Typed Feature Structures As Descriptions

A descr ip t ion is an enti ty tha t can be interpre ted as t rue or false of an object , and using fea ture s t ruc tures as descr ipt ions accrues several compu ta t i ona l benefits. In this paper , 1 create an explicit in te rpre ta t ion of a typed feature s t ruc ture used as a descript ion, define the not ion of a satisfiable fe.ature s t ructure , and create a s imple and effective a lgor i thm to decide if a fe.ature s t ruc ture is satisfiable. 1. I N T R O D U C T I O N Describing ob jec t s is one of several purposes for which l inguists use fe.at, ul:e s t ructures . A descr ipt ion is an er, t i ty t ha t can be in terpre ted as t rue or false of an object . For example , the convent ional in te rp re ta t ion of the descr ipt ion ' i t is black ' is t rue of a soot particle, bu t false of ~ snowtlake, q 'herefore, any use of a feature s t ruc ture to describe an object de lnands tha t the feature s t ruc ture can be in terpre ted as t rue or false of the object . In this paper, I tailor the semant ics of [K~NG 1 989] to suit the typed feature s t ruc tures of [CAII.I'F, NTFA{ 1 992], and so create an explici t in te rpre ta t ion of a typed feature s t ruc tu re used as a descript, ion. I then use this in t e rp re ta t ion to define the not ion of a sat isf iable feature s t ructure . T h o u g h no featm'e s t ruc ture algebra provides descr ip t ions as expressive as those provided by a feature logic, using feature s t ruc tures to describe objec ts profits f rom a large stock of available compu ta t iona l techniques to represent, test and process feature s t ructures . In this paper , I d e m o n s t r a t e the compu ta t iona l benefi ts of ma r ry ing a t rac table syn tax and an explicit semant ics by creat ing a s imple and effective a lgor i thm to decide the satisfiabil i ty *The research presented in this paper was sl)onsored by ']'eilprojekt B4 "Constraints on (h'ammar for Efficient Generation" cff the Sonderforschungsbereich 340 of the Deutsche ForschungsgemeinschafL I also wish to thank Bob Carpenter, Dale (lerdemmm, q'hile) GStz and Jennifer King for their invalualAe hel l) with this paper. tWilhehnst.r. 113, 72t17,1 ~l~{ilfingen, (]el 'Ill[ |fly. Einaih klng®sfs.nphil.unlt uebingen.de. of a feature s t ructure . G e r d e m a n n and Ggtz ' s 'Doll type resolut ion sys t em implemen t s bo th the sen,ant ics and an efficient ref inement of the satisfiabil i ty a lgor i thm I present here (see [C,5TZ 1993], [GEItDF, MANN AND I(ING 1994] and [G~m)EMA~N (VC)]). 2. A F E A T U R E S T R U C T U R E S E M A N T I C S A signatm.e provides the symbols from which to construc.t typed fea ture s t ructures , and an in terpre ta t ion gives those symbols meaning . D e f i n i t i o n 1. E is a siguature iff E is a sextuple (~, %, ~, G, ffl, ~), i s a set, (%,-<} is a partial order, { fo reachrE72 , } = crC72. | f a i r t h c n a = r '