From 5b03c5de2eeb9cdffbc9271eb0fe05c5e84520ba Mon Sep 17 00:00:00 2001
From: Kevin Brubeck Unhammer <pixiemotion@gmail.com>
Date: Sat, 27 Sep 2008 10:12:18 +0200
Subject: [PATCH] Update to remove faulty gold parses. Seems Done now :)

---
 report/report.pdf | Bin 170226 -> 170384 bytes
 report/report.tex |  68 +++++++++++++++++++++++++++++++++---------------------
 src/main.py       |  55 ++++++++++++++++++++++---------------------
 src/wsjdep.py     |   4 ++--
 4 files changed, 73 insertions(+), 54 deletions(-)

diff --git a/report/report.pdf b/report/report.pdf
index 6d37c40cf64c1239c24cd9e9fd49bb7e5d89c99b..5d071330ee43bc30b0d9c993ac2de1507a70012d 100644
GIT binary patch
delta 8026
zcwU89c|4Te+c#rvkZgmBhU_uTea}7j+_#WDLI~MXwk(ycvP29DAqr_!B)cqyilRbN
zktIoyEfgY!$gX!h&-;7c_c8D1^UgnW&G%gAe6Q;|*PQb`GiA6d_i?2?BUltn4)0W;
zZ8Z+)(fje^XlBQ?ptc*$A~i{socRVvkJ`TK>*|q6+Ai;V(syT~qCw}x+7delLAUY2
zQ1R;T$xIRc^L5P*Bo+?VvANF1xE)Spv;8@a1#(s&Szb!DH%jl1H}zG0+$6Vyb?4AG
zBLju=e0RI=esWz}92dW2At<A(Xdb!qt4kg&GZ%_q^t|!9*4Rxpr7p|XyZqDF_s8iH
zB@v$}=E4h$=c=pvcV5vf>o4y8l;ZgTZ)xzTC(B;oYP(#(Hmi}lhm9;nd)tjlZ9A0b
zIfv}kI>A)Qf!76~6i-as%eLa=_q*>UU%0mKg*W}D!^aDUE#_Q%wyY)de8uW@htA+d
zv}xrNRknisZAoks#l5aYo#L7I%ZnZMh^t*50LnaLXIx+B8ONCPD(laRa7q<fy?4j^
zm3d$iIvhgxJvy4d{|><}{C#XvRlo3`CD0NqsE{eW`{E2f*Tcu=D)&CNM<Y$`3w9<p
zqrI#hVea<z>32sYOSxUghEBmRy>IsQF6^Y0N;YNr=d<iPLX6)DwQ-pK%I!LNcyn5$
zhiF>KjpO@_h2{^4`K&6i_NJ%gD~rY$C6rb)n~Ler68uUc_yVtqh%5x_rj|F^|H3(B
zN90NMmgIk^ld@B;*rl#h>wUaZet|3VddekEXuE3NFI?w`HZ1P2r`;`FYfgjUynSZ{
zO=378V~_KYgXPtuTZPP0-<`VX@`lBoS_Cv&MFl^_9KJTHD)pm{tFCZg`$=kN?8je2
z-g-vk{@?fxMT1=y#^%kMl`7HvanQB?qdJ+UvbRJN^&q9KzZ4I3y*S{yM{!UjEi#d$
z!bW%2ap^KnufJYWy8LCbv+<`Kt}989f}!mZlzdc90=4!{yOgOfF4!D6HOAH&X^`Al
zB4@u7Z49A2R$YA6c;e@d`2|2xD8|UK+*-8$*$?`+=DacS!EKR>3z~hSS{_~Z;#Eb!
z{RzH~#li=3S@yn;0{6>%$~n4Uz+le8p`ojT-&g1NK3kX??zvCh_r<~7(P2rhv2L6u
z)VinFt+v+j?erML|Dk|YO+EO}$j}Oxv#gNsiqX027Sd6|@ANYhrW;NzFYY?kVH=jB
zE=W@mP;pHE;M2cvVfmZ)=zy{IRFaFicPmZpd5)2e(1e@QK@a|b569|)^io~;kA_W*
zKaTU9nsLN?;;!9p>9GFJQ!6Wpgk4Bopt;!`wwk*ks?e0BTrNzR$?}+RN(iM*ILAak
zHWWIS@jCrL$;Eos9}fhQ9@VjE$w<BI5oIlca8Th`srh6;RB-x-fTZe6(cEnoE|Okx
zzlFz^Z=D$pb(WMt4zaJ4;cxM$;GzXzN*$PU(U_YDcE=MgHVD49a=w(S6|VZFa*B2u
zeQCnR`kp$Sy%rqWEV1|JtA%iLi|NY7=uJtIn(|840_IP)-^6<Hmef3x`{fC$@8}T>
zZ2e+pmsIFBc2QGzQ>j-*<83Epm!#ib<nHV251wbXEbKEJ@c3QQ60v(*f(1?N{ZWoN
z+A-6ai@K3pgkKGkZ7jHwPpxnOjkT_yX;JY}u;Aw1l3aIw7b|Z)a+GvXy-2g+GFY%s
zhEw%2Yl_v+yKFJIU?wZPckM|=aoBYBrSUQlG=XOp01p1Wmo=>ko$Z3>j|6rb9CHA7
z9n>k)O;vjuip;#*WG(Xqi87o1GkeQ)u%mP7tXFK`o$I4BMSO*Rvn2_0kG<}V&d_F;
zQk^EOQ|0!?>}k-OdzkL@s-Ek0?VAeoljjB+v}95bw?9|>{njeAt%a0A?l2P=qhEEZ
zAC(|D8I2s26$uVeR2nMgms+w=Bn-yrz22Lr!c$&VaL!3MXrb}YO5<X`c>D2*RE}=t
z%8AECXG^lS8)!OKJ)ZSY3WeP4^~U{a-}XmvJf=#&KAq9V{nZiPLQ;49z^xxH6E+$l
z;rKnY!Y*GWEIKb^Q|+pFo-2*2p;2C|X($n5!a+K2mSTJ}MOjul9>ml*pJkzi48g;G
z*KAsp9I$*f89DEd9pLQsbRd)#e|b?zZ*Drak~ZMtIdS}&sP9B%>O&s0s;*HwZO>wh
zeMhXC0q?z~Oeq7g+xY#$&>NiY>7FguUSMhult>G6sh+aCcE;!6QqG;9YQ^}4)}*DS
z#US0QNvD_c=a0QBt17Z{@BC`TpViHRH1FSe_s#wj59$4T8-(sT@?*P=!<S_u%ClX+
zrYdGkV6F~K`!ul5i8kWro22r{7X@hNcV#?IfdhjR21wTUB1=2Zf_7w!&4JRkx#s~t
