WYSIWYM applications

WYSIWYM has been applied in a number of systems at ITRI (University of Brighton) and the CRC (Open University):

Drafter-II: Origins

WYSIWYM developed from the GIST and DRAFTER projects which were in progress during 1993-96. The first WYSIWYM system was a re-implementation of DRAFTER by Richard Power in 1996, which generates software instructions in English, French and Italian.

 

Drafter-III: Coreference

A serious limitation of Drafter-II was the lack of any control over coreference. The knowledge author could not specify that two phrases in the feedback text referred to two different tokens of the same type, rather than to the same token. In 1998 Kees van Deemter and Richard Power designed a further version of DRAFTER in which the editing operations were extended to allow control over coreference.

 

Iconoclast: Logical scope

To meet the requirements of the Iconoclast project, in which we generate Patient Information Leaflets (which include much logically complex material such as conditionals), the editing operations in WYSIWYM were extended further to allow some control over logical scope. During 1998-99 Richard Power, with help from Rodger Kibble and Kees van Deemter, developed a demonstration system in which the editing choices made by the author lead to the construction of a Discourse Representation Structure, entities being assigned logical scopes by binding them to DRT boxes.

 

CLIME: Query definition

Complex queries to a database or an expert system usually have to be expressed in a special formalism. In the CLIME system, which answers questions about maritime law, a WYSIWYM interface allows users to define their queries through a feedback text, either in English or Italian. The system has been developed during 1998-99 by Paul Piwek.

 

CLIME: Plurals

In the latest version of the CLIME query interface, Paul Piwek has added new WYSIWYM editing operations which allow the user to construct plural entities (e.g. the pumps) and to define relations between individual entities and collections (one of the pumps) or between collections and sub-collections (some of the pumps). These operations are related to a DRT treatment of plurals.

 

PILLS: Patient Information Leaflets

A web-based application was developed in 1997 to illustrate the potential of WYSIWYM as the basis for a system that could generate patient information leaflets in many languages and styles. In addition to the Drafter-II languages the PILLS system generates output in German, Dutch and Portuguese. Intended as a demonstration of the concept rather than a full NLG system, PILLS does not allow authors to control the meaning of the output texts in detail; most choices result in the creation of a prefabricated proposition rather than an individual entity.

 

CLEF: Querying electronic patient records

CLEF aims to provide an information repository, enhanced by information extracted from texts (mainly letters to GPs) that can be queried by medical researchers. To be effective, such a system requires a easily used query interface. The CLEF query interface uses WYSIWYM editing to construct a query in unambiguous natural language; the underlying semantic representation is converted automatically to a database query in SQL. The CLEF project began in 2002 and has been extended through a second project CLEF-Services to 2007.

 

HALO: Querying a scientific knowledge base

HALO aims to create tools for formulating scientific knowledge and querying it by posing problems. The knowledge comes from scientific textbooks; the target questions are taken from American AP examinations (similar to A-levels in the UK). In the current phase (HALO-2), natural language technologies are used for both tasks. The OU is contributing to the task of question formulation, using WYSIWYM editing in a way analogous to its use in CLIME and CLEF.

 

CROCODIAL: CROss-lingual COmputer-mediated DIALogue

We are developing an application in which two authors hold a dialogue in a chat interface, each contribution being constructed by WYSIWYM editing. In this way they create a knowledge-base encoding the content of the dialogue; this can be used in order to provide services like a generated report, or automatic implementation of any actions agreed during its course. We think the application would be suitable for contexts in which experts speaking different languages need to confer on a restricted technical issue in which precise understanding is very important. We have in mind situations like financial transactions, legal or medical problems, and international policing.

 

CAKE: Conceptual Authoring plus Knowledge Extraction

As part of CLEF-SERVICES we are developing an application in which WYSIWYM is used for post-editing of records resulting from Information Extraction. The technique will be applied to letters from hospital specialists to the patient's doctor (GP). From the IE records, the system will create a feedback text showing (in English) the extracted content, linked to the original letter, along with any gaps. By normal WYSIWYM editing, a human data manager can correct and complete the IE records, so providing a gold-standard through which the IE rules can be improved.