Parsing Agglutinative Word Structures and Its Application to Spelling Checking for Turkish

PARSING AGGLUTINATIVE WORD STRUCTURES AND ITS APPLICATION TO SPELLING CHECKING FOR TURKISH by AY,~IN SOI,AK and KEMAL OFI,AZER l)el>arl, In<'nt, of ('onlputer Engineering and llfformation Sciences Bilkent Uniw~rsity Ililkenl. Ankara, 06533 Tilrkiye ABSTRACT Most of tile research on parsing natnral [allguages has beetl concerned with I",nglish, or wil, h other languages nlOrl)hologically similar Io English. Parsing agglntinat.ive word st, ructures ha.s altracted relatively little attcnl;ion most probal~ly becanse agghlfinatiw? lallgllages COlll~aill word s/ructtlres of considerable complexity, and parsing WOrdS ill Stlch languages I'(?(llliros morphok~gical analysis techniques. Ill this pal)er, we pi'eSell(r the design and implementation of a morphological root-driven parser tor Turkish word structures which has been mcorporatoed into a spelling checking kerllel for on-line Tiirkish texl, The agghltmative Ilatllre of the language and the resulting ('Olll[)l<?x Wol'd ['ornlatiollS, V;ll'iOllS pholleLic llall/lOlly l'tlleS alld sill)tie eKcepLiOllS [)reselll, cel'taill difficulties llOl usually on('ountered in the spelling checking of laagua,ges like English and make this a very challenging probhnH. 1. Introduction Morphological cbussilicat, ion of natttral languages ac- cording to their word Stl'tI('ttlrt+s idaces languages like Turkish, Finnish, and lhmgarian Io a class called "ag ghfl+inalive langua.ges". [n sllch hmguages, words are COlllbillaLiOll of several Iilorphel]les. There is a root and several suffixes are conlbined lo this root in order to modil},' or extend its meaning. Whal characterizes agglut, inative languages is thai stem fornlation hy at" fixation 1o previously derived st.oms is extremely pro- ductive. A given stellL ew'n Ihough itself qlnt0 corn ph+x, call generally serve as basis for evell lllol'l' ('o111 l)lex words. Consequ.ntly, agglutinative languages contain words of considerable COnll~lexity, and parsing such languages necessitates a thorough morphological analysis. Morphological parsing has allracted relatively ]itl,le attention ill con'tputational linguistics. Tile reason is that nearly all parsing research has been concerned wMl English, or wit.h languages morl)hologicaII ) sim- ilar to English. ,qillce in such languages words con- tain only a fi~w nalldJer of affixes, or none at all. alhnost all of the parsing mod<+ls [br Ill(!lll consider recognizing those affix<+s +Is being trivial and thus do nol require a mOt'l+hological nnalysis, hi agghni native langaages, words C(/lll,ail111o direct indication Of t/lOrl;llel/le bOtlltdarios whMi at,, i. gellela[ (IOpOll dent on tit(? inorpho]ogieal and pllon(Jh)gical conlex[+ A morphological parser requires a nlorphold/OaOlog- lest[ COlllpollellt which l/lediat, es I)olwl?ell I[he Sill[kl('t • 1['o1'111 of a [llorp]lellll! as ellco/llllel'l?d ill Ihe il/ptll text aud the lexical form in which the t]torl)h<~me is stor<.d ill tile lllOl'phellle illVelltory, ie, a i[WallS of i'e('oglliZ- ing variallt forms of [l/Ol'phelllOS as tll~! SaltlO. alld a nlorl)hotactic component which specilies which corn hi.rot,ions of Inorl)henws at," Iwrn:itt,'d [7] \lorphotogical parsing algorithms ma+x he divided it/to Ix',() classes as ollir .slrtpl~la 9 ;llt(I rool-df'iv~ It ;nlal+ ysis met.hods. FIolh approaches hawr beell Ilse/l frOlll very early on in l.he history of morphologicM parsing, For instance, I)ackal'd's parser flw ancien| Greek [15). aud Brodda and Karlsson's for Finnish [3] used affix slripping. Sagval[, on tile other hand, devised a root- driwnl morpllological analyzer for Russian [17]. In addition, …

Content truncated to 500 words