z#_Xz3mVH;bKk<Ps@rW|7aFQ)P#~Twgbp3e2WI${!tBvP-0VC=BYBGGS*Ua29x9CoC
z9$>G)!g2^Ll$zc8JcW1aMxB^R=Vf<;zAZ{kPpn-^wiS=uh_@56$oht_`f27*Q{M-g
zbDfJImlPHHp&EGKO^-0N(lghNNWJb+IAP=)ul8~Cs$ff*p@o6cqX%Qjf$6%3-laKQ
zK{KDI)zd8}5^R?{67;Lp<$EW+;B>B=shSe%q4Ip*dNWfF@WFgee8Q<ppP8r4rGsjv
zubjTUZhC$5W7EUU?3cw|yv`izdP!4hY!P7Nlob0=fWI`4H(@Q1kM<c|lYmCJh}=qN
zQ|-+rt`^xk@F!G-+M-7(tXIYrGkIIdzu}9kTnY7|c?q993#(+#oqVXCx+?rLIC5@h
z&Tc)EJPk{>HVt74OV_d#7hU*6YXQczj?iC)Q|)Tb#IFPok2_7yHS&!Uv9yz6Mbcro
zIEm&8&xafhhf7;oam`NEK8x0Xn-$;tK(b4Gjd|Jh^r^BH7pI=JZ*jr0n=Nic@dN=_
zGdVLBgDjR(oy3E=nEkU?V-2@yvNgJm&s5xNj31^eZ4Lj<p>)hBttYQK>Pc>GfN5Am
z@Gb6Aqx&vlKiGkTm4gwF_6yPiD<tAxT%|nMn=gb#VP~A^ka~KoP_cgQL7hzX;vg9Z
zyTsi0I<bRv{wmt_6S`kjh*g-hD;Miq$=IPIi4V425$qK)2=~7Zl>1i473fjF<QI?W
z?LK?w<b6!8%PxI`Y99d~L*C4)a#uGRZa83Shoe<Zf6MKXEDiQ&;%}5`mA)nmx#uyr
z8$a+Yrr+N7aLa&cWcw*I{I`M&=SWq8^!%LHe7En0^_;G^s#Yst3mB(#MO|i_z6Hb{
zm(c6ED~t2LK=r<r_^_*Wmj7spJvaQxl=H}*Gi93E9ydYDQ5E_75$R`)SZd?-b)LT9
zYO_2#%jZ(4y4hsp%t_9}y`40bA?%Y6(;7ixBYOdsGk+AdSyO^Ueq!4lQ8ihsDEolB
zlkS0DLRkK@zt}F#n-0hE_rKC0IdEGAOyWIX<%zsFc{0u&8+k&<+g5wq1-e8RUTm%|
zB%m3)`^n1c(Oo10$C|&Zwz%$(DU+@HgVJ@3I`Gzi_W8)?%v38}@K^SuO`WJSdHZ(i
z?8|3ceeHe1c>i=|#T2zfc@XsO6`h55AJX_H%)=!Q&3bp$<<<UqEk7=C>umd#*(pi6
z2bVU_O4oDGKXZpvy|dTmTC1=U56q8vJ+$e2#-gh>S?FtQ2=G8FF`J+6jxzg|)jYmu
zb9QORt7tCqsHU-oQDJzKoF0Bhx8xHKVRhF*gI2kclcNKFSi8;tDBV9XEr!_vAOI1C
zQ33^_5@7%!00=^aAPNB{0Yoy0COaw$CvT3!B|o1zCQL?&FqJ|8U@DOeBLp%9AVOfW
zqyj$q;I9+HREP*5AOT>YR1_e92!#ldliTm%|C&GZ>zFVE5fL&#Ad?UxnF13)3QVLR
z$rv$w@{|%+^0PU?<n&pKf4s<yfMD?d@}kaJ2m@3il?)Qd5K5##jI}_N2qDRiZ~x=v
z`i}>d%9sx@dk7_;_<frL1c?+Xkjy`iaG>j9Q5mLSveUdw@{w}U<c#@5RuBRhDqfzJ
zVr`S$0}F&^9L+2&`DAK|k3s^CZ=I92EE^8=cYR`6ov?#|6SzKw!?>z0PgHK>JuV&D
z>bgTZOoMf7c!D)>&c|;x<fWCWtM1{c&ic@XTqm0w#pTujv7@uzYffP))6QEbQ_4zg
z-1;LurQ$Ae+D<Rptg=KxxkY09p?(%^;G=4ck@e${s;<n*fV|9xpr6h9*#l2$zM28c
zgSQHr^z5qFR;KVs>jCuI6Nke&MX7<Sqe1n$_@_aF?rk|e`q#^mlp<cbk{P@5B6?s5
zlVekVh9B*)K@Jxp7fD{>VI3~H0pW!>$TFm$`H8>ia!LOO<H=j@6$2eFKD?*(g;#w-
z?GL%`W4p`8X7awj#Br*gbA@K@rWmB@viz+3wg%y=*o5)r!q69eI*Efs?yU!tHc1kN
zXMf7DZ)!-I)sKx${H+npmtA&ZGu|&}^Y)HDuWz+VZjX(Kd{0)J!WB$R_`WZBoeq*u
z^Qg}phyZoAs~z&QX+PI7Ula5`msYS`QE}UKMEK#Lm>c#H((#c`+<~{5-KC{N#n`k)
z>P@zU?~P4we?e8TOWN6n+I_Fy+6t;%F!}EO1xjv@ABeFSavM?hTy3x%OwEYA!gGP8
zYL9dNOPeeYj7~yT%8@v+vxfMAtuuF2KXnE}YZ7eZ+j-SK+*-`K_+3Y_3FAnsxqFt!
z|Ng9JpEVz=r0nJ2w<)&#hl!V)LbvB>7`^*UgzQ+%`zr@csLB3*11&w{KX|e>O;>je
zZ_YJwdMEWDU&?==82x@gu=e^WceSPGoq3mDenf0ndFs$rd23sjU5&9@a%N-#74l9^
z{dOx9+pg-YW&a|)GI@U3TF=;6g_eCUfOqKGkFkm5i^FVvNvtEH@n0L<##zo4WZ%@@
z8$q#M5i+xxgTGghS?Vf&@I)DIpOR$xR3d4MQzn^;J-r`3a8(%G<0j(FtynOT-^w9;
zdF!Fe_i=vI`G}Ea$1%@I9`357U3XJ0K2P7?ReIA%bhOR~*9#{MkNFLxw7AgP-*V%k
zbU#0ToJAQ3h79gTNZmHC)ONPjD0mgid%nPpw0FULTU3}Gb^g*v;WKslhR!9Mpj?Sw
zC<^VV%p5(h^j$BuULjv3JGTfp=Fz-o(`C{F>2}BCkZ2)%P4C1#GCyVOrG9bQ5Yle<
z3$MEq4YD``#(@XUv?yb9tB@L6Hc&3GcSJzNej(Mh3j97~ygDh+W_fX{O~LQQmoU|@
zVs~u5Oh?$+-Y9isU6N9EZrii-L3Pl+^t<!?1s84m!rz8J%lC)0S#O+nT(BSR!&KAq
zM^+=}+r9^E@u;zpb2zT`nZ;7ec<H%)2oFz}rx6!x)wgOEYmK(9?Dvuc0a|Q^W8a#v
z^nlpKL{V4Mf#0<3h4}fKkPOg$Fv^h4P)9w#gp-fMdaCSxhr`Q(6BEOtmsZ?oyAo@s
