% % The above mentioned file should become obsolete, % and the author of the original file should preferaby modify this file instead. % % Modificatios were needed in order to support native UTF-8 engines, % but functionality (hopefully) didn't change in any way, at least not intentionally. % This file is no longer stand-alone; at least for 8-bit engines % you probably want to use loadhyph-foo.tex (which will load this file) instead. % % Modifications were done by Jonathan Kew, Mojca Miklavec & Arthur Reutenauer % with help & support from: % - Karl Berry, who gave us free hands and all resources % - Taco Hoekwater, with useful macros % - Hans Hagen, who did the unicodifisation of patterns already long before % and helped with testing, suggestions and bug reports % - Norbert Preining, who tested & integrated patterns into TeX Live % % However, the 'copyright/copyleft' owner of patterns remains the original author. % % The copyright statement of this file is thus: % % Do with this file whatever needs to be done in future for the sake of % 'a better world' as long as you respect the copyright of original file. % If you're the original author of patterns or taking over a new revolution, % plese remove all of the TUG comments & credits that we added here - % you are the Queen / the King, we are only the servants. % % If you want to change this file, rather than uploading directly to CTAN, % we would be grateful if you could send it to us (http://tug.org/tex-hyphen) % or ask for credentials for SVN repository and commit it yourself; % we will then upload the whole 'package' to CTAN. % % Before a new 'pattern-revolution' starts, % please try to follow some guidelines if possible: % % - \lccode is *forbidden*, and I really mean it % - all the patterns should be in UTF-8 % - the only 'allowed' TeX commands in this file are: \patterns, \hyphenation, % and if you really cannot do without, also \input and \message % - in particular, please no \catcode or \lccode changes, % they belong to loadhyph-foo.tex, % and no \lefthyphenmin and \righthyphenmin, % they have no influence here and belong elsewhere % - \begingroup and/or \endinput is not needed % - feel free to do whatever you want inside comments % % We know that TeX is extremely powerful, but give a stupid parser % at least a chance to read your patterns. % % For more unformation see % % http://tug.org/tex-hyphen % %------------------------------------------------------------------------------ % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% ROHYPHEN.TEX, version 1.1 <29.10.1996> R [7.11.1996] %% %% (C) 1995-1996 Adrian Rezus [adriaan@{sci,cs}.kun.nl] %% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% %% Romanian TeX hyphenation table: NFSS 2 encoding, medium. %% Contents: 647 Romanian hyphen patterns, with diacritics. %% %% This file is part of the Romanian TeX system. