<Iconoclast logotype>

ICONOCLAST

Integrating Constraints on Layout and Style

Overview In the ICONOCLAST project, we developed a framework for integrating constraints on the style and layout of the output documents in an NLG system. By interacting with the system, authors are able to determine the optimal set of constraints whose interaction will lead to the production of documents in the desired style and layout. The main contributions of the project include the following:

  • Document Structure Theory - We have developed a formal theory to describe the interaction between the linguistic requirements of a text (e.g., wording, syntax) and its graphical presentation (e.g., punctuation, orthography, layout): Document Structure Theory.

  • Constraint satisfaction - We defined the generation of documents as a constraint satisfaction problem. This allows to have full control on the quality of the output documents. We distinguished between hard and soft constraints. Hard constraints always hold and can never be violated whereas soft constraints, typically non-fatal defects of the texts, can be switched on and off and prioritised by users.

    In addition, this framework allows to generate a large number of paraphrases which can be evaluated and ranked (automatically or by the user) according to preferences. This proves a great research tool for evaluating the impact and interaction of various constraints.

  • Corpus - A corpus consisting of about 500 Patient Information Leaflets was put online and SGML-annotated with logical structure. We hope to make this electronic corpus soon available to the research community.

  • Document planning - We investigated the issue of mapping the rhetorical input onto a text. This was done via the (abstract) document structure, which specifies the segmentation of the message into units such as paragraphs, sentences and lists. This level of representation (i.e., the abstract document structure) has been adoped as a basic data-type in the RAGS project specifications.

  • Coherence - Together with the GNOME project, we implemented a constraint-based algorithm based on centering theory. This algorithm allows the text planner to rank the document structures produced according to the centering transitions licensed by the ordering of propositions, giving a high score to the structures that ensure the best coherence and salience.

  • Logical complexity - In order to give authors control over the logical scope of their input during knowledge editing, we developed in the WYSIWYM knowledge editor a DRT-based representation in which referents are bound to logical contexts.

We organised the AAAI Fall Symposium on Using Layout for the Generation, Understanding or Retrieval of Documents in November 1999 in Cape Cod, during which researchers with a strong interest on the topic shared their views and experiences. A technical report is available [abstract].

Publications Have a look here at our list of publications

A closer look The following pages give more details of the project:

An illustrated walkthrough of the system

Interface of the web based demo
(For technical reasons, the full demo is not yet available on the web)

Document structuring module

The project's final report (ps format)

Staff The ICONOCLAST team included: In carrying out this work, we also working very closely with Kees van Deemter on the treatment of diagrams within documents.
Partners We held discussions for determining the system's requirements and evaluation with Information Design Unit and Multilingual Technology Ltd.
Financial Support ICONOCLAST was a 3-year project (1997-2000) supported by the EPSRC under grant L77102.