Abstracts
Dr Gavin Bierman
This talk will be a non-technical introduction to object databases. I will give some reasons why i think it is worth reconsidering relational database technology, and show how object databases can help. In particular I will discuss the work of the ODMG (Object Data Management Group) in defining a standard for persistance of objects in databases. I will also mention work in thw Opera Group in Cambridge involving the ODMG standard.
Don Syme
An Overview of .NET Common Language Runtime Research at MSR Cambridge
The .NET Common Language Runtime (CLR) provides services for the execution
of software, e.g. JIT compilation and garbage collection. As the name
suggests, it has been designed to support a range of programming languages
and to help them interoperate. MSR Cambridge have been involved with the
platform since mid 1998, analysing the design, writing compilers for the
platform and proposing extensions to the design. Several significant
research projects are currently underway, and this talk will give an
overview of these. The projects include:
o A formal description and analysis of the verification rules for
the CLR's intermediary language, MS-IL (Gordon and Syme);
o The design and implementation of support for polymorphism in the
CLR (Kennedy & Syme);
o The implementation of compilers for Haskell (Syme & Thomas) and
Standard ML
(Benton & Kennedy), along with language-specific extensions
for better interoperability;
o The design and implementation of MS-ILX, an extended version of
MS-IL with support for closures, polymorphism and algebraic datatypes
(Syme).
This talk will outline what we're covering in these projects from at a
high-level, explaining how they fit together in the wider scheme of
things.
Ann Copestake
Defaults and Probabilities
Many approaches to natural language processing use probabilities, some
use defaults and a few involve both. I'll discuss some applications
of default inheritance and nonmonotonic logics in formalizing models
of language and contrast these with cases where probabilistic models
are more appropriate. I'll aim to show that, although it isn't always
straightforward to decide which sort of approach to use, there are
clear differences in applicability, and that we really do need both
probabilities and defaults.