zzSgl!nr!vCBybiYT-Pg*4BRX^^?B^+wtNAFV|tCB`MSyB@~H*a`D*iqRI5t*-v*g(
z5o3S8$!!J{`xJC&RgLXu$lfSg*K*k&h0nEzf21ln3S<v#*Ksx#W0S!&ef}2n!zHbC
zOQNrS4D!2bYaQ!yU9^1K*?M>E+m6NNrv^fokG%U_r*Vl*e(x`>VHt;>PA@FyCyjx#
zH7i5z`wDj@KD}OT9;2xzRb87hx;x`dm&UtEjm6LUw{}J9PGPqM-q7$u@g&;rj{;I7
zmklc-K#H2w2zxpZmTjYQGj4m~`KyK*b{8~s<9udcRhRBO^hKnh%3|Lyt5#lWUB8d5
zSsFJj<Y^Dgdb28s6)^~@5btiSouZ8S(n{WN(AGXTJ!~0x!aMNFi9bm9Ibmmmn}=|7
zJ=@U;og@*mbm@d-csg1CGm91t+vBG1!`}Stw{D|r(Nj0cx_0&1n7ffBA-Hh&#x>Kk
z%X_%5J`{cQ^z`+cbF!m0vl5BUz6s#|q0@5?RpJ+Tbs7a*_$Nq<m+DEp2O_?oDj2#}
z7oif@F|<2EX4iqY5psgI5g~&&@@7tdvlx-gnR&5%Ja)^$XT3%>xJ)dLJB5}ZIrgVE
z2ItRLohElzknN6)7`ZOzk&Y=*tnED-)*sHJwTUaq*;q}{Fy+#Z`u9C=*G{xZp3~C&
z@(bEx>K-MgwAAzXNTm0U*i5zRCPnW~H0$K=6U72N#%aC2j@@7GcUk$uB8UClCepiS
zlk5EtcxOU~o0Y7)UUpB<%vO2Rq8nRe`c@q-aoP!2jBCAK(_XWFeia+}M$l+b%WtP3
zn{iSU=7^AQY2<76^9R2^zZrkx((XsGV|anUm|HgkzsQ+#KFJZIqF4CuZwt+Am1!Hr
zmfSu#{>EW{A&y$IcCd1e7IF5<(1XyCvG1nmf2<jO4UN0gcld(?&S~P#<a7_M$XIiQ
zqRjiq?BmtkY3|JM$!XYRtX5Ko^9YA)7(oSJcuQ>8n}@9Pz>NFn>gtAtSU%yE_qKtL
zdp}LQJMt<vE>(_NICboHnz%UYSF6*JP1Z{%*Y-}&I-HZgh-+>8nnjr6wn$I+3QGyy
zzPba{QD8-Kp6(xzHT&#m^CGRGk!CQRWLR7yX;3MF2L8#YFcv?asuB9q^p1iyYh7mm
zpH+>0=F1`B?W`hCwktxn<lp*NDfSeZWDIw1maVfj!Y_`uCs*7SYZP*wa`mIlT`sa3
zG~>~-mHpK%6SsX=v(S&UhOd6)S6Ge3e0hT5RE5Gw&+aPimu*M=9Hckj(<q_UJl+{u
z!=@Fi-BI2Dk)8B*eAp8DwOew{>#E~ij|t~FYg3@lb_MP<D(KOezxZaZL-k&L_>CLo
zoVNuW125FJm!L=XJf4X>l9(1*A@$n|KfY*cyP!NRc<u}R&P_u(cJ#;jnp~lk@7J(i
zvn!#stIL>Y!fz$2Jc1`i)c0?BVOk1t0PU^$lrtYS_ZtSccm(H3a(t6e4W5Y6*QOlb
zQ)7I^;%$9pLFi8-YP9WfT2-$9-f``1{kdIEk6NVP3?9MC-G6fBj1Y>|nya~xeLdh|
zDh9*|lgUI9d0k~OzDW=PhF~I<@*mYl|AWCC!vpKWiA(@clp)L*%8(2YK{Aw_Ikzp@
zA$lwQ9u{NyA0H4z7-9=$@-bn>=!%e03Xuwv2@pe(K~$6gkr>Jhfarr97(IFo8%CKf
z!G_ThAtN9WB{POW3Xudcgdj?0NIc3vZl~5`CWDlJ)*&+#Bbh?~%z@drKF&u^W5+lP
zg8)O0G9qH=SQKSAkx4`{@^2>zdNc<{7r(w775ry85TXzPki5?KVm&VeBr{@Qh)Iwk
zA{lv6sDF*o=qg+oCr(DTL@I>Pv$-$`2MiJc0HhCcVW{gih(L6SO&BUW2oph&Znp^o
zbCRhr5hlU(R0hrgkQrJVq<5|lfDC#v6`>2@Flrcbv?&f#X^O)Zp5mmEPzWTUbW8|#
z`+t5^N{3)&F|1Sym2Mo0HD-m$5Iud}pui;h<hlVdqEw?X3^D@K<JS!kMd^Ki86btO
z9ELSyXJ~v9i5~K|NdjSdOBiDqC4qGDaE5_E0Nr!lpi&_E)9}A&NOazabs7pxw_i6P
z7^Rm){6z!OSN}3d2;KPnUo-$c{XEu)4Fwp{O`gZ9;HdzCwEiJ@5pe6@00kv60#>_#
zMMV&X*55FQpa`Wh{sPtmD}o}VjiXeAQAXZqth5pWP{>SkDJ1H@F%nE&U(Y@oi^0$U
z1l~9UX3#NVR0@+H6=ud@7$9$CLm2NQCX59AHJ9FZ5o^y%CIfV}7_1{>vq+VhF<471
zD-4n=7vr!(7&a=S_v!rcSZOu{Ai<SrJT?=<s3}74OTczBrjjZjU&gv(uw<AnnTYkH
zQ6bVsHV{CWG4lW62cZxX215uFhLAue4E?(r^oL1UJ_V+P$q>M_41ho=-8cy=Ljy=C
z(*%&r#KCxnGQ~$GZ`=hE<FyDf<pz^Cb`Arh{9Oauzd)42L`)^aOvH>M5ZH)O5N1Cz
z1!7(gA^q=QXn%t!$ke@5h|07eOof;-rZO6x38NrP#0ZF>8#!PSjm(q@0>R9Ui~!74
zKp+Y;Mr9&q941UXfs!bHR{{PPh%!6E%)N@D5dG0rtSpVn)L2x2#1s@2V7&h~Vj%Mt
zGHT0|B;#1$STz!LJ*bVnLuFhI8@GmvLQD-xMX5{`N60WU29YQmgFz_Bzc~Kt9RwyZ
zkuY|Pvat#XLPj@|Fm?vnNW$2ejfnsxVEtm>h#|C%<p)R<3b^qICs7#Py#WIlRb|3Z
z5ZVw7Kt?Fw#$d?*v<dBhKnml?VCrrV-q^^DvQSaxdjcUdV<@okqJl}tzd2~@Ai&h!
z42;oTOc=xzAk4V?nEe=c3loM?m^eUyN@E70jW-?xgPF2}LF&ftXJ81#?1wUKCj*23
z`yaXae?Z2KPo<lsVl^GKD7q*`6M|3*prb|7)CMWK2;*KtQ9W1()df*?g6jY4Voy+4
c$DKaoc<M~xsbj7<#_|ve#L39$n&{#F3%P1R)&Kwi

