I) derived_from v(V) provided recoverIfromV(I,V).
rule neg(Neg) derived_from v(V) provided negFromV(Neg,V).

% Coping with antecedent government of head traces
rule head_adjoined _ add_feature transparent(ecp).
Set language parameter:
agr(strong)
Rules are selected and compiled into the bottom-up parser with an environment stack for passing HM traces. For details, see "Verbal Inflection, Head Movement and LR Parsing", Fong S. To appear in Principle-Based Parsing: from Theory to Practice, ed. R.C. Berwick, Kluwer Academic.

2.3 Empty Expletive

PAPPI has overt expletives, e.g. it and there.
lex(there,n,[agr([3,[],[]]),nonarg(+),linkTo(np)]).
lex(it,   n,[agr([3,sg,n]), nonarg(+),linkTo(c2)]).
Crucial feature is nonarg(+).

linkTo for expletive linking there is a man in... and it is important that....

Empty categories are inserted essentially as featureless (modulo positional features) at phrase structure recovery time:

Features instantiated as constraints are applied over the course of the derivation:

Typology: PRO, pro, wh-trace, NP-trace, parasitic gap, etc.

In particular, in the standard implementation, BT-relevant features a(_) and p(_) are instantiated by Functional Determination.

Other principles: ECP, AGR Criterion (see AGR Criterion notes)

2.4 VP-Internal Subject

[See detailed notes here.]

2.5 Case Assignment for Spec-VP

In PAPPI's VP-Internal Subject implementation, Spec-VP is not a Case position.

Structural Case assignment occurs under government. Extend definition of government to allow INFL to govern into Spec-VP (bit like ECM):

governs(I,Subj,Proj) :-				  % governing into Spec-VP
	\+ adjoined(Proj),
	I head_of Proj,
	I has_feature agr(AGR),
	poorAGR(AGR),
	XP complement_of Proj,
	Subj specifier_of XP.
In the following tree, Proj = I1/I2, I = I(AGR)[1], XP = VP, Subj = NP[3]:

[For poorAGR/1, see AGR Criterion notes.]

2.6 Case Particles as Markers

% Case
lex(u, mrkr,[left(n,[],[suffix(nom),morphC(nom)])]).
lex(un,mrkr,[left(n,[],[suffix(nom),morphC(nom)])]).
lex(a, mrkr,[left(n,[],[suffix(acc),morphC(acc)])]).
lex(an,mrkr,[left(n,[],[suffix(acc),morphC(acc)])]).

left(C,MatchFeatures,NewFeatures)

Parameter Description
C category to which the marker attaches
MatchFeatures list of features the matching category must have
NewFeatures the features represented by the marker

morphC(nom/acc) is added to the noun to which the marker (mrkr) attaches.

[suffix/1 is for tree display purposes only.]

Example:

2.7 Determiners as Markers

Similar to Case particles, but using feature right/3:
lex(r,mrkr,[right(n,[],[prefix(r),def(+)])]).
for r-raju-u and r-risaalat-a in (5).

[prefix/1 is for tree display purposes only. Doesn't work at the moment.]

Verb agreement morphemes are also implemented this way. For example:

% Agreement
lex(at,mrkr,[left(v,[],[suffix(f),     agr([[],[],f])])]). % f
lex(na,mrkr,[left(v,[],[suffix('f.pl'),agr([[],pl,f])])]). % f.pl
at encodes GEN (f) and na encodes GEN + NUM for daxal-at (entered-f) and daxal-na (entered-f.pl), respectively.

What to do about Qayy-a rajul-in (which-acc man-gen) and kull-a rajul-in (every-acc man-gen)?

2.8 Pronominal Clitics

Example:
[39a] baqarat-an Saahad-tu
      cow-acc    saw-I
      A cow, I saw

Also see parse for (6).

lex(tu,pf(cl),[a(-),p(+),agr([1,sg,[m,f]]),eng(i)|Fs]) :- subjClFeatures(Fs).

subjClFeatures([adjoin(v,right),morphC(nom)]).

2.9 NegP

Example:
[25] Qa laa y-aQtii zayd-un
     Q  not 3-comes Zayd-nom
     Isn't Zayd coming?
Redefine compl/2 relation:
compl(i,vp). compl(neg,i2). compl(c,i2). compl(c,negp).


Back to the Arabic PAPPI homepage