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Romanian TeX, version 1.3R <29.10.1996> %% %% (C) 1994-1996 Adrian Rezus %% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% History: %% ROHYPHEN.TEX 1.0 <10.02.1995>: Plain TeX and LaTeX 2.09. %% ROHYPHEN.TEX 1.1 <29.10.1996>: Plain TeX and LaTeX2e. % % ------------------------------------------------------------------- % TODO: fix the notice below - it only holds for the old patterns % NB This file must be used in conjunction with either one of % % (1) ROMANIAN.TEX v1.2(R) [1994-1995] [(La)TeX] or % (2) ROMANIAN.STY v1.3R [1996] [(La)TeX(2e)] % % NB Romanian has LR-HYPHEN-MINs [2 2] (like German)! % NB Romanian has STRUCTURAL HYPHEN-AMBIGUA: % i.e., words that canNOT be hyphenated correctly without % additional (e.g., semantic, stress-mark) information. % -------------------------------------------------------- % The Romanian TeX encoding of the Romanian diacritics: % -------------------------------------------------------- % Romanian TeX DQ-macro encodings = (La)TeX macros % -------------------------------------------------------- % ă = \u{a} [-] \u{A} [not encoded] % â = \^{a} [-] \^{A} [not encoded] % î = \^{\i} 'I = \^{I} % ș = \c{s} 'S = \c{S} % ț = \c{t} 'T = \c{T} % ------------------------------------------------------------- % NB Romanian \^{a} behaves like \^{\i} as regards hyphenation. % NB The capital \u{A} and \^{A} are rare in script; as such, % they occur only in records of the Romanian substandard. % ------------------------------------------------------------------- % % original patterns generated by PatGen2-output hyphen-level 9: do NOT modify the list by hand! //============================================================================================================ */ $patgenLanguage = 'Romanian'; $patgenExceptions = array(); $patgenMaxSeg = 7; $patgen = array( 'begin'=>array( 'aic'=>'0300', 'anis'=>'04300', 'az'=>'020', 'cre'=>'0001', 'deaj'=>'00200', 'dez'=>'0021', 'g'=>'04', 'ia'=>'020', 'ie'=>'020', 'iț'=>'030', 'iu'=>'043', 'iv'=>'030', 'îm'=>'040', 'n'=>'02', 'ni'=>'002', 'p'=>'04', 'preș'=>'00030', 's'=>'04', 'ș'=>'04', 'ui'=>'040', 'uni'=>'0500', 'z'=>'02' ), 'end'=>array( 'an'=>'200', 'ăti'=>'0200', 'b'=>'20', 'bia'=>'0020', 'bține'=>'000400', 'c'=>'40', 'chi'=>'2000', 'ci'=>'200', 'd'=>'40', 'f'=>'20', 'fi'=>'200', 'g'=>'20', 'ghi'=>'2000', 'gi'=>'200', 'h'=>'20', 'hi'=>'200', 'i'=>'40', 'j'=>'20', 'ji'=>'200', 'l'=>'40', 'li'=>'400', 'm'=>'20', 'mi'=>'400', 'n'=>'40', 'ni'=>'400', 'obi'=>'0200', 'omedie'=>'0000020', 'orte'=>'00200', 'p'=>'20', 'pi'=>'200', 'pie'=>'0030', 'pți'=>'0400', 'r'=>'40', 'ri'=>'400', 's'=>'40', 'sc'=>'400', 'see'=>'0040', 'ș'=>'40', 'și'=>'400', 'ști'=>'4000', 't'=>'40', 'ti'=>'400', 'tii'=>'3000', 'tî'=>'200', 'tru'=>'3000', 'ț'=>'20', 