UIMA TC Call This Week (May 25th)

[David Ferrucci <ferrucci@us.ibm.com>]

Hi All,

The telecon is ON at our usual time
(11AM Eastern on Friday). 

I will not be able to make this week's
call, however. Eric Nyberg will fill in for me as acting Chair.

Let me remind everyone about where we
are....

We are developing a platform independent
specification for text and multi-modal analytics that should support interoperability
among different UIMA applications and framework implementations. IBM submitted
a strawman for such a specification to the TC at is inception.

To date, the TC has raised issues, taken
votes and produced notes/documents describing what and how to revise the
following sections of the strawman spec provided by IBM to meet its requirements:

1. CAS Specification
2. Type-System Base Model
3. Type-System Language

The remaining sections are: Behavioral
Meta-Data Specification,  WSDL Service Descriptions, Processing Element
Meta-data Specification,  Aggregate Analytic Descriptor Specification,
SOAP Bindings, Java Bindings. 

Abstract Interfaces and Constant IDs
------------------------------------------------------
We are mostly through the section on
Abstract Interfaces. This is a critical section since it provides the logical
foundation for the WSDL descriptions, the SOAP and Java Bindings.

We got hung-up on the issue of requiring
that a UIMA service return a CAS such that the XMI ids are kept constant
across the interface (e.g,, the object id referring to a particular annotation
in the input CAS necessarily refers to the same annotation in the output
CAS). 

A delta interface that allowed analytics
to respond with just updates to objects in the input CAS, would require
that object id's are used to consistently across the interface. But this
still does not require that the return of a complete CAS would respect
"constant ids". We have not yet decided to require constant id's.
It was suggested that describing or enforcing aspects of an analytics behavior
may require  "constant ids".  Based on this, it was
recommended that we defer the vote on constant ids until after we discussed
behavioral metadata.

Behavioral Metadata
---------------------------------
Adam sent out a document summarizing
this section for the TC's consideration.

A key perspective, in my opinion, is
that the behavioral metadata specification should allow for extensibility
but NOT burden the framework, application or component developer beyond
their specific interest for reusability, validation and discoverability
of their analytics.

While we have proposed OCL as a bedrock
for expressing constraints over the content of the CAS, we did NOT intend
that developers would be required to provide OCL constraints to declaratively
express the behavior of their analytics. 

Rather we intended that a higher level
language, ultimately expressible and enforceable in OCL, be provided to
facilitate developers to express relatively simple things about their analytics
-- like required input types, expected output types. But that more expressive
things may be declared by extending this language as per the needs of any
given framework. Similarly, a framework may choose NOT to require any behavior
metadata be available for a compliant analytic.

Ultimately, I think we would like to
be in a situation, where if analytic developers want to provide some level
of behavior meta-data, it is expressed with respect to the standard we
provide and therefore how to interpret it for reuse, discoverability, validation
and/or planning would be common and knowable.

As we discuss this section particularly
in tomorrow's call we should keep focus on how the elements of the behavioral
meta-data spec might or might not require "constant ids" (see
above).


-Dave

------------------------------------------------------------------------
David A. Ferrucci, PhD
Senior Manager, Semantic Analysis & Integration 
Chief Architect,  UIMA
IBM T.J. Watson Research Center
19 Skyline Drive, Hawthorne, NY 10532
Tel: 914-784-7847, 8/863-7847
ferrucci@us.ibm.com
------------------------------------------------------------------------
http://www.ibm.com/research/uima