Glue logic vs . Spreading Architecture in LFG

Although the f-structures of LFG look like an intuitively clear representation of many aspects of meaning, it has seemed somewhat harder than one might expect to connect them to conventional logical forms, over which entailment and other semantic properties and relations can be defined by standard methods. In this talk I will present what I believe to be a relatively easy formulation of the 'glue logic' approach (Dalrym-ple 1999, 2000), and use it to eliminate the 'restriction projections' used by Andrews and Manning (1993, 1999) in their analyses of scoping modifiers and complex predicates , allowing the analyses to work with the conventional 'locational projections' of Kaplan (1995). The basic idea is to apply recent work by de Groote (1999) and Perrier (1999) to formulate semantic assembly as a system of 'content flow', whereby the 'meaning constructors' introduced by lexical items (and perhaps grammatical constructions) are construed as 'content processors' which collect inputs from and deliver outputs to various locations in the f-structure. If the content-processors can shift content between different locations in the f-structure, it is natural to ask whether they could shift content between different semantically relevant projections, as required by the Andrews and Manning analyses, and the answer turns out to be that they can. As a theory of semantic assembly, glue logic is agnostic about the exact nature of the meaning-elements that are being assembled; there is a tradition, of using Mon-tague's Intentional Logic without the intensions in introductory expositions. I will follow this tradition here. First I will talk about basic predicate-argument structure, then quantifiers, and finally, scoping modifiers. I will close with some more general remarks. In the semantic format we'll be using, the simplest kind of meaning-constructor is one for a proper name. All meaning-constructors consist of two components, on the left, the 'meaning-side', a specification of a meaning, on the right the 'glue-side', a specification of the information needed to control assembly. This will consist of a specification of (Montogovian) semantic type (based on e and t, in the introductory version), combined with f-structural 'locational' information (where the input or output will be collected from/delivered to). So if a proper name such as Art is inserted under an N with f-structure g, the resulting constructor will look like this, where the semantic type is subscripted to the location (but is however often omitted, although not in this paper): (1) art : g …