'ți'=>'200', 'ția'=>'0030', 'u'=>'60', 'ua'=>'020', 'v'=>'20', 'vi'=>'200', 'x'=>'20', 'z'=>'20', 'zi'=>'200' ), 'all'=>array( 'a'=>'01', 'acă'=>'2000', 'achi'=>'00005', 'ae'=>'030', 'afo'=>'0003', 'aia'=>'0320', 'aie'=>'0320', 'ail'=>'0300', 'ais'=>'0032', 'aiu'=>'0300', 'alie'=>'00006', 'alt'=>'2000', 'am'=>'020', 'an'=>'020', 'ane'=>'0520', 'anie'=>'00020', 'aniș'=>'00034', 'ans'=>'0040', 'anu'=>'2000', 'anz'=>'0020', 'aog'=>'0020', 'atia'=>'00040', 'atr'=>'2000', 'atu'=>'0540', 'ața'=>'2000', 'ață'=>'2000', 'au'=>'200', 'aua'=>'0300', 'aud'=>'0300', 'aug'=>'0300', 'aul'=>'0300', 'aun'=>'0300', 'aur'=>'0300', 'aus'=>'0300', 'aute'=>'03000', 'auț'=>'0320', 'auz'=>'0300', 'ă'=>'21', 'ăi'=>'030', 'ăie'=>'0020', 'ăm'=>'022', 'ănu'=>'0003', 'ărgi'=>'00005', 'ăș'=>'030', 'ășt'=>'0430', 'ătie'=>'00040', 'ău'=>'030', 'ăv'=>'030', 'ăzi'=>'0200', 'b'=>'10', 'baț'=>'0020', 'bănu'=>'00005', 'bc'=>'200', 'bd'=>'200', 'biat'=>'00200', 'bie'=>'0020', 'bii'=>'3000', 'bl'=>'020', 'blim'=>'34000', 'blu'=>'0400', 'bo'=>'001', 'boric'=>'003000', 'bs'=>'200', 'bt'=>'200', 'bț'=>'200', 'bu'=>'003', 'c'=>'10', 'caut'=>'00300', 'căc'=>'0020', 'cătu'=>'00005', 'cc'=>'200', 'cea'=>'0020', 'ceț'=>'0020', 'ciale'=>'003000', 'cio'=>'0020', 'cis'=>'0002', 'cisp'=>'00300', 'ciza'=>'00002', 'cl'=>'040', 'cm'=>'200', 'cn'=>'250', 'copiată'=>'00000200', 'coț'=>'0020', 'cs'=>'200', 'ct'=>'200', 'cț'=>'200', 'cuim'=>'00300', 'cul'=>'3000', 'cuț'=>'0020', 'cv'=>'200', 'd'=>'10', 'dam'=>'0040', 'daț'=>'0020', 'dc'=>'200', 'desc'=>'00400', 'dezin'=>'000300', 'dian'=>'00200', 'diată'=>'000200', 'dj'=>'200', 'dm'=>'200', 'dn'=>'210', 'doil'=>'00400', 'du'=>'300', 'eac'=>'0100', 'eaj'=>'0100', 'eal'=>'0100', 'eaș'=>'0100', 'eat'=>'0100', 'eaț'=>'0020', 'eav'=>'0100', 'ebui'=>'00050', 'ec'=>'200', 'ecia'=>'00020', 'eclare'=>'0000200', 'ediulu'=>'0004000', 'ee'=>'030', 'eea'=>'0020', 'efa'=>'1000', 'eh'=>'010', 'eia'=>'0320', 'eie'=>'0320', 'eii'=>'0300', 'eil'=>'0300', 'eim'=>'0300', 'ein'=>'0300', 'eio'=>'0320', 'eis'=>'0332', 'eit'=>'0300', 'eiu'=>'0340', 'eî'=>'010', 'el'=>'200', 'em'=>'020', 'emon'=>'00005', 'en'=>'200', 'ene'=>'0500', 'eo'=>'011', 'eon'=>'0300', 'er'=>'010', 'era'=>'2000', 'eră'=>'2000', 'erc'=>'2000', 'es'=>'220', 'esco'=>'00300', 'esti'=>'00500', 'eș'=>'200', 'eși'=>'0300', 'etanț'=>'000040', 'eț'=>'200', 'eu'=>'030', 'euș'=>'0050', 'evit'=>'10000', 'ex'=>'020', 'ez'=>'200', 'eză'=>'0005', 'ezia'=>'00030', 'ezo'=>'0210', 'f'=>'14', 'fa'=>'300', 'făș'=>'3000', 'fie'=>'0030', 'fo'=>'300', 'ft'=>'200', 'ftu'=>'0500', 'g'=>'12', 'găț'=>'0030', 'gl'=>'040', 'gm'=>'230', 'gn'=>'230', 'gon'=>'0050', 'gu'=>'303', 'gv'=>'230', 'hia'=>'0020', 'hic'=>'0030', 'hiu'=>'0040', 'hn'=>'210', 'i'=>'21', 'iac'=>'3200', 'iag'=>'0034', 'iai'=>'0200', 'iaș'=>'0200', 'iaț'=>'0020', 'ică'=>'0300', 'ied'=>'0200', 