Multi-Tape Two-Level Morphology: A Case Study in Semitic Non-linear Morphology

This I)aper presents an implemented multi-tal)e twolevel model capable of describing Semitie non-linear morphology. The computational fl'arnework behind the ettrrcnt work is motivated by [Kay 1987]; the fimnalism presented here is an extension to the formalism reported by [Puhnan art(1 Hepl)le. 1993]. The objectives of the current work are: to stay as close as possible, in spirit, to standard two-level morl)hology, to stay close to the linguistic description of Semitic stems, and to present a model which can be used with ease by the Semitist. The. Imper illustrates that if finite-state transducers (FSTs) in a standard two-level morphology model are replaced with multi-tape attxiliary versions (AFSTs) , one can account for Semitic root-andq)attern morphology using high level notation. 1 I N T R O D U C T I O N This paper aims at presenting a computational morphology model which can handle the non-linear phenomenon of Semitic morphology. The approach presented here builds on two-level mori)hology [Koskennienfi 1983], extending it to achieve the desired objective. Tit('. contril)ution of this l)almr tnay ])e Slllllmarised as follows: With regards to the two-level model, we extend this model by allowing it to have multiI)le tapes on the lexical level and retaining the one tape on the surface level; hence, 'multitape two-level morphology'. Feasible pairs in the standard two-level model become 'feasible tuple pairs' in our multi-tape model. With regards to the formalism, we have. chosen a twodevel formalism and extended it to be al)le to write multi-tape two-level grammars which involve non-linear operations. To achieve this, we made all lexieal expressions n-tuple regular expressions. In addition, we introduced the notion of 'ellipsis', which in*Suppor t ed by a Benefitctor S tudentsh ip from SI+ Jo lm ' s College. q~llis research was done tllld(!r the SUlmrvision tion 5 al)i)lies our model on the Arabic verb. Section 6 I)resents an auxiliary automaton into which multi-tape two-level rules can/)e compiled. Finally, section 7 giw;s eonchtding r e m a r k s . 2 1~10 ( ) T A N D PATTI, ; I )~N M O R P I t O L O G Y Non-linear r o o t a n d p a t t e r n morphology is best il+ lustrated in Semitic. A Setnitic stem (:onsists of a r o o t and a vowel tne lody , ;u'rattged according to a canonic.al i)atte.rn. For examph~, Arahic/Iv'uttib/ 'caused t.o write' is composed front the root murphenm {ktb} 'notion of wril.inp;' and the vowel melody morpheme {ui} 'pertlwt lmssive'; the two are arr:mged act:ording to the pattern morpheme {CVCCVC} 'causative'. Table ] (next page) gives the Arabic perfeetive vethal forms (from [McCarthy 1981]). l t As indicated by [McCar thy 1981], the d a t a i n q'a|fle 1 provi(les s tems [n urtdtwlyhlg morphl)h)i;i(:al forms. Ilence, it, should he noted that : tlICTCld~ C3+S(++ l,~tHt[t(}r gLrld lllli+111)t!l ' Hl3.t'k[llg. i,q IU2)~ shown'~ llh+ttly sl, etns t!xperhmce l~holxcd()gicaJ l)roc,~!ssitlg t.<) give am'face forms, (!.~i. /nkatab/ -+ /?inkatab/ (ffn'm 7); the root, morphemes .shown ar,'+ iwd; +fit++d lit tlm litm+ature in all forms, e.g. Lhere is llo such verb as */tal~attab/ ( form 5), bu t there is /takassab/ from the root m o r p h e m e {ksb}; the qua.lity of the Sl!COlld VOWel ill forth I iS ([iflerent, frm+t ()lie roo£ t() tl+tlOI,hol'+ 1!.+~, /qalal/ %o k i l l ' , / qab i l / %0 accept ' , /kabur/ ' to become I)i~,', front the m e t m o r p h e m e s {qtl}, {qbl} and {kbr}, reSlmctiv(dy. Some ['orNflS do llol. ()(:cut' ill the passive.