gideon%edai.edinburgh.ac.uk@CS.UCL.AC.UK (Gideon Sahar) (06/17/86)
EDINBURGH AI SEMINARS Date: Wednesday 28th May l986 Place: Department of Artificial Intelligence Seminar Room Forrest Hill EDINBURGH. Dr. M. Steedman, Centre for Cognitive Sciences and Department of Artificial Intelligence will give a seminar entitled - "Combinators, Universals and Natural Language Processing". Combinators are primitive elements in terms of which we can define the notion of defining a function, as with the lambda operator of LISP, without the use of the bound variables which are associated with that operator, and which are so expensive for interpreters of LISP and related functional programming languages. For some time, my colleagues and I have been arguing that the syntax and semantics of certain problematic "unbounded dependencies" and "reduced" constituents in natural language constructions such as English relative clauses and coordinate constructions can be elegantly captured by extending Categorial Grammars (discussed by Ewan Klein here a couple of months ago) with operations corresponding to certain simple combinators. Such grammars hold out the promise of a theory according to which natural language syntax is a very direct reflection of a computational efficient applicative semantics which minimises the use of bound variables. The paper concerns some implications for processing and the prediction of certain contrasts between the grammars of Spanish and English. Date: Wednesday, 4th June l986 Time: 2.00 p.m. Place: Department of Artificial Intelligence, Seminar Room, Forrest Hill, EDINBURGH. Professor Colin Bell, University of Iowa will give a seminar entitled - ``A Point-Based Representation of Temporal Knowledge in Automated Project Planning". A point-based temporal reasoning system is presented as an alternative to existing interval-based temporal logics. It appears to be especially applicable in nonlinear hierarchical planning where such temporal quantities as activity durations and scheduling delays are uncertain. Temporal constraints representable in this system fall into a very restricted class. However, it is argued that representing more general constraints results in computational intractability. Details of implementation are discussed. Date: Wednesday, 11th June l986 Time: 2.00 p.m. Place: Department of Artificial Intelligence, Seminar Room F10, 80 South Bridge, EDINBURGH. Mr. Peter Jackson, Department of Artificial Intelligence, University of Edinburgh will give a seminar entitled - ``Towards a Methodology for Designing Problem Solving Architectures in the Object-Oriented Style". Although current object-oriented systems provide the programmer with both software modules (such as production rule interpreters and theorem provers) and software tools (such as browsers and debuggers), they fail to provide a set of guidelines as to how to select and combine modules to create a particular architecture. Too often, one is given some combination of Flavors, OPS and Prolog (or their look-alikes), and then left to get on with it. A further criticism is that the modules provided do not lend themselves to adaptation by specialization in the spirit of the object-oriented environment in which they are embedded. A methodology for creating 'abstract architectures', which can be instantiated via a process of specialization, is described in the context of a new object-oriented programming language called SLOOP. A detailed example is given of how to create a generic production rule architecture whose behaviour is easy to modify incrementally, together with a sample problem solving program. It is suggested that certain features of SLOOP, namely its transparency and the fact that it is mostly implemented in itself, make it particularly useful as a vehicle for tasks of this kind, while some of the facilities offered, such as pattern-matched parameter-passing and the ability to compile SLOOP into Lisp and thence into native code, encourage a functional style of programming without extracting too high a price in terms of efficiency.