'iia'=>'0300', 'iie'=>'0300', 'iii'=>'0300', 'iil'=>'0300', 'iin'=>'0300', 'iir'=>'0300', 'iit'=>'0300', 'iitură'=>'0000200', 'iî'=>'020', 'ila'=>'4000', 'ile'=>'0300', 'ilo'=>'0300', 'imateri'=>'00000006', 'in'=>'020', 'ined'=>'04100', 'ingă'=>'00200', 'inții'=>'000040', 'inv'=>'3000', 'iod'=>'0300', 'ioni'=>'03000', 'ioț'=>'0020', 'ipă'=>'0005', 'is'=>'020', 'isf'=>'0030', 'isp'=>'4000', 'ișt'=>'0030', 'iti'=>'0500', 'iția'=>'00020', 'ițio'=>'03020', 'iua'=>'0300', 'iul'=>'0300', 'ium'=>'0300', 'iund'=>'03000', 'iunu'=>'03000', 'ius'=>'0300', 'iut'=>'0300', 'izv'=>'0030', 'î'=>'02', 'îd'=>'030', 'îe'=>'030', 'îlo'=>'0300', 'îna'=>'0003', 'înș'=>'0050', 'îri'=>'0300', 'îrî'=>'0300', 'îrș'=>'0050', 'îșt'=>'0030', 'ît'=>'030', 'îti'=>'0400', 'îț'=>'030', 'îți'=>'0400', 'îții'=>'05000', 'îz'=>'030', 'j'=>'10', 'jd'=>'200', 'jiț'=>'0020', 'jl'=>'200', 'ju'=>'040', 'jut'=>'0030', 'k'=>'10', 'l'=>'10', 'larați'=>'0000002', 'lăti'=>'00200', 'lătu'=>'00005', 'lb'=>'200', 'lc'=>'200', 'ld'=>'200', 'lea'=>'0020', 'lf'=>'200', 'lg'=>'200', 'lia'=>'0030', 'lie'=>'0030', 'lio'=>'0030', 'lm'=>'200', 'ln'=>'250', 'lp'=>'200', 'ls'=>'200', 'lș'=>'230', 'lt'=>'200', 'lț'=>'200', 'lu'=>'300', 'lv'=>'200', 'm'=>'10', 'ma'=>'300', 'mă'=>'300', 'mb'=>'200', 'mblîn'=>'000003', 'me'=>'300', 'mez'=>'0020', 'mf'=>'200', 'mi'=>'300', 'miț'=>'0020', 'mî'=>'300', 'mn'=>'210', 'mo'=>'300', 'mon'=>'0004', 'mp'=>'200', 'ms'=>'232', 'mt'=>'200', 'mț'=>'200', 'mu'=>'300', 'muț'=>'0020', 'mv'=>'200', 'na'=>'300', 'nad'=>'4100', 'nain'=>'00300', 'nă'=>'300', 'nc'=>'200', 'ncis'=>'02000', 'nciz'=>'02000', 'nd'=>'200', 'ne'=>'300', 'neab'=>'00100', 'nean'=>'00100', 'neap'=>'00100', 'nef'=>'4000', 'neg'=>'4100', 'nes'=>'0032', 'nevi'=>'40000', 'nex'=>'4100', 'ng'=>'200', 'ngăt'=>'00300', 'ni'=>'300', 'niez'=>'00300', 'nî'=>'300', 'nj'=>'030', 'nn'=>'010', 'no'=>'300', 'noș'=>'0040', 'nr'=>'010', 'ns'=>'232', 'nsf'=>'0030', 'nsî'=>'0400', 'nspo'=>'00300', 'nș'=>'032', 'nși'=>'0400', 'nt'=>'200', 'nti'=>'0500', 'ntu'=>'0540', 'nț'=>'200', 'nu'=>'500', 'nua'=>'0030', 'nuă'=>'0030', 'num'=>'0050', 'nus'=>'0032', 'nz'=>'200', 'oag'=>'0100', 'oal'=>'0200', 'oca'=>'2000', 'ocui'=>'00050', 'od'=>'200', 'odia'=>'00020', 'oe'=>'030', 'oi'=>'032', 'oiecti'=>'0000002', 'oisp'=>'00320', 'omn'=>'0040', 'on'=>'200', 'oo'=>'010', 'opie'=>'00030', 'opla'=>'00002', 'oplagi'=>'0000002', 'ora'=>'0100', 'oră'=>'0100', 'orc'=>'0020', 'ore'=>'0100', 'ori'=>'0100', 'oric'=>'02000', 'orî'=>'0100', 'oro'=>'0100', 'oru'=>'0100', 'osti'=>'00500', 'oși'=>'0300', 'otați'=>'000004', 'oti'=>'0500', 'otod'=>'00300', 'ou'=>'030', 'p'=>'12', 'pa'=>'300', 'părț'=>'00030', 'pc'=>'230', 'pecți'=>'000002', 'peț'=>'0020', 'pie'=>'0020', 'piez'=>'00300', 'pio'=>'0030', 'piț'=>'0020', 'piz'=>'0020', 'pl'=>'040', 'poș'=>'0040', 'poț'=>'0020', 'ps'=>'230', 'pș'=>'230', 