(Ara) Add a distinction in CN and NP whether it's heavy

This makes a difference in word order; default is VSO, but if the
subject NP is complex, e.g. coming from RelNP or SentCN,
the more natural word order is SVO.

Example that triggered the change comes from an application grammar:

Gold standard by informant: مَا تَوَصْلّنَا إلَيهِ يَبْدُو مَنْطِقِيَّاً
Grammar before change: يَشْعُرُ مَاً نَتَوَصَّلُ إِلَىهُ مَنْطِقِيّاً
Grammar after change:  مَاً نَتَوَصَّلُ إِلَىهُ يَشْعُرُ مَنْطِقِيّاً
This commit is contained in:
Inari Listenmaa
2019-01-04 13:01:03 +02:00
parent 19dca60b6e
commit 4141375976
4 changed files with 67 additions and 42 deletions

View File

@@ -5,7 +5,7 @@ lincat
[S] = {s1,s2 : Order => Str} ;
[Adv] = {s1,s2 : Str} ;
[NP] = {s1,s2 : Case => Str ; a : Agr ; empty : Str} ;
[NP] = {s1,s2 : Case => Str ; a : Agr ; empty : Str; isHeavy : Bool} ;
[AP] = {s1,s2 : Species => Gender => Number => State => Case => Str} ;
lin
@@ -21,16 +21,19 @@ lin
BaseNP x y = twoTable Case x y ** {
a = conjAgr x.a y.a ;
empty = []
empty = [] ;
isHeavy = True ;
} ;
ConsNP xs x = consrTable Case comma xs x ** {
a = conjAgr xs.a x.a ;
empty = []
empty = [] ;
isHeavy = True ;
} ;
ConjNP conj ss = conjunctDistrTable Case conj ss ** {
a = let gn = pgn2gn ss.a.pgn in
{pgn = Per3 gn.g (conjNumber conj.n gn.n) ; isPron = False} ;
empty = []
empty = [] ;
isHeavy = True ;
} ;
BaseAP = twoTable5 Species Gender Number State Case ;

View File

@@ -22,7 +22,7 @@ lin
cn.s2 ! number
! (definite ! det.d) -- Indef remains Indef, rest become Def
! c
} in {
} in emptyNP ** {
s = \\c => -- Dat is just a hack for liPrep
let c' = case c of {Dat => Gen ; x => x} in
case cnB4det det of {
@@ -37,13 +37,12 @@ lin
};
a = { pgn = agrP3 cn.h cn.g number;
isPron = False } ;
empty = []
};
isHeavy = cn.isHeavy
} ;
UsePN pn = {
UsePN pn = emptyNP ** {
s = pn.s;
a = {pgn = Per3 pn.g Sg ; isPron = False} ;
empty = []
a = {pgn = Per3 pn.g Sg ; isPron = False}
};
UsePron p = p ;
@@ -70,7 +69,8 @@ lin
-}
AdvNP np adv = np ** {
s = \\c => np.s ! c ++ adv.s
s = \\c => np.s ! c ++ adv.s ;
isHeavy = True ;
};
DetQuantOrd quant num ord = quant ** {
@@ -169,10 +169,11 @@ lin
isEmpty = True
} ;
MassNP cn =
{s = \\c => cn2str cn Sg Indef c ;
MassNP cn = emptyNP ** {
s = \\c => cn2str cn Sg Indef c ;
a = {pgn = Per3 cn.g Sg ; isPron = False} ;
empty = []} ;
isHeavy = cn.isHeavy ;
} ;
UseN,
UseN2 = useN ;
@@ -190,12 +191,22 @@ lin
};
RelCN cn rs = cn ** {
s2 = \\n,s,c => cn.s2 ! n ! s ! c ++ rs.s ! {pgn=Per3 cn.g n ; isPron=False} ! c};
s2 = \\n,s,c => cn.s2 ! n ! s ! c
++ rs.s ! {pgn=Per3 cn.g n ; isPron=False} ! c ;
isHeavy = True
} ;
RelNP np rs = np ** {s = \\c => np.s ! c ++ rs.s ! np.a ! c} ;
RelNP np rs = np ** {
s = \\c => np.s ! c ++ rs.s ! np.a ! c ;
isHeavy = True
} ;
AdvCN,
SentCN = \cn,ss -> cn ** {s2 = \\n,d,c => cn.s2 ! n ! d ! c ++ ss.s} ;
SentCN = \cn,ss -> cn ** {
s2 = \\n,d,c => cn.s2 ! n ! d ! c ++ ss.s ;
isHeavy = True
} ;
ApposCN cn np = cn ** {
np = \\c => cn.np ! c ++ np.s ! c