delta 7855
zcwU89c|25a+n*U@-(`u&$d;vPmNPSFkS$A+QtF2kQV8*rHHkyWzD}8xU5adp?7J5G
zQb{FCNho`k3VDy7_xXLEXXgEU-udTT^S!_KbzS#;U-xysC!vgMsg&!{&=(XLpal=?
zg$5+fdJd77h8uN{nL9t%MF=3;FU-j}y_|mHofaK17%s<u6K$1x;@;sCE4huoeM%Af
z9Wmo)fRqo>RZZBSXLgN21JU2g6t(0Qk8mVA?Q<NYAQV(@O?O?u?=T|NTKA3^;Elg<
zH&f$B-ul@1%x-M19K)UHLOY80qT@c7g+Zao3BZ?1zYbY-n?p3cf~&QiyK3+MRk=7E
zo6$x3dRO*R>F_Ndmjdj3R84k6ZT1l6rUBKvaJJ_!7ptN#x`xtu{K?L(ry;vW*VC!U
zD+UVsUF8b9FL#40E_)V+297;CcHP{&Y^W05IDV(Zmg`w-j8t(@-rf5ndoJ*mUHa|}
zWzZ+sohMG6YTb?643fQl>e0_ViF<04AKsGk$RpU5?t3^SG&$kqMG@0@xGOb6@i=#O
zaO}_<S=aNN$j;+~-#QEQ>7yexxmmnC<W)&qB|*0X0XtI+X6K)|ISw|_4v$ayy?ev{
zRq4x|k<A14>2HN%imN5uqtkC&(@iw#I_Xg8)euxe-P2K*?EJ4!Tr)meK0j_{Hky_@
zue{%UJxMM)$2az@$O}a&i*A(%f<ia>p5H?6cR8UrEnG0-Ti*9@#9XQ^&2*(fW-LN|
zN-6s5+N$Ka01M1iUWOxfkJhP;MRA3l)@puBdu>XbG@5?*Nm{I(MSplRaR=jz+Sly6
z3-YZhu5}x-vwT0|vm;slLrSsm9r+WzcFE$WhmTlR*=35BJqwPaK5w3hEn5$ZvFxbd
zv}tO;6`|$t9kBW$C1`Uo<#sz@FR>AIl!vI6UR2@u_vrU`5(yZTsN?iZPoi65>gVG<
z#XZ0MR$HhAf(P*nX-k*xoAn5NOdD|N8{vV@id>5o?dSibDW>bJaP#t+yzuIXTXttI
zAfTJPg4_!C?l_&da8GHhyYhWwa2~q0^yP(luZ<>#P1dqyMJy`V@kmPw&tN#O#G_%f
zmum8Xy=Dua=eqQgu0wQLUk=H}y1FYtmbdU(-I2Ek@HA_4?LZCOUA%}ksr?czfnwbP
zHD1$g!5_M=sAwngmKfy0bFojR*STe#cjqEzP-|=UQ*HH|#L8Z}*j8YWx03ka1G}>l
zrYH8h+h4ua7sQu&;S=)B{xsR;BZ|sbe<3C%8#FcvJt3C)s^=ve;v%6_&rF6dMJ8q|
z*X-G!Uy6NStdzXdY&9(&_3A_Z5uFBFfH>qUq+qlx==)NM;1MDof^ccw8DDMYu%|8V
zDTmVYLkDz~!{zp;@`zb;<czs@{cvxXH{y3Ya4uB^b6#Q9`Xlu?=Ydq*IU@x}mWPeU
zs1RC5H7-c&t{qQW2$x*^_@r}`G>1afvg3m6Z{NUAf?AD77il$bmn@eS(bEaq2k-Sm
z!JnpI3*+}DZ7ipM>ZL6{d!0^5z90WyNbJBt{+%b4Zwo0tejtp#^&>)VxLekOwc>_C
zj)j`G^8FWwsJc-v-q8YYUhK~4YQIMEB*YUGzYlb^YjmV`ZQY`#ytqyCtD`Qm&$Ktm
ze@N}f6sj8Tv^90OXewaa-RH23V|#Y(k}2e{vmSR|-6=?|NI%kvCd753dFH#<4{ldy
z_pt{rn|4^9x~8nMr7W5y68$V+Rrdjp(8$Q1AEDN3$K`feqPNPcujPnlH6DnLXv9Yg
zjNZ|23s=>>%to)+VHtY;%wYxTNWpfC>Fz_*-Jn5R{~>WzRLRZEJwiti*e4w}2TK-X
zA@Bu7Y3;sBMvMMQC*MYmK+Msc1-2RKh*z!O_IS0VC8%ws8cR)>9G~B;ecUK33a#|E
z1?PRRIu`Qbul3oN<iH}kC0@bNM=EP<lYh_qU9qYidK;D5JS1~q#Bxj7ccY?xdbUI?
z0kNn}ZqP1i8R*}2RYA6Ds3spFIggKiu?`_oNd^k_g@`J}FrEy>f{M-pzESh@f`-4p
z&anR|@%QZ6RJ&WZs9H5?_@iO`%wH~0OCWm)pjPsWeonx|>AQl)ai28Ty0!=_#b<Tj
z1?s)F;xaHzxvuJDp@D@<Hy%Fq3w3VOf9>Wh+<n%#UXBmJ7saYqycCqS{xF*s!d4Ic
zLakbhqY_0v{`z_K)aMU_ygl=sa*x#VZbUTPleume9hrS&r!YxRO;u)Wyey`kBY38w
zX;Vl?B(L71E6m^fI<|8BW0c`neAasXW82G^{{AN_KJFs*HBBXPWPMAYl{kqa*2E%C
zgHz4(FJD%c{&e1v@Y5vx8mZxl*WU{q&!N-?)~U$z4|6Mqlu%c+&3eCN%4-Wah7;pN
ze&$pZs|*Pw-A4vRkr7vJdFl%exJ2yl4{D7~U(f#~D3!nKy8aXFUw)441@C+u&BOMy
z{0y5j&*ZZX_rEW2h28P}Yj4Lvgwl+0GjCN$Zkd;m&PY%Fb74wEb?H>*#kkg?oxo{m
zD9dc=Tj5gF7j(!OGkTfhkE;W}w2KzYcf53q>j(}>()}968h%=5i7LX1+l4KY<u<HV
z0e@OMymiOEU9qz?x;)-hP@bmWEPK5xwufuTVyb#LXbdxT&d+sXJ?+^mgz|=r<cCqi
z*mv4hK`d_;vv!A*%7p7JKfY!$)uM~;fudN>rtE#AT+>!x**sD?2EO*!1$@jYQY(sw
zJ96G-{0yL5@E!bRw{(H)r2YcWNgi~QzYSXE^f{#j353Y*r&TIVA^TLFn<1YHrHZ2&
zA2J0v9wCd9f&z^F5x7y`6;WZ&3GdxCO{%LZwdsB9I!b1C7uu}EPL?jNIzRCfQ-mI_
zcMd3z%L!TDKQQ%_Lr*6gk)(%Q5Yj)KsK56(WpAuc49Cm0P>+iJ?Xlx@Rf*sYQC0iH
zsq5s>EYmCM!hUBn*YrnnvL&B|#TynuRi=fyscZo=Uk4Qa)`=3n{P*2!wcTkUh?!E~
zaICq4WX~gSZwh5+n#;lZ5^PWWl(h|H{S!chY2Es9J@0l<Ot)fA0*y08Z@=|XaYCE;
zkvC7rXS1X~Of<`c{AQc~I^31yVW{WUw?j&j(03_n$Yj3D*QCCSu5alj&GGH<T9C!*
z6WR!w2ONp|{ZuMy9y0uCo}KNvr#8RynO;EO7Psy__rLugY3x1Ob+g)B>ghMrRH!D#
z+Ch~?G5lu<Clm+XG&ge(FzEL@mF&D6b|T#Qy52`y`yq3$^Jz(}Qy4|nLTP*M%^*_q
zCM0pDeb~2RHPeXKfRkcx!f8H1@9<S$5FatJzpw#v>cuA4=)TjaG7MfL%tfi(+>&xg
zIBdF`bF|50a;QpiBgq5R_r;388cNHYQ$D1XpUhkD8yxrhkvmYcKS);I{tF_kk$d^W
zk1p=aL^t#|=}WElUkkO$K7MSO5a5d~pV-@Dvb$&vqKl-bBNA`k=3p@o|Blfst)S|D
za254$QW=vL6OjDvo&C#=PwA{72W0HNqa3)F;x8Q=AfwcDqt>DS>5AK=(+{ZWHP-`A
zZEdBj@BASq!FHanYQw$zv4{A_xL(LntV69;6V!`mLNCgFh%kYMA85VGoa9;TsjR7+
zP7mry4R5g-IzE@(E)r9Ry??ATyW=juW`OQJPx-lcA;Y7>^=prxE3A~Z`&TC);d*O@
zqPt}jaEM60fRK-FBhg&;ZWlbuVitvwyNN^smPDdp$T&O}PsU@2L@E|XBw)yR01HxR
z&HSP?zGyDm$dbJXiGT&E6bznB!4k<JhDZjmu#(+oxP5mK1u0kn2Vn3dJeESmV+d3b
zOC|xdbIKxr9Pll}4nQo9M8ptrAeKlXV+dp_7R1rc^@-9JR5@tcE5fvzWwU=+L<*Kn
zCNQ%cMTs~pl>}gj02NCDU_Sr>OTm%gP!crvRWaJ!%87p{6bhCE;u$EFY(x?~Os4Hy
z1KB_TOT?3DF0cuwzHR2O<*?xK_&*x`mc=k_s6r5GY}6&DIpv9S8w!jKPP9vRADIDt
zdMA6z<D@Y{LMaL8gF%`$TfgS)bke-ChL&|uRThZo;n6_E{<-2}sO{&C7lkp#4b|$G
zz+THl;kz2iG0~Si&HvWaPkX1}<W=%*7OV9A;?|?hYikk?UnYsUwnj;p?CvZ<@w0J?