'pt'=>'230', 'pț'=>'230', 'pub'=>'0034', 'purie'=>'000020', 'puș'=>'0040', 'rb'=>'200', 'rc'=>'200', 'rd'=>'200', 're'=>'020', 'rebi'=>'00200', 'recizi'=>'0000002', 'rescr'=>'003200', 'reși'=>'00400', 'rf'=>'200', 'rg'=>'200', 'rh'=>'210', 'ria'=>'0030', 'riali'=>'004000', 'rieț'=>'00300', 'riez'=>'00300', 'rimi'=>'00500', 'riun'=>'20300', 'riv'=>'0030', 'rk'=>'200', 'rl'=>'200', 'rm'=>'200', 'rn'=>'210', 'rnaț'=>'00020', 'rografi'=>'00000006', 'rp'=>'200', 'rr'=>'210', 'rs'=>'202', 'rsp'=>'0300', 'rst'=>'0300', 'rș'=>'230', 'rt'=>'200', 'rtuale'=>'0000200', 'rț'=>'200', 'ruil'=>'00300', 'rusp'=>'00300', 'rv'=>'200', 'rz'=>'200', 's'=>'10', 'sa'=>'500', 'să'=>'500', 'săm'=>'0040', 'săș'=>'0040', 'sc'=>'200', 'sco'=>'3200', 'se'=>'300', 'sea'=>'0020', 'ses'=>'0002', 'sesp'=>'00300', 'seș'=>'0040', 'sf'=>'420', 'sfî'=>'5000', 'si'=>'300', 'sip'=>'0030', 'sî'=>'300', 'sl'=>'340', 'sm'=>'400', 'sn'=>'010', 'so'=>'300', 'soric'=>'003000', 'sp'=>'200', 'st'=>'200', 'sto'=>'0003', 'su'=>'500', 'suț'=>'0020', 'ș'=>'20', 'șa'=>'300', 'șaț'=>'0020', 'șă'=>'302', 'șe'=>'300', 'și'=>'100', 'șii'=>'5000', 'șil'=>'5000', 'șin'=>'3000', 'șî'=>'300', 'șn'=>'450', 'șnu'=>'0005', 'șo'=>'300', 'șp'=>'020', 'ști'=>'0200', 'ștr'=>'4300', 'șu'=>'300', 't'=>'12', 'taut'=>'00300', 'tc'=>'230', 'td'=>'230', 'tea'=>'0020', 'teni'=>'00500', 'terială'=>'00006000', 'tesp'=>'00320', 'tf'=>'230', 'tia'=>'0030', 'tie'=>'0030', 'til'=>'3000', 'tin'=>'3000', 'tiț'=>'0020', 'tl'=>'040', 'tm'=>'230', 'tol'=>'3000', 'tor'=>'3000', 'toto'=>'00200', 'trul'=>'30000', 'truo'=>'30000', 'ts'=>'432', 'tt'=>'230', 'tua'=>'0030', 'tuim'=>'00300', 'tun'=>'4300', 'tuș'=>'0040', 'tz'=>'430', 'ț'=>'10', 'ța'=>'300', 'ță'=>'300', 'țeț'=>'0020', 'ția'=>'3000', 'ție'=>'3000', 'ții'=>'3000', 'țil'=>'3000', 'țiț'=>'0020', 'țiu'=>'3000', 'țu'=>'003', 'țui'=>'0050', 'u'=>'21', 'uad'=>'0200', 'uau'=>'0300', 'uă'=>'003', 'uăs'=>'0002', 'ubia'=>'02000', 'ubl'=>'0230', 'ubo'=>'0210', 'ubs'=>'0032', 'ue'=>'030', 'ugu'=>'4000', 'uia'=>'0320', 'uie'=>'0320', 'uin'=>'0300', 'uir'=>'0300', 'uis'=>'0300', 'uit'=>'0300', 'uiț'=>'0320', 'uiz'=>'0300', 'ul'=>'020', 'ula'=>'0300', 'ulă'=>'0300', 'ule'=>'0300', 'ulii'=>'03000', 'ulî'=>'0300', 'ulo'=>'0300', 'umir'=>'00050', 'urz'=>'0020', 'us'=>'020', 'uspr'=>'00200', 'ust'=>'0400', 'uș'=>'030', 'ușt'=>'0400', 'uto'=>'0200', 'utor'=>'30000', 'uui'=>'0300', 'uum'=>'0300', 'v'=>'10', 'veni'=>'00500', 'veț'=>'0020', 'vez'=>'0020', 'viț'=>'0020', 'vn'=>'210', 'vorbito'=>'00000002', 'vr'=>'300', 'x'=>'10', 'xa'=>'300', 'xă'=>'300', 'xe'=>'300', 'xez'=>'0020', 'xi'=>'300', 'xo'=>'300', 'xu'=>'300', 'z'=>'10', 'zaț'=>'0020', 'zb'=>'200', 'zg'=>'220', 'zian'=>'00200', 'ziar'=>'00200', 'zii'=>'3000', 'zil'=>'3000', 'zm'=>'040', 'zn'=>'210', 'zol'=>'3200', 'zon'=>'3000', 'zuț'=>'0020', 'zv'=>'220', 'zvă'=>'0300' ) ); ?>