The New Grammar Formalisms - A Tutorial Survey (Abstract)

'lliE NE.W GRAMMAR FOOMALISMS - A 'lUIORIAL SURVEY

(Abstract)

Gerald Gazdar

SChcx:ll of Cognitive sciences, university of SUssex, Brighton m1 ~. united Kin;Jdan

Take two or three ~ case letters ani glan them together with a "G" on the en:i, ani, pres­ to, you have the name for a new gramnar formal­ ism. or, at least, so it nust seem in a decade that has seen the arrival of cx:x;, ax;, rx:x;, ru;, FUG, GPSG, HPSG, JPSG, IFG, MCX;, MIG, R:;PSG, uo:;, ani XG. But the plethora of acronyms (ani nonacronyns - PATR II), disguises nuch oamoonal­ ity, shared prqlerties which set them apart fran the mainstream li.rguistic ani ocrrp.Itational lirguistic activity of the 1970s. '!he key characteristics of the new gramnar formalisms, characteristics eJdlibited by JOOst of them, are (i) the availability, at least in principle ani sanetimes in practice, of a declarative seman­ tics for the notation used, (ii) the use of a basically type 2 rule fonrat (sin;Jle IOOther, \.IDOrdered, no explicit context sensitivity) un::ler (iii) a node admissibility rather than a strin;J rewritin;J interpretation, with (iv) a re­ cursively defined tree or directed acyclic gra};h based category set, ani (v) l.Dlification as the primary ~tion for cc:rrbinin;J syntactic infor­ mation. It is primarily the shared prqlerties, rather than the myriad differences of notation, that this survey fcx::usses on, althcu;Jh sane of the xrore interestin;J differences between them emerge fran the historical aoc::o.mt of conceptual links ani antecedents.

Declarative gramnar formalisms -were not in vogue in the 1970s. In li.rguistics, the daninant transfonrational tradition had essentially ceased to use aey rule formalism at all in research \«<rk (thcu;Jh the t:extl:looks of the period still provided expositions of the rule formalisms used in 1960s research) · All that you will see in papers of the time is a widely used informal notation for structural descrip­ tions ani a taxonany of syntactic constructions whose taxa have the names of the grammatical rules prq;xJSed by li.rguists in the 1960s - Rais­ in;J, Gappin;J, Equi, Passive, ani so on. In na­ tural lanJUage processin;J (NIP) , gramnars ani grammatical rules were equally hard to fin:l. '!he augmented transition network (A'IN) apparatus provided a paradigm for designin;J piogz:aniS in which gramnar, parser ani, often, semantics -were inextricably intertwined. Even prrx;m is far closer to a procedural lan::Jila<Je for programnin;J Marcus-parsers than it is to a declarative gram­ mar formalism.

'!he inspirations for the gramnar formalisms of

the 1980s are manifold. Within li.rguistics, many of those who had been profoun:lly ilrpressed by the technical rigour ani formal elegance of Montague's \«<rk on semantics began to react against the informal rut still baroque character of mainstream 1970s \«<rk on syntax. In search of s:inplicity ani precision, they looked again at pretransfonrational gramnar formalisms such as deperrl~ granmar, };hrase structure granmar, ani categorial grammar (the latter havin;J been used by ~ in his fragments of ED;Jlish syntax). As a result we see the appearance of Depen:le1'x:y unification Granmar (J:U;) , General­ ized Rlrase Structure Granmar (GPSG), Canbinato­ ry categorial Granmar (cx:x;) ani their various descen:lants ani hybrids such as HPSG (Head­ driven Rlrase Structure Granmar), JPSG (used by the NIP group at IOJI') , R-GPSG (fran MIT) , ax; (categorial unification Granmar), MCX; (Meta­ categorial Grammar) ani uo:; (unification categorial Granmar) ·

Within the NIP OCI'll'l'lD1ity, two develcpnents, one :Elll:qlean, the other North American, began to converge. '!he first of these was Prolog. Ccm­ J;Utational lirguistic considerations provided the primary rrotivation for COl.merauer' s design of the lan::Jila<Je ani his Metaxror};hosis Grammars are thus nuch xrore intilllately related to Prolog than A'!Ns are to Lisp, say. Metanoqilosis Gram­ mar was to be the first of a whole series of logic grammar formalisms, a series that has in­ cluded eXtraposition Granmar (XG) , Gappin;J Gram­ mar, Slot Granmar, ani M::xiul.ar logic Granmar (MIG) · urx:Joobtedly the JOOst influential trember of this group, however, is the Definite Clause Granmar (IXX;) notation. '!he convergent North American develcpnent was the appeararre of Kay's Fllnct:ional Grammar, subsequently renamed F\mc­ tional unification Granmar (FUG) ani the the ap­ pearan:::e of the closely related lexical F\mc­ tional Granmar (IFG). FUG was directly influen­ tial, especially in its advocacy of l.Dlification, in many of the granmar types listed in the precedin;J paragra};h. IFG, like GPSG, was in­ terned to embody enpirical claims about the na­ ture of all natural languages in the details of its formal design. By contrast, granmar formal­ isms like rx:x;, FUG ani PATR II (itself the pro­ geny of rx:x;, FUG ani GPSG) are not interned to make li.rguistic claims, rut merely to provide li.rguistically natural ani CCI!p.ltationally tractable languages for writin;J gramnars in. A recent trern has seen carpilation fran formal­ isms of the fonner type into fonralisms of the latter type.