ze#@V1f>-QldXEA1(%(W4N*@`X_D?MzUOL3j*Oq$GId72M#r;>r^7M`Q#?o`o%f9on
z6K8rTVFz5_F0aoXmAmXUT7K%_cpARZa>k*_`PZ9@*2IsRccpr%#@bO*Vyd^~T^Gj3
zfYer`)!L)(ms8?LdZOES-YWDrcR@6lJT0GxFJIU57Y$^M-^tA}eR)QY5-5KnQra@P
zeAaw0)HTpZsCi7F$+uwk1Jzg~!2gn@_pT>lvVrfGfS129ixZryr^j3D1qWu4u}}~T
zZ``4^BC!TlS#?@eR9nhug$7%#<*y1i=NE|rW_s*>65W}y$b7;pkljIzw;0mRk>_Cb
z4{X*N=<aypV3B;P)y-t=v<YdqR;*EMUwX#nA8wuJ{3H<lb+&I^PaD^>&mL`Sw@i8Q
zna`;GF=3SBFE*RKZx-ZJRG&JmwLQsXsZ&b)bmH}38LN(-YId=FZvWt2zTi>5<rSsK
zap9wKU{IW3(OxYx_JxB|15({+=*!e1G4kQ_w~tO2AwvSxr`m;@#husgT&XWk?7Mhs
z2=n*z!-=#iB_kKUdl}7LA-odl!bj3Yc^{seW_@<|GEm`thxJq>Fa5}&g)62aZ%yvS
zcp7yYyAOeJIdq4-ns?b*=F8l)rm$$-BzwF<vwa_n*ZaU3FMf^{i_+i-XEx|th}KFl
z`ejY-gZhV=+?r>dT)fBfUr&8eim`kZzFQe(E?fUG!K}d4Yu}Z?T(y&I@woDP{g>y2
z-{ub_=M@cF3VCx~(v)=ce%%|PJ2N9Ew-}<H^6}6nE9vaePO}}cg-==DKR7U({^R!j
zaOJ{J`!+YLyPDffI2Jy82Sygf1VC?<Jzjd8c&cLU<kWP7bt?L1%z^o+zBjHWY0W<n
zCk^T)#=3$>P1UDMS)NsE2fp%iY?`yl_gm+kl*aT0R+==wL^mBb*_q8^p@e3e@=-%0
z2U29lM4E`=8r~7%=ud_x*C4jGN(o&#Ht~~{o1>V(iD8A#soR(KHRo~~%ZNbXh2E2;
zh#i`IKD$?zbDwBNAtw}z?g#Vn*zD%+`(225eq|3o+Xb@-;Asy&RO()!`>QiD589N(
zHe~$n3$U0~w;}s(rW(%QH3y`H^$db-TT(1M&S@vRigiOleE~Q#jymb^Rh~-%`;Phv
zvEi5Z2?A`fI(mn-#WW6>rA3rMtTKD}I0c>d-{susjnUNp*rsB@f=F+koIR`Tx11Vb
z$cq$8*x1Y6uM$+@GR!kq7SizD`sYA@RDHhp<TI;#>g=+qiEq=JIy%kuUKp*^pSHLm
z;l&^LIXB4^Ec_Vcs3ezc_H*f)3dZ_(xq~uavKsP>_$>Z<hUNRaimb2M#33=wQaMg#
zHBKw2j#tHQz_;O^c3Vc_C`cJWd$zjS`<<Kp_U-Cu?4{lBr84Edr})@CVOt{QExH#y
z<bwi7pRWwt)NzhJZ&zOExT!;q+J8dXB3%9&ub|U$Y(!JBA=(Ek5RBGCSHwq*355ns
zt0WCu`!(v`p!rShkm9yl7l2y&?)*%<E-r%8b$HVJWVzeNd_m)CbW<s4FfF&@clGr_
zUL6*lCGGQaYCM`PIG&LB%qq#lRnmCR!e-V1QOAjlP-E-#kE|5MIyAR&a_6U(r-Zeg
zN%~*YNC$b0Y+M~ipPF*ohpA#Q%MeDNtuy=-wmcwCrrPJJ#<@LyQCW~KF1^#3qdu4@
zM+nAHpS;d_crt40;LfX3?;$s8=&`);Yj@b#xVmxINl@d>f`!?G_;N(!!{cvQSj;uJ
zm1<i>qtPZTGHMog$m^=B!cID4X4M0H6-6sOU3^3GwP7{xq*B{cyCxqLyfS_jUuMNs
z!=LkJx2uVTGl$GN|MYo?_qfpVz8-<F9TTyK1c^?f9CfPWwo&&E?G#o%_QA6D=(<Ni
z%*est#eLpp{F}lZL6g8I2X@D&pKsXYc{E4oYN+HV{8IZG<=k?%q@f~Ueco2;Gq^vU
zdR6^o=E|MtV&y;OzXYz8qi6zFy(aOa)~-UIO6DTR6b4QtJ>Rd#RskjKS#=)u*%!=W
z#KBj`srp0Py-@f7_sKWL>{Yikt*_YJ85$Zu+||iE)Ke5@-dSH?JmanEC-i;UAi04{
zI6C9}y$y%r1ZwB*?kyY113RU^W+i>+wlL%PA>!Pl<Bf?-QP>J!GVMSho$@;>orLwy
zYMpETv|Q+=gcco5$nocgB<nVM*Nb^Yj@UN#dFJlbVQC#0ohuG({PgHd-Ny$HgwhY2
z`c&utDma!gZo(U4tE$jp=Q(fbZyIAO)ZOo@d1iC-tNs<=;GpDayT!K0O}_}|e}DD!
z8&Jh<8D9uhaY+1RLNuBX8kJdHniAz+^_^eX;!_%{;`2EEYTZTM;|IizDW?EucG-v`
zvdka2T%60$*?6}1#rwf7!)vD!@MkSI2;D7tznXmGUk$Vpid$|(-RVFajl|b!W4ud_
znIGMoRobKSYDweqt<uI4(_Z$MFZV}gjO}UcX+1$TI^;gTA(i`C%vX3J`^5{EIcffd
zw4gVpXa6p=`gP^HJC8BIVG`-{t|;d&pU3IbE=EJ#{G5mAW6vym;{Y=4ac3<$jqB<U
z0eilq{KofkS0~qA?^f`hY0VL~imj-&W&Lp2*6}aZWBi{vI0m*5eeT#Bm1HE6fD|R;
zuy`V@Nk~L24pvNL3K0te)PE}_dMXNOF9MKZ-9o_Nsj&DVk}zag6cLHE{*qlu1(I|V
z7Uao)GYA9_eo`42vn)tGF@OLIHdsLcum}PuR183b1sk3~4`N3e(Bsg^%c1}t2a7!r
z0}x199Ds+PI9SrrA$FwUHenw<k`<}Hoi9KD{&7R3;IRNe{l^VIoyvwhF9LvA!gfBe
zj-*my771p9|4rIWmuE-n3vc_Q{Ns*5Az%pvdMO(mGm`^JrEkF;QFwJ!0<1=<6fA|X
zy;rcBr2I)8qQB!n+Od-HWGqNS#_yn;aUvDjKq{7i$J2v2k(6zl2omY_+cvz%L_A3U
z$%!Ph12`fq0s%Ui3myP)uy(}LJ-OfkA`XC4A=4jmArBzQ;d5Nb$~i97Ll1U59wdP{
z5<MdTCHe0^l`R1%1tbfJN~M1bM47N`PgIAt4IF_Uy=@TjB>K>{0iO+BB?va~a3ue2
z14pLNJA+WhtnhSzK;ISo#{|hB-6a?vrT_rFcH1Ds`Ey@~4HA`1x4Ql(41r#-Z4jv-
zef^ICuh%%_PZ)rHciX^|aP)5>f5H&y>Y=D3Xjre4>CvGmHBo?w!EOIAZWylZCmt>u
zIN_mC6jh9ZV;%$vluDH_loLt}BoG)!DP;I)3&K%ysuU`f#55Nc=Kqbs%LWPL%Jpy*
z5_tee02ybHDFBWMgZG6gE<hKIMD0)kNF>GpAP)C`1O+K{-$>Ls7CePWAKEsEctWLC
z6zUX;1t60u=`ko_B$^DuyIm5Cl0#Fd09ZK~i^@W>;3zn{e?00PJe5#so`7;dqHrL+
zF%jhrZC^WvNCZ5Y8G~anVE~!Tgpmn2CX7M>m@w+UHjeK<K@ycNNJH%qBT|`s;6VbO
z{w)b558>c1AcjFA0ml@8NMKG5{yYR3F)|D@VFUs*2H(PM4EisS!dNiyIT3J7fk8Z#
z$svit%p;QUObda_^FJ31{R@Oqrh0-u_n5NY?xu_mVKNg2Fs~Tg?&}OONFYRECK7SX
z6-*{E?*bX5Fkuwt6@ws|DQOU-{;U2D|0hT#G7;gTXDU=G1!N2Yw@w^g_ZCV40-2Jc
z;0bs}zi{7XIzqU?F)fk;P#AX&M}hA%6A%7?2N~;$LM1S^2^b?W?Ks?%{tpbKG6n(R
zzQgDbZu|cg<4-+-aOY#(1-NMv8G}%9cmiV(Dvrn)1dqcL35=Z-hllSPBSs=ZOofai
zk%^2L9;7g1R2;~V3?9CuWX5EO%w>Wn5%83M9TW7QAc?V%;fjYZB_jrsDa;s!#Q5HU
zTMLEp29a?L#SLwPc(@)IiFg9@WIRA*=8<vC7<`_L7_4QOF#-TFg8<X&$OIB&^}`q$
zWacsVB{BiTF}`!R+yDQ@rLWvZsoUxS@PPmX9RM^S5{P7y9{Hewz7C0i(<cE`Jql41
kqw)Vz&SEq*xjenjd3as%uy^3X!><4U;F6cuKW4!7KZC>G_W%F@

