#how to calculate the pcfg values for English monosyllables with R
#mike hammond, LSA, 2011

#this file includes how to get counts for the other rules in the pcfg
#this intermingles R code, results, and discussion

#copy the dictionary file
#http://dingo.sbs.arizona.edu/~hammond/lsasummer11/newdic.txt

#read in dictionary
nd <- read.table('newdic',sep='\t',quote='')

#counts for different numbers of syllables
sum(nd[nd$V3=='S1',5])
sum(nd[nd$V3=='S2',5])
sum(nd[nd$V3=='S3',5])
sum(nd[nd$V3=='S4',5])
sum(nd[nd$V3=='S5',5])
sum(nd[nd$V3=='S6',5])

#there are two rules for syllables
w -> s
w -> s w

#here are counts and which rules apply for each
s1 477416	w->s
s2	120440	w->sw	w->s
s3	40745		w->sw	w->sw	w->s
s4	17727		w->sw	w->sw	w->sw	w->s
s5	4045		w->sw	w->sw	w->sw	w->sw	w->s
s6	328		w->sw	w->sw	w->sw	w->sw	w->sw	w->s

#total number of tokens; count for w->s
sum(nd$V5)
w->s  = 660716

#count for w->sw
w->sw = 272931 = (1*120440) + (2*40745) + (3*17727) + (4*4045) + (5*328)

#calculate probability for w->s
p(w->s)  = 0.7076722 = 660716/(660716+272931)

#calculate probability for w->sw
p(w->sw) = 0.2923278 = 272931/(660716+272931)

#actual and then estimated probabilities for different word counts
p(s1) = .72 = 477416/660716
	.71 = p(w->s)
p(s2) = .18 = 120440/660716
	.20 = p(w->sw) * p(w->s)
p(s3) = .06 = 40745/660716
	.06 = p(w->sw) * p(w->sw) * p(w->s)
p(s4) = .03 = 17727/660716
	.02 = p(w->sw) * p(w->sw) * p(w->sw) * p(w->s)
p(s5) = .006 = 4045/660716
	.005 = p(w->sw) * p(w->sw) * p(w->sw) * p(w->sw) * p(w->s)
p(s6) = .0005 = 328/660716
	.001 = p(w->sw) * p(w->sw) * p(w->sw) * p(w->sw) * p(w->sw) * p(w->s)

#calculate values for different syllable expansion rules
#limited to (C)V(C)
#estimated solely on monosyllables

#all monosyllables
nd1 <- nd[nd$V3=='S1',c(1,5)]
#only (C)V(C)
nd1[grep('^.?[aeiouIE@UOWY\\^RcxX].?$',nd1$V1),]

#total number of CVC,CV,VC,V monosyllable tokens
sum(nd1[grep('^.?[aeiouIE@UOWY\\^RcxX].?$',nd1$V1),2])

#V-initial items
sum(nd1[grep('^[aeiouIE@UOWY\\^RcxX].?$',nd1$V1),2])
87512

#obstruent-initial items
sum(nd1[grep('^[ptkbdgfsSvzZCJ][aeiouIE@UOWY\\^RcxX].?$',nd1$V1),2])
116665

#liquid-initial items
sum(nd1[grep('^[lr][aeiouIE@UOWY\\^RcxX].?$',nd1$V1),2])
12681

#nasal-initial items
sum(nd1[grep('^[mn][aeiouIE@UOWY\\^RcxX].?$',nd1$V1),2])
27639

#probabilities for onset rules
p(O -> 0) = .36 = 87512/(87512+116665+12681+27638)
p(O -> obs) = .48 = 116665/(87512+116665+12681+27638)
p(O -> liq) = .05 = 12681/(87512+116665+12681+27638)
p(O -> nas) = .11 = 27639/(87512+116665+12681+27638)

#vowel-final items
sum(nd1[grep('^.?[aeiouIE@UOWY\\^RcxX]$',nd1$V1),2])
126728

#obstruent-final items
sum(nd1[grep('^.?[aeiouIE@UOWY\\^RcxX][ptkbdgfsSvzZCJ]$',nd1$V1),2])
120301

#liquid-final items
sum(nd1[grep('^.?[aeiouIE@UOWY\\^RcxX][lr]$',nd1$V1),2])
46276

#nasal-final items
sum(nd1[grep('^.?[aeiouIE@UOWY\\^RcxX][mnG]$',nd1$V1),2])
67350

#probability values for rhyme rules
p(R -> V) = .35 = 126728/(126728+120301+46276+67350)
p(R -> Vobs) = .33 = 120301/(126728+120301+46276+67350)
p(R -> Vliq) = .13 = 46276/(126728+120301+46276+67350)
p(R -> Vnas) = .19 = 67350/(126728+120301+46276+67350)

#for VCV sequences we don't need all the rules, just
#the onset and rhyme rules

#V-obstruent-V sequences
VO.V = p(R -> VO) * p(O -> 0)
			.12 (41%) = .33 * .36
V.OV = p(R -> V) * p(O -> O)
			.17 (59%) = .35 * .48 

#V-nasal-V sequences
VN.V = p(R -> VN) * p(O -> 0)
			.07 (58%) = .19 * .36
V.NV = p(R -> V) * p(O -> N)
			.05 (42%) = .35 * .11

#V-liquid-V sequences
VL.V = p(R -> VL) * p(O -> 0)
			.05 (71%) = .13 * .36
V.LV = p(R -> V) * p(O -> L)
			.02 (29%) = .35 * .05