View File

@@ -173,7 +173,7 @@ resource ResAra = PatternsAra ** open Prelude, Predef, OrthoAra, ParamX in {
uttAP : AP -> (Gender => Str) ;
uttAP ap = \\g => ap.s ! NoHum ! g ! Sg ! Const ! Bare ; ----IL
CN : Type = Noun ** {np : Case => Str};
CN : Type = Noun ** {np : Case => Str ; isHeavy : Bool};
-- All fields of NP
cn2str : CN -> Number -> State -> Case -> Str = \cn,n,s,c ->
@@ -181,7 +181,9 @@ resource ResAra = PatternsAra ** open Prelude, Predef, OrthoAra, ParamX in {
cn.s2 ! n ! s ! c ++
cn.np ! c ;
useN : Noun -> CN = \n -> n ** {np = \\_ => []} ;
useN : Noun -> CN = \n -> n ** {
np = \\_ => [] ;
isHeavy = False } ;
uttCN : CN -> (Gender => Str) ;
uttCN cn = \\_ => cn2str cn Sg Indef Bare ;
@@ -1507,6 +1509,7 @@ patGeminateImp : (_,_ :Str) -> Gender => Number => Str = \facc,facic ->
NP : Type = {
s : Case => Str ;
a : Agr ;
isHeavy : Bool ; -- overrides verbal word order, if the subject is very complicated; e.g. built out of RelNP or similar
empty : Str -- to prevent ambiguities with prodrop
} ;
@@ -1516,15 +1519,14 @@ patGeminateImp : (_,_ :Str) -> Gender => Number => Str = \facc,facic ->
} ;
mkPron : (_,_,_ : Str) -> PerGenNum -> NP = \ana,nI,I,pgn ->
{ s =
emptyNP ** {s =
table {
(Nom|Bare) => ana;
Acc => nI ; -- object suffix
Gen => I ; -- possessive suffix
Dat => I -- will only be used with preposition لِ
};
a = {pgn = pgn; isPron = True };
empty = []
a = {pgn = pgn; isPron = True}
};
proDrop : NP -> NP = \np ->
@@ -1536,6 +1538,7 @@ patGeminateImp : (_,_ :Str) -> Gender => Number => Str = \facc,facic ->
emptyNP : NP = {
s = \\_ => [] ;
a = {pgn = Per3 Masc Sg ; isPron = False} ;
isHeavy = False ;
empty = [] } ;
agrNP : Agr -> NP = \agr -> emptyNP ** {a = agr} ;
@@ -1653,7 +1656,7 @@ patGeminateImp : (_,_ :Str) -> Gender => Number => Str = \facc,facic ->
}
} ;
ip2np : IP -> Bool -> NP = \ip,isPred -> ip ** { s = ip.s ! isPred ! Masc ! Def ; empty = [] } ;
ip2np : IP -> Bool -> NP = \ip,isPred -> emptyNP ** ip ** {s = ip.s ! isPred ! Masc ! Def} ;
np2ip : NP -> IP = \np -> np ** {s = \\_,_,_ => np.s} ;
IDet : Type = {
@@ -1748,7 +1751,7 @@ patGeminateImp : (_,_ :Str) -> Gender => Number => Str = \facc,facic ->
True => np.s ! sc.c
} ;
} in wordOrder o
vp.obj.a.isPron np.a.isPron
vp.obj.a.isPron np.a.isPron np.isHeavy
(vStr vp pgn t p o)
vp.obj.s
(pred vp pgn t p)
@@ -1757,13 +1760,23 @@ patGeminateImp : (_,_ :Str) -> Gender => Number => Str = \facc,facic ->
} ;
-- seems complicated, but this is to share code with VPS and other similar structures
wordOrder : Order -> (objIsPron,subjIsPron : Bool) -> (verb,obj,pred,adv,subj : Str) -> Str =
\o,objIsPron,subjIsPron,verb,obj,pred,adv,subj ->
wordOrder : Order -> (objIsPron,subjIsPron,subjIsHeavy : Bool) -> (verb,obj,pred,adv,subj : Str) -> Str =
\o,objIsPron,subjIsPron,subjIsHeavy,verb,obj,pred,adv,subj ->
let cl = wordOrderNoSubj o objIsPron verb obj pred adv in
case o of {
-- If subject is pronoun, affix it in Subord word order.
Subord =>
let bind = if_then_Str subjIsPron BIND [] -- in subord. clause, subj. pronoun binds to the main verb
let bind = if_then_Str subjIsPron BIND []
in cl.before ++ bind ++ subj ++ cl.after ;
-- If subject is "heavy" (e.g. contains a relative clause),
Verbal => -- then override Verbal word order.
case subjIsHeavy of {
True => subj ++ cl.before ++ cl.after ;
False => cl.before ++ subj ++ cl.after
} ;
-- Any other word order, no special checks.
_ => cl.before ++ subj ++ cl.after
} ;
@@ -1833,7 +1846,7 @@ patGeminateImp : (_,_ :Str) -> Gender => Number => Str = \facc,facic ->
Subj : Type = {s : Case => Str ; isPron : Bool} ;
np2subj : NP -> Subj = \np -> np ** {isPron = np.a.isPron} ;
subj2np : Subj -> NP = \su -> su ** {a = {pgn = emptyNP.a.pgn ; isPron = su.isPron} ; empty=[]} ;
subj2np : Subj -> NP = \su -> emptyNP ** su ** {a = {pgn = emptyNP.a.pgn ; isPron = su.isPron}} ;
emptyObj : Obj = {a = {gn = {g=Masc ; n=Sg} ; isPron = False}; s = []} ;
insertObj : NP -> VPSlash -> VP = \np,vp -> vp ** {

View File

@@ -14,15 +14,13 @@ lin
-- } ;
--IL TODO: check out some opers regarding state in ResAra. These are just dummy values.
CNSymbNP det cn xs =
let g = cn.g ; n = sizeToNumber det.n in {
let g = cn.g ; n = sizeToNumber det.n in emptyNP ** {
s = \\c => det.s ! NoHum ! g ! c ++ cn2str cn n Def c ++ xs.s; ----IL word order?? Seems to be nontrivial according to ResAra comments.
a = dummyAgrP3 n ;
empty = []
a = dummyAgrP3 n
} ;
CNNumNP cn i = {
CNNumNP cn i = emptyNP ** {
s = \\c => cn2str cn Sg Def c ++ uttNum i ! cn.g ;
a = dummyAgrP3 Sg ;
empty = []
a = dummyAgrP3 Sg
} ;
SymbS sy = {s = \\_ => sy.s} ;