diff --git a/report/report.tex b/report/report.tex
index d3b94ac..d630da1 100644
--- a/report/report.tex
+++ b/report/report.tex
@@ -1,4 +1,4 @@
-% merged Kevin's & Emily's reports 2008-09-21 Sun 10:39 with Emerge
+% merged Kevin's & Emily's reports 2008-09-21 Sun 19:04 with Emerge
 
 \documentclass[11pt,a4paper]{article}
 \usepackage[utf8]{inputenc}
@@ -534,44 +534,39 @@ next section discusses the effect of these.
 \subsection{Results}
 We compared the results of the implementation with a dependency parsed
 version of the WSJ-10 corpus (converted from the manually annotated
-version). Since single word sentences were not POS-tagged there, these
-were skipped. Also, the dependency parsed WSJ-10 did not have ROOT
-nodes; so we calculated precision and recall both without counting our
-ROOT links, and with counting ROOT links, by adding these to the gold
-parses where possible\footnote{221 parses in the dependency parsed
-  WSJ-10 had several tokens appearing as heads without appearing as
-  dependents in the same parse, here we skipped the parses when
-  calculating with ROOT links.}.
+version using a Perl script by Valentin Jijkoun). Since single word
+sentences were not POS-tagged there, these were skipped. Also, the
+dependency parsed WSJ-10 did not have ROOT nodes; so we calculated
+precision and recall both without counting our ROOT links, and with
+counting ROOT links, by adding these to the gold parses where
+possible\footnote{221 parses in the dependency parsed WSJ-10 had
+  several tokens appearing as heads without appearing as dependents in
+  the same parse, here we skipped the parses when calculating with
+  ROOT links. Our gold standard also sometimes (for 1249 sentences)
+  had one dependent with two heads, these were skipped from
+  evaluation. We have not yet had time to run evaluation of undirected
+  dependencies.}.
 
 \begin{table*}[hb]
   \centering
   \begin{tabular}{l|ccc}
     Model                          & P             & R              & F1          \\
     \hline 
-    LBRANCH/RHEAD                  & 25.6          & 32.6           & 28.7        \\
-    RANDOM                         & 31.0          & 39.4           & 34.7        \\ 
-    RBRANCH/LHEAD                  & 55.1          & 70.0           & 61.7        \\
-    K\&M's DMV                     & 46.6          & 59.2           & 52.1        \\
+    LBRANCH/RHEAD                  &               &                & 33.6        \\
+    RANDOM                         &               &                & 30.1        \\ 
+    RBRANCH/LHEAD                  &               &                & 24.0        \\
+    K\&M's DMV                     &               &                & 43.2        \\
     Our DMV:                                                                      \\
     Uniform initial distribution   & 21.0 (18.7)   & 20.1 (18.1)    & 20.5 (18.4) \\
-    $C_A=0; C_S=1;C_N=0.1;C_M=1$    & 23.7 (23.7)   & 24.8 (24.5)    & 24.2 (24.1) \\ 
-    $C_A=0; C_S=1;C_N=0.1;C_M=10$   & 26.7 (31.6)   & 25.5 (30.5)    & 26.1 (31.0) \\ 
+    $C_A=0; C_S=1;C_N=0.1;C_M=1$    & 24.8 (24.5)   & 23.7 (23.7)    & 24.2 (24.1) \\ % local screen
+    $C_A=0; C_S=1;C_N=0.1;C_M=10$   & 26.7 (31.6)   & 25.5 (30.5)    & 26.1 (31.0) \\ % uib screen
 %    $C_A=10;C_S=1;C_N=0.1;C_M=10$   & 25.6 (30.6)    & 24.5 (29.6)   & 25.0 (30.1) \\
-    $C_A=10;C_S=1;C_N=3  ;C_M=10$   & 26.0 (31.0)   & 24.9 (30.0)    & 25.5 (30.5) \\
-    $C_A=15;C_S=3;C_N=1  ;C_M=20$   & 26.7 (31.7)   & 25.6 (30.6)    & 26.2 (31.2) \\ 
+    $C_A=10;C_S=1;C_N=3  ;C_M=10$   &               &                & 26.3 (31.4) \\ 
+    $C_A=15;C_S=3;C_N=1  ;C_M=20$   &               &                & 27.2 (32.2) \\
   \end{tabular}
   \caption{DMV results on the WSJ-10 for various initialization values (numbers in parentheses are when counting added ROOT links)}
   \label{tab:dmv-wsj}
 \end{table*}
-% trying locally: ?
-% HARMONIC_C: 115.854497176, STOP_C: 1.47684590293, NSTOP_C:
-% 4.27464793921, FSTOP_MIN: 6.9710245489
-% P:  11627/45444 = 0.255853357979 | P_r:  15710/51712 = 0.303797957921
-% R:  11627/47419 = 0.245197072903 | R_r:  15710/53466 = 0.29383159391
-% F1:               0.250411897096 | F1_r:               0.298731673924 
-% 
-% trying remotely: uniform distribution
-% todo: fix results when done
 
 We tried various values for the initialization constants; but it was
 hard to find any clear pattern for what worked best. 
@@ -596,6 +591,27 @@ not clear whether their DMV-experiment was run using automatically
 induced word classes \citep[Schütze, 1995, in][p.~8]{km-dmv} or on the
 tagset used to manually annotate the WSJ-10.
 
+% underproposed =     # $C_A=10;C_S=1;C_N=3  ;C_M=10$
+% {(('NN', 1), ('DT', 0)): 347,
+%  (('NN', 2), ('DT', 0)): 148,
+%  (('NN', 3), ('DT', 2)): 136,
+%  (('NN', 4), ('DT', 3)): 144,
+%  (('NN', 5), ('DT', 4)): 128,
+%  (('NN', 6), ('DT', 5)): 124,
+%  (('NNP', 1), ('NNP', 0)): 358,
+%  (('NNP', 2), ('NNP', 0)): 125,
+%  (('NNP', 2), ('NNP', 1)): 174,
+%  (('NNS', 1), ('JJ', 0)): 124,
+%  (('ROOT', -1), ('NN', 0)): 100,
+%  (('ROOT', -1), ('NN', 1)): 106,
+%  (('ROOT', -1), ('NNP', 1)): 140,
+%  (('ROOT', -1), ('NNP', 2)): 104,
+%  (('VBD', 2), ('NN', 1)): 145,
+%  (('VBP', 2), ('NNS', 1)): 111,
+%  (('VBZ', 2), ('NN', 1)): 152,
+%  (('VBZ', 2), ('NNP', 1)): 107,
+%  (('VBZ', 3), ('NN', 2)): 109, }
+
 In the stochastic grammars given by the model after EM, the POS class
 $VBD$ (past tense verb) had the highest probability of being attached
 to by ROOT, followed by $VBZ$ (3sg present) and $VBP$ (non-3sg
diff --git a/src/main.py b/src/main.py
index 7b90b7d..a4384ba 100644
--- a/src/main.py
+++ b/src/main.py
@@ -16,11 +16,11 @@ def initialize_loc_h(tagonlys):
 #     loc_h_harmonic.HARMONIC_C = 380.111684914
 #     loc_h_harmonic.FSTOP_MIN = 13.5744632704
 #     loc_h_harmonic.FNONSTOP_MIN = 34.8939452454
-    loc_h_harmonic.HARMONIC_C = 0.0 #120.0 * random.random() # 509.63
-    loc_h_harmonic.STOP_C = 1.0 #3.0 * random.random() 
-    loc_h_harmonic.NSTOP_C = 0.1 #5.0 * random.random() # 0.1
-    loc_h_harmonic.FSTOP_MIN = 1.0 #20.0 * random.random() # 13.08 
-
+    loc_h_harmonic.HARMONIC_C =  0.0 #120.0 * random.random() # 509.63
+    loc_h_harmonic.STOP_C =      1.0 #3.0 * random.random() 
+    loc_h_harmonic.NSTOP_C =     0.1 #5.0 * random.random() # 0.1
+    loc_h_harmonic.FSTOP_MIN =  10.0 #20.0 * random.random() # 13.08 
+#   $C_A=0; C_S=1;C_N=0.1;C_M=10$
     loc_h_harmonic.RIGHT_FIRST = 1.0
     loc_h_harmonic.OLD_STOP_CALC = False
     print '''
@@ -53,14 +53,14 @@ def test_likelihood(reestimate, initialize, inner_sent,
         sumlog,msg = corpus_likelihood(g, tagonlys)
         print msg
         if EVAL:
-            E = evaluate(g, tags_and_parses)
-            print E
+            g.E = evaluate(g, tags_and_parses)
+            print g.E
         for i in range(iterations):
             g = reestimate(g, tagonlys)
             print "reestimation number %d done\n"%i
             if EVAL:
-                E = evaluate(g, tags_and_parses)
-                print E
+                g.E = evaluate(g, tags_and_parses)
+                print g.E
             prev_sumlog = sumlog
             sumlog,msg = corpus_likelihood(g, tagonlys)
             if sumlog < prev_sumlog:
@@ -70,14 +70,8 @@ def test_likelihood(reestimate, initialize, inner_sent,
             from pickle import dump # let us say g = pickle.load(open('..','rb'))
             filehandler = open('current_grammar.obj','w')
             dump(g, filehandler)
-            
-        if EVAL:
-            import pprint
-            print "underproposed:"
-            pprint.pprint(E.underproposed)
-            print "overproposed:"
-            pprint.pprint(E.overproposed)
-        
+            filehandler.close()
+                    
         return g
 
     def corpus_likelihood(g, tagsonly):
@@ -95,16 +89,20 @@ def test_likelihood(reestimate, initialize, inner_sent,
     reader = WSJDepCorpusReader(None)
     tagonlys = reader.tagonly_sents()[corpus_offset:corpus_offset+corpus_size]
     tags_and_parses = reader.tagged_and_parsed_sents()[corpus_offset:corpus_offset+corpus_size]
-
-#     from loc_h_dmv import testcorpus
-#     tagonlys = testcorpus
     
     print "\ninitializing %d sentences..." % corpus_size,
     g = initialize(tagonlys)
     print "initialized"
     
     g = run_IO(g, iterations, tagonlys, tags_and_parses) # make iterations argument, todo
-    
+
+    if EVAL:
+        import pprint
+        print "underproposed:"
+        pprint.pprint(g.E.underproposed)
+        print "overproposed:"
+        pprint.pprint(g.E.overproposed)
+
     return g
 
 
@@ -117,6 +115,7 @@ class Evaluation():
             self.R[nd], self.R_r[nd], self.P[nd], self.P_r[nd] = 0, 0, 0, 0
 
         self.unrooted = 0 # parses where we couldn't add_root
+        self.double_heads = 0 # parses w/ two heads to one argument
         self._precision, self._recall, self._precision_r, self._recall_r = 0.0, 0.0, 0.0, 0.0
         self._F1, self._F1_r = 0.0, 0.0
         
@@ -142,15 +141,15 @@ class Evaluation():
         P_rden = self.P['den']+self.P_r['den']
         str_vals = (self.P['num'],self.P['den'],self._precision, P_rnum,P_rden,self._precision_r,
                     self.R['num'],self.R['den'],self._recall,    R_rnum,R_rden,self._recall_r,
-                    self._F1, self._F1_r, self.unrooted)
+                    self._F1, self._F1_r, self.unrooted, self.double_heads)
         regular_str = '''P:  %5d/%5d = %s | P_r:  %5d/%5d = %s
 R:  %5d/%5d = %s | R_r:  %5d/%5d = %s
-F1:               %s | F1_r:               %s (unrooted gold parses: %d)'''%str_vals
+F1:               %s | F1_r:               %s (unrooted gold parses: %d, double-headed: %d)'''%str_vals
 
         tex_str_vals = tuple([p * 100 for p in (self._precision,self._precision_r,self._recall,self._recall_r,self._F1,self._F1_r)])
         tex_str = "$C_A=; C_S=;C_N=;C_M=$   & %.1f (%.1f)   & %.1f (%.1f)    & %.1f (%.1f) \\"%tex_str_vals
 
-        return tex_str # todo
+        return tex_str # todo make variable
                 
         
 
@@ -168,9 +167,12 @@ def evaluate(g, tagged_and_parsed_sents):
     E = Evaluation()
 
     for sent, gold_parse in tagged_and_parsed_sents:
+        if len(sent)-1 != len(gold_parse):
+            E.double_heads += 1
+            continue
         mpp_sent = mpp(g, sent)
         try: gold_parse = add_root(gold_parse)
-        except ValueError: E.unrooted += 1
+        except RuntimeError: E.unrooted += 1
 
         for pair in gold_parse:
             dict = E.R
@@ -274,6 +276,7 @@ def rnd_grammars_test():
                       H,S,N,M)
                     for g,H,S,N,M in rnd_grammars1]
 
+
 if __name__ == "__main__":
     print "main.py:"
     
@@ -287,7 +290,7 @@ if __name__ == "__main__":
                             initialize_loc_h,
                             loc_h_dmv.inner_sent,
                             corpus_size=6268,
-                            iterations=50,
+                            iterations=30,
                             corpus_offset=0,
                             EVAL=True)
         print g 
diff --git a/src/wsjdep.py b/src/wsjdep.py
index 8dd59bf..1f66590 100644
--- a/src/wsjdep.py
+++ b/src/wsjdep.py
@@ -161,12 +161,12 @@ def add_root(parse):
     for (head,loc_h) in set([h for h,a in parse]):
         if (head,loc_h) not in set([a for h,a in parse]):
             if rooted:
-                raise ValueError, "Several possible roots in parse: %s"%(list(parse),)
+                raise RuntimeError, "Several possible roots in parse: %s"%(list(parse),)
             else:
                 rooted = (head,loc_h)
 
     if not rooted:
-        raise ValueError, "No root in parse!"
+        raise RuntimeError, "No root in parse!"
     
     rootpair = (MPPROOT, rooted)
     return parse.union(set([ rootpair ]))
-- 
2